fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         BUG_ON(ret);
 154         if (ret) {
 155                 err = ret;
 156                 goto fail;
 157         }
 158         leaf = path->nodes[0];
 159         ei = btrfs_item_ptr(leaf, path->slots[0],
 160                             struct btrfs_file_extent_item);
 161         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 162         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 163         btrfs_set_file_extent_encryption(leaf, ei, 0);
 164         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 165         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 166         ptr = btrfs_file_extent_inline_start(ei);
 167
 168         if (compress_type != BTRFS_COMPRESS_NONE) {
 169                 struct page *cpage;
 170                 int i = 0;
 171                 while (compressed_size > 0) {
 172                         cpage = compressed_pages[i];
 173                         cur_size = min_t(unsigned long, compressed_size,
 174                                        PAGE_CACHE_SIZE);
 175
 176                         kaddr = kmap_atomic(cpage, KM_USER0);
 177                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 178                         kunmap_atomic(kaddr, KM_USER0);
 179
 180                         i++;
 181                         ptr += cur_size;
 182                         compressed_size -= cur_size;
 183                 }
 184                 btrfs_set_file_extent_compression(leaf, ei,
 185                                                   compress_type);
 186         } else {
 187                 page = find_get_page(inode->i_mapping,
 188                                      start >> PAGE_CACHE_SHIFT);
 189                 btrfs_set_file_extent_compression(leaf, ei, 0);
 190                 kaddr = kmap_atomic(page, KM_USER0);
 191                 offset = start & (PAGE_CACHE_SIZE - 1);
 192                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 193                 kunmap_atomic(kaddr, KM_USER0);
 194                 page_cache_release(page);
 195         }
 196         btrfs_mark_buffer_dirty(leaf);
 197         btrfs_free_path(path);
 198
 199         /*
 200          * we're an inline extent, so nobody can
 201          * extend the file past i_size without locking
 202          * a page we already have locked.
 203          *
 204          * We must do any isize and inode updates
 205          * before we unlock the pages.  Otherwise we
 206          * could end up racing with unlink.
 207          */
 208         BTRFS_I(inode)->disk_i_size = inode->i_size;
 209         btrfs_update_inode(trans, root, inode);
 210
 211         return 0;
 212 fail:
 213         btrfs_free_path(path);
 214         return err;
 215 }
 216
 217
 218 /*
 219  * conditionally insert an inline extent into the file.  This
 220  * does the checks required to make sure the data is small enough
 221  * to fit as an inline extent.
 222  */
 223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 224                                  struct btrfs_root *root,
 225                                  struct inode *inode, u64 start, u64 end,
 226                                  size_t compressed_size, int compress_type,
 227                                  struct page **compressed_pages)
 228 {
 229         u64 isize = i_size_read(inode);
 230         u64 actual_end = min(end + 1, isize);
 231         u64 inline_len = actual_end - start;
 232         u64 aligned_end = (end + root->sectorsize - 1) &
 233                         ~((u64)root->sectorsize - 1);
 234         u64 hint_byte;
 235         u64 data_len = inline_len;
 236         int ret;
 237
 238         if (compressed_size)
 239                 data_len = compressed_size;
 240
 241         if (start > 0 ||
 242             actual_end >= PAGE_CACHE_SIZE ||
 243             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 244             (!compressed_size &&
 245             (actual_end & (root->sectorsize - 1)) == 0) ||
 246             end + 1 < isize ||
 247             data_len > root->fs_info->max_inline) {
 248                 return 1;
 249         }
 250
 251         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 252                                  &hint_byte, 1);
 253         BUG_ON(ret);
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         BUG_ON(ret);
 261         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 262         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 263         return 0;
 264 }
 265
 266 struct async_extent {
 267         u64 start;
 268         u64 ram_size;
 269         u64 compressed_size;
 270         struct page **pages;
 271         unsigned long nr_pages;
 272         int compress_type;
 273         struct list_head list;
 274 };
 275
 276 struct async_cow {
 277         struct inode *inode;
 278         struct btrfs_root *root;
 279         struct page *locked_page;
 280         u64 start;
 281         u64 end;
 282         struct list_head extents;
 283         struct btrfs_work work;
 284 };
 285
 286 static noinline int add_async_extent(struct async_cow *cow,
 287                                      u64 start, u64 ram_size,
 288                                      u64 compressed_size,
 289                                      struct page **pages,
 290                                      unsigned long nr_pages,
 291                                      int compress_type)
 292 {
 293         struct async_extent *async_extent;
 294
 295         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 296         BUG_ON(!async_extent);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         async_extent->compress_type = compress_type;
 303         list_add_tail(&async_extent->list, &cow->extents);
 304         return 0;
 305 }
 306
 307 /*
 308  * we create compressed extents in two phases.  The first
 309  * phase compresses a range of pages that have already been
 310  * locked (both pages and state bits are locked).
 311  *
 312  * This is done inside an ordered work queue, and the compression
 313  * is spread across many cpus.  The actual IO submission is step
 314  * two, and the ordered work queue takes care of making sure that
 315  * happens in the same order things were put onto the queue by
 316  * writepages and friends.
 317  *
 318  * If this code finds it can't get good compression, it puts an
 319  * entry onto the work queue to write the uncompressed bytes.  This
 320  * makes sure that both compressed inodes and uncompressed inodes
 321  * are written in the same order that pdflush sent them down.
 322  */
 323 static noinline int compress_file_range(struct inode *inode,
 324                                         struct page *locked_page,
 325                                         u64 start, u64 end,
 326                                         struct async_cow *async_cow,
 327                                         int *num_added)
 328 {
 329         struct btrfs_root *root = BTRFS_I(inode)->root;
 330         struct btrfs_trans_handle *trans;
 331         u64 num_bytes;
 332         u64 blocksize = root->sectorsize;
 333         u64 actual_end;
 334         u64 isize = i_size_read(inode);
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345         int compress_type = root->fs_info->compress_type;
 346
 347         /* if this is a small write inside eof, kick off a defragbot */
 348         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 349                 btrfs_add_inode_defrag(NULL, inode);
 350
 351         actual_end = min_t(u64, isize, end + 1);
 352 again:
 353         will_compress = 0;
 354         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 355         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 356
 357         /*
 358          * we don't want to send crud past the end of i_size through
 359          * compression, that's just a waste of CPU time.  So, if the
 360          * end of the file is before the start of our current
 361          * requested range of bytes, we bail out to the uncompressed
 362          * cleanup code that can deal with all of this.
 363          *
 364          * It isn't really the fastest way to fix things, but this is a
 365          * very uncommon corner.
 366          */
 367         if (actual_end <= start)
 368                 goto cleanup_and_bail_uncompressed;
 369
 370         total_compressed = actual_end - start;
 371
 372         /* we want to make sure that amount of ram required to uncompress
 373          * an extent is reasonable, so we limit the total size in ram
 374          * of a compressed extent to 128k.  This is a crucial number
 375          * because it also controls how easily we can spread reads across
 376          * cpus for decompression.
 377          *
 378          * We also want to make sure the amount of IO required to do
 379          * a random read is reasonably small, so we limit the size of
 380          * a compressed extent to 128k.
 381          */
 382         total_compressed = min(total_compressed, max_uncompressed);
 383         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 384         num_bytes = max(blocksize,  num_bytes);
 385         total_in = 0;
 386         ret = 0;
 387
 388         /*
 389          * we do compression for mount -o compress and when the
 390          * inode has not been flagged as nocompress.  This flag can
 391          * change at any time if we discover bad compression ratios.
 392          */
 393         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 394             (btrfs_test_opt(root, COMPRESS) ||
 395              (BTRFS_I(inode)->force_compress) ||
 396              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 397                 WARN_ON(pages);
 398                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 399                 if (!pages) {
 400                         /* just bail out to the uncompressed code */
 401                         goto cont;
 402                 }
 403
 404                 if (BTRFS_I(inode)->force_compress)
 405                         compress_type = BTRFS_I(inode)->force_compress;
 406
 407                 ret = btrfs_compress_pages(compress_type,
 408                                            inode->i_mapping, start,
 409                                            total_compressed, pages,
 410                                            nr_pages, &nr_pages_ret,
 411                                            &total_in,
 412                                            &total_compressed,
 413                                            max_compressed);
 414
 415                 if (!ret) {
 416                         unsigned long offset = total_compressed &
 417                                 (PAGE_CACHE_SIZE - 1);
 418                         struct page *page = pages[nr_pages_ret - 1];
 419                         char *kaddr;
 420
 421                         /* zero the tail end of the last page, we might be
 422                          * sending it down to disk
 423                          */
 424                         if (offset) {
 425                                 kaddr = kmap_atomic(page, KM_USER0);
 426                                 memset(kaddr + offset, 0,
 427                                        PAGE_CACHE_SIZE - offset);
 428                                 kunmap_atomic(kaddr, KM_USER0);
 429                         }
 430                         will_compress = 1;
 431                 }
 432         }
 433 cont:
 434         if (start == 0) {
 435                 trans = btrfs_join_transaction(root);
 436                 BUG_ON(IS_ERR(trans));
 437                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 438
 439                 /* lets try to make an inline extent */
 440                 if (ret || total_in < (actual_end - start)) {
 441                         /* we didn't compress the entire range, try
 442                          * to make an uncompressed inline extent.
 443                          */
 444                         ret = cow_file_range_inline(trans, root, inode,
 445                                                     start, end, 0, 0, NULL);
 446                 } else {
 447                         /* try making a compressed inline extent */
 448                         ret = cow_file_range_inline(trans, root, inode,
 449                                                     start, end,
 450                                                     total_compressed,
 451                                                     compress_type, pages);
 452                 }
 453                 if (ret == 0) {
 454                         /*
 455                          * inline extent creation worked, we don't need
 456                          * to create any more async work items.  Unlock
 457                          * and free up our temp pages.
 458                          */
 459                         extent_clear_unlock_delalloc(inode,
 460                              &BTRFS_I(inode)->io_tree,
 461                              start, end, NULL,
 462                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 463                              EXTENT_CLEAR_DELALLOC |
 464                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 465
 466                         btrfs_end_transaction(trans, root);
 467                         goto free_pages_out;
 468                 }
 469                 btrfs_end_transaction(trans, root);
 470         }
 471
 472         if (will_compress) {
 473                 /*
 474                  * we aren't doing an inline extent round the compressed size
 475                  * up to a block size boundary so the allocator does sane
 476                  * things
 477                  */
 478                 total_compressed = (total_compressed + blocksize - 1) &
 479                         ~(blocksize - 1);
 480
 481                 /*
 482                  * one last check to make sure the compression is really a
 483                  * win, compare the page count read with the blocks on disk
 484                  */
 485                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 486                         ~(PAGE_CACHE_SIZE - 1);
 487                 if (total_compressed >= total_in) {
 488                         will_compress = 0;
 489                 } else {
 490                         num_bytes = total_in;
 491                 }
 492         }
 493         if (!will_compress && pages) {
 494                 /*
 495                  * the compression code ran but failed to make things smaller,
 496                  * free any pages it allocated and our page pointer array
 497                  */
 498                 for (i = 0; i < nr_pages_ret; i++) {
 499                         WARN_ON(pages[i]->mapping);
 500                         page_cache_release(pages[i]);
 501                 }
 502                 kfree(pages);
 503                 pages = NULL;
 504                 total_compressed = 0;
 505                 nr_pages_ret = 0;
 506
 507                 /* flag the file so we don't compress in the future */
 508                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 509                     !(BTRFS_I(inode)->force_compress)) {
 510                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 511                 }
 512         }
 513         if (will_compress) {
 514                 *num_added += 1;
 515
 516                 /* the async work queues will take care of doing actual
 517                  * allocation on disk for these compressed pages,
 518                  * and will submit them to the elevator.
 519                  */
 520                 add_async_extent(async_cow, start, num_bytes,
 521                                  total_compressed, pages, nr_pages_ret,
 522                                  compress_type);
 523
 524                 if (start + num_bytes < end) {
 525                         start += num_bytes;
 526                         pages = NULL;
 527                         cond_resched();
 528                         goto again;
 529                 }
 530         } else {
 531 cleanup_and_bail_uncompressed:
 532                 /*
 533                  * No compression, but we still need to write the pages in
 534                  * the file we've been given so far.  redirty the locked
 535                  * page if it corresponds to our extent and set things up
 536                  * for the async work queue to run cow_file_range to do
 537                  * the normal delalloc dance
 538                  */
 539                 if (page_offset(locked_page) >= start &&
 540                     page_offset(locked_page) <= end) {
 541                         __set_page_dirty_nobuffers(locked_page);
 542                         /* unlocked later on in the async handlers */
 543                 }
 544                 add_async_extent(async_cow, start, end - start + 1,
 545                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 546                 *num_added += 1;
 547         }
 548
 549 out:
 550         return 0;
 551
 552 free_pages_out:
 553         for (i = 0; i < nr_pages_ret; i++) {
 554                 WARN_ON(pages[i]->mapping);
 555                 page_cache_release(pages[i]);
 556         }
 557         kfree(pages);
 558
 559         goto out;
 560 }
 561
 562 /*
 563  * phase two of compressed writeback.  This is the ordered portion
 564  * of the code, which only gets called in the order the work was
 565  * queued.  We walk all the async extents created by compress_file_range
 566  * and send them down to the disk.
 567  */
 568 static noinline int submit_compressed_extents(struct inode *inode,
 569                                               struct async_cow *async_cow)
 570 {
 571         struct async_extent *async_extent;
 572         u64 alloc_hint = 0;
 573         struct btrfs_trans_handle *trans;
 574         struct btrfs_key ins;
 575         struct extent_map *em;
 576         struct btrfs_root *root = BTRFS_I(inode)->root;
 577         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 578         struct extent_io_tree *io_tree;
 579         int ret = 0;
 580
 581         if (list_empty(&async_cow->extents))
 582                 return 0;
 583
 584
 585         while (!list_empty(&async_cow->extents)) {
 586                 async_extent = list_entry(async_cow->extents.next,
 587                                           struct async_extent, list);
 588                 list_del(&async_extent->list);
 589
 590                 io_tree = &BTRFS_I(inode)->io_tree;
 591
 592 retry:
 593                 /* did the compression code fall back to uncompressed IO? */
 594                 if (!async_extent->pages) {
 595                         int page_started = 0;
 596                         unsigned long nr_written = 0;
 597
 598                         lock_extent(io_tree, async_extent->start,
 599                                          async_extent->start +
 600                                          async_extent->ram_size - 1, GFP_NOFS);
 601
 602                         /* allocate blocks */
 603                         ret = cow_file_range(inode, async_cow->locked_page,
 604                                              async_extent->start,
 605                                              async_extent->start +
 606                                              async_extent->ram_size - 1,
 607                                              &page_started, &nr_written, 0);
 608
 609                         /*
 610                          * if page_started, cow_file_range inserted an
 611                          * inline extent and took care of all the unlocking
 612                          * and IO for us.  Otherwise, we need to submit
 613                          * all those pages down to the drive.
 614                          */
 615                         if (!page_started && !ret)
 616                                 extent_write_locked_range(io_tree,
 617                                                   inode, async_extent->start,
 618                                                   async_extent->start +
 619                                                   async_extent->ram_size - 1,
 620                                                   btrfs_get_extent,
 621                                                   WB_SYNC_ALL);
 622                         kfree(async_extent);
 623                         cond_resched();
 624                         continue;
 625                 }
 626
 627                 lock_extent(io_tree, async_extent->start,
 628                             async_extent->start + async_extent->ram_size - 1,
 629                             GFP_NOFS);
 630
 631                 trans = btrfs_join_transaction(root);
 632                 BUG_ON(IS_ERR(trans));
 633                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 634                 ret = btrfs_reserve_extent(trans, root,
 635                                            async_extent->compressed_size,
 636                                            async_extent->compressed_size,
 637                                            0, alloc_hint,
 638                                            (u64)-1, &ins, 1);
 639                 btrfs_end_transaction(trans, root);
 640
 641                 if (ret) {
 642                         int i;
 643                         for (i = 0; i < async_extent->nr_pages; i++) {
 644                                 WARN_ON(async_extent->pages[i]->mapping);
 645                                 page_cache_release(async_extent->pages[i]);
 646                         }
 647                         kfree(async_extent->pages);
 648                         async_extent->nr_pages = 0;
 649                         async_extent->pages = NULL;
 650                         unlock_extent(io_tree, async_extent->start,
 651                                       async_extent->start +
 652                                       async_extent->ram_size - 1, GFP_NOFS);
 653                         goto retry;
 654                 }
 655
 656                 /*
 657                  * here we're doing allocation and writeback of the
 658                  * compressed pages
 659                  */
 660                 btrfs_drop_extent_cache(inode, async_extent->start,
 661                                         async_extent->start +
 662                                         async_extent->ram_size - 1, 0);
 663
 664                 em = alloc_extent_map();
 665                 BUG_ON(!em);
 666                 em->start = async_extent->start;
 667                 em->len = async_extent->ram_size;
 668                 em->orig_start = em->start;
 669
 670                 em->block_start = ins.objectid;
 671                 em->block_len = ins.offset;
 672                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 673                 em->compress_type = async_extent->compress_type;
 674                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 675                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 676
 677                 while (1) {
 678                         write_lock(&em_tree->lock);
 679                         ret = add_extent_mapping(em_tree, em);
 680                         write_unlock(&em_tree->lock);
 681                         if (ret != -EEXIST) {
 682                                 free_extent_map(em);
 683                                 break;
 684                         }
 685                         btrfs_drop_extent_cache(inode, async_extent->start,
 686                                                 async_extent->start +
 687                                                 async_extent->ram_size - 1, 0);
 688                 }
 689
 690                 ret = btrfs_add_ordered_extent_compress(inode,
 691                                                 async_extent->start,
 692                                                 ins.objectid,
 693                                                 async_extent->ram_size,
 694                                                 ins.offset,
 695                                                 BTRFS_ORDERED_COMPRESSED,
 696                                                 async_extent->compress_type);
 697                 BUG_ON(ret);
 698
 699                 /*
 700                  * clear dirty, set writeback and unlock the pages.
 701                  */
 702                 extent_clear_unlock_delalloc(inode,
 703                                 &BTRFS_I(inode)->io_tree,
 704                                 async_extent->start,
 705                                 async_extent->start +
 706                                 async_extent->ram_size - 1,
 707                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 708                                 EXTENT_CLEAR_UNLOCK |
 709                                 EXTENT_CLEAR_DELALLOC |
 710                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 711
 712                 ret = btrfs_submit_compressed_write(inode,
 713                                     async_extent->start,
 714                                     async_extent->ram_size,
 715                                     ins.objectid,
 716                                     ins.offset, async_extent->pages,
 717                                     async_extent->nr_pages);
 718
 719                 BUG_ON(ret);
 720                 alloc_hint = ins.objectid + ins.offset;
 721                 kfree(async_extent);
 722                 cond_resched();
 723         }
 724
 725         return 0;
 726 }
 727
 728 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 729                                       u64 num_bytes)
 730 {
 731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 732         struct extent_map *em;
 733         u64 alloc_hint = 0;
 734
 735         read_lock(&em_tree->lock);
 736         em = search_extent_mapping(em_tree, start, num_bytes);
 737         if (em) {
 738                 /*
 739                  * if block start isn't an actual block number then find the
 740                  * first block in this inode and use that as a hint.  If that
 741                  * block is also bogus then just don't worry about it.
 742                  */
 743                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 744                         free_extent_map(em);
 745                         em = search_extent_mapping(em_tree, 0, 0);
 746                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 747                                 alloc_hint = em->block_start;
 748                         if (em)
 749                                 free_extent_map(em);
 750                 } else {
 751                         alloc_hint = em->block_start;
 752                         free_extent_map(em);
 753                 }
 754         }
 755         read_unlock(&em_tree->lock);
 756
 757         return alloc_hint;
 758 }
 759
 760 /*
 761  * when extent_io.c finds a delayed allocation range in the file,
 762  * the call backs end up in this code.  The basic idea is to
 763  * allocate extents on disk for the range, and create ordered data structs
 764  * in ram to track those extents.
 765  *
 766  * locked_page is the page that writepage had locked already.  We use
 767  * it to make sure we don't do extra locks or unlocks.
 768  *
 769  * *page_started is set to one if we unlock locked_page and do everything
 770  * required to start IO on it.  It may be clean and already done with
 771  * IO when we return.
 772  */
 773 static noinline int cow_file_range(struct inode *inode,
 774                                    struct page *locked_page,
 775                                    u64 start, u64 end, int *page_started,
 776                                    unsigned long *nr_written,
 777                                    int unlock)
 778 {
 779         struct btrfs_root *root = BTRFS_I(inode)->root;
 780         struct btrfs_trans_handle *trans;
 781         u64 alloc_hint = 0;
 782         u64 num_bytes;
 783         unsigned long ram_size;
 784         u64 disk_num_bytes;
 785         u64 cur_alloc_size;
 786         u64 blocksize = root->sectorsize;
 787         struct btrfs_key ins;
 788         struct extent_map *em;
 789         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 790         int ret = 0;
 791
 792         BUG_ON(btrfs_is_free_space_inode(root, inode));
 793         trans = btrfs_join_transaction(root);
 794         BUG_ON(IS_ERR(trans));
 795         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 796
 797         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 798         num_bytes = max(blocksize,  num_bytes);
 799         disk_num_bytes = num_bytes;
 800         ret = 0;
 801
 802         /* if this is a small write inside eof, kick off defrag */
 803         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 804                 btrfs_add_inode_defrag(trans, inode);
 805
 806         if (start == 0) {
 807                 /* lets try to make an inline extent */
 808                 ret = cow_file_range_inline(trans, root, inode,
 809                                             start, end, 0, 0, NULL);
 810                 if (ret == 0) {
 811                         extent_clear_unlock_delalloc(inode,
 812                                      &BTRFS_I(inode)->io_tree,
 813                                      start, end, NULL,
 814                                      EXTENT_CLEAR_UNLOCK_PAGE |
 815                                      EXTENT_CLEAR_UNLOCK |
 816                                      EXTENT_CLEAR_DELALLOC |
 817                                      EXTENT_CLEAR_DIRTY |
 818                                      EXTENT_SET_WRITEBACK |
 819                                      EXTENT_END_WRITEBACK);
 820
 821                         *nr_written = *nr_written +
 822                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 823                         *page_started = 1;
 824                         ret = 0;
 825                         goto out;
 826                 }
 827         }
 828
 829         BUG_ON(disk_num_bytes >
 830                btrfs_super_total_bytes(root->fs_info->super_copy));
 831
 832         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 833         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 834
 835         while (disk_num_bytes > 0) {
 836                 unsigned long op;
 837
 838                 cur_alloc_size = disk_num_bytes;
 839                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 840                                            root->sectorsize, 0, alloc_hint,
 841                                            (u64)-1, &ins, 1);
 842                 BUG_ON(ret);
 843
 844                 em = alloc_extent_map();
 845                 BUG_ON(!em);
 846                 em->start = start;
 847                 em->orig_start = em->start;
 848                 ram_size = ins.offset;
 849                 em->len = ins.offset;
 850
 851                 em->block_start = ins.objectid;
 852                 em->block_len = ins.offset;
 853                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 854                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 855
 856                 while (1) {
 857                         write_lock(&em_tree->lock);
 858                         ret = add_extent_mapping(em_tree, em);
 859                         write_unlock(&em_tree->lock);
 860                         if (ret != -EEXIST) {
 861                                 free_extent_map(em);
 862                                 break;
 863                         }
 864                         btrfs_drop_extent_cache(inode, start,
 865                                                 start + ram_size - 1, 0);
 866                 }
 867
 868                 cur_alloc_size = ins.offset;
 869                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 870                                                ram_size, cur_alloc_size, 0);
 871                 BUG_ON(ret);
 872
 873                 if (root->root_key.objectid ==
 874                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 875                         ret = btrfs_reloc_clone_csums(inode, start,
 876                                                       cur_alloc_size);
 877                         BUG_ON(ret);
 878                 }
 879
 880                 if (disk_num_bytes < cur_alloc_size)
 881                         break;
 882
 883                 /* we're not doing compressed IO, don't unlock the first
 884                  * page (which the caller expects to stay locked), don't
 885                  * clear any dirty bits and don't set any writeback bits
 886                  *
 887                  * Do set the Private2 bit so we know this page was properly
 888                  * setup for writepage
 889                  */
 890                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 891                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 892                         EXTENT_SET_PRIVATE2;
 893
 894                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 895                                              start, start + ram_size - 1,
 896                                              locked_page, op);
 897                 disk_num_bytes -= cur_alloc_size;
 898                 num_bytes -= cur_alloc_size;
 899                 alloc_hint = ins.objectid + ins.offset;
 900                 start += cur_alloc_size;
 901         }
 902 out:
 903         ret = 0;
 904         btrfs_end_transaction(trans, root);
 905
 906         return ret;
 907 }
 908
 909 /*
 910  * work queue call back to started compression on a file and pages
 911  */
 912 static noinline void async_cow_start(struct btrfs_work *work)
 913 {
 914         struct async_cow *async_cow;
 915         int num_added = 0;
 916         async_cow = container_of(work, struct async_cow, work);
 917
 918         compress_file_range(async_cow->inode, async_cow->locked_page,
 919                             async_cow->start, async_cow->end, async_cow,
 920                             &num_added);
 921         if (num_added == 0)
 922                 async_cow->inode = NULL;
 923 }
 924
 925 /*
 926  * work queue call back to submit previously compressed pages
 927  */
 928 static noinline void async_cow_submit(struct btrfs_work *work)
 929 {
 930         struct async_cow *async_cow;
 931         struct btrfs_root *root;
 932         unsigned long nr_pages;
 933
 934         async_cow = container_of(work, struct async_cow, work);
 935
 936         root = async_cow->root;
 937         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 938                 PAGE_CACHE_SHIFT;
 939
 940         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 941
 942         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 943             5 * 1042 * 1024 &&
 944             waitqueue_active(&root->fs_info->async_submit_wait))
 945                 wake_up(&root->fs_info->async_submit_wait);
 946
 947         if (async_cow->inode)
 948                 submit_compressed_extents(async_cow->inode, async_cow);
 949 }
 950
 951 static noinline void async_cow_free(struct btrfs_work *work)
 952 {
 953         struct async_cow *async_cow;
 954         async_cow = container_of(work, struct async_cow, work);
 955         kfree(async_cow);
 956 }
 957
 958 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 959                                 u64 start, u64 end, int *page_started,
 960                                 unsigned long *nr_written)
 961 {
 962         struct async_cow *async_cow;
 963         struct btrfs_root *root = BTRFS_I(inode)->root;
 964         unsigned long nr_pages;
 965         u64 cur_end;
 966         int limit = 10 * 1024 * 1042;
 967
 968         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 969                          1, 0, NULL, GFP_NOFS);
 970         while (start < end) {
 971                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 972                 BUG_ON(!async_cow);
 973                 async_cow->inode = inode;
 974                 async_cow->root = root;
 975                 async_cow->locked_page = locked_page;
 976                 async_cow->start = start;
 977
 978                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 979                         cur_end = end;
 980                 else
 981                         cur_end = min(end, start + 512 * 1024 - 1);
 982
 983                 async_cow->end = cur_end;
 984                 INIT_LIST_HEAD(&async_cow->extents);
 985
 986                 async_cow->work.func = async_cow_start;
 987                 async_cow->work.ordered_func = async_cow_submit;
 988                 async_cow->work.ordered_free = async_cow_free;
 989                 async_cow->work.flags = 0;
 990
 991                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 992                         PAGE_CACHE_SHIFT;
 993                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 994
 995                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 996                                    &async_cow->work);
 997
 998                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1001                             limit));
1002                 }
1003
1004                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1005                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1006                         wait_event(root->fs_info->async_submit_wait,
1007                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008                            0));
1009                 }
1010
1011                 *nr_written += nr_pages;
1012                 start = cur_end + 1;
1013         }
1014         *page_started = 1;
1015         return 0;
1016 }
1017
1018 static noinline int csum_exist_in_range(struct btrfs_root *root,
1019                                         u64 bytenr, u64 num_bytes)
1020 {
1021         int ret;
1022         struct btrfs_ordered_sum *sums;
1023         LIST_HEAD(list);
1024
1025         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1026                                        bytenr + num_bytes - 1, &list, 0);
1027         if (ret == 0 && list_empty(&list))
1028                 return 0;
1029
1030         while (!list_empty(&list)) {
1031                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1032                 list_del(&sums->list);
1033                 kfree(sums);
1034         }
1035         return 1;
1036 }
1037
1038 /*
1039  * when nowcow writeback call back.  This checks for snapshots or COW copies
1040  * of the extents that exist in the file, and COWs the file as required.
1041  *
1042  * If no cow copies or snapshots exist, we write directly to the existing
1043  * blocks on disk
1044  */
1045 static noinline int run_delalloc_nocow(struct inode *inode,
1046                                        struct page *locked_page,
1047                               u64 start, u64 end, int *page_started, int force,
1048                               unsigned long *nr_written)
1049 {
1050         struct btrfs_root *root = BTRFS_I(inode)->root;
1051         struct btrfs_trans_handle *trans;
1052         struct extent_buffer *leaf;
1053         struct btrfs_path *path;
1054         struct btrfs_file_extent_item *fi;
1055         struct btrfs_key found_key;
1056         u64 cow_start;
1057         u64 cur_offset;
1058         u64 extent_end;
1059         u64 extent_offset;
1060         u64 disk_bytenr;
1061         u64 num_bytes;
1062         int extent_type;
1063         int ret;
1064         int type;
1065         int nocow;
1066         int check_prev = 1;
1067         bool nolock;
1068         u64 ino = btrfs_ino(inode);
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         nolock = btrfs_is_free_space_inode(root, inode);
1075
1076         if (nolock)
1077                 trans = btrfs_join_transaction_nolock(root);
1078         else
1079                 trans = btrfs_join_transaction(root);
1080
1081         BUG_ON(IS_ERR(trans));
1082         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083
1084         cow_start = (u64)-1;
1085         cur_offset = start;
1086         while (1) {
1087                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1088                                                cur_offset, 0);
1089                 BUG_ON(ret < 0);
1090                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1091                         leaf = path->nodes[0];
1092                         btrfs_item_key_to_cpu(leaf, &found_key,
1093                                               path->slots[0] - 1);
1094                         if (found_key.objectid == ino &&
1095                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1096                                 path->slots[0]--;
1097                 }
1098                 check_prev = 0;
1099 next_slot:
1100                 leaf = path->nodes[0];
1101                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1102                         ret = btrfs_next_leaf(root, path);
1103                         if (ret < 0)
1104                                 BUG_ON(1);
1105                         if (ret > 0)
1106                                 break;
1107                         leaf = path->nodes[0];
1108                 }
1109
1110                 nocow = 0;
1111                 disk_bytenr = 0;
1112                 num_bytes = 0;
1113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114
1115                 if (found_key.objectid > ino ||
1116                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1117                     found_key.offset > end)
1118                         break;
1119
1120                 if (found_key.offset > cur_offset) {
1121                         extent_end = found_key.offset;
1122                         extent_type = 0;
1123                         goto out_check;
1124                 }
1125
1126                 fi = btrfs_item_ptr(leaf, path->slots[0],
1127                                     struct btrfs_file_extent_item);
1128                 extent_type = btrfs_file_extent_type(leaf, fi);
1129
1130                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1131                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1132                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1133                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1134                         extent_end = found_key.offset +
1135                                 btrfs_file_extent_num_bytes(leaf, fi);
1136                         if (extent_end <= start) {
1137                                 path->slots[0]++;
1138                                 goto next_slot;
1139                         }
1140                         if (disk_bytenr == 0)
1141                                 goto out_check;
1142                         if (btrfs_file_extent_compression(leaf, fi) ||
1143                             btrfs_file_extent_encryption(leaf, fi) ||
1144                             btrfs_file_extent_other_encoding(leaf, fi))
1145                                 goto out_check;
1146                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147                                 goto out_check;
1148                         if (btrfs_extent_readonly(root, disk_bytenr))
1149                                 goto out_check;
1150                         if (btrfs_cross_ref_exist(trans, root, ino,
1151                                                   found_key.offset -
1152                                                   extent_offset, disk_bytenr))
1153                                 goto out_check;
1154                         disk_bytenr += extent_offset;
1155                         disk_bytenr += cur_offset - found_key.offset;
1156                         num_bytes = min(end + 1, extent_end) - cur_offset;
1157                         /*
1158                          * force cow if csum exists in the range.
1159                          * this ensure that csum for a given extent are
1160                          * either valid or do not exist.
1161                          */
1162                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1163                                 goto out_check;
1164                         nocow = 1;
1165                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1166                         extent_end = found_key.offset +
1167                                 btrfs_file_extent_inline_len(leaf, fi);
1168                         extent_end = ALIGN(extent_end, root->sectorsize);
1169                 } else {
1170                         BUG_ON(1);
1171                 }
1172 out_check:
1173                 if (extent_end <= start) {
1174                         path->slots[0]++;
1175                         goto next_slot;
1176                 }
1177                 if (!nocow) {
1178                         if (cow_start == (u64)-1)
1179                                 cow_start = cur_offset;
1180                         cur_offset = extent_end;
1181                         if (cur_offset > end)
1182                                 break;
1183                         path->slots[0]++;
1184                         goto next_slot;
1185                 }
1186
1187                 btrfs_release_path(path);
1188                 if (cow_start != (u64)-1) {
1189                         ret = cow_file_range(inode, locked_page, cow_start,
1190                                         found_key.offset - 1, page_started,
1191                                         nr_written, 1);
1192                         BUG_ON(ret);
1193                         cow_start = (u64)-1;
1194                 }
1195
1196                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1197                         struct extent_map *em;
1198                         struct extent_map_tree *em_tree;
1199                         em_tree = &BTRFS_I(inode)->extent_tree;
1200                         em = alloc_extent_map();
1201                         BUG_ON(!em);
1202                         em->start = cur_offset;
1203                         em->orig_start = em->start;
1204                         em->len = num_bytes;
1205                         em->block_len = num_bytes;
1206                         em->block_start = disk_bytenr;
1207                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1208                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209                         while (1) {
1210                                 write_lock(&em_tree->lock);
1211                                 ret = add_extent_mapping(em_tree, em);
1212                                 write_unlock(&em_tree->lock);
1213                                 if (ret != -EEXIST) {
1214                                         free_extent_map(em);
1215                                         break;
1216                                 }
1217                                 btrfs_drop_extent_cache(inode, em->start,
1218                                                 em->start + em->len - 1, 0);
1219                         }
1220                         type = BTRFS_ORDERED_PREALLOC;
1221                 } else {
1222                         type = BTRFS_ORDERED_NOCOW;
1223                 }
1224
1225                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1226                                                num_bytes, num_bytes, type);
1227                 BUG_ON(ret);
1228
1229                 if (root->root_key.objectid ==
1230                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1231                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1232                                                       num_bytes);
1233                         BUG_ON(ret);
1234                 }
1235
1236                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1237                                 cur_offset, cur_offset + num_bytes - 1,
1238                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1239                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1240                                 EXTENT_SET_PRIVATE2);
1241                 cur_offset = extent_end;
1242                 if (cur_offset > end)
1243                         break;
1244         }
1245         btrfs_release_path(path);
1246
1247         if (cur_offset <= end && cow_start == (u64)-1)
1248                 cow_start = cur_offset;
1249         if (cow_start != (u64)-1) {
1250                 ret = cow_file_range(inode, locked_page, cow_start, end,
1251                                      page_started, nr_written, 1);
1252                 BUG_ON(ret);
1253         }
1254
1255         if (nolock) {
1256                 ret = btrfs_end_transaction_nolock(trans, root);
1257                 BUG_ON(ret);
1258         } else {
1259                 ret = btrfs_end_transaction(trans, root);
1260                 BUG_ON(ret);
1261         }
1262         btrfs_free_path(path);
1263         return 0;
1264 }
1265
1266 /*
1267  * extent_io.c call back to do delayed allocation processing
1268  */
1269 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1270                               u64 start, u64 end, int *page_started,
1271                               unsigned long *nr_written)
1272 {
1273         int ret;
1274         struct btrfs_root *root = BTRFS_I(inode)->root;
1275
1276         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 1, nr_written);
1279         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1280                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1281                                          page_started, 0, nr_written);
1282         else if (!btrfs_test_opt(root, COMPRESS) &&
1283                  !(BTRFS_I(inode)->force_compress) &&
1284                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1285                 ret = cow_file_range(inode, locked_page, start, end,
1286                                       page_started, nr_written, 1);
1287         else
1288                 ret = cow_file_range_async(inode, locked_page, start, end,
1289                                            page_started, nr_written);
1290         return ret;
1291 }
1292
1293 static void btrfs_split_extent_hook(struct inode *inode,
1294                                     struct extent_state *orig, u64 split)
1295 {
1296         /* not delalloc, ignore it */
1297         if (!(orig->state & EXTENT_DELALLOC))
1298                 return;
1299
1300         spin_lock(&BTRFS_I(inode)->lock);
1301         BTRFS_I(inode)->outstanding_extents++;
1302         spin_unlock(&BTRFS_I(inode)->lock);
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static void btrfs_merge_extent_hook(struct inode *inode,
1312                                     struct extent_state *new,
1313                                     struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return;
1318
1319         spin_lock(&BTRFS_I(inode)->lock);
1320         BTRFS_I(inode)->outstanding_extents--;
1321         spin_unlock(&BTRFS_I(inode)->lock);
1322 }
1323
1324 /*
1325  * extent_io.c set_bit_hook, used to track delayed allocation
1326  * bytes in this file, and to maintain the list of inodes that
1327  * have pending delalloc work to be done.
1328  */
1329 static void btrfs_set_bit_hook(struct inode *inode,
1330                                struct extent_state *state, int *bits)
1331 {
1332
1333         /*
1334          * set_bit and clear bit hooks normally require _irqsave/restore
1335          * but in this case, we are only testing for the DELALLOC
1336          * bit, which is only set or cleared with irqs on
1337          */
1338         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1339                 struct btrfs_root *root = BTRFS_I(inode)->root;
1340                 u64 len = state->end + 1 - state->start;
1341                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342
1343                 if (*bits & EXTENT_FIRST_DELALLOC) {
1344                         *bits &= ~EXTENT_FIRST_DELALLOC;
1345                 } else {
1346                         spin_lock(&BTRFS_I(inode)->lock);
1347                         BTRFS_I(inode)->outstanding_extents++;
1348                         spin_unlock(&BTRFS_I(inode)->lock);
1349                 }
1350
1351                 spin_lock(&root->fs_info->delalloc_lock);
1352                 BTRFS_I(inode)->delalloc_bytes += len;
1353                 root->fs_info->delalloc_bytes += len;
1354                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1355                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1356                                       &root->fs_info->delalloc_inodes);
1357                 }
1358                 spin_unlock(&root->fs_info->delalloc_lock);
1359         }
1360 }
1361
1362 /*
1363  * extent_io.c clear_bit_hook, see set_bit_hook for why
1364  */
1365 static void btrfs_clear_bit_hook(struct inode *inode,
1366                                  struct extent_state *state, int *bits)
1367 {
1368         /*
1369          * set_bit and clear bit hooks normally require _irqsave/restore
1370          * but in this case, we are only testing for the DELALLOC
1371          * bit, which is only set or cleared with irqs on
1372          */
1373         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1374                 struct btrfs_root *root = BTRFS_I(inode)->root;
1375                 u64 len = state->end + 1 - state->start;
1376                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377
1378                 if (*bits & EXTENT_FIRST_DELALLOC) {
1379                         *bits &= ~EXTENT_FIRST_DELALLOC;
1380                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1381                         spin_lock(&BTRFS_I(inode)->lock);
1382                         BTRFS_I(inode)->outstanding_extents--;
1383                         spin_unlock(&BTRFS_I(inode)->lock);
1384                 }
1385
1386                 if (*bits & EXTENT_DO_ACCOUNTING)
1387                         btrfs_delalloc_release_metadata(inode, len);
1388
1389                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390                     && do_list)
1391                         btrfs_free_reserved_data_space(inode, len);
1392
1393                 spin_lock(&root->fs_info->delalloc_lock);
1394                 root->fs_info->delalloc_bytes -= len;
1395                 BTRFS_I(inode)->delalloc_bytes -= len;
1396
1397                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1398                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1399                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400                 }
1401                 spin_unlock(&root->fs_info->delalloc_lock);
1402         }
1403 }
1404
1405 /*
1406  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407  * we don't create bios that span stripes or chunks
1408  */
1409 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1410                          size_t size, struct bio *bio,
1411                          unsigned long bio_flags)
1412 {
1413         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1414         struct btrfs_mapping_tree *map_tree;
1415         u64 logical = (u64)bio->bi_sector << 9;
1416         u64 length = 0;
1417         u64 map_length;
1418         int ret;
1419
1420         if (bio_flags & EXTENT_BIO_COMPRESSED)
1421                 return 0;
1422
1423         length = bio->bi_size;
1424         map_tree = &root->fs_info->mapping_tree;
1425         map_length = length;
1426         ret = btrfs_map_block(map_tree, READ, logical,
1427                               &map_length, NULL, 0);
1428
1429         if (map_length < length + size)
1430                 return 1;
1431         return ret;
1432 }
1433
1434 /*
1435  * in order to insert checksums into the metadata in large chunks,
1436  * we wait until bio submission time.   All the pages in the bio are
1437  * checksummed and sums are attached onto the ordered extent record.
1438  *
1439  * At IO completion time the cums attached on the ordered extent record
1440  * are inserted into the btree
1441  */
1442 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443                                     struct bio *bio, int mirror_num,
1444                                     unsigned long bio_flags,
1445                                     u64 bio_offset)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         int ret = 0;
1449
1450         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1451         BUG_ON(ret);
1452         return 0;
1453 }
1454
1455 /*
1456  * in order to insert checksums into the metadata in large chunks,
1457  * we wait until bio submission time.   All the pages in the bio are
1458  * checksummed and sums are attached onto the ordered extent record.
1459  *
1460  * At IO completion time the cums attached on the ordered extent record
1461  * are inserted into the btree
1462  */
1463 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1464                           int mirror_num, unsigned long bio_flags,
1465                           u64 bio_offset)
1466 {
1467         struct btrfs_root *root = BTRFS_I(inode)->root;
1468         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1469 }
1470
1471 /*
1472  * extent_io.c submission hook. This does the right thing for csum calculation
1473  * on write, or reading the csums from the tree before a read
1474  */
1475 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1476                           int mirror_num, unsigned long bio_flags,
1477                           u64 bio_offset)
1478 {
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480         int ret = 0;
1481         int skip_sum;
1482
1483         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484
1485         if (btrfs_is_free_space_inode(root, inode))
1486                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487         else
1488                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1489         BUG_ON(ret);
1490
1491         if (!(rw & REQ_WRITE)) {
1492                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1493                         return btrfs_submit_compressed_read(inode, bio,
1494                                                     mirror_num, bio_flags);
1495                 } else if (!skip_sum) {
1496                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1497                         if (ret)
1498                                 return ret;
1499                 }
1500                 goto mapit;
1501         } else if (!skip_sum) {
1502                 /* csum items have already been cloned */
1503                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504                         goto mapit;
1505                 /* we're doing a write, do the async checksumming */
1506                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1507                                    inode, rw, bio, mirror_num,
1508                                    bio_flags, bio_offset,
1509                                    __btrfs_submit_bio_start,
1510                                    __btrfs_submit_bio_done);
1511         }
1512
1513 mapit:
1514         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1515 }
1516
1517 /*
1518  * given a list of ordered sums record them in the inode.  This happens
1519  * at IO completion time based on sums calculated at bio submission time.
1520  */
1521 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1522                              struct inode *inode, u64 file_offset,
1523                              struct list_head *list)
1524 {
1525         struct btrfs_ordered_sum *sum;
1526
1527         list_for_each_entry(sum, list, list) {
1528                 btrfs_csum_file_blocks(trans,
1529                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1530         }
1531         return 0;
1532 }
1533
1534 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1535                               struct extent_state **cached_state)
1536 {
1537         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538                 WARN_ON(1);
1539         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1540                                    cached_state, GFP_NOFS);
1541 }
1542
1543 /* see btrfs_writepage_start_hook for details on why this is required */
1544 struct btrfs_writepage_fixup {
1545         struct page *page;
1546         struct btrfs_work work;
1547 };
1548
1549 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 {
1551         struct btrfs_writepage_fixup *fixup;
1552         struct btrfs_ordered_extent *ordered;
1553         struct extent_state *cached_state = NULL;
1554         struct page *page;
1555         struct inode *inode;
1556         u64 page_start;
1557         u64 page_end;
1558
1559         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560         page = fixup->page;
1561 again:
1562         lock_page(page);
1563         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564                 ClearPageChecked(page);
1565                 goto out_page;
1566         }
1567
1568         inode = page->mapping->host;
1569         page_start = page_offset(page);
1570         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571
1572         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573                          &cached_state, GFP_NOFS);
1574
1575         /* already ordered? We're done */
1576         if (PagePrivate2(page))
1577                 goto out;
1578
1579         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580         if (ordered) {
1581                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582                                      page_end, &cached_state, GFP_NOFS);
1583                 unlock_page(page);
1584                 btrfs_start_ordered_extent(inode, ordered, 1);
1585                 goto again;
1586         }
1587
1588         BUG();
1589         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1590         ClearPageChecked(page);
1591 out:
1592         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1593                              &cached_state, GFP_NOFS);
1594 out_page:
1595         unlock_page(page);
1596         page_cache_release(page);
1597         kfree(fixup);
1598 }
1599
1600 /*
1601  * There are a few paths in the higher layers of the kernel that directly
1602  * set the page dirty bit without asking the filesystem if it is a
1603  * good idea.  This causes problems because we want to make sure COW
1604  * properly happens and the data=ordered rules are followed.
1605  *
1606  * In our case any range that doesn't have the ORDERED bit set
1607  * hasn't been properly setup for IO.  We kick off an async process
1608  * to fix it up.  The async helper will wait for ordered extents, set
1609  * the delalloc bit and make it safe to write the page.
1610  */
1611 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1612 {
1613         struct inode *inode = page->mapping->host;
1614         struct btrfs_writepage_fixup *fixup;
1615         struct btrfs_root *root = BTRFS_I(inode)->root;
1616
1617         /* this page is properly in the ordered list */
1618         if (TestClearPagePrivate2(page))
1619                 return 0;
1620
1621         if (PageChecked(page))
1622                 return -EAGAIN;
1623
1624         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1625         if (!fixup)
1626                 return -EAGAIN;
1627
1628         SetPageChecked(page);
1629         page_cache_get(page);
1630         fixup->work.func = btrfs_writepage_fixup_worker;
1631         fixup->page = page;
1632         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1633         return -EAGAIN;
1634 }
1635
1636 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1637                                        struct inode *inode, u64 file_pos,
1638                                        u64 disk_bytenr, u64 disk_num_bytes,
1639                                        u64 num_bytes, u64 ram_bytes,
1640                                        u8 compression, u8 encryption,
1641                                        u16 other_encoding, int extent_type)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         struct btrfs_file_extent_item *fi;
1645         struct btrfs_path *path;
1646         struct extent_buffer *leaf;
1647         struct btrfs_key ins;
1648         u64 hint;
1649         int ret;
1650
1651         path = btrfs_alloc_path();
1652         if (!path)
1653                 return -ENOMEM;
1654
1655         path->leave_spinning = 1;
1656
1657         /*
1658          * we may be replacing one extent in the tree with another.
1659          * The new extent is pinned in the extent map, and we don't want
1660          * to drop it from the cache until it is completely in the btree.
1661          *
1662          * So, tell btrfs_drop_extents to leave this extent in the cache.
1663          * the caller is expected to unpin it and allow it to be merged
1664          * with the others.
1665          */
1666         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1667                                  &hint, 0);
1668         BUG_ON(ret);
1669
1670         ins.objectid = btrfs_ino(inode);
1671         ins.offset = file_pos;
1672         ins.type = BTRFS_EXTENT_DATA_KEY;
1673         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1674         BUG_ON(ret);
1675         leaf = path->nodes[0];
1676         fi = btrfs_item_ptr(leaf, path->slots[0],
1677                             struct btrfs_file_extent_item);
1678         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1679         btrfs_set_file_extent_type(leaf, fi, extent_type);
1680         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1681         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1682         btrfs_set_file_extent_offset(leaf, fi, 0);
1683         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1684         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1685         btrfs_set_file_extent_compression(leaf, fi, compression);
1686         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1687         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1688
1689         btrfs_unlock_up_safe(path, 1);
1690         btrfs_set_lock_blocking(leaf);
1691
1692         btrfs_mark_buffer_dirty(leaf);
1693
1694         inode_add_bytes(inode, num_bytes);
1695
1696         ins.objectid = disk_bytenr;
1697         ins.offset = disk_num_bytes;
1698         ins.type = BTRFS_EXTENT_ITEM_KEY;
1699         ret = btrfs_alloc_reserved_file_extent(trans, root,
1700                                         root->root_key.objectid,
1701                                         btrfs_ino(inode), file_pos, &ins);
1702         BUG_ON(ret);
1703         btrfs_free_path(path);
1704
1705         return 0;
1706 }
1707
1708 /*
1709  * helper function for btrfs_finish_ordered_io, this
1710  * just reads in some of the csum leaves to prime them into ram
1711  * before we start the transaction.  It limits the amount of btree
1712  * reads required while inside the transaction.
1713  */
1714 /* as ordered data IO finishes, this gets called so we can finish
1715  * an ordered extent if the range of bytes in the file it covers are
1716  * fully written.
1717  */
1718 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1719 {
1720         struct btrfs_root *root = BTRFS_I(inode)->root;
1721         struct btrfs_trans_handle *trans = NULL;
1722         struct btrfs_ordered_extent *ordered_extent = NULL;
1723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1724         struct extent_state *cached_state = NULL;
1725         int compress_type = 0;
1726         int ret;
1727         bool nolock;
1728
1729         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1730                                              end - start + 1);
1731         if (!ret)
1732                 return 0;
1733         BUG_ON(!ordered_extent);
1734
1735         nolock = btrfs_is_free_space_inode(root, inode);
1736
1737         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1738                 BUG_ON(!list_empty(&ordered_extent->list));
1739                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1740                 if (!ret) {
1741                         if (nolock)
1742                                 trans = btrfs_join_transaction_nolock(root);
1743                         else
1744                                 trans = btrfs_join_transaction(root);
1745                         BUG_ON(IS_ERR(trans));
1746                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1747                         ret = btrfs_update_inode_fallback(trans, root, inode);
1748                         BUG_ON(ret);
1749                 }
1750                 goto out;
1751         }
1752
1753         lock_extent_bits(io_tree, ordered_extent->file_offset,
1754                          ordered_extent->file_offset + ordered_extent->len - 1,
1755                          0, &cached_state, GFP_NOFS);
1756
1757         if (nolock)
1758                 trans = btrfs_join_transaction_nolock(root);
1759         else
1760                 trans = btrfs_join_transaction(root);
1761         BUG_ON(IS_ERR(trans));
1762         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1763
1764         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1765                 compress_type = ordered_extent->compress_type;
1766         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1767                 BUG_ON(compress_type);
1768                 ret = btrfs_mark_extent_written(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->file_offset +
1771                                                 ordered_extent->len);
1772                 BUG_ON(ret);
1773         } else {
1774                 BUG_ON(root == root->fs_info->tree_root);
1775                 ret = insert_reserved_file_extent(trans, inode,
1776                                                 ordered_extent->file_offset,
1777                                                 ordered_extent->start,
1778                                                 ordered_extent->disk_len,
1779                                                 ordered_extent->len,
1780                                                 ordered_extent->len,
1781                                                 compress_type, 0, 0,
1782                                                 BTRFS_FILE_EXTENT_REG);
1783                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1784                                    ordered_extent->file_offset,
1785                                    ordered_extent->len);
1786                 BUG_ON(ret);
1787         }
1788         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1789                              ordered_extent->file_offset +
1790                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1791
1792         add_pending_csums(trans, inode, ordered_extent->file_offset,
1793                           &ordered_extent->list);
1794
1795         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1796         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1797                 ret = btrfs_update_inode_fallback(trans, root, inode);
1798                 BUG_ON(ret);
1799         }
1800         ret = 0;
1801 out:
1802         if (root != root->fs_info->tree_root)
1803                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1804         if (trans) {
1805                 if (nolock)
1806                         btrfs_end_transaction_nolock(trans, root);
1807                 else
1808                         btrfs_end_transaction(trans, root);
1809         }
1810
1811         /* once for us */
1812         btrfs_put_ordered_extent(ordered_extent);
1813         /* once for the tree */
1814         btrfs_put_ordered_extent(ordered_extent);
1815
1816         return 0;
1817 }
1818
1819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1820                                 struct extent_state *state, int uptodate)
1821 {
1822         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1823
1824         ClearPagePrivate2(page);
1825         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1826 }
1827
1828 /*
1829  * when reads are done, we need to check csums to verify the data is correct
1830  * if there's a match, we allow the bio to finish.  If not, the code in
1831  * extent_io.c will try to find good copies for us.
1832  */
1833 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1834                                struct extent_state *state)
1835 {
1836         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1837         struct inode *inode = page->mapping->host;
1838         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1839         char *kaddr;
1840         u64 private = ~(u32)0;
1841         int ret;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843         u32 csum = ~(u32)0;
1844
1845         if (PageChecked(page)) {
1846                 ClearPageChecked(page);
1847                 goto good;
1848         }
1849
1850         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1851                 goto good;
1852
1853         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1854             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1855                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1856                                   GFP_NOFS);
1857                 return 0;
1858         }
1859
1860         if (state && state->start == start) {
1861                 private = state->private;
1862                 ret = 0;
1863         } else {
1864                 ret = get_state_private(io_tree, start, &private);
1865         }
1866         kaddr = kmap_atomic(page, KM_USER0);
1867         if (ret)
1868                 goto zeroit;
1869
1870         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1871         btrfs_csum_final(csum, (char *)&csum);
1872         if (csum != private)
1873                 goto zeroit;
1874
1875         kunmap_atomic(kaddr, KM_USER0);
1876 good:
1877         return 0;
1878
1879 zeroit:
1880         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1881                        "private %llu\n",
1882                        (unsigned long long)btrfs_ino(page->mapping->host),
1883                        (unsigned long long)start, csum,
1884                        (unsigned long long)private);
1885         memset(kaddr + offset, 1, end - start + 1);
1886         flush_dcache_page(page);
1887         kunmap_atomic(kaddr, KM_USER0);
1888         if (private == 0)
1889                 return 0;
1890         return -EIO;
1891 }
1892
1893 struct delayed_iput {
1894         struct list_head list;
1895         struct inode *inode;
1896 };
1897
1898 void btrfs_add_delayed_iput(struct inode *inode)
1899 {
1900         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1901         struct delayed_iput *delayed;
1902
1903         if (atomic_add_unless(&inode->i_count, -1, 1))
1904                 return;
1905
1906         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1907         delayed->inode = inode;
1908
1909         spin_lock(&fs_info->delayed_iput_lock);
1910         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1911         spin_unlock(&fs_info->delayed_iput_lock);
1912 }
1913
1914 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1915 {
1916         LIST_HEAD(list);
1917         struct btrfs_fs_info *fs_info = root->fs_info;
1918         struct delayed_iput *delayed;
1919         int empty;
1920
1921         spin_lock(&fs_info->delayed_iput_lock);
1922         empty = list_empty(&fs_info->delayed_iputs);
1923         spin_unlock(&fs_info->delayed_iput_lock);
1924         if (empty)
1925                 return;
1926
1927         down_read(&root->fs_info->cleanup_work_sem);
1928         spin_lock(&fs_info->delayed_iput_lock);
1929         list_splice_init(&fs_info->delayed_iputs, &list);
1930         spin_unlock(&fs_info->delayed_iput_lock);
1931
1932         while (!list_empty(&list)) {
1933                 delayed = list_entry(list.next, struct delayed_iput, list);
1934                 list_del(&delayed->list);
1935                 iput(delayed->inode);
1936                 kfree(delayed);
1937         }
1938         up_read(&root->fs_info->cleanup_work_sem);
1939 }
1940
1941 enum btrfs_orphan_cleanup_state {
1942         ORPHAN_CLEANUP_STARTED  = 1,
1943         ORPHAN_CLEANUP_DONE     = 2,
1944 };
1945
1946 /*
1947  * This is called in transaction commmit time. If there are no orphan
1948  * files in the subvolume, it removes orphan item and frees block_rsv
1949  * structure.
1950  */
1951 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1952                               struct btrfs_root *root)
1953 {
1954         int ret;
1955
1956         if (!list_empty(&root->orphan_list) ||
1957             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1958                 return;
1959
1960         if (root->orphan_item_inserted &&
1961             btrfs_root_refs(&root->root_item) > 0) {
1962                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1963                                             root->root_key.objectid);
1964                 BUG_ON(ret);
1965                 root->orphan_item_inserted = 0;
1966         }
1967
1968         if (root->orphan_block_rsv) {
1969                 WARN_ON(root->orphan_block_rsv->size > 0);
1970                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1971                 root->orphan_block_rsv = NULL;
1972         }
1973 }
1974
1975 /*
1976  * This creates an orphan entry for the given inode in case something goes
1977  * wrong in the middle of an unlink/truncate.
1978  *
1979  * NOTE: caller of this function should reserve 5 units of metadata for
1980  *       this function.
1981  */
1982 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1983 {
1984         struct btrfs_root *root = BTRFS_I(inode)->root;
1985         struct btrfs_block_rsv *block_rsv = NULL;
1986         int reserve = 0;
1987         int insert = 0;
1988         int ret;
1989
1990         if (!root->orphan_block_rsv) {
1991                 block_rsv = btrfs_alloc_block_rsv(root);
1992                 if (!block_rsv)
1993                         return -ENOMEM;
1994         }
1995
1996         spin_lock(&root->orphan_lock);
1997         if (!root->orphan_block_rsv) {
1998                 root->orphan_block_rsv = block_rsv;
1999         } else if (block_rsv) {
2000                 btrfs_free_block_rsv(root, block_rsv);
2001                 block_rsv = NULL;
2002         }
2003
2004         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2005                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2006 #if 0
2007                 /*
2008                  * For proper ENOSPC handling, we should do orphan
2009                  * cleanup when mounting. But this introduces backward
2010                  * compatibility issue.
2011                  */
2012                 if (!xchg(&root->orphan_item_inserted, 1))
2013                         insert = 2;
2014                 else
2015                         insert = 1;
2016 #endif
2017                 insert = 1;
2018         }
2019
2020         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2021                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2022                 reserve = 1;
2023         }
2024         spin_unlock(&root->orphan_lock);
2025
2026         /* grab metadata reservation from transaction handle */
2027         if (reserve) {
2028                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2029                 BUG_ON(ret);
2030         }
2031
2032         /* insert an orphan item to track this unlinked/truncated file */
2033         if (insert >= 1) {
2034                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2035                 BUG_ON(ret);
2036         }
2037
2038         /* insert an orphan item to track subvolume contains orphan files */
2039         if (insert >= 2) {
2040                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2041                                                root->root_key.objectid);
2042                 BUG_ON(ret);
2043         }
2044         return 0;
2045 }
2046
2047 /*
2048  * We have done the truncate/delete so we can go ahead and remove the orphan
2049  * item for this particular inode.
2050  */
2051 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2052 {
2053         struct btrfs_root *root = BTRFS_I(inode)->root;
2054         int delete_item = 0;
2055         int release_rsv = 0;
2056         int ret = 0;
2057
2058         spin_lock(&root->orphan_lock);
2059         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2060                 list_del_init(&BTRFS_I(inode)->i_orphan);
2061                 delete_item = 1;
2062         }
2063
2064         if (BTRFS_I(inode)->orphan_meta_reserved) {
2065                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2066                 release_rsv = 1;
2067         }
2068         spin_unlock(&root->orphan_lock);
2069
2070         if (trans && delete_item) {
2071                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2072                 BUG_ON(ret);
2073         }
2074
2075         if (release_rsv)
2076                 btrfs_orphan_release_metadata(inode);
2077
2078         return 0;
2079 }
2080
2081 /*
2082  * this cleans up any orphans that may be left on the list from the last use
2083  * of this root.
2084  */
2085 int btrfs_orphan_cleanup(struct btrfs_root *root)
2086 {
2087         struct btrfs_path *path;
2088         struct extent_buffer *leaf;
2089         struct btrfs_key key, found_key;
2090         struct btrfs_trans_handle *trans;
2091         struct inode *inode;
2092         u64 last_objectid = 0;
2093         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2094
2095         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2096                 return 0;
2097
2098         path = btrfs_alloc_path();
2099         if (!path) {
2100                 ret = -ENOMEM;
2101                 goto out;
2102         }
2103         path->reada = -1;
2104
2105         key.objectid = BTRFS_ORPHAN_OBJECTID;
2106         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2107         key.offset = (u64)-1;
2108
2109         while (1) {
2110                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2111                 if (ret < 0)
2112                         goto out;
2113
2114                 /*
2115                  * if ret == 0 means we found what we were searching for, which
2116                  * is weird, but possible, so only screw with path if we didn't
2117                  * find the key and see if we have stuff that matches
2118                  */
2119                 if (ret > 0) {
2120                         ret = 0;
2121                         if (path->slots[0] == 0)
2122                                 break;
2123                         path->slots[0]--;
2124                 }
2125
2126                 /* pull out the item */
2127                 leaf = path->nodes[0];
2128                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2129
2130                 /* make sure the item matches what we want */
2131                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2132                         break;
2133                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2134                         break;
2135
2136                 /* release the path since we're done with it */
2137                 btrfs_release_path(path);
2138
2139                 /*
2140                  * this is where we are basically btrfs_lookup, without the
2141                  * crossing root thing.  we store the inode number in the
2142                  * offset of the orphan item.
2143                  */
2144
2145                 if (found_key.offset == last_objectid) {
2146                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2147                                "stopping orphan cleanup\n");
2148                         ret = -EINVAL;
2149                         goto out;
2150                 }
2151
2152                 last_objectid = found_key.offset;
2153
2154                 found_key.objectid = found_key.offset;
2155                 found_key.type = BTRFS_INODE_ITEM_KEY;
2156                 found_key.offset = 0;
2157                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2158                 ret = PTR_RET(inode);
2159                 if (ret && ret != -ESTALE)
2160                         goto out;
2161
2162                 /*
2163                  * Inode is already gone but the orphan item is still there,
2164                  * kill the orphan item.
2165                  */
2166                 if (ret == -ESTALE) {
2167                         trans = btrfs_start_transaction(root, 1);
2168                         if (IS_ERR(trans)) {
2169                                 ret = PTR_ERR(trans);
2170                                 goto out;
2171                         }
2172                         ret = btrfs_del_orphan_item(trans, root,
2173                                                     found_key.objectid);
2174                         BUG_ON(ret);
2175                         btrfs_end_transaction(trans, root);
2176                         continue;
2177                 }
2178
2179                 /*
2180                  * add this inode to the orphan list so btrfs_orphan_del does
2181                  * the proper thing when we hit it
2182                  */
2183                 spin_lock(&root->orphan_lock);
2184                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2185                 spin_unlock(&root->orphan_lock);
2186
2187                 /* if we have links, this was a truncate, lets do that */
2188                 if (inode->i_nlink) {
2189                         if (!S_ISREG(inode->i_mode)) {
2190                                 WARN_ON(1);
2191                                 iput(inode);
2192                                 continue;
2193                         }
2194                         nr_truncate++;
2195                         /*
2196                          * Need to hold the imutex for reservation purposes, not
2197                          * a huge deal here but I have a WARN_ON in
2198                          * btrfs_delalloc_reserve_space to catch offenders.
2199                          */
2200                         mutex_lock(&inode->i_mutex);
2201                         ret = btrfs_truncate(inode);
2202                         mutex_unlock(&inode->i_mutex);
2203                 } else {
2204                         nr_unlink++;
2205                 }
2206
2207                 /* this will do delete_inode and everything for us */
2208                 iput(inode);
2209                 if (ret)
2210                         goto out;
2211         }
2212         /* release the path since we're done with it */
2213         btrfs_release_path(path);
2214
2215         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2216
2217         if (root->orphan_block_rsv)
2218                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2219                                         (u64)-1);
2220
2221         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2222                 trans = btrfs_join_transaction(root);
2223                 if (!IS_ERR(trans))
2224                         btrfs_end_transaction(trans, root);
2225         }
2226
2227         if (nr_unlink)
2228                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2229         if (nr_truncate)
2230                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2231
2232 out:
2233         if (ret)
2234                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2235         btrfs_free_path(path);
2236         return ret;
2237 }
2238
2239 /*
2240  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2241  * don't find any xattrs, we know there can't be any acls.
2242  *
2243  * slot is the slot the inode is in, objectid is the objectid of the inode
2244  */
2245 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2246                                           int slot, u64 objectid)
2247 {
2248         u32 nritems = btrfs_header_nritems(leaf);
2249         struct btrfs_key found_key;
2250         int scanned = 0;
2251
2252         slot++;
2253         while (slot < nritems) {
2254                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2255
2256                 /* we found a different objectid, there must not be acls */
2257                 if (found_key.objectid != objectid)
2258                         return 0;
2259
2260                 /* we found an xattr, assume we've got an acl */
2261                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2262                         return 1;
2263
2264                 /*
2265                  * we found a key greater than an xattr key, there can't
2266                  * be any acls later on
2267                  */
2268                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2269                         return 0;
2270
2271                 slot++;
2272                 scanned++;
2273
2274                 /*
2275                  * it goes inode, inode backrefs, xattrs, extents,
2276                  * so if there are a ton of hard links to an inode there can
2277                  * be a lot of backrefs.  Don't waste time searching too hard,
2278                  * this is just an optimization
2279                  */
2280                 if (scanned >= 8)
2281                         break;
2282         }
2283         /* we hit the end of the leaf before we found an xattr or
2284          * something larger than an xattr.  We have to assume the inode
2285          * has acls
2286          */
2287         return 1;
2288 }
2289
2290 /*
2291  * read an inode from the btree into the in-memory inode
2292  */
2293 static void btrfs_read_locked_inode(struct inode *inode)
2294 {
2295         struct btrfs_path *path;
2296         struct extent_buffer *leaf;
2297         struct btrfs_inode_item *inode_item;
2298         struct btrfs_timespec *tspec;
2299         struct btrfs_root *root = BTRFS_I(inode)->root;
2300         struct btrfs_key location;
2301         int maybe_acls;
2302         u32 rdev;
2303         int ret;
2304         bool filled = false;
2305
2306         ret = btrfs_fill_inode(inode, &rdev);
2307         if (!ret)
2308                 filled = true;
2309
2310         path = btrfs_alloc_path();
2311         if (!path)
2312                 goto make_bad;
2313
2314         path->leave_spinning = 1;
2315         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2316
2317         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2318         if (ret)
2319                 goto make_bad;
2320
2321         leaf = path->nodes[0];
2322
2323         if (filled)
2324                 goto cache_acl;
2325
2326         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2327                                     struct btrfs_inode_item);
2328         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2329         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2330         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2331         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2332         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2333
2334         tspec = btrfs_inode_atime(inode_item);
2335         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2336         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2337
2338         tspec = btrfs_inode_mtime(inode_item);
2339         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2340         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2341
2342         tspec = btrfs_inode_ctime(inode_item);
2343         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2344         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2345
2346         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2347         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2348         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2349         inode->i_generation = BTRFS_I(inode)->generation;
2350         inode->i_rdev = 0;
2351         rdev = btrfs_inode_rdev(leaf, inode_item);
2352
2353         BTRFS_I(inode)->index_cnt = (u64)-1;
2354         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2355 cache_acl:
2356         /*
2357          * try to precache a NULL acl entry for files that don't have
2358          * any xattrs or acls
2359          */
2360         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2361                                            btrfs_ino(inode));
2362         if (!maybe_acls)
2363                 cache_no_acl(inode);
2364
2365         btrfs_free_path(path);
2366
2367         switch (inode->i_mode & S_IFMT) {
2368         case S_IFREG:
2369                 inode->i_mapping->a_ops = &btrfs_aops;
2370                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2371                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2372                 inode->i_fop = &btrfs_file_operations;
2373                 inode->i_op = &btrfs_file_inode_operations;
2374                 break;
2375         case S_IFDIR:
2376                 inode->i_fop = &btrfs_dir_file_operations;
2377                 if (root == root->fs_info->tree_root)
2378                         inode->i_op = &btrfs_dir_ro_inode_operations;
2379                 else
2380                         inode->i_op = &btrfs_dir_inode_operations;
2381                 break;
2382         case S_IFLNK:
2383                 inode->i_op = &btrfs_symlink_inode_operations;
2384                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2385                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2386                 break;
2387         default:
2388                 inode->i_op = &btrfs_special_inode_operations;
2389                 init_special_inode(inode, inode->i_mode, rdev);
2390                 break;
2391         }
2392
2393         btrfs_update_iflags(inode);
2394         return;
2395
2396 make_bad:
2397         btrfs_free_path(path);
2398         make_bad_inode(inode);
2399 }
2400
2401 /*
2402  * given a leaf and an inode, copy the inode fields into the leaf
2403  */
2404 static void fill_inode_item(struct btrfs_trans_handle *trans,
2405                             struct extent_buffer *leaf,
2406                             struct btrfs_inode_item *item,
2407                             struct inode *inode)
2408 {
2409         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2410         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2411         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2412         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2413         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2414
2415         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2416                                inode->i_atime.tv_sec);
2417         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2418                                 inode->i_atime.tv_nsec);
2419
2420         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2421                                inode->i_mtime.tv_sec);
2422         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2423                                 inode->i_mtime.tv_nsec);
2424
2425         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2426                                inode->i_ctime.tv_sec);
2427         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2428                                 inode->i_ctime.tv_nsec);
2429
2430         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2431         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2432         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2433         btrfs_set_inode_transid(leaf, item, trans->transid);
2434         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2435         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2436         btrfs_set_inode_block_group(leaf, item, 0);
2437 }
2438
2439 /*
2440  * copy everything in the in-memory inode into the btree.
2441  */
2442 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2443                                 struct btrfs_root *root, struct inode *inode)
2444 {
2445         struct btrfs_inode_item *inode_item;
2446         struct btrfs_path *path;
2447         struct extent_buffer *leaf;
2448         int ret;
2449
2450         path = btrfs_alloc_path();
2451         if (!path)
2452                 return -ENOMEM;
2453
2454         path->leave_spinning = 1;
2455         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2456                                  1);
2457         if (ret) {
2458                 if (ret > 0)
2459                         ret = -ENOENT;
2460                 goto failed;
2461         }
2462
2463         btrfs_unlock_up_safe(path, 1);
2464         leaf = path->nodes[0];
2465         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2466                                     struct btrfs_inode_item);
2467
2468         fill_inode_item(trans, leaf, inode_item, inode);
2469         btrfs_mark_buffer_dirty(leaf);
2470         btrfs_set_inode_last_trans(trans, inode);
2471         ret = 0;
2472 failed:
2473         btrfs_free_path(path);
2474         return ret;
2475 }
2476
2477 /*
2478  * copy everything in the in-memory inode into the btree.
2479  */
2480 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2481                                 struct btrfs_root *root, struct inode *inode)
2482 {
2483         int ret;
2484
2485         /*
2486          * If the inode is a free space inode, we can deadlock during commit
2487          * if we put it into the delayed code.
2488          *
2489          * The data relocation inode should also be directly updated
2490          * without delay
2491          */
2492         if (!btrfs_is_free_space_inode(root, inode)
2493             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2494                 ret = btrfs_delayed_update_inode(trans, root, inode);
2495                 if (!ret)
2496                         btrfs_set_inode_last_trans(trans, inode);
2497                 return ret;
2498         }
2499
2500         return btrfs_update_inode_item(trans, root, inode);
2501 }
2502
2503 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2504                                 struct btrfs_root *root, struct inode *inode)
2505 {
2506         int ret;
2507
2508         ret = btrfs_update_inode(trans, root, inode);
2509         if (ret == -ENOSPC)
2510                 return btrfs_update_inode_item(trans, root, inode);
2511         return ret;
2512 }
2513
2514 /*
2515  * unlink helper that gets used here in inode.c and in the tree logging
2516  * recovery code.  It remove a link in a directory with a given name, and
2517  * also drops the back refs in the inode to the directory
2518  */
2519 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2520                                 struct btrfs_root *root,
2521                                 struct inode *dir, struct inode *inode,
2522                                 const char *name, int name_len)
2523 {
2524         struct btrfs_path *path;
2525         int ret = 0;
2526         struct extent_buffer *leaf;
2527         struct btrfs_dir_item *di;
2528         struct btrfs_key key;
2529         u64 index;
2530         u64 ino = btrfs_ino(inode);
2531         u64 dir_ino = btrfs_ino(dir);
2532
2533         path = btrfs_alloc_path();
2534         if (!path) {
2535                 ret = -ENOMEM;
2536                 goto out;
2537         }
2538
2539         path->leave_spinning = 1;
2540         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2541                                     name, name_len, -1);
2542         if (IS_ERR(di)) {
2543                 ret = PTR_ERR(di);
2544                 goto err;
2545         }
2546         if (!di) {
2547                 ret = -ENOENT;
2548                 goto err;
2549         }
2550         leaf = path->nodes[0];
2551         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2552         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2553         if (ret)
2554                 goto err;
2555         btrfs_release_path(path);
2556
2557         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2558                                   dir_ino, &index);
2559         if (ret) {
2560                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2561                        "inode %llu parent %llu\n", name_len, name,
2562                        (unsigned long long)ino, (unsigned long long)dir_ino);
2563                 goto err;
2564         }
2565
2566         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2567         if (ret)
2568                 goto err;
2569
2570         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2571                                          inode, dir_ino);
2572         BUG_ON(ret != 0 && ret != -ENOENT);
2573
2574         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2575                                            dir, index);
2576         if (ret == -ENOENT)
2577                 ret = 0;
2578 err:
2579         btrfs_free_path(path);
2580         if (ret)
2581                 goto out;
2582
2583         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2584         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2585         btrfs_update_inode(trans, root, dir);
2586 out:
2587         return ret;
2588 }
2589
2590 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2591                        struct btrfs_root *root,
2592                        struct inode *dir, struct inode *inode,
2593                        const char *name, int name_len)
2594 {
2595         int ret;
2596         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2597         if (!ret) {
2598                 btrfs_drop_nlink(inode);
2599                 ret = btrfs_update_inode(trans, root, inode);
2600         }
2601         return ret;
2602 }
2603
2604
2605 /* helper to check if there is any shared block in the path */
2606 static int check_path_shared(struct btrfs_root *root,
2607                              struct btrfs_path *path)
2608 {
2609         struct extent_buffer *eb;
2610         int level;
2611         u64 refs = 1;
2612
2613         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2614                 int ret;
2615
2616                 if (!path->nodes[level])
2617                         break;
2618                 eb = path->nodes[level];
2619                 if (!btrfs_block_can_be_shared(root, eb))
2620                         continue;
2621                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2622                                                &refs, NULL);
2623                 if (refs > 1)
2624                         return 1;
2625         }
2626         return 0;
2627 }
2628
2629 /*
2630  * helper to start transaction for unlink and rmdir.
2631  *
2632  * unlink and rmdir are special in btrfs, they do not always free space.
2633  * so in enospc case, we should make sure they will free space before
2634  * allowing them to use the global metadata reservation.
2635  */
2636 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2637                                                        struct dentry *dentry)
2638 {
2639         struct btrfs_trans_handle *trans;
2640         struct btrfs_root *root = BTRFS_I(dir)->root;
2641         struct btrfs_path *path;
2642         struct btrfs_inode_ref *ref;
2643         struct btrfs_dir_item *di;
2644         struct inode *inode = dentry->d_inode;
2645         u64 index;
2646         int check_link = 1;
2647         int err = -ENOSPC;
2648         int ret;
2649         u64 ino = btrfs_ino(inode);
2650         u64 dir_ino = btrfs_ino(dir);
2651
2652         /*
2653          * 1 for the possible orphan item
2654          * 1 for the dir item
2655          * 1 for the dir index
2656          * 1 for the inode ref
2657          * 1 for the inode ref in the tree log
2658          * 2 for the dir entries in the log
2659          * 1 for the inode
2660          */
2661         trans = btrfs_start_transaction(root, 8);
2662         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2663                 return trans;
2664
2665         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2666                 return ERR_PTR(-ENOSPC);
2667
2668         /* check if there is someone else holds reference */
2669         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2670                 return ERR_PTR(-ENOSPC);
2671
2672         if (atomic_read(&inode->i_count) > 2)
2673                 return ERR_PTR(-ENOSPC);
2674
2675         if (xchg(&root->fs_info->enospc_unlink, 1))
2676                 return ERR_PTR(-ENOSPC);
2677
2678         path = btrfs_alloc_path();
2679         if (!path) {
2680                 root->fs_info->enospc_unlink = 0;
2681                 return ERR_PTR(-ENOMEM);
2682         }
2683
2684         /* 1 for the orphan item */
2685         trans = btrfs_start_transaction(root, 1);
2686         if (IS_ERR(trans)) {
2687                 btrfs_free_path(path);
2688                 root->fs_info->enospc_unlink = 0;
2689                 return trans;
2690         }
2691
2692         path->skip_locking = 1;
2693         path->search_commit_root = 1;
2694
2695         ret = btrfs_lookup_inode(trans, root, path,
2696                                 &BTRFS_I(dir)->location, 0);
2697         if (ret < 0) {
2698                 err = ret;
2699                 goto out;
2700         }
2701         if (ret == 0) {
2702                 if (check_path_shared(root, path))
2703                         goto out;
2704         } else {
2705                 check_link = 0;
2706         }
2707         btrfs_release_path(path);
2708
2709         ret = btrfs_lookup_inode(trans, root, path,
2710                                 &BTRFS_I(inode)->location, 0);
2711         if (ret < 0) {
2712                 err = ret;
2713                 goto out;
2714         }
2715         if (ret == 0) {
2716                 if (check_path_shared(root, path))
2717                         goto out;
2718         } else {
2719                 check_link = 0;
2720         }
2721         btrfs_release_path(path);
2722
2723         if (ret == 0 && S_ISREG(inode->i_mode)) {
2724                 ret = btrfs_lookup_file_extent(trans, root, path,
2725                                                ino, (u64)-1, 0);
2726                 if (ret < 0) {
2727                         err = ret;
2728                         goto out;
2729                 }
2730                 BUG_ON(ret == 0);
2731                 if (check_path_shared(root, path))
2732                         goto out;
2733                 btrfs_release_path(path);
2734         }
2735
2736         if (!check_link) {
2737                 err = 0;
2738                 goto out;
2739         }
2740
2741         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2742                                 dentry->d_name.name, dentry->d_name.len, 0);
2743         if (IS_ERR(di)) {
2744                 err = PTR_ERR(di);
2745                 goto out;
2746         }
2747         if (di) {
2748                 if (check_path_shared(root, path))
2749                         goto out;
2750         } else {
2751                 err = 0;
2752                 goto out;
2753         }
2754         btrfs_release_path(path);
2755
2756         ref = btrfs_lookup_inode_ref(trans, root, path,
2757                                 dentry->d_name.name, dentry->d_name.len,
2758                                 ino, dir_ino, 0);
2759         if (IS_ERR(ref)) {
2760                 err = PTR_ERR(ref);
2761                 goto out;
2762         }
2763         BUG_ON(!ref);
2764         if (check_path_shared(root, path))
2765                 goto out;
2766         index = btrfs_inode_ref_index(path->nodes[0], ref);
2767         btrfs_release_path(path);
2768
2769         /*
2770          * This is a commit root search, if we can lookup inode item and other
2771          * relative items in the commit root, it means the transaction of
2772          * dir/file creation has been committed, and the dir index item that we
2773          * delay to insert has also been inserted into the commit root. So
2774          * we needn't worry about the delayed insertion of the dir index item
2775          * here.
2776          */
2777         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2778                                 dentry->d_name.name, dentry->d_name.len, 0);
2779         if (IS_ERR(di)) {
2780                 err = PTR_ERR(di);
2781                 goto out;
2782         }
2783         BUG_ON(ret == -ENOENT);
2784         if (check_path_shared(root, path))
2785                 goto out;
2786
2787         err = 0;
2788 out:
2789         btrfs_free_path(path);
2790         /* Migrate the orphan reservation over */
2791         if (!err)
2792                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2793                                 &root->fs_info->global_block_rsv,
2794                                 trans->bytes_reserved);
2795
2796         if (err) {
2797                 btrfs_end_transaction(trans, root);
2798                 root->fs_info->enospc_unlink = 0;
2799                 return ERR_PTR(err);
2800         }
2801
2802         trans->block_rsv = &root->fs_info->global_block_rsv;
2803         return trans;
2804 }
2805
2806 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2807                                struct btrfs_root *root)
2808 {
2809         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2810                 btrfs_block_rsv_release(root, trans->block_rsv,
2811                                         trans->bytes_reserved);
2812                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2813                 BUG_ON(!root->fs_info->enospc_unlink);
2814                 root->fs_info->enospc_unlink = 0;
2815         }
2816         btrfs_end_transaction_throttle(trans, root);
2817 }
2818
2819 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2820 {
2821         struct btrfs_root *root = BTRFS_I(dir)->root;
2822         struct btrfs_trans_handle *trans;
2823         struct inode *inode = dentry->d_inode;
2824         int ret;
2825         unsigned long nr = 0;
2826
2827         trans = __unlink_start_trans(dir, dentry);
2828         if (IS_ERR(trans))
2829                 return PTR_ERR(trans);
2830
2831         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2832
2833         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2834                                  dentry->d_name.name, dentry->d_name.len);
2835         if (ret)
2836                 goto out;
2837
2838         if (inode->i_nlink == 0) {
2839                 ret = btrfs_orphan_add(trans, inode);
2840                 if (ret)
2841                         goto out;
2842         }
2843
2844 out:
2845         nr = trans->blocks_used;
2846         __unlink_end_trans(trans, root);
2847         btrfs_btree_balance_dirty(root, nr);
2848         return ret;
2849 }
2850
2851 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2852                         struct btrfs_root *root,
2853                         struct inode *dir, u64 objectid,
2854                         const char *name, int name_len)
2855 {
2856         struct btrfs_path *path;
2857         struct extent_buffer *leaf;
2858         struct btrfs_dir_item *di;
2859         struct btrfs_key key;
2860         u64 index;
2861         int ret;
2862         u64 dir_ino = btrfs_ino(dir);
2863
2864         path = btrfs_alloc_path();
2865         if (!path)
2866                 return -ENOMEM;
2867
2868         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2869                                    name, name_len, -1);
2870         BUG_ON(IS_ERR_OR_NULL(di));
2871
2872         leaf = path->nodes[0];
2873         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2874         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2875         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2876         BUG_ON(ret);
2877         btrfs_release_path(path);
2878
2879         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2880                                  objectid, root->root_key.objectid,
2881                                  dir_ino, &index, name, name_len);
2882         if (ret < 0) {
2883                 BUG_ON(ret != -ENOENT);
2884                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2885                                                  name, name_len);
2886                 BUG_ON(IS_ERR_OR_NULL(di));
2887
2888                 leaf = path->nodes[0];
2889                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2890                 btrfs_release_path(path);
2891                 index = key.offset;
2892         }
2893         btrfs_release_path(path);
2894
2895         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2896         BUG_ON(ret);
2897
2898         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2899         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2900         ret = btrfs_update_inode(trans, root, dir);
2901         BUG_ON(ret);
2902
2903         btrfs_free_path(path);
2904         return 0;
2905 }
2906
2907 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2908 {
2909         struct inode *inode = dentry->d_inode;
2910         int err = 0;
2911         struct btrfs_root *root = BTRFS_I(dir)->root;
2912         struct btrfs_trans_handle *trans;
2913         unsigned long nr = 0;
2914
2915         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2916             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2917                 return -ENOTEMPTY;
2918
2919         trans = __unlink_start_trans(dir, dentry);
2920         if (IS_ERR(trans))
2921                 return PTR_ERR(trans);
2922
2923         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2924                 err = btrfs_unlink_subvol(trans, root, dir,
2925                                           BTRFS_I(inode)->location.objectid,
2926                                           dentry->d_name.name,
2927                                           dentry->d_name.len);
2928                 goto out;
2929         }
2930
2931         err = btrfs_orphan_add(trans, inode);
2932         if (err)
2933                 goto out;
2934
2935         /* now the directory is empty */
2936         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2937                                  dentry->d_name.name, dentry->d_name.len);
2938         if (!err)
2939                 btrfs_i_size_write(inode, 0);
2940 out:
2941         nr = trans->blocks_used;
2942         __unlink_end_trans(trans, root);
2943         btrfs_btree_balance_dirty(root, nr);
2944
2945         return err;
2946 }
2947
2948 /*
2949  * this can truncate away extent items, csum items and directory items.
2950  * It starts at a high offset and removes keys until it can't find
2951  * any higher than new_size
2952  *
2953  * csum items that cross the new i_size are truncated to the new size
2954  * as well.
2955  *
2956  * min_type is the minimum key type to truncate down to.  If set to 0, this
2957  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2958  */
2959 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2960                                struct btrfs_root *root,
2961                                struct inode *inode,
2962                                u64 new_size, u32 min_type)
2963 {
2964         struct btrfs_path *path;
2965         struct extent_buffer *leaf;
2966         struct btrfs_file_extent_item *fi;
2967         struct btrfs_key key;
2968         struct btrfs_key found_key;
2969         u64 extent_start = 0;
2970         u64 extent_num_bytes = 0;
2971         u64 extent_offset = 0;
2972         u64 item_end = 0;
2973         u64 mask = root->sectorsize - 1;
2974         u32 found_type = (u8)-1;
2975         int found_extent;
2976         int del_item;
2977         int pending_del_nr = 0;
2978         int pending_del_slot = 0;
2979         int extent_type = -1;
2980         int encoding;
2981         int ret;
2982         int err = 0;
2983         u64 ino = btrfs_ino(inode);
2984
2985         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2986
2987         path = btrfs_alloc_path();
2988         if (!path)
2989                 return -ENOMEM;
2990         path->reada = -1;
2991
2992         if (root->ref_cows || root == root->fs_info->tree_root)
2993                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2994
2995         /*
2996          * This function is also used to drop the items in the log tree before
2997          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2998          * it is used to drop the loged items. So we shouldn't kill the delayed
2999          * items.
3000          */
3001         if (min_type == 0 && root == BTRFS_I(inode)->root)
3002                 btrfs_kill_delayed_inode_items(inode);
3003
3004         key.objectid = ino;
3005         key.offset = (u64)-1;
3006         key.type = (u8)-1;
3007
3008 search_again:
3009         path->leave_spinning = 1;
3010         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3011         if (ret < 0) {
3012                 err = ret;
3013                 goto out;
3014         }
3015
3016         if (ret > 0) {
3017                 /* there are no items in the tree for us to truncate, we're
3018                  * done
3019                  */
3020                 if (path->slots[0] == 0)
3021                         goto out;
3022                 path->slots[0]--;
3023         }
3024
3025         while (1) {
3026                 fi = NULL;
3027                 leaf = path->nodes[0];
3028                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3029                 found_type = btrfs_key_type(&found_key);
3030                 encoding = 0;
3031
3032                 if (found_key.objectid != ino)
3033                         break;
3034
3035                 if (found_type < min_type)
3036                         break;
3037
3038                 item_end = found_key.offset;
3039                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3040                         fi = btrfs_item_ptr(leaf, path->slots[0],
3041                                             struct btrfs_file_extent_item);
3042                         extent_type = btrfs_file_extent_type(leaf, fi);
3043                         encoding = btrfs_file_extent_compression(leaf, fi);
3044                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3045                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3046
3047                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3048                                 item_end +=
3049                                     btrfs_file_extent_num_bytes(leaf, fi);
3050                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3051                                 item_end += btrfs_file_extent_inline_len(leaf,
3052                                                                          fi);
3053                         }
3054                         item_end--;
3055                 }
3056                 if (found_type > min_type) {
3057                         del_item = 1;
3058                 } else {
3059                         if (item_end < new_size)
3060                                 break;
3061                         if (found_key.offset >= new_size)
3062                                 del_item = 1;
3063                         else
3064                                 del_item = 0;
3065                 }
3066                 found_extent = 0;
3067                 /* FIXME, shrink the extent if the ref count is only 1 */
3068                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3069                         goto delete;
3070
3071                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3072                         u64 num_dec;
3073                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3074                         if (!del_item && !encoding) {
3075                                 u64 orig_num_bytes =
3076                                         btrfs_file_extent_num_bytes(leaf, fi);
3077                                 extent_num_bytes = new_size -
3078                                         found_key.offset + root->sectorsize - 1;
3079                                 extent_num_bytes = extent_num_bytes &
3080                                         ~((u64)root->sectorsize - 1);
3081                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3082                                                          extent_num_bytes);
3083                                 num_dec = (orig_num_bytes -
3084                                            extent_num_bytes);
3085                                 if (root->ref_cows && extent_start != 0)
3086                                         inode_sub_bytes(inode, num_dec);
3087                                 btrfs_mark_buffer_dirty(leaf);
3088                         } else {
3089                                 extent_num_bytes =
3090                                         btrfs_file_extent_disk_num_bytes(leaf,
3091                                                                          fi);
3092                                 extent_offset = found_key.offset -
3093                                         btrfs_file_extent_offset(leaf, fi);
3094
3095                                 /* FIXME blocksize != 4096 */
3096                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3097                                 if (extent_start != 0) {
3098                                         found_extent = 1;
3099                                         if (root->ref_cows)
3100                                                 inode_sub_bytes(inode, num_dec);
3101                                 }
3102                         }
3103                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3104                         /*
3105                          * we can't truncate inline items that have had
3106                          * special encodings
3107                          */
3108                         if (!del_item &&
3109                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3110                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3111                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3112                                 u32 size = new_size - found_key.offset;
3113
3114                                 if (root->ref_cows) {
3115                                         inode_sub_bytes(inode, item_end + 1 -
3116                                                         new_size);
3117                                 }
3118                                 size =
3119                                     btrfs_file_extent_calc_inline_size(size);
3120                                 ret = btrfs_truncate_item(trans, root, path,
3121                                                           size, 1);
3122                         } else if (root->ref_cows) {
3123                                 inode_sub_bytes(inode, item_end + 1 -
3124                                                 found_key.offset);
3125                         }
3126                 }
3127 delete:
3128                 if (del_item) {
3129                         if (!pending_del_nr) {
3130                                 /* no pending yet, add ourselves */
3131                                 pending_del_slot = path->slots[0];
3132                                 pending_del_nr = 1;
3133                         } else if (pending_del_nr &&
3134                                    path->slots[0] + 1 == pending_del_slot) {
3135                                 /* hop on the pending chunk */
3136                                 pending_del_nr++;
3137                                 pending_del_slot = path->slots[0];
3138                         } else {
3139                                 BUG();
3140                         }
3141                 } else {
3142                         break;
3143                 }
3144                 if (found_extent && (root->ref_cows ||
3145                                      root == root->fs_info->tree_root)) {
3146                         btrfs_set_path_blocking(path);
3147                         ret = btrfs_free_extent(trans, root, extent_start,
3148                                                 extent_num_bytes, 0,
3149                                                 btrfs_header_owner(leaf),
3150                                                 ino, extent_offset);
3151                         BUG_ON(ret);
3152                 }
3153
3154                 if (found_type == BTRFS_INODE_ITEM_KEY)
3155                         break;
3156
3157                 if (path->slots[0] == 0 ||
3158                     path->slots[0] != pending_del_slot) {
3159                         if (root->ref_cows &&
3160                             BTRFS_I(inode)->location.objectid !=
3161                                                 BTRFS_FREE_INO_OBJECTID) {
3162                                 err = -EAGAIN;
3163                                 goto out;
3164                         }
3165                         if (pending_del_nr) {
3166                                 ret = btrfs_del_items(trans, root, path,
3167                                                 pending_del_slot,
3168                                                 pending_del_nr);
3169                                 BUG_ON(ret);
3170                                 pending_del_nr = 0;
3171                         }
3172                         btrfs_release_path(path);
3173                         goto search_again;
3174                 } else {
3175                         path->slots[0]--;
3176                 }
3177         }
3178 out:
3179         if (pending_del_nr) {
3180                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3181                                       pending_del_nr);
3182                 BUG_ON(ret);
3183         }
3184         btrfs_free_path(path);
3185         return err;
3186 }
3187
3188 /*
3189  * taken from block_truncate_page, but does cow as it zeros out
3190  * any bytes left in the last page in the file.
3191  */
3192 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3193 {
3194         struct inode *inode = mapping->host;
3195         struct btrfs_root *root = BTRFS_I(inode)->root;
3196         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3197         struct btrfs_ordered_extent *ordered;
3198         struct extent_state *cached_state = NULL;
3199         char *kaddr;
3200         u32 blocksize = root->sectorsize;
3201         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3202         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3203         struct page *page;
3204         gfp_t mask = btrfs_alloc_write_mask(mapping);
3205         int ret = 0;
3206         u64 page_start;
3207         u64 page_end;
3208
3209         if ((offset & (blocksize - 1)) == 0)
3210                 goto out;
3211         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3212         if (ret)
3213                 goto out;
3214
3215         ret = -ENOMEM;
3216 again:
3217         page = find_or_create_page(mapping, index, mask);
3218         if (!page) {
3219                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3220                 goto out;
3221         }
3222
3223         page_start = page_offset(page);
3224         page_end = page_start + PAGE_CACHE_SIZE - 1;
3225
3226         if (!PageUptodate(page)) {
3227                 ret = btrfs_readpage(NULL, page);
3228                 lock_page(page);
3229                 if (page->mapping != mapping) {
3230                         unlock_page(page);
3231                         page_cache_release(page);
3232                         goto again;
3233                 }
3234                 if (!PageUptodate(page)) {
3235                         ret = -EIO;
3236                         goto out_unlock;
3237                 }
3238         }
3239         wait_on_page_writeback(page);
3240
3241         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3242                          GFP_NOFS);
3243         set_page_extent_mapped(page);
3244
3245         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3246         if (ordered) {
3247                 unlock_extent_cached(io_tree, page_start, page_end,
3248                                      &cached_state, GFP_NOFS);
3249                 unlock_page(page);
3250                 page_cache_release(page);
3251                 btrfs_start_ordered_extent(inode, ordered, 1);
3252                 btrfs_put_ordered_extent(ordered);
3253                 goto again;
3254         }
3255
3256         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3257                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3258                           0, 0, &cached_state, GFP_NOFS);
3259
3260         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3261                                         &cached_state);
3262         if (ret) {
3263                 unlock_extent_cached(io_tree, page_start, page_end,
3264                                      &cached_state, GFP_NOFS);
3265                 goto out_unlock;
3266         }
3267
3268         ret = 0;
3269         if (offset != PAGE_CACHE_SIZE) {
3270                 kaddr = kmap(page);
3271                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3272                 flush_dcache_page(page);
3273                 kunmap(page);
3274         }
3275         ClearPageChecked(page);
3276         set_page_dirty(page);
3277         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3278                              GFP_NOFS);
3279
3280 out_unlock:
3281         if (ret)
3282                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3283         unlock_page(page);
3284         page_cache_release(page);
3285 out:
3286         return ret;
3287 }
3288
3289 /*
3290  * This function puts in dummy file extents for the area we're creating a hole
3291  * for.  So if we are truncating this file to a larger size we need to insert
3292  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3293  * the range between oldsize and size
3294  */
3295 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3296 {
3297         struct btrfs_trans_handle *trans;
3298         struct btrfs_root *root = BTRFS_I(inode)->root;
3299         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3300         struct extent_map *em = NULL;
3301         struct extent_state *cached_state = NULL;
3302         u64 mask = root->sectorsize - 1;
3303         u64 hole_start = (oldsize + mask) & ~mask;
3304         u64 block_end = (size + mask) & ~mask;
3305         u64 last_byte;
3306         u64 cur_offset;
3307         u64 hole_size;
3308         int err = 0;
3309
3310         if (size <= hole_start)
3311                 return 0;
3312
3313         while (1) {
3314                 struct btrfs_ordered_extent *ordered;
3315                 btrfs_wait_ordered_range(inode, hole_start,
3316                                          block_end - hole_start);
3317                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3318                                  &cached_state, GFP_NOFS);
3319                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3320                 if (!ordered)
3321                         break;
3322                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3323                                      &cached_state, GFP_NOFS);
3324                 btrfs_put_ordered_extent(ordered);
3325         }
3326
3327         cur_offset = hole_start;
3328         while (1) {
3329                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3330                                 block_end - cur_offset, 0);
3331                 BUG_ON(IS_ERR_OR_NULL(em));
3332                 last_byte = min(extent_map_end(em), block_end);
3333                 last_byte = (last_byte + mask) & ~mask;
3334                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3335                         u64 hint_byte = 0;
3336                         hole_size = last_byte - cur_offset;
3337
3338                         trans = btrfs_start_transaction(root, 2);
3339                         if (IS_ERR(trans)) {
3340                                 err = PTR_ERR(trans);
3341                                 break;
3342                         }
3343
3344                         err = btrfs_drop_extents(trans, inode, cur_offset,
3345                                                  cur_offset + hole_size,
3346                                                  &hint_byte, 1);
3347                         if (err) {
3348                                 btrfs_end_transaction(trans, root);
3349                                 break;
3350                         }
3351
3352                         err = btrfs_insert_file_extent(trans, root,
3353                                         btrfs_ino(inode), cur_offset, 0,
3354                                         0, hole_size, 0, hole_size,
3355                                         0, 0, 0);
3356                         if (err) {
3357                                 btrfs_end_transaction(trans, root);
3358                                 break;
3359                         }
3360
3361                         btrfs_drop_extent_cache(inode, hole_start,
3362                                         last_byte - 1, 0);
3363
3364                         btrfs_end_transaction(trans, root);
3365                 }
3366                 free_extent_map(em);
3367                 em = NULL;
3368                 cur_offset = last_byte;
3369                 if (cur_offset >= block_end)
3370                         break;
3371         }
3372
3373         free_extent_map(em);
3374         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3375                              GFP_NOFS);
3376         return err;
3377 }
3378
3379 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3380 {
3381         loff_t oldsize = i_size_read(inode);
3382         int ret;
3383
3384         if (newsize == oldsize)
3385                 return 0;
3386
3387         if (newsize > oldsize) {
3388                 i_size_write(inode, newsize);
3389                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3390                 truncate_pagecache(inode, oldsize, newsize);
3391                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3392                 if (ret) {
3393                         btrfs_setsize(inode, oldsize);
3394                         return ret;
3395                 }
3396
3397                 ret = btrfs_dirty_inode(inode);
3398         } else {
3399
3400                 /*
3401                  * We're truncating a file that used to have good data down to
3402                  * zero. Make sure it gets into the ordered flush list so that
3403                  * any new writes get down to disk quickly.
3404                  */
3405                 if (newsize == 0)
3406                         BTRFS_I(inode)->ordered_data_close = 1;
3407
3408                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3409                 truncate_setsize(inode, newsize);
3410                 ret = btrfs_truncate(inode);
3411         }
3412
3413         return ret;
3414 }
3415
3416 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3417 {
3418         struct inode *inode = dentry->d_inode;
3419         struct btrfs_root *root = BTRFS_I(inode)->root;
3420         int err;
3421
3422         if (btrfs_root_readonly(root))
3423                 return -EROFS;
3424
3425         err = inode_change_ok(inode, attr);
3426         if (err)
3427                 return err;
3428
3429         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3430                 err = btrfs_setsize(inode, attr->ia_size);
3431                 if (err)
3432                         return err;
3433         }
3434
3435         if (attr->ia_valid) {
3436                 setattr_copy(inode, attr);
3437                 err = btrfs_dirty_inode(inode);
3438
3439                 if (!err && attr->ia_valid & ATTR_MODE)
3440                         err = btrfs_acl_chmod(inode);
3441         }
3442
3443         return err;
3444 }
3445
3446 void btrfs_evict_inode(struct inode *inode)
3447 {
3448         struct btrfs_trans_handle *trans;
3449         struct btrfs_root *root = BTRFS_I(inode)->root;
3450         struct btrfs_block_rsv *rsv, *global_rsv;
3451         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3452         unsigned long nr;
3453         int ret;
3454
3455         trace_btrfs_inode_evict(inode);
3456
3457         truncate_inode_pages(&inode->i_data, 0);
3458         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3459                                btrfs_is_free_space_inode(root, inode)))
3460                 goto no_delete;
3461
3462         if (is_bad_inode(inode)) {
3463                 btrfs_orphan_del(NULL, inode);
3464                 goto no_delete;
3465         }
3466         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3467         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3468
3469         if (root->fs_info->log_root_recovering) {
3470                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3471                 goto no_delete;
3472         }
3473
3474         if (inode->i_nlink > 0) {
3475                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3476                 goto no_delete;
3477         }
3478
3479         rsv = btrfs_alloc_block_rsv(root);
3480         if (!rsv) {
3481                 btrfs_orphan_del(NULL, inode);
3482                 goto no_delete;
3483         }
3484         rsv->size = min_size;
3485         global_rsv = &root->fs_info->global_block_rsv;
3486
3487         btrfs_i_size_write(inode, 0);
3488
3489         /*
3490          * This is a bit simpler than btrfs_truncate since
3491          *
3492          * 1) We've already reserved our space for our orphan item in the
3493          *    unlink.
3494          * 2) We're going to delete the inode item, so we don't need to update
3495          *    it at all.
3496          *
3497          * So we just need to reserve some slack space in case we add bytes when
3498          * doing the truncate.
3499          */
3500         while (1) {
3501                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3502
3503                 /*
3504                  * Try and steal from the global reserve since we will
3505                  * likely not use this space anyway, we want to try as
3506                  * hard as possible to get this to work.
3507                  */
3508                 if (ret)
3509                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3510
3511                 if (ret) {
3512                         printk(KERN_WARNING "Could not get space for a "
3513                                "delete, will truncate on mount %d\n", ret);
3514                         btrfs_orphan_del(NULL, inode);
3515                         btrfs_free_block_rsv(root, rsv);
3516                         goto no_delete;
3517                 }
3518
3519                 trans = btrfs_start_transaction(root, 0);
3520                 if (IS_ERR(trans)) {
3521                         btrfs_orphan_del(NULL, inode);
3522                         btrfs_free_block_rsv(root, rsv);
3523                         goto no_delete;
3524                 }
3525
3526                 trans->block_rsv = rsv;
3527
3528                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3529                 if (ret != -EAGAIN)
3530                         break;
3531
3532                 nr = trans->blocks_used;
3533                 btrfs_end_transaction(trans, root);
3534                 trans = NULL;
3535                 btrfs_btree_balance_dirty(root, nr);
3536         }
3537
3538         btrfs_free_block_rsv(root, rsv);
3539
3540         if (ret == 0) {
3541                 trans->block_rsv = root->orphan_block_rsv;
3542                 ret = btrfs_orphan_del(trans, inode);
3543                 BUG_ON(ret);
3544         }
3545
3546         trans->block_rsv = &root->fs_info->trans_block_rsv;
3547         if (!(root == root->fs_info->tree_root ||
3548               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3549                 btrfs_return_ino(root, btrfs_ino(inode));
3550
3551         nr = trans->blocks_used;
3552         btrfs_end_transaction(trans, root);
3553         btrfs_btree_balance_dirty(root, nr);
3554 no_delete:
3555         end_writeback(inode);
3556         return;
3557 }
3558
3559 /*
3560  * this returns the key found in the dir entry in the location pointer.
3561  * If no dir entries were found, location->objectid is 0.
3562  */
3563 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3564                                struct btrfs_key *location)
3565 {
3566         const char *name = dentry->d_name.name;
3567         int namelen = dentry->d_name.len;
3568         struct btrfs_dir_item *di;
3569         struct btrfs_path *path;
3570         struct btrfs_root *root = BTRFS_I(dir)->root;
3571         int ret = 0;
3572
3573         path = btrfs_alloc_path();
3574         if (!path)
3575                 return -ENOMEM;
3576
3577         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3578                                     namelen, 0);
3579         if (IS_ERR(di))
3580                 ret = PTR_ERR(di);
3581
3582         if (IS_ERR_OR_NULL(di))
3583                 goto out_err;
3584
3585         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3586 out:
3587         btrfs_free_path(path);
3588         return ret;
3589 out_err:
3590         location->objectid = 0;
3591         goto out;
3592 }
3593
3594 /*
3595  * when we hit a tree root in a directory, the btrfs part of the inode
3596  * needs to be changed to reflect the root directory of the tree root.  This
3597  * is kind of like crossing a mount point.
3598  */
3599 static int fixup_tree_root_location(struct btrfs_root *root,
3600                                     struct inode *dir,
3601                                     struct dentry *dentry,
3602                                     struct btrfs_key *location,
3603                                     struct btrfs_root **sub_root)
3604 {
3605         struct btrfs_path *path;
3606         struct btrfs_root *new_root;
3607         struct btrfs_root_ref *ref;
3608         struct extent_buffer *leaf;
3609         int ret;
3610         int err = 0;
3611
3612         path = btrfs_alloc_path();
3613         if (!path) {
3614                 err = -ENOMEM;
3615                 goto out;
3616         }
3617
3618         err = -ENOENT;
3619         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3620                                   BTRFS_I(dir)->root->root_key.objectid,
3621                                   location->objectid);
3622         if (ret) {
3623                 if (ret < 0)
3624                         err = ret;
3625                 goto out;
3626         }
3627
3628         leaf = path->nodes[0];
3629         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3630         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3631             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3632                 goto out;
3633
3634         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3635                                    (unsigned long)(ref + 1),
3636                                    dentry->d_name.len);
3637         if (ret)
3638                 goto out;
3639
3640         btrfs_release_path(path);
3641
3642         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3643         if (IS_ERR(new_root)) {
3644                 err = PTR_ERR(new_root);
3645                 goto out;
3646         }
3647
3648         if (btrfs_root_refs(&new_root->root_item) == 0) {
3649                 err = -ENOENT;
3650                 goto out;
3651         }
3652
3653         *sub_root = new_root;
3654         location->objectid = btrfs_root_dirid(&new_root->root_item);
3655         location->type = BTRFS_INODE_ITEM_KEY;
3656         location->offset = 0;
3657         err = 0;
3658 out:
3659         btrfs_free_path(path);
3660         return err;
3661 }
3662
3663 static void inode_tree_add(struct inode *inode)
3664 {
3665         struct btrfs_root *root = BTRFS_I(inode)->root;
3666         struct btrfs_inode *entry;
3667         struct rb_node **p;
3668         struct rb_node *parent;
3669         u64 ino = btrfs_ino(inode);
3670 again:
3671         p = &root->inode_tree.rb_node;
3672         parent = NULL;
3673
3674         if (inode_unhashed(inode))
3675                 return;
3676
3677         spin_lock(&root->inode_lock);
3678         while (*p) {
3679                 parent = *p;
3680                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3681
3682                 if (ino < btrfs_ino(&entry->vfs_inode))
3683                         p = &parent->rb_left;
3684                 else if (ino > btrfs_ino(&entry->vfs_inode))
3685                         p = &parent->rb_right;
3686                 else {
3687                         WARN_ON(!(entry->vfs_inode.i_state &
3688                                   (I_WILL_FREE | I_FREEING)));
3689                         rb_erase(parent, &root->inode_tree);
3690                         RB_CLEAR_NODE(parent);
3691                         spin_unlock(&root->inode_lock);
3692                         goto again;
3693                 }
3694         }
3695         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3696         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3697         spin_unlock(&root->inode_lock);
3698 }
3699
3700 static void inode_tree_del(struct inode *inode)
3701 {
3702         struct btrfs_root *root = BTRFS_I(inode)->root;
3703         int empty = 0;
3704
3705         spin_lock(&root->inode_lock);
3706         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3707                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3708                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3709                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3710         }
3711         spin_unlock(&root->inode_lock);
3712
3713         /*
3714          * Free space cache has inodes in the tree root, but the tree root has a
3715          * root_refs of 0, so this could end up dropping the tree root as a
3716          * snapshot, so we need the extra !root->fs_info->tree_root check to
3717          * make sure we don't drop it.
3718          */
3719         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3720             root != root->fs_info->tree_root) {
3721                 synchronize_srcu(&root->fs_info->subvol_srcu);
3722                 spin_lock(&root->inode_lock);
3723                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3724                 spin_unlock(&root->inode_lock);
3725                 if (empty)
3726                         btrfs_add_dead_root(root);
3727         }
3728 }
3729
3730 int btrfs_invalidate_inodes(struct btrfs_root *root)
3731 {
3732         struct rb_node *node;
3733         struct rb_node *prev;
3734         struct btrfs_inode *entry;
3735         struct inode *inode;
3736         u64 objectid = 0;
3737
3738         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3739
3740         spin_lock(&root->inode_lock);
3741 again:
3742         node = root->inode_tree.rb_node;
3743         prev = NULL;
3744         while (node) {
3745                 prev = node;
3746                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3747
3748                 if (objectid < btrfs_ino(&entry->vfs_inode))
3749                         node = node->rb_left;
3750                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3751                         node = node->rb_right;
3752                 else
3753                         break;
3754         }
3755         if (!node) {
3756                 while (prev) {
3757                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3758                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3759                                 node = prev;
3760                                 break;
3761                         }
3762                         prev = rb_next(prev);
3763                 }
3764         }
3765         while (node) {
3766                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3767                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3768                 inode = igrab(&entry->vfs_inode);
3769                 if (inode) {
3770                         spin_unlock(&root->inode_lock);
3771                         if (atomic_read(&inode->i_count) > 1)
3772                                 d_prune_aliases(inode);
3773                         /*
3774                          * btrfs_drop_inode will have it removed from
3775                          * the inode cache when its usage count
3776                          * hits zero.
3777                          */
3778                         iput(inode);
3779                         cond_resched();
3780                         spin_lock(&root->inode_lock);
3781                         goto again;
3782                 }
3783
3784                 if (cond_resched_lock(&root->inode_lock))
3785                         goto again;
3786
3787                 node = rb_next(node);
3788         }
3789         spin_unlock(&root->inode_lock);
3790         return 0;
3791 }
3792
3793 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3794 {
3795         struct btrfs_iget_args *args = p;
3796         inode->i_ino = args->ino;
3797         BTRFS_I(inode)->root = args->root;
3798         btrfs_set_inode_space_info(args->root, inode);
3799         return 0;
3800 }
3801
3802 static int btrfs_find_actor(struct inode *inode, void *opaque)
3803 {
3804         struct btrfs_iget_args *args = opaque;
3805         return args->ino == btrfs_ino(inode) &&
3806                 args->root == BTRFS_I(inode)->root;
3807 }
3808
3809 static struct inode *btrfs_iget_locked(struct super_block *s,
3810                                        u64 objectid,
3811                                        struct btrfs_root *root)
3812 {
3813         struct inode *inode;
3814         struct btrfs_iget_args args;
3815         args.ino = objectid;
3816         args.root = root;
3817
3818         inode = iget5_locked(s, objectid, btrfs_find_actor,
3819                              btrfs_init_locked_inode,
3820                              (void *)&args);
3821         return inode;
3822 }
3823
3824 /* Get an inode object given its location and corresponding root.
3825  * Returns in *is_new if the inode was read from disk
3826  */
3827 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3828                          struct btrfs_root *root, int *new)
3829 {
3830         struct inode *inode;
3831
3832         inode = btrfs_iget_locked(s, location->objectid, root);
3833         if (!inode)
3834                 return ERR_PTR(-ENOMEM);
3835
3836         if (inode->i_state & I_NEW) {
3837                 BTRFS_I(inode)->root = root;
3838                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3839                 btrfs_read_locked_inode(inode);
3840                 if (!is_bad_inode(inode)) {
3841                         inode_tree_add(inode);
3842                         unlock_new_inode(inode);
3843                         if (new)
3844                                 *new = 1;
3845                 } else {
3846                         unlock_new_inode(inode);
3847                         iput(inode);
3848                         inode = ERR_PTR(-ESTALE);
3849                 }
3850         }
3851
3852         return inode;
3853 }
3854
3855 static struct inode *new_simple_dir(struct super_block *s,
3856                                     struct btrfs_key *key,
3857                                     struct btrfs_root *root)
3858 {
3859         struct inode *inode = new_inode(s);
3860
3861         if (!inode)
3862                 return ERR_PTR(-ENOMEM);
3863
3864         BTRFS_I(inode)->root = root;
3865         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3866         BTRFS_I(inode)->dummy_inode = 1;
3867
3868         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3869         inode->i_op = &simple_dir_inode_operations;
3870         inode->i_fop = &simple_dir_operations;
3871         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3872         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3873
3874         return inode;
3875 }
3876
3877 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3878 {
3879         struct inode *inode;
3880         struct btrfs_root *root = BTRFS_I(dir)->root;
3881         struct btrfs_root *sub_root = root;
3882         struct btrfs_key location;
3883         int index;
3884         int ret = 0;
3885
3886         if (dentry->d_name.len > BTRFS_NAME_LEN)
3887                 return ERR_PTR(-ENAMETOOLONG);
3888
3889         if (unlikely(d_need_lookup(dentry))) {
3890                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3891                 kfree(dentry->d_fsdata);
3892                 dentry->d_fsdata = NULL;
3893                 /* This thing is hashed, drop it for now */
3894                 d_drop(dentry);
3895         } else {
3896                 ret = btrfs_inode_by_name(dir, dentry, &location);
3897         }
3898
3899         if (ret < 0)
3900                 return ERR_PTR(ret);
3901
3902         if (location.objectid == 0)
3903                 return NULL;
3904
3905         if (location.type == BTRFS_INODE_ITEM_KEY) {
3906                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3907                 return inode;
3908         }
3909
3910         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3911
3912         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3913         ret = fixup_tree_root_location(root, dir, dentry,
3914                                        &location, &sub_root);
3915         if (ret < 0) {
3916                 if (ret != -ENOENT)
3917                         inode = ERR_PTR(ret);
3918                 else
3919                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3920         } else {
3921                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3922         }
3923         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3924
3925         if (!IS_ERR(inode) && root != sub_root) {
3926                 down_read(&root->fs_info->cleanup_work_sem);
3927                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3928                         ret = btrfs_orphan_cleanup(sub_root);
3929                 up_read(&root->fs_info->cleanup_work_sem);
3930                 if (ret)
3931                         inode = ERR_PTR(ret);
3932         }
3933
3934         return inode;
3935 }
3936
3937 static int btrfs_dentry_delete(const struct dentry *dentry)
3938 {
3939         struct btrfs_root *root;
3940
3941         if (!dentry->d_inode && !IS_ROOT(dentry))
3942                 dentry = dentry->d_parent;
3943
3944         if (dentry->d_inode) {
3945                 root = BTRFS_I(dentry->d_inode)->root;
3946                 if (btrfs_root_refs(&root->root_item) == 0)
3947                         return 1;
3948         }
3949         return 0;
3950 }
3951
3952 static void btrfs_dentry_release(struct dentry *dentry)
3953 {
3954         if (dentry->d_fsdata)
3955                 kfree(dentry->d_fsdata);
3956 }
3957
3958 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3959                                    struct nameidata *nd)
3960 {
3961         struct dentry *ret;
3962
3963         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3964         if (unlikely(d_need_lookup(dentry))) {
3965                 spin_lock(&dentry->d_lock);
3966                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3967                 spin_unlock(&dentry->d_lock);
3968         }
3969         return ret;
3970 }
3971
3972 unsigned char btrfs_filetype_table[] = {
3973         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3974 };
3975
3976 static int btrfs_real_readdir(struct file *filp, void *dirent,
3977                               filldir_t filldir)
3978 {
3979         struct inode *inode = filp->f_dentry->d_inode;
3980         struct btrfs_root *root = BTRFS_I(inode)->root;
3981         struct btrfs_item *item;
3982         struct btrfs_dir_item *di;
3983         struct btrfs_key key;
3984         struct btrfs_key found_key;
3985         struct btrfs_path *path;
3986         struct list_head ins_list;
3987         struct list_head del_list;
3988         struct qstr q;
3989         int ret;
3990         struct extent_buffer *leaf;
3991         int slot;
3992         unsigned char d_type;
3993         int over = 0;
3994         u32 di_cur;
3995         u32 di_total;
3996         u32 di_len;
3997         int key_type = BTRFS_DIR_INDEX_KEY;
3998         char tmp_name[32];
3999         char *name_ptr;
4000         int name_len;
4001         int is_curr = 0;        /* filp->f_pos points to the current index? */
4002
4003         /* FIXME, use a real flag for deciding about the key type */
4004         if (root->fs_info->tree_root == root)
4005                 key_type = BTRFS_DIR_ITEM_KEY;
4006
4007         /* special case for "." */
4008         if (filp->f_pos == 0) {
4009                 over = filldir(dirent, ".", 1,
4010                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4011                 if (over)
4012                         return 0;
4013                 filp->f_pos = 1;
4014         }
4015         /* special case for .., just use the back ref */
4016         if (filp->f_pos == 1) {
4017                 u64 pino = parent_ino(filp->f_path.dentry);
4018                 over = filldir(dirent, "..", 2,
4019                                filp->f_pos, pino, DT_DIR);
4020                 if (over)
4021                         return 0;
4022                 filp->f_pos = 2;
4023         }
4024         path = btrfs_alloc_path();
4025         if (!path)
4026                 return -ENOMEM;
4027
4028         path->reada = 1;
4029
4030         if (key_type == BTRFS_DIR_INDEX_KEY) {
4031                 INIT_LIST_HEAD(&ins_list);
4032                 INIT_LIST_HEAD(&del_list);
4033                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4034         }
4035
4036         btrfs_set_key_type(&key, key_type);
4037         key.offset = filp->f_pos;
4038         key.objectid = btrfs_ino(inode);
4039
4040         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4041         if (ret < 0)
4042                 goto err;
4043
4044         while (1) {
4045                 leaf = path->nodes[0];
4046                 slot = path->slots[0];
4047                 if (slot >= btrfs_header_nritems(leaf)) {
4048                         ret = btrfs_next_leaf(root, path);
4049                         if (ret < 0)
4050                                 goto err;
4051                         else if (ret > 0)
4052                                 break;
4053                         continue;
4054                 }
4055
4056                 item = btrfs_item_nr(leaf, slot);
4057                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4058
4059                 if (found_key.objectid != key.objectid)
4060                         break;
4061                 if (btrfs_key_type(&found_key) != key_type)
4062                         break;
4063                 if (found_key.offset < filp->f_pos)
4064                         goto next;
4065                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4066                     btrfs_should_delete_dir_index(&del_list,
4067                                                   found_key.offset))
4068                         goto next;
4069
4070                 filp->f_pos = found_key.offset;
4071                 is_curr = 1;
4072
4073                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4074                 di_cur = 0;
4075                 di_total = btrfs_item_size(leaf, item);
4076
4077                 while (di_cur < di_total) {
4078                         struct btrfs_key location;
4079                         struct dentry *tmp;
4080
4081                         if (verify_dir_item(root, leaf, di))
4082                                 break;
4083
4084                         name_len = btrfs_dir_name_len(leaf, di);
4085                         if (name_len <= sizeof(tmp_name)) {
4086                                 name_ptr = tmp_name;
4087                         } else {
4088                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4089                                 if (!name_ptr) {
4090                                         ret = -ENOMEM;
4091                                         goto err;
4092                                 }
4093                         }
4094                         read_extent_buffer(leaf, name_ptr,
4095                                            (unsigned long)(di + 1), name_len);
4096
4097                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4098                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4099
4100                         q.name = name_ptr;
4101                         q.len = name_len;
4102                         q.hash = full_name_hash(q.name, q.len);
4103                         tmp = d_lookup(filp->f_dentry, &q);
4104                         if (!tmp) {
4105                                 struct btrfs_key *newkey;
4106
4107                                 newkey = kzalloc(sizeof(struct btrfs_key),
4108                                                  GFP_NOFS);
4109                                 if (!newkey)
4110                                         goto no_dentry;
4111                                 tmp = d_alloc(filp->f_dentry, &q);
4112                                 if (!tmp) {
4113                                         kfree(newkey);
4114                                         dput(tmp);
4115                                         goto no_dentry;
4116                                 }
4117                                 memcpy(newkey, &location,
4118                                        sizeof(struct btrfs_key));
4119                                 tmp->d_fsdata = newkey;
4120                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4121                                 d_rehash(tmp);
4122                                 dput(tmp);
4123                         } else {
4124                                 dput(tmp);
4125                         }
4126 no_dentry:
4127                         /* is this a reference to our own snapshot? If so
4128                          * skip it
4129                          */
4130                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4131                             location.objectid == root->root_key.objectid) {
4132                                 over = 0;
4133                                 goto skip;
4134                         }
4135                         over = filldir(dirent, name_ptr, name_len,
4136                                        found_key.offset, location.objectid,
4137                                        d_type);
4138
4139 skip:
4140                         if (name_ptr != tmp_name)
4141                                 kfree(name_ptr);
4142
4143                         if (over)
4144                                 goto nopos;
4145                         di_len = btrfs_dir_name_len(leaf, di) +
4146                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4147                         di_cur += di_len;
4148                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4149                 }
4150 next:
4151                 path->slots[0]++;
4152         }
4153
4154         if (key_type == BTRFS_DIR_INDEX_KEY) {
4155                 if (is_curr)
4156                         filp->f_pos++;
4157                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4158                                                       &ins_list);
4159                 if (ret)
4160                         goto nopos;
4161         }
4162
4163         /* Reached end of directory/root. Bump pos past the last item. */
4164         if (key_type == BTRFS_DIR_INDEX_KEY)
4165                 /*
4166                  * 32-bit glibc will use getdents64, but then strtol -
4167                  * so the last number we can serve is this.
4168                  */
4169                 filp->f_pos = 0x7fffffff;
4170         else
4171                 filp->f_pos++;
4172 nopos:
4173         ret = 0;
4174 err:
4175         if (key_type == BTRFS_DIR_INDEX_KEY)
4176                 btrfs_put_delayed_items(&ins_list, &del_list);
4177         btrfs_free_path(path);
4178         return ret;
4179 }
4180
4181 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4182 {
4183         struct btrfs_root *root = BTRFS_I(inode)->root;
4184         struct btrfs_trans_handle *trans;
4185         int ret = 0;
4186         bool nolock = false;
4187
4188         if (BTRFS_I(inode)->dummy_inode)
4189                 return 0;
4190
4191         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4192                 nolock = true;
4193
4194         if (wbc->sync_mode == WB_SYNC_ALL) {
4195                 if (nolock)
4196                         trans = btrfs_join_transaction_nolock(root);
4197                 else
4198                         trans = btrfs_join_transaction(root);
4199                 if (IS_ERR(trans))
4200                         return PTR_ERR(trans);
4201                 if (nolock)
4202                         ret = btrfs_end_transaction_nolock(trans, root);
4203                 else
4204                         ret = btrfs_commit_transaction(trans, root);
4205         }
4206         return ret;
4207 }
4208
4209 /*
4210  * This is somewhat expensive, updating the tree every time the
4211  * inode changes.  But, it is most likely to find the inode in cache.
4212  * FIXME, needs more benchmarking...there are no reasons other than performance
4213  * to keep or drop this code.
4214  */
4215 int btrfs_dirty_inode(struct inode *inode)
4216 {
4217         struct btrfs_root *root = BTRFS_I(inode)->root;
4218         struct btrfs_trans_handle *trans;
4219         int ret;
4220
4221         if (BTRFS_I(inode)->dummy_inode)
4222                 return 0;
4223
4224         trans = btrfs_join_transaction(root);
4225         if (IS_ERR(trans))
4226                 return PTR_ERR(trans);
4227
4228         ret = btrfs_update_inode(trans, root, inode);
4229         if (ret && ret == -ENOSPC) {
4230                 /* whoops, lets try again with the full transaction */
4231                 btrfs_end_transaction(trans, root);
4232                 trans = btrfs_start_transaction(root, 1);
4233                 if (IS_ERR(trans))
4234                         return PTR_ERR(trans);
4235
4236                 ret = btrfs_update_inode(trans, root, inode);
4237         }
4238         btrfs_end_transaction(trans, root);
4239         if (BTRFS_I(inode)->delayed_node)
4240                 btrfs_balance_delayed_items(root);
4241
4242         return ret;
4243 }
4244
4245 /*
4246  * This is a copy of file_update_time.  We need this so we can return error on
4247  * ENOSPC for updating the inode in the case of file write and mmap writes.
4248  */
4249 int btrfs_update_time(struct file *file)
4250 {
4251         struct inode *inode = file->f_path.dentry->d_inode;
4252         struct timespec now;
4253         int ret;
4254         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4255
4256         /* First try to exhaust all avenues to not sync */
4257         if (IS_NOCMTIME(inode))
4258                 return 0;
4259
4260         now = current_fs_time(inode->i_sb);
4261         if (!timespec_equal(&inode->i_mtime, &now))
4262                 sync_it = S_MTIME;
4263
4264         if (!timespec_equal(&inode->i_ctime, &now))
4265                 sync_it |= S_CTIME;
4266
4267         if (IS_I_VERSION(inode))
4268                 sync_it |= S_VERSION;
4269
4270         if (!sync_it)
4271                 return 0;
4272
4273         /* Finally allowed to write? Takes lock. */
4274         if (mnt_want_write_file(file))
4275                 return 0;
4276
4277         /* Only change inode inside the lock region */
4278         if (sync_it & S_VERSION)
4279                 inode_inc_iversion(inode);
4280         if (sync_it & S_CTIME)
4281                 inode->i_ctime = now;
4282         if (sync_it & S_MTIME)
4283                 inode->i_mtime = now;
4284         ret = btrfs_dirty_inode(inode);
4285         if (!ret)
4286                 mark_inode_dirty_sync(inode);
4287         mnt_drop_write(file->f_path.mnt);
4288         return ret;
4289 }
4290
4291 /*
4292  * find the highest existing sequence number in a directory
4293  * and then set the in-memory index_cnt variable to reflect
4294  * free sequence numbers
4295  */
4296 static int btrfs_set_inode_index_count(struct inode *inode)
4297 {
4298         struct btrfs_root *root = BTRFS_I(inode)->root;
4299         struct btrfs_key key, found_key;
4300         struct btrfs_path *path;
4301         struct extent_buffer *leaf;
4302         int ret;
4303
4304         key.objectid = btrfs_ino(inode);
4305         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4306         key.offset = (u64)-1;
4307
4308         path = btrfs_alloc_path();
4309         if (!path)
4310                 return -ENOMEM;
4311
4312         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4313         if (ret < 0)
4314                 goto out;
4315         /* FIXME: we should be able to handle this */
4316         if (ret == 0)
4317                 goto out;
4318         ret = 0;
4319
4320         /*
4321          * MAGIC NUMBER EXPLANATION:
4322          * since we search a directory based on f_pos we have to start at 2
4323          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4324          * else has to start at 2
4325          */
4326         if (path->slots[0] == 0) {
4327                 BTRFS_I(inode)->index_cnt = 2;
4328                 goto out;
4329         }
4330
4331         path->slots[0]--;
4332
4333         leaf = path->nodes[0];
4334         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4335
4336         if (found_key.objectid != btrfs_ino(inode) ||
4337             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4338                 BTRFS_I(inode)->index_cnt = 2;
4339                 goto out;
4340         }
4341
4342         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4343 out:
4344         btrfs_free_path(path);
4345         return ret;
4346 }
4347
4348 /*
4349  * helper to find a free sequence number in a given directory.  This current
4350  * code is very simple, later versions will do smarter things in the btree
4351  */
4352 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4353 {
4354         int ret = 0;
4355
4356         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4357                 ret = btrfs_inode_delayed_dir_index_count(dir);
4358                 if (ret) {
4359                         ret = btrfs_set_inode_index_count(dir);
4360                         if (ret)
4361                                 return ret;
4362                 }
4363         }
4364
4365         *index = BTRFS_I(dir)->index_cnt;
4366         BTRFS_I(dir)->index_cnt++;
4367
4368         return ret;
4369 }
4370
4371 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4372                                      struct btrfs_root *root,
4373                                      struct inode *dir,
4374                                      const char *name, int name_len,
4375                                      u64 ref_objectid, u64 objectid, int mode,
4376                                      u64 *index)
4377 {
4378         struct inode *inode;
4379         struct btrfs_inode_item *inode_item;
4380         struct btrfs_key *location;
4381         struct btrfs_path *path;
4382         struct btrfs_inode_ref *ref;
4383         struct btrfs_key key[2];
4384         u32 sizes[2];
4385         unsigned long ptr;
4386         int ret;
4387         int owner;
4388
4389         path = btrfs_alloc_path();
4390         if (!path)
4391                 return ERR_PTR(-ENOMEM);
4392
4393         inode = new_inode(root->fs_info->sb);
4394         if (!inode) {
4395                 btrfs_free_path(path);
4396                 return ERR_PTR(-ENOMEM);
4397         }
4398
4399         /*
4400          * we have to initialize this early, so we can reclaim the inode
4401          * number if we fail afterwards in this function.
4402          */
4403         inode->i_ino = objectid;
4404
4405         if (dir) {
4406                 trace_btrfs_inode_request(dir);
4407
4408                 ret = btrfs_set_inode_index(dir, index);
4409                 if (ret) {
4410                         btrfs_free_path(path);
4411                         iput(inode);
4412                         return ERR_PTR(ret);
4413                 }
4414         }
4415         /*
4416          * index_cnt is ignored for everything but a dir,
4417          * btrfs_get_inode_index_count has an explanation for the magic
4418          * number
4419          */
4420         BTRFS_I(inode)->index_cnt = 2;
4421         BTRFS_I(inode)->root = root;
4422         BTRFS_I(inode)->generation = trans->transid;
4423         inode->i_generation = BTRFS_I(inode)->generation;
4424         btrfs_set_inode_space_info(root, inode);
4425
4426         if (S_ISDIR(mode))
4427                 owner = 0;
4428         else
4429                 owner = 1;
4430
4431         key[0].objectid = objectid;
4432         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4433         key[0].offset = 0;
4434
4435         key[1].objectid = objectid;
4436         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4437         key[1].offset = ref_objectid;
4438
4439         sizes[0] = sizeof(struct btrfs_inode_item);
4440         sizes[1] = name_len + sizeof(*ref);
4441
4442         path->leave_spinning = 1;
4443         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4444         if (ret != 0)
4445                 goto fail;
4446
4447         inode_init_owner(inode, dir, mode);
4448         inode_set_bytes(inode, 0);
4449         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4450         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4451                                   struct btrfs_inode_item);
4452         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4453
4454         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4455                              struct btrfs_inode_ref);
4456         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4457         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4458         ptr = (unsigned long)(ref + 1);
4459         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4460
4461         btrfs_mark_buffer_dirty(path->nodes[0]);
4462         btrfs_free_path(path);
4463
4464         location = &BTRFS_I(inode)->location;
4465         location->objectid = objectid;
4466         location->offset = 0;
4467         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4468
4469         btrfs_inherit_iflags(inode, dir);
4470
4471         if (S_ISREG(mode)) {
4472                 if (btrfs_test_opt(root, NODATASUM))
4473                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4474                 if (btrfs_test_opt(root, NODATACOW) ||
4475                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4476                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4477         }
4478
4479         insert_inode_hash(inode);
4480         inode_tree_add(inode);
4481
4482         trace_btrfs_inode_new(inode);
4483         btrfs_set_inode_last_trans(trans, inode);
4484
4485         return inode;
4486 fail:
4487         if (dir)
4488                 BTRFS_I(dir)->index_cnt--;
4489         btrfs_free_path(path);
4490         iput(inode);
4491         return ERR_PTR(ret);
4492 }
4493
4494 static inline u8 btrfs_inode_type(struct inode *inode)
4495 {
4496         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4497 }
4498
4499 /*
4500  * utility function to add 'inode' into 'parent_inode' with
4501  * a give name and a given sequence number.
4502  * if 'add_backref' is true, also insert a backref from the
4503  * inode to the parent directory.
4504  */
4505 int btrfs_add_link(struct btrfs_trans_handle *trans,
4506                    struct inode *parent_inode, struct inode *inode,
4507                    const char *name, int name_len, int add_backref, u64 index)
4508 {
4509         int ret = 0;
4510         struct btrfs_key key;
4511         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4512         u64 ino = btrfs_ino(inode);
4513         u64 parent_ino = btrfs_ino(parent_inode);
4514
4515         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4516                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4517         } else {
4518                 key.objectid = ino;
4519                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4520                 key.offset = 0;
4521         }
4522
4523         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4524                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4525                                          key.objectid, root->root_key.objectid,
4526                                          parent_ino, index, name, name_len);
4527         } else if (add_backref) {
4528                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4529                                              parent_ino, index);
4530         }
4531
4532         if (ret == 0) {
4533                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4534                                             parent_inode, &key,
4535                                             btrfs_inode_type(inode), index);
4536                 BUG_ON(ret);
4537
4538                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4539                                    name_len * 2);
4540                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4541                 ret = btrfs_update_inode(trans, root, parent_inode);
4542         }
4543         return ret;
4544 }
4545
4546 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4547                             struct inode *dir, struct dentry *dentry,
4548                             struct inode *inode, int backref, u64 index)
4549 {
4550         int err = btrfs_add_link(trans, dir, inode,
4551                                  dentry->d_name.name, dentry->d_name.len,
4552                                  backref, index);
4553         if (!err) {
4554                 d_instantiate(dentry, inode);
4555                 return 0;
4556         }
4557         if (err > 0)
4558                 err = -EEXIST;
4559         return err;
4560 }
4561
4562 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4563                         int mode, dev_t rdev)
4564 {
4565         struct btrfs_trans_handle *trans;
4566         struct btrfs_root *root = BTRFS_I(dir)->root;
4567         struct inode *inode = NULL;
4568         int err;
4569         int drop_inode = 0;
4570         u64 objectid;
4571         unsigned long nr = 0;
4572         u64 index = 0;
4573
4574         if (!new_valid_dev(rdev))
4575                 return -EINVAL;
4576
4577         /*
4578          * 2 for inode item and ref
4579          * 2 for dir items
4580          * 1 for xattr if selinux is on
4581          */
4582         trans = btrfs_start_transaction(root, 5);
4583         if (IS_ERR(trans))
4584                 return PTR_ERR(trans);
4585
4586         err = btrfs_find_free_ino(root, &objectid);
4587         if (err)
4588                 goto out_unlock;
4589
4590         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4591                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4592                                 mode, &index);
4593         if (IS_ERR(inode)) {
4594                 err = PTR_ERR(inode);
4595                 goto out_unlock;
4596         }
4597
4598         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4599         if (err) {
4600                 drop_inode = 1;
4601                 goto out_unlock;
4602         }
4603
4604         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4605         if (err)
4606                 drop_inode = 1;
4607         else {
4608                 inode->i_op = &btrfs_special_inode_operations;
4609                 init_special_inode(inode, inode->i_mode, rdev);
4610                 btrfs_update_inode(trans, root, inode);
4611         }
4612 out_unlock:
4613         nr = trans->blocks_used;
4614         btrfs_end_transaction_throttle(trans, root);
4615         btrfs_btree_balance_dirty(root, nr);
4616         if (drop_inode) {
4617                 inode_dec_link_count(inode);
4618                 iput(inode);
4619         }
4620         return err;
4621 }
4622
4623 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4624                         int mode, struct nameidata *nd)
4625 {
4626         struct btrfs_trans_handle *trans;
4627         struct btrfs_root *root = BTRFS_I(dir)->root;
4628         struct inode *inode = NULL;
4629         int drop_inode = 0;
4630         int err;
4631         unsigned long nr = 0;
4632         u64 objectid;
4633         u64 index = 0;
4634
4635         /*
4636          * 2 for inode item and ref
4637          * 2 for dir items
4638          * 1 for xattr if selinux is on
4639          */
4640         trans = btrfs_start_transaction(root, 5);
4641         if (IS_ERR(trans))
4642                 return PTR_ERR(trans);
4643
4644         err = btrfs_find_free_ino(root, &objectid);
4645         if (err)
4646                 goto out_unlock;
4647
4648         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4649                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4650                                 mode, &index);
4651         if (IS_ERR(inode)) {
4652                 err = PTR_ERR(inode);
4653                 goto out_unlock;
4654         }
4655
4656         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4657         if (err) {
4658                 drop_inode = 1;
4659                 goto out_unlock;
4660         }
4661
4662         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4663         if (err)
4664                 drop_inode = 1;
4665         else {
4666                 inode->i_mapping->a_ops = &btrfs_aops;
4667                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4668                 inode->i_fop = &btrfs_file_operations;
4669                 inode->i_op = &btrfs_file_inode_operations;
4670                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4671         }
4672 out_unlock:
4673         nr = trans->blocks_used;
4674         btrfs_end_transaction_throttle(trans, root);
4675         if (drop_inode) {
4676                 inode_dec_link_count(inode);
4677                 iput(inode);
4678         }
4679         btrfs_btree_balance_dirty(root, nr);
4680         return err;
4681 }
4682
4683 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4684                       struct dentry *dentry)
4685 {
4686         struct btrfs_trans_handle *trans;
4687         struct btrfs_root *root = BTRFS_I(dir)->root;
4688         struct inode *inode = old_dentry->d_inode;
4689         u64 index;
4690         unsigned long nr = 0;
4691         int err;
4692         int drop_inode = 0;
4693
4694         /* do not allow sys_link's with other subvols of the same device */
4695         if (root->objectid != BTRFS_I(inode)->root->objectid)
4696                 return -EXDEV;
4697
4698         if (inode->i_nlink == ~0U)
4699                 return -EMLINK;
4700
4701         err = btrfs_set_inode_index(dir, &index);
4702         if (err)
4703                 goto fail;
4704
4705         /*
4706          * 2 items for inode and inode ref
4707          * 2 items for dir items
4708          * 1 item for parent inode
4709          */
4710         trans = btrfs_start_transaction(root, 5);
4711         if (IS_ERR(trans)) {
4712                 err = PTR_ERR(trans);
4713                 goto fail;
4714         }
4715
4716         btrfs_inc_nlink(inode);
4717         inode->i_ctime = CURRENT_TIME;
4718         ihold(inode);
4719
4720         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4721
4722         if (err) {
4723                 drop_inode = 1;
4724         } else {
4725                 struct dentry *parent = dentry->d_parent;
4726                 err = btrfs_update_inode(trans, root, inode);
4727                 BUG_ON(err);
4728                 btrfs_log_new_name(trans, inode, NULL, parent);
4729         }
4730
4731         nr = trans->blocks_used;
4732         btrfs_end_transaction_throttle(trans, root);
4733 fail:
4734         if (drop_inode) {
4735                 inode_dec_link_count(inode);
4736                 iput(inode);
4737         }
4738         btrfs_btree_balance_dirty(root, nr);
4739         return err;
4740 }
4741
4742 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4743 {
4744         struct inode *inode = NULL;
4745         struct btrfs_trans_handle *trans;
4746         struct btrfs_root *root = BTRFS_I(dir)->root;
4747         int err = 0;
4748         int drop_on_err = 0;
4749         u64 objectid = 0;
4750         u64 index = 0;
4751         unsigned long nr = 1;
4752
4753         /*
4754          * 2 items for inode and ref
4755          * 2 items for dir items
4756          * 1 for xattr if selinux is on
4757          */
4758         trans = btrfs_start_transaction(root, 5);
4759         if (IS_ERR(trans))
4760                 return PTR_ERR(trans);
4761
4762         err = btrfs_find_free_ino(root, &objectid);
4763         if (err)
4764                 goto out_fail;
4765
4766         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4767                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4768                                 S_IFDIR | mode, &index);
4769         if (IS_ERR(inode)) {
4770                 err = PTR_ERR(inode);
4771                 goto out_fail;
4772         }
4773
4774         drop_on_err = 1;
4775
4776         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4777         if (err)
4778                 goto out_fail;
4779
4780         inode->i_op = &btrfs_dir_inode_operations;
4781         inode->i_fop = &btrfs_dir_file_operations;
4782
4783         btrfs_i_size_write(inode, 0);
4784         err = btrfs_update_inode(trans, root, inode);
4785         if (err)
4786                 goto out_fail;
4787
4788         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4789                              dentry->d_name.len, 0, index);
4790         if (err)
4791                 goto out_fail;
4792
4793         d_instantiate(dentry, inode);
4794         drop_on_err = 0;
4795
4796 out_fail:
4797         nr = trans->blocks_used;
4798         btrfs_end_transaction_throttle(trans, root);
4799         if (drop_on_err)
4800                 iput(inode);
4801         btrfs_btree_balance_dirty(root, nr);
4802         return err;
4803 }
4804
4805 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4806  * and an extent that you want to insert, deal with overlap and insert
4807  * the new extent into the tree.
4808  */
4809 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4810                                 struct extent_map *existing,
4811                                 struct extent_map *em,
4812                                 u64 map_start, u64 map_len)
4813 {
4814         u64 start_diff;
4815
4816         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4817         start_diff = map_start - em->start;
4818         em->start = map_start;
4819         em->len = map_len;
4820         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4821             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4822                 em->block_start += start_diff;
4823                 em->block_len -= start_diff;
4824         }
4825         return add_extent_mapping(em_tree, em);
4826 }
4827
4828 static noinline int uncompress_inline(struct btrfs_path *path,
4829                                       struct inode *inode, struct page *page,
4830                                       size_t pg_offset, u64 extent_offset,
4831                                       struct btrfs_file_extent_item *item)
4832 {
4833         int ret;
4834         struct extent_buffer *leaf = path->nodes[0];
4835         char *tmp;
4836         size_t max_size;
4837         unsigned long inline_size;
4838         unsigned long ptr;
4839         int compress_type;
4840
4841         WARN_ON(pg_offset != 0);
4842         compress_type = btrfs_file_extent_compression(leaf, item);
4843         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4844         inline_size = btrfs_file_extent_inline_item_len(leaf,
4845                                         btrfs_item_nr(leaf, path->slots[0]));
4846         tmp = kmalloc(inline_size, GFP_NOFS);
4847         if (!tmp)
4848                 return -ENOMEM;
4849         ptr = btrfs_file_extent_inline_start(item);
4850
4851         read_extent_buffer(leaf, tmp, ptr, inline_size);
4852
4853         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4854         ret = btrfs_decompress(compress_type, tmp, page,
4855                                extent_offset, inline_size, max_size);
4856         if (ret) {
4857                 char *kaddr = kmap_atomic(page, KM_USER0);
4858                 unsigned long copy_size = min_t(u64,
4859                                   PAGE_CACHE_SIZE - pg_offset,
4860                                   max_size - extent_offset);
4861                 memset(kaddr + pg_offset, 0, copy_size);
4862                 kunmap_atomic(kaddr, KM_USER0);
4863         }
4864         kfree(tmp);
4865         return 0;
4866 }
4867
4868 /*
4869  * a bit scary, this does extent mapping from logical file offset to the disk.
4870  * the ugly parts come from merging extents from the disk with the in-ram
4871  * representation.  This gets more complex because of the data=ordered code,
4872  * where the in-ram extents might be locked pending data=ordered completion.
4873  *
4874  * This also copies inline extents directly into the page.
4875  */
4876
4877 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4878                                     size_t pg_offset, u64 start, u64 len,
4879                                     int create)
4880 {
4881         int ret;
4882         int err = 0;
4883         u64 bytenr;
4884         u64 extent_start = 0;
4885         u64 extent_end = 0;
4886         u64 objectid = btrfs_ino(inode);
4887         u32 found_type;
4888         struct btrfs_path *path = NULL;
4889         struct btrfs_root *root = BTRFS_I(inode)->root;
4890         struct btrfs_file_extent_item *item;
4891         struct extent_buffer *leaf;
4892         struct btrfs_key found_key;
4893         struct extent_map *em = NULL;
4894         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4895         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4896         struct btrfs_trans_handle *trans = NULL;
4897         int compress_type;
4898
4899 again:
4900         read_lock(&em_tree->lock);
4901         em = lookup_extent_mapping(em_tree, start, len);
4902         if (em)
4903                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4904         read_unlock(&em_tree->lock);
4905
4906         if (em) {
4907                 if (em->start > start || em->start + em->len <= start)
4908                         free_extent_map(em);
4909                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4910                         free_extent_map(em);
4911                 else
4912                         goto out;
4913         }
4914         em = alloc_extent_map();
4915         if (!em) {
4916                 err = -ENOMEM;
4917                 goto out;
4918         }
4919         em->bdev = root->fs_info->fs_devices->latest_bdev;
4920         em->start = EXTENT_MAP_HOLE;
4921         em->orig_start = EXTENT_MAP_HOLE;
4922         em->len = (u64)-1;
4923         em->block_len = (u64)-1;
4924
4925         if (!path) {
4926                 path = btrfs_alloc_path();
4927                 if (!path) {
4928                         err = -ENOMEM;
4929                         goto out;
4930                 }
4931                 /*
4932                  * Chances are we'll be called again, so go ahead and do
4933                  * readahead
4934                  */
4935                 path->reada = 1;
4936         }
4937
4938         ret = btrfs_lookup_file_extent(trans, root, path,
4939                                        objectid, start, trans != NULL);
4940         if (ret < 0) {
4941                 err = ret;
4942                 goto out;
4943         }
4944
4945         if (ret != 0) {
4946                 if (path->slots[0] == 0)
4947                         goto not_found;
4948                 path->slots[0]--;
4949         }
4950
4951         leaf = path->nodes[0];
4952         item = btrfs_item_ptr(leaf, path->slots[0],
4953                               struct btrfs_file_extent_item);
4954         /* are we inside the extent that was found? */
4955         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4956         found_type = btrfs_key_type(&found_key);
4957         if (found_key.objectid != objectid ||
4958             found_type != BTRFS_EXTENT_DATA_KEY) {
4959                 goto not_found;
4960         }
4961
4962         found_type = btrfs_file_extent_type(leaf, item);
4963         extent_start = found_key.offset;
4964         compress_type = btrfs_file_extent_compression(leaf, item);
4965         if (found_type == BTRFS_FILE_EXTENT_REG ||
4966             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4967                 extent_end = extent_start +
4968                        btrfs_file_extent_num_bytes(leaf, item);
4969         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4970                 size_t size;
4971                 size = btrfs_file_extent_inline_len(leaf, item);
4972                 extent_end = (extent_start + size + root->sectorsize - 1) &
4973                         ~((u64)root->sectorsize - 1);
4974         }
4975
4976         if (start >= extent_end) {
4977                 path->slots[0]++;
4978                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4979                         ret = btrfs_next_leaf(root, path);
4980                         if (ret < 0) {
4981                                 err = ret;
4982                                 goto out;
4983                         }
4984                         if (ret > 0)
4985                                 goto not_found;
4986                         leaf = path->nodes[0];
4987                 }
4988                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4989                 if (found_key.objectid != objectid ||
4990                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4991                         goto not_found;
4992                 if (start + len <= found_key.offset)
4993                         goto not_found;
4994                 em->start = start;
4995                 em->len = found_key.offset - start;
4996                 goto not_found_em;
4997         }
4998
4999         if (found_type == BTRFS_FILE_EXTENT_REG ||
5000             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5001                 em->start = extent_start;
5002                 em->len = extent_end - extent_start;
5003                 em->orig_start = extent_start -
5004                                  btrfs_file_extent_offset(leaf, item);
5005                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5006                 if (bytenr == 0) {
5007                         em->block_start = EXTENT_MAP_HOLE;
5008                         goto insert;
5009                 }
5010                 if (compress_type != BTRFS_COMPRESS_NONE) {
5011                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5012                         em->compress_type = compress_type;
5013                         em->block_start = bytenr;
5014                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5015                                                                          item);
5016                 } else {
5017                         bytenr += btrfs_file_extent_offset(leaf, item);
5018                         em->block_start = bytenr;
5019                         em->block_len = em->len;
5020                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5021                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5022                 }
5023                 goto insert;
5024         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5025                 unsigned long ptr;
5026                 char *map;
5027                 size_t size;
5028                 size_t extent_offset;
5029                 size_t copy_size;
5030
5031                 em->block_start = EXTENT_MAP_INLINE;
5032                 if (!page || create) {
5033                         em->start = extent_start;
5034                         em->len = extent_end - extent_start;
5035                         goto out;
5036                 }
5037
5038                 size = btrfs_file_extent_inline_len(leaf, item);
5039                 extent_offset = page_offset(page) + pg_offset - extent_start;
5040                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5041                                 size - extent_offset);
5042                 em->start = extent_start + extent_offset;
5043                 em->len = (copy_size + root->sectorsize - 1) &
5044                         ~((u64)root->sectorsize - 1);
5045                 em->orig_start = EXTENT_MAP_INLINE;
5046                 if (compress_type) {
5047                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5048                         em->compress_type = compress_type;
5049                 }
5050                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5051                 if (create == 0 && !PageUptodate(page)) {
5052                         if (btrfs_file_extent_compression(leaf, item) !=
5053                             BTRFS_COMPRESS_NONE) {
5054                                 ret = uncompress_inline(path, inode, page,
5055                                                         pg_offset,
5056                                                         extent_offset, item);
5057                                 BUG_ON(ret);
5058                         } else {
5059                                 map = kmap(page);
5060                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5061                                                    copy_size);
5062                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5063                                         memset(map + pg_offset + copy_size, 0,
5064                                                PAGE_CACHE_SIZE - pg_offset -
5065                                                copy_size);
5066                                 }
5067                                 kunmap(page);
5068                         }
5069                         flush_dcache_page(page);
5070                 } else if (create && PageUptodate(page)) {
5071                         WARN_ON(1);
5072                         if (!trans) {
5073                                 kunmap(page);
5074                                 free_extent_map(em);
5075                                 em = NULL;
5076
5077                                 btrfs_release_path(path);
5078                                 trans = btrfs_join_transaction(root);
5079
5080                                 if (IS_ERR(trans))
5081                                         return ERR_CAST(trans);
5082                                 goto again;
5083                         }
5084                         map = kmap(page);
5085                         write_extent_buffer(leaf, map + pg_offset, ptr,
5086                                             copy_size);
5087                         kunmap(page);
5088                         btrfs_mark_buffer_dirty(leaf);
5089                 }
5090                 set_extent_uptodate(io_tree, em->start,
5091                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5092                 goto insert;
5093         } else {
5094                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5095                 WARN_ON(1);
5096         }
5097 not_found:
5098         em->start = start;
5099         em->len = len;
5100 not_found_em:
5101         em->block_start = EXTENT_MAP_HOLE;
5102         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5103 insert:
5104         btrfs_release_path(path);
5105         if (em->start > start || extent_map_end(em) <= start) {
5106                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5107                        "[%llu %llu]\n", (unsigned long long)em->start,
5108                        (unsigned long long)em->len,
5109                        (unsigned long long)start,
5110                        (unsigned long long)len);
5111                 err = -EIO;
5112                 goto out;
5113         }
5114
5115         err = 0;
5116         write_lock(&em_tree->lock);
5117         ret = add_extent_mapping(em_tree, em);
5118         /* it is possible that someone inserted the extent into the tree
5119          * while we had the lock dropped.  It is also possible that
5120          * an overlapping map exists in the tree
5121          */
5122         if (ret == -EEXIST) {
5123                 struct extent_map *existing;
5124
5125                 ret = 0;
5126
5127                 existing = lookup_extent_mapping(em_tree, start, len);
5128                 if (existing && (existing->start > start ||
5129                     existing->start + existing->len <= start)) {
5130                         free_extent_map(existing);
5131                         existing = NULL;
5132                 }
5133                 if (!existing) {
5134                         existing = lookup_extent_mapping(em_tree, em->start,
5135                                                          em->len);
5136                         if (existing) {
5137                                 err = merge_extent_mapping(em_tree, existing,
5138                                                            em, start,
5139                                                            root->sectorsize);
5140                                 free_extent_map(existing);
5141                                 if (err) {
5142                                         free_extent_map(em);
5143                                         em = NULL;
5144                                 }
5145                         } else {
5146                                 err = -EIO;
5147                                 free_extent_map(em);
5148                                 em = NULL;
5149                         }
5150                 } else {
5151                         free_extent_map(em);
5152                         em = existing;
5153                         err = 0;
5154                 }
5155         }
5156         write_unlock(&em_tree->lock);
5157 out:
5158
5159         trace_btrfs_get_extent(root, em);
5160
5161         if (path)
5162                 btrfs_free_path(path);
5163         if (trans) {
5164                 ret = btrfs_end_transaction(trans, root);
5165                 if (!err)
5166                         err = ret;
5167         }
5168         if (err) {
5169                 free_extent_map(em);
5170                 return ERR_PTR(err);
5171         }
5172         return em;
5173 }
5174
5175 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5176                                            size_t pg_offset, u64 start, u64 len,
5177                                            int create)
5178 {
5179         struct extent_map *em;
5180         struct extent_map *hole_em = NULL;
5181         u64 range_start = start;
5182         u64 end;
5183         u64 found;
5184         u64 found_end;
5185         int err = 0;
5186
5187         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5188         if (IS_ERR(em))
5189                 return em;
5190         if (em) {
5191                 /*
5192                  * if our em maps to a hole, there might
5193                  * actually be delalloc bytes behind it
5194                  */
5195                 if (em->block_start != EXTENT_MAP_HOLE)
5196                         return em;
5197                 else
5198                         hole_em = em;
5199         }
5200
5201         /* check to see if we've wrapped (len == -1 or similar) */
5202         end = start + len;
5203         if (end < start)
5204                 end = (u64)-1;
5205         else
5206                 end -= 1;
5207
5208         em = NULL;
5209
5210         /* ok, we didn't find anything, lets look for delalloc */
5211         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5212                                  end, len, EXTENT_DELALLOC, 1);
5213         found_end = range_start + found;
5214         if (found_end < range_start)
5215                 found_end = (u64)-1;
5216
5217         /*
5218          * we didn't find anything useful, return
5219          * the original results from get_extent()
5220          */
5221         if (range_start > end || found_end <= start) {
5222                 em = hole_em;
5223                 hole_em = NULL;
5224                 goto out;
5225         }
5226
5227         /* adjust the range_start to make sure it doesn't
5228          * go backwards from the start they passed in
5229          */
5230         range_start = max(start,range_start);
5231         found = found_end - range_start;
5232
5233         if (found > 0) {
5234                 u64 hole_start = start;
5235                 u64 hole_len = len;
5236
5237                 em = alloc_extent_map();
5238                 if (!em) {
5239                         err = -ENOMEM;
5240                         goto out;
5241                 }
5242                 /*
5243                  * when btrfs_get_extent can't find anything it
5244                  * returns one huge hole
5245                  *
5246                  * make sure what it found really fits our range, and
5247                  * adjust to make sure it is based on the start from
5248                  * the caller
5249                  */
5250                 if (hole_em) {
5251                         u64 calc_end = extent_map_end(hole_em);
5252
5253                         if (calc_end <= start || (hole_em->start > end)) {
5254                                 free_extent_map(hole_em);
5255                                 hole_em = NULL;
5256                         } else {
5257                                 hole_start = max(hole_em->start, start);
5258                                 hole_len = calc_end - hole_start;
5259                         }
5260                 }
5261                 em->bdev = NULL;
5262                 if (hole_em && range_start > hole_start) {
5263                         /* our hole starts before our delalloc, so we
5264                          * have to return just the parts of the hole
5265                          * that go until  the delalloc starts
5266                          */
5267                         em->len = min(hole_len,
5268                                       range_start - hole_start);
5269                         em->start = hole_start;
5270                         em->orig_start = hole_start;
5271                         /*
5272                          * don't adjust block start at all,
5273                          * it is fixed at EXTENT_MAP_HOLE
5274                          */
5275                         em->block_start = hole_em->block_start;
5276                         em->block_len = hole_len;
5277                 } else {
5278                         em->start = range_start;
5279                         em->len = found;
5280                         em->orig_start = range_start;
5281                         em->block_start = EXTENT_MAP_DELALLOC;
5282                         em->block_len = found;
5283                 }
5284         } else if (hole_em) {
5285                 return hole_em;
5286         }
5287 out:
5288
5289         free_extent_map(hole_em);
5290         if (err) {
5291                 free_extent_map(em);
5292                 return ERR_PTR(err);
5293         }
5294         return em;
5295 }
5296
5297 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5298                                                   struct extent_map *em,
5299                                                   u64 start, u64 len)
5300 {
5301         struct btrfs_root *root = BTRFS_I(inode)->root;
5302         struct btrfs_trans_handle *trans;
5303         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5304         struct btrfs_key ins;
5305         u64 alloc_hint;
5306         int ret;
5307         bool insert = false;
5308
5309         /*
5310          * Ok if the extent map we looked up is a hole and is for the exact
5311          * range we want, there is no reason to allocate a new one, however if
5312          * it is not right then we need to free this one and drop the cache for
5313          * our range.
5314          */
5315         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5316             em->len != len) {
5317                 free_extent_map(em);
5318                 em = NULL;
5319                 insert = true;
5320                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5321         }
5322
5323         trans = btrfs_join_transaction(root);
5324         if (IS_ERR(trans))
5325                 return ERR_CAST(trans);
5326
5327         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5328                 btrfs_add_inode_defrag(trans, inode);
5329
5330         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5331
5332         alloc_hint = get_extent_allocation_hint(inode, start, len);
5333         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5334                                    alloc_hint, (u64)-1, &ins, 1);
5335         if (ret) {
5336                 em = ERR_PTR(ret);
5337                 goto out;
5338         }
5339
5340         if (!em) {
5341                 em = alloc_extent_map();
5342                 if (!em) {
5343                         em = ERR_PTR(-ENOMEM);
5344                         goto out;
5345                 }
5346         }
5347
5348         em->start = start;
5349         em->orig_start = em->start;
5350         em->len = ins.offset;
5351
5352         em->block_start = ins.objectid;
5353         em->block_len = ins.offset;
5354         em->bdev = root->fs_info->fs_devices->latest_bdev;
5355
5356         /*
5357          * We need to do this because if we're using the original em we searched
5358          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5359          */
5360         em->flags = 0;
5361         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5362
5363         while (insert) {
5364                 write_lock(&em_tree->lock);
5365                 ret = add_extent_mapping(em_tree, em);
5366                 write_unlock(&em_tree->lock);
5367                 if (ret != -EEXIST)
5368                         break;
5369                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5370         }
5371
5372         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5373                                            ins.offset, ins.offset, 0);
5374         if (ret) {
5375                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5376                 em = ERR_PTR(ret);
5377         }
5378 out:
5379         btrfs_end_transaction(trans, root);
5380         return em;
5381 }
5382
5383 /*
5384  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5385  * block must be cow'd
5386  */
5387 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5388                                       struct inode *inode, u64 offset, u64 len)
5389 {
5390         struct btrfs_path *path;
5391         int ret;
5392         struct extent_buffer *leaf;
5393         struct btrfs_root *root = BTRFS_I(inode)->root;
5394         struct btrfs_file_extent_item *fi;
5395         struct btrfs_key key;
5396         u64 disk_bytenr;
5397         u64 backref_offset;
5398         u64 extent_end;
5399         u64 num_bytes;
5400         int slot;
5401         int found_type;
5402
5403         path = btrfs_alloc_path();
5404         if (!path)
5405                 return -ENOMEM;
5406
5407         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5408                                        offset, 0);
5409         if (ret < 0)
5410                 goto out;
5411
5412         slot = path->slots[0];
5413         if (ret == 1) {
5414                 if (slot == 0) {
5415                         /* can't find the item, must cow */
5416                         ret = 0;
5417                         goto out;
5418                 }
5419                 slot--;
5420         }
5421         ret = 0;
5422         leaf = path->nodes[0];
5423         btrfs_item_key_to_cpu(leaf, &key, slot);
5424         if (key.objectid != btrfs_ino(inode) ||
5425             key.type != BTRFS_EXTENT_DATA_KEY) {
5426                 /* not our file or wrong item type, must cow */
5427                 goto out;
5428         }
5429
5430         if (key.offset > offset) {
5431                 /* Wrong offset, must cow */
5432                 goto out;
5433         }
5434
5435         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5436         found_type = btrfs_file_extent_type(leaf, fi);
5437         if (found_type != BTRFS_FILE_EXTENT_REG &&
5438             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5439                 /* not a regular extent, must cow */
5440                 goto out;
5441         }
5442         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5443         backref_offset = btrfs_file_extent_offset(leaf, fi);
5444
5445         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5446         if (extent_end < offset + len) {
5447                 /* extent doesn't include our full range, must cow */
5448                 goto out;
5449         }
5450
5451         if (btrfs_extent_readonly(root, disk_bytenr))
5452                 goto out;
5453
5454         /*
5455          * look for other files referencing this extent, if we
5456          * find any we must cow
5457          */
5458         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5459                                   key.offset - backref_offset, disk_bytenr))
5460                 goto out;
5461
5462         /*
5463          * adjust disk_bytenr and num_bytes to cover just the bytes
5464          * in this extent we are about to write.  If there
5465          * are any csums in that range we have to cow in order
5466          * to keep the csums correct
5467          */
5468         disk_bytenr += backref_offset;
5469         disk_bytenr += offset - key.offset;
5470         num_bytes = min(offset + len, extent_end) - offset;
5471         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5472                                 goto out;
5473         /*
5474          * all of the above have passed, it is safe to overwrite this extent
5475          * without cow
5476          */
5477         ret = 1;
5478 out:
5479         btrfs_free_path(path);
5480         return ret;
5481 }
5482
5483 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5484                                    struct buffer_head *bh_result, int create)
5485 {
5486         struct extent_map *em;
5487         struct btrfs_root *root = BTRFS_I(inode)->root;
5488         u64 start = iblock << inode->i_blkbits;
5489         u64 len = bh_result->b_size;
5490         struct btrfs_trans_handle *trans;
5491
5492         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5493         if (IS_ERR(em))
5494                 return PTR_ERR(em);
5495
5496         /*
5497          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5498          * io.  INLINE is special, and we could probably kludge it in here, but
5499          * it's still buffered so for safety lets just fall back to the generic
5500          * buffered path.
5501          *
5502          * For COMPRESSED we _have_ to read the entire extent in so we can
5503          * decompress it, so there will be buffering required no matter what we
5504          * do, so go ahead and fallback to buffered.
5505          *
5506          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5507          * to buffered IO.  Don't blame me, this is the price we pay for using
5508          * the generic code.
5509          */
5510         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5511             em->block_start == EXTENT_MAP_INLINE) {
5512                 free_extent_map(em);
5513                 return -ENOTBLK;
5514         }
5515
5516         /* Just a good old fashioned hole, return */
5517         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5518                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5519                 free_extent_map(em);
5520                 /* DIO will do one hole at a time, so just unlock a sector */
5521                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5522                               start + root->sectorsize - 1, GFP_NOFS);
5523                 return 0;
5524         }
5525
5526         /*
5527          * We don't allocate a new extent in the following cases
5528          *
5529          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5530          * existing extent.
5531          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5532          * just use the extent.
5533          *
5534          */
5535         if (!create) {
5536                 len = em->len - (start - em->start);
5537                 goto map;
5538         }
5539
5540         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5541             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5542              em->block_start != EXTENT_MAP_HOLE)) {
5543                 int type;
5544                 int ret;
5545                 u64 block_start;
5546
5547                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5548                         type = BTRFS_ORDERED_PREALLOC;
5549                 else
5550                         type = BTRFS_ORDERED_NOCOW;
5551                 len = min(len, em->len - (start - em->start));
5552                 block_start = em->block_start + (start - em->start);
5553
5554                 /*
5555                  * we're not going to log anything, but we do need
5556                  * to make sure the current transaction stays open
5557                  * while we look for nocow cross refs
5558                  */
5559                 trans = btrfs_join_transaction(root);
5560                 if (IS_ERR(trans))
5561                         goto must_cow;
5562
5563                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5564                         ret = btrfs_add_ordered_extent_dio(inode, start,
5565                                            block_start, len, len, type);
5566                         btrfs_end_transaction(trans, root);
5567                         if (ret) {
5568                                 free_extent_map(em);
5569                                 return ret;
5570                         }
5571                         goto unlock;
5572                 }
5573                 btrfs_end_transaction(trans, root);
5574         }
5575 must_cow:
5576         /*
5577          * this will cow the extent, reset the len in case we changed
5578          * it above
5579          */
5580         len = bh_result->b_size;
5581         em = btrfs_new_extent_direct(inode, em, start, len);
5582         if (IS_ERR(em))
5583                 return PTR_ERR(em);
5584         len = min(len, em->len - (start - em->start));
5585 unlock:
5586         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5587                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5588                           0, NULL, GFP_NOFS);
5589 map:
5590         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5591                 inode->i_blkbits;
5592         bh_result->b_size = len;
5593         bh_result->b_bdev = em->bdev;
5594         set_buffer_mapped(bh_result);
5595         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5596                 set_buffer_new(bh_result);
5597
5598         free_extent_map(em);
5599
5600         return 0;
5601 }
5602
5603 struct btrfs_dio_private {
5604         struct inode *inode;
5605         u64 logical_offset;
5606         u64 disk_bytenr;
5607         u64 bytes;
5608         u32 *csums;
5609         void *private;
5610
5611         /* number of bios pending for this dio */
5612         atomic_t pending_bios;
5613
5614         /* IO errors */
5615         int errors;
5616
5617         struct bio *orig_bio;
5618 };
5619
5620 static void btrfs_endio_direct_read(struct bio *bio, int err)
5621 {
5622         struct btrfs_dio_private *dip = bio->bi_private;
5623         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5624         struct bio_vec *bvec = bio->bi_io_vec;
5625         struct inode *inode = dip->inode;
5626         struct btrfs_root *root = BTRFS_I(inode)->root;
5627         u64 start;
5628         u32 *private = dip->csums;
5629
5630         start = dip->logical_offset;
5631         do {
5632                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5633                         struct page *page = bvec->bv_page;
5634                         char *kaddr;
5635                         u32 csum = ~(u32)0;
5636                         unsigned long flags;
5637
5638                         local_irq_save(flags);
5639                         kaddr = kmap_atomic(page, KM_IRQ0);
5640                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5641                                                csum, bvec->bv_len);
5642                         btrfs_csum_final(csum, (char *)&csum);
5643                         kunmap_atomic(kaddr, KM_IRQ0);
5644                         local_irq_restore(flags);
5645
5646                         flush_dcache_page(bvec->bv_page);
5647                         if (csum != *private) {
5648                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5649                                       " %llu csum %u private %u\n",
5650                                       (unsigned long long)btrfs_ino(inode),
5651                                       (unsigned long long)start,
5652                                       csum, *private);
5653                                 err = -EIO;
5654                         }
5655                 }
5656
5657                 start += bvec->bv_len;
5658                 private++;
5659                 bvec++;
5660         } while (bvec <= bvec_end);
5661
5662         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5663                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5664         bio->bi_private = dip->private;
5665
5666         kfree(dip->csums);
5667         kfree(dip);
5668
5669         /* If we had a csum failure make sure to clear the uptodate flag */
5670         if (err)
5671                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5672         dio_end_io(bio, err);
5673 }
5674
5675 static void btrfs_endio_direct_write(struct bio *bio, int err)
5676 {
5677         struct btrfs_dio_private *dip = bio->bi_private;
5678         struct inode *inode = dip->inode;
5679         struct btrfs_root *root = BTRFS_I(inode)->root;
5680         struct btrfs_trans_handle *trans;
5681         struct btrfs_ordered_extent *ordered = NULL;
5682         struct extent_state *cached_state = NULL;
5683         u64 ordered_offset = dip->logical_offset;
5684         u64 ordered_bytes = dip->bytes;
5685         int ret;
5686
5687         if (err)
5688                 goto out_done;
5689 again:
5690         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5691                                                    &ordered_offset,
5692                                                    ordered_bytes);
5693         if (!ret)
5694                 goto out_test;
5695
5696         BUG_ON(!ordered);
5697
5698         trans = btrfs_join_transaction(root);
5699         if (IS_ERR(trans)) {
5700                 err = -ENOMEM;
5701                 goto out;
5702         }
5703         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5704
5705         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5706                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5707                 if (!ret)
5708                         err = btrfs_update_inode_fallback(trans, root, inode);
5709                 goto out;
5710         }
5711
5712         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5713                          ordered->file_offset + ordered->len - 1, 0,
5714                          &cached_state, GFP_NOFS);
5715
5716         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5717                 ret = btrfs_mark_extent_written(trans, inode,
5718                                                 ordered->file_offset,
5719                                                 ordered->file_offset +
5720                                                 ordered->len);
5721                 if (ret) {
5722                         err = ret;
5723                         goto out_unlock;
5724                 }
5725         } else {
5726                 ret = insert_reserved_file_extent(trans, inode,
5727                                                   ordered->file_offset,
5728                                                   ordered->start,
5729                                                   ordered->disk_len,
5730                                                   ordered->len,
5731                                                   ordered->len,
5732                                                   0, 0, 0,
5733                                                   BTRFS_FILE_EXTENT_REG);
5734                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5735                                    ordered->file_offset, ordered->len);
5736                 if (ret) {
5737                         err = ret;
5738                         WARN_ON(1);
5739                         goto out_unlock;
5740                 }
5741         }
5742
5743         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5744         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5745         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5746                 btrfs_update_inode_fallback(trans, root, inode);
5747         ret = 0;
5748 out_unlock:
5749         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5750                              ordered->file_offset + ordered->len - 1,
5751                              &cached_state, GFP_NOFS);
5752 out:
5753         btrfs_delalloc_release_metadata(inode, ordered->len);
5754         btrfs_end_transaction(trans, root);
5755         ordered_offset = ordered->file_offset + ordered->len;
5756         btrfs_put_ordered_extent(ordered);
5757         btrfs_put_ordered_extent(ordered);
5758
5759 out_test:
5760         /*
5761          * our bio might span multiple ordered extents.  If we haven't
5762          * completed the accounting for the whole dio, go back and try again
5763          */
5764         if (ordered_offset < dip->logical_offset + dip->bytes) {
5765                 ordered_bytes = dip->logical_offset + dip->bytes -
5766                         ordered_offset;
5767                 goto again;
5768         }
5769 out_done:
5770         bio->bi_private = dip->private;
5771
5772         kfree(dip->csums);
5773         kfree(dip);
5774
5775         /* If we had an error make sure to clear the uptodate flag */
5776         if (err)
5777                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5778         dio_end_io(bio, err);
5779 }
5780
5781 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5782                                     struct bio *bio, int mirror_num,
5783                                     unsigned long bio_flags, u64 offset)
5784 {
5785         int ret;
5786         struct btrfs_root *root = BTRFS_I(inode)->root;
5787         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5788         BUG_ON(ret);
5789         return 0;
5790 }
5791
5792 static void btrfs_end_dio_bio(struct bio *bio, int err)
5793 {
5794         struct btrfs_dio_private *dip = bio->bi_private;
5795
5796         if (err) {
5797                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5798                       "sector %#Lx len %u err no %d\n",
5799                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5800                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5801                 dip->errors = 1;
5802
5803                 /*
5804                  * before atomic variable goto zero, we must make sure
5805                  * dip->errors is perceived to be set.
5806                  */
5807                 smp_mb__before_atomic_dec();
5808         }
5809
5810         /* if there are more bios still pending for this dio, just exit */
5811         if (!atomic_dec_and_test(&dip->pending_bios))
5812                 goto out;
5813
5814         if (dip->errors)
5815                 bio_io_error(dip->orig_bio);
5816         else {
5817                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5818                 bio_endio(dip->orig_bio, 0);
5819         }
5820 out:
5821         bio_put(bio);
5822 }
5823
5824 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5825                                        u64 first_sector, gfp_t gfp_flags)
5826 {
5827         int nr_vecs = bio_get_nr_vecs(bdev);
5828         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5829 }
5830
5831 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5832                                          int rw, u64 file_offset, int skip_sum,
5833                                          u32 *csums, int async_submit)
5834 {
5835         int write = rw & REQ_WRITE;
5836         struct btrfs_root *root = BTRFS_I(inode)->root;
5837         int ret;
5838
5839         bio_get(bio);
5840         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5841         if (ret)
5842                 goto err;
5843
5844         if (skip_sum)
5845                 goto map;
5846
5847         if (write && async_submit) {
5848                 ret = btrfs_wq_submit_bio(root->fs_info,
5849                                    inode, rw, bio, 0, 0,
5850                                    file_offset,
5851                                    __btrfs_submit_bio_start_direct_io,
5852                                    __btrfs_submit_bio_done);
5853                 goto err;
5854         } else if (write) {
5855                 /*
5856                  * If we aren't doing async submit, calculate the csum of the
5857                  * bio now.
5858                  */
5859                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5860                 if (ret)
5861                         goto err;
5862         } else if (!skip_sum) {
5863                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5864                                           file_offset, csums);
5865                 if (ret)
5866                         goto err;
5867         }
5868
5869 map:
5870         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5871 err:
5872         bio_put(bio);
5873         return ret;
5874 }
5875
5876 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5877                                     int skip_sum)
5878 {
5879         struct inode *inode = dip->inode;
5880         struct btrfs_root *root = BTRFS_I(inode)->root;
5881         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5882         struct bio *bio;
5883         struct bio *orig_bio = dip->orig_bio;
5884         struct bio_vec *bvec = orig_bio->bi_io_vec;
5885         u64 start_sector = orig_bio->bi_sector;
5886         u64 file_offset = dip->logical_offset;
5887         u64 submit_len = 0;
5888         u64 map_length;
5889         int nr_pages = 0;
5890         u32 *csums = dip->csums;
5891         int ret = 0;
5892         int async_submit = 0;
5893         int write = rw & REQ_WRITE;
5894
5895         map_length = orig_bio->bi_size;
5896         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5897                               &map_length, NULL, 0);
5898         if (ret) {
5899                 bio_put(orig_bio);
5900                 return -EIO;
5901         }
5902
5903         if (map_length >= orig_bio->bi_size) {
5904                 bio = orig_bio;
5905                 goto submit;
5906         }
5907
5908         async_submit = 1;
5909         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5910         if (!bio)
5911                 return -ENOMEM;
5912         bio->bi_private = dip;
5913         bio->bi_end_io = btrfs_end_dio_bio;
5914         atomic_inc(&dip->pending_bios);
5915
5916         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5917                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5918                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5919                                  bvec->bv_offset) < bvec->bv_len)) {
5920                         /*
5921                          * inc the count before we submit the bio so
5922                          * we know the end IO handler won't happen before
5923                          * we inc the count. Otherwise, the dip might get freed
5924                          * before we're done setting it up
5925                          */
5926                         atomic_inc(&dip->pending_bios);
5927                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5928                                                      file_offset, skip_sum,
5929                                                      csums, async_submit);
5930                         if (ret) {
5931                                 bio_put(bio);
5932                                 atomic_dec(&dip->pending_bios);
5933                                 goto out_err;
5934                         }
5935
5936                         /* Write's use the ordered csums */
5937                         if (!write && !skip_sum)
5938                                 csums = csums + nr_pages;
5939                         start_sector += submit_len >> 9;
5940                         file_offset += submit_len;
5941
5942                         submit_len = 0;
5943                         nr_pages = 0;
5944
5945                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5946                                                   start_sector, GFP_NOFS);
5947                         if (!bio)
5948                                 goto out_err;
5949                         bio->bi_private = dip;
5950                         bio->bi_end_io = btrfs_end_dio_bio;
5951
5952                         map_length = orig_bio->bi_size;
5953                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5954                                               &map_length, NULL, 0);
5955                         if (ret) {
5956                                 bio_put(bio);
5957                                 goto out_err;
5958                         }
5959                 } else {
5960                         submit_len += bvec->bv_len;
5961                         nr_pages ++;
5962                         bvec++;
5963                 }
5964         }
5965
5966 submit:
5967         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5968                                      csums, async_submit);
5969         if (!ret)
5970                 return 0;
5971
5972         bio_put(bio);
5973 out_err:
5974         dip->errors = 1;
5975         /*
5976          * before atomic variable goto zero, we must
5977          * make sure dip->errors is perceived to be set.
5978          */
5979         smp_mb__before_atomic_dec();
5980         if (atomic_dec_and_test(&dip->pending_bios))
5981                 bio_io_error(dip->orig_bio);
5982
5983         /* bio_end_io() will handle error, so we needn't return it */
5984         return 0;
5985 }
5986
5987 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5988                                 loff_t file_offset)
5989 {
5990         struct btrfs_root *root = BTRFS_I(inode)->root;
5991         struct btrfs_dio_private *dip;
5992         struct bio_vec *bvec = bio->bi_io_vec;
5993         int skip_sum;
5994         int write = rw & REQ_WRITE;
5995         int ret = 0;
5996
5997         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5998
5999         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6000         if (!dip) {
6001                 ret = -ENOMEM;
6002                 goto free_ordered;
6003         }
6004         dip->csums = NULL;
6005
6006         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6007         if (!write && !skip_sum) {
6008                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6009                 if (!dip->csums) {
6010                         kfree(dip);
6011                         ret = -ENOMEM;
6012                         goto free_ordered;
6013                 }
6014         }
6015
6016         dip->private = bio->bi_private;
6017         dip->inode = inode;
6018         dip->logical_offset = file_offset;
6019
6020         dip->bytes = 0;
6021         do {
6022                 dip->bytes += bvec->bv_len;
6023                 bvec++;
6024         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6025
6026         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6027         bio->bi_private = dip;
6028         dip->errors = 0;
6029         dip->orig_bio = bio;
6030         atomic_set(&dip->pending_bios, 0);
6031
6032         if (write)
6033                 bio->bi_end_io = btrfs_endio_direct_write;
6034         else
6035                 bio->bi_end_io = btrfs_endio_direct_read;
6036
6037         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6038         if (!ret)
6039                 return;
6040 free_ordered:
6041         /*
6042          * If this is a write, we need to clean up the reserved space and kill
6043          * the ordered extent.
6044          */
6045         if (write) {
6046                 struct btrfs_ordered_extent *ordered;
6047                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6048                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6049                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6050                         btrfs_free_reserved_extent(root, ordered->start,
6051                                                    ordered->disk_len);
6052                 btrfs_put_ordered_extent(ordered);
6053                 btrfs_put_ordered_extent(ordered);
6054         }
6055         bio_endio(bio, ret);
6056 }
6057
6058 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6059                         const struct iovec *iov, loff_t offset,
6060                         unsigned long nr_segs)
6061 {
6062         int seg;
6063         int i;
6064         size_t size;
6065         unsigned long addr;
6066         unsigned blocksize_mask = root->sectorsize - 1;
6067         ssize_t retval = -EINVAL;
6068         loff_t end = offset;
6069
6070         if (offset & blocksize_mask)
6071                 goto out;
6072
6073         /* Check the memory alignment.  Blocks cannot straddle pages */
6074         for (seg = 0; seg < nr_segs; seg++) {
6075                 addr = (unsigned long)iov[seg].iov_base;
6076                 size = iov[seg].iov_len;
6077                 end += size;
6078                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6079                         goto out;
6080
6081                 /* If this is a write we don't need to check anymore */
6082                 if (rw & WRITE)
6083                         continue;
6084
6085                 /*
6086                  * Check to make sure we don't have duplicate iov_base's in this
6087                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6088                  * when reading back.
6089                  */
6090                 for (i = seg + 1; i < nr_segs; i++) {
6091                         if (iov[seg].iov_base == iov[i].iov_base)
6092                                 goto out;
6093                 }
6094         }
6095         retval = 0;
6096 out:
6097         return retval;
6098 }
6099 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6100                         const struct iovec *iov, loff_t offset,
6101                         unsigned long nr_segs)
6102 {
6103         struct file *file = iocb->ki_filp;
6104         struct inode *inode = file->f_mapping->host;
6105         struct btrfs_ordered_extent *ordered;
6106         struct extent_state *cached_state = NULL;
6107         u64 lockstart, lockend;
6108         ssize_t ret;
6109         int writing = rw & WRITE;
6110         int write_bits = 0;
6111         size_t count = iov_length(iov, nr_segs);
6112
6113         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6114                             offset, nr_segs)) {
6115                 return 0;
6116         }
6117
6118         lockstart = offset;
6119         lockend = offset + count - 1;
6120
6121         if (writing) {
6122                 ret = btrfs_delalloc_reserve_space(inode, count);
6123                 if (ret)
6124                         goto out;
6125         }
6126
6127         while (1) {
6128                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6129                                  0, &cached_state, GFP_NOFS);
6130                 /*
6131                  * We're concerned with the entire range that we're going to be
6132                  * doing DIO to, so we need to make sure theres no ordered
6133                  * extents in this range.
6134                  */
6135                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6136                                                      lockend - lockstart + 1);
6137                 if (!ordered)
6138                         break;
6139                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6140                                      &cached_state, GFP_NOFS);
6141                 btrfs_start_ordered_extent(inode, ordered, 1);
6142                 btrfs_put_ordered_extent(ordered);
6143                 cond_resched();
6144         }
6145
6146         /*
6147          * we don't use btrfs_set_extent_delalloc because we don't want
6148          * the dirty or uptodate bits
6149          */
6150         if (writing) {
6151                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6152                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6153                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6154                                      GFP_NOFS);
6155                 if (ret) {
6156                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6157                                          lockend, EXTENT_LOCKED | write_bits,
6158                                          1, 0, &cached_state, GFP_NOFS);
6159                         goto out;
6160                 }
6161         }
6162
6163         free_extent_state(cached_state);
6164         cached_state = NULL;
6165
6166         ret = __blockdev_direct_IO(rw, iocb, inode,
6167                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6168                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6169                    btrfs_submit_direct, 0);
6170
6171         if (ret < 0 && ret != -EIOCBQUEUED) {
6172                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6173                               offset + iov_length(iov, nr_segs) - 1,
6174                               EXTENT_LOCKED | write_bits, 1, 0,
6175                               &cached_state, GFP_NOFS);
6176         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6177                 /*
6178                  * We're falling back to buffered, unlock the section we didn't
6179                  * do IO on.
6180                  */
6181                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6182                               offset + iov_length(iov, nr_segs) - 1,
6183                               EXTENT_LOCKED | write_bits, 1, 0,
6184                               &cached_state, GFP_NOFS);
6185         }
6186 out:
6187         free_extent_state(cached_state);
6188         return ret;
6189 }
6190
6191 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6192                 __u64 start, __u64 len)
6193 {
6194         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6195 }
6196
6197 int btrfs_readpage(struct file *file, struct page *page)
6198 {
6199         struct extent_io_tree *tree;
6200         tree = &BTRFS_I(page->mapping->host)->io_tree;
6201         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6202 }
6203
6204 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6205 {
6206         struct extent_io_tree *tree;
6207
6208
6209         if (current->flags & PF_MEMALLOC) {
6210                 redirty_page_for_writepage(wbc, page);
6211                 unlock_page(page);
6212                 return 0;
6213         }
6214         tree = &BTRFS_I(page->mapping->host)->io_tree;
6215         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6216 }
6217
6218 int btrfs_writepages(struct address_space *mapping,
6219                      struct writeback_control *wbc)
6220 {
6221         struct extent_io_tree *tree;
6222
6223         tree = &BTRFS_I(mapping->host)->io_tree;
6224         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6225 }
6226
6227 static int
6228 btrfs_readpages(struct file *file, struct address_space *mapping,
6229                 struct list_head *pages, unsigned nr_pages)
6230 {
6231         struct extent_io_tree *tree;
6232         tree = &BTRFS_I(mapping->host)->io_tree;
6233         return extent_readpages(tree, mapping, pages, nr_pages,
6234                                 btrfs_get_extent);
6235 }
6236 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6237 {
6238         struct extent_io_tree *tree;
6239         struct extent_map_tree *map;
6240         int ret;
6241
6242         tree = &BTRFS_I(page->mapping->host)->io_tree;
6243         map = &BTRFS_I(page->mapping->host)->extent_tree;
6244         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6245         if (ret == 1) {
6246                 ClearPagePrivate(page);
6247                 set_page_private(page, 0);
6248                 page_cache_release(page);
6249         }
6250         return ret;
6251 }
6252
6253 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6254 {
6255         if (PageWriteback(page) || PageDirty(page))
6256                 return 0;
6257         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6258 }
6259
6260 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6261 {
6262         struct extent_io_tree *tree;
6263         struct btrfs_ordered_extent *ordered;
6264         struct extent_state *cached_state = NULL;
6265         u64 page_start = page_offset(page);
6266         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6267
6268
6269         /*
6270          * we have the page locked, so new writeback can't start,
6271          * and the dirty bit won't be cleared while we are here.
6272          *
6273          * Wait for IO on this page so that we can safely clear
6274          * the PagePrivate2 bit and do ordered accounting
6275          */
6276         wait_on_page_writeback(page);
6277
6278         tree = &BTRFS_I(page->mapping->host)->io_tree;
6279         if (offset) {
6280                 btrfs_releasepage(page, GFP_NOFS);
6281                 return;
6282         }
6283         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6284                          GFP_NOFS);
6285         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6286                                            page_offset(page));
6287         if (ordered) {
6288                 /*
6289                  * IO on this page will never be started, so we need
6290                  * to account for any ordered extents now
6291                  */
6292                 clear_extent_bit(tree, page_start, page_end,
6293                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6294                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6295                                  &cached_state, GFP_NOFS);
6296                 /*
6297                  * whoever cleared the private bit is responsible
6298                  * for the finish_ordered_io
6299                  */
6300                 if (TestClearPagePrivate2(page)) {
6301                         btrfs_finish_ordered_io(page->mapping->host,
6302                                                 page_start, page_end);
6303                 }
6304                 btrfs_put_ordered_extent(ordered);
6305                 cached_state = NULL;
6306                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6307                                  GFP_NOFS);
6308         }
6309         clear_extent_bit(tree, page_start, page_end,
6310                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6311                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6312         __btrfs_releasepage(page, GFP_NOFS);
6313
6314         ClearPageChecked(page);
6315         if (PagePrivate(page)) {
6316                 ClearPagePrivate(page);
6317                 set_page_private(page, 0);
6318                 page_cache_release(page);
6319         }
6320 }
6321
6322 /*
6323  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6324  * called from a page fault handler when a page is first dirtied. Hence we must
6325  * be careful to check for EOF conditions here. We set the page up correctly
6326  * for a written page which means we get ENOSPC checking when writing into
6327  * holes and correct delalloc and unwritten extent mapping on filesystems that
6328  * support these features.
6329  *
6330  * We are not allowed to take the i_mutex here so we have to play games to
6331  * protect against truncate races as the page could now be beyond EOF.  Because
6332  * vmtruncate() writes the inode size before removing pages, once we have the
6333  * page lock we can determine safely if the page is beyond EOF. If it is not
6334  * beyond EOF, then the page is guaranteed safe against truncation until we
6335  * unlock the page.
6336  */
6337 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6338 {
6339         struct page *page = vmf->page;
6340         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6341         struct btrfs_root *root = BTRFS_I(inode)->root;
6342         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6343         struct btrfs_ordered_extent *ordered;
6344         struct extent_state *cached_state = NULL;
6345         char *kaddr;
6346         unsigned long zero_start;
6347         loff_t size;
6348         int ret;
6349         u64 page_start;
6350         u64 page_end;
6351
6352         /* Need this to keep space reservations serialized */
6353         mutex_lock(&inode->i_mutex);
6354         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6355         mutex_unlock(&inode->i_mutex);
6356         if (!ret)
6357                 ret = btrfs_update_time(vma->vm_file);
6358         if (ret) {
6359                 if (ret == -ENOMEM)
6360                         ret = VM_FAULT_OOM;
6361                 else /* -ENOSPC, -EIO, etc */
6362                         ret = VM_FAULT_SIGBUS;
6363                 goto out;
6364         }
6365
6366         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6367 again:
6368         lock_page(page);
6369         size = i_size_read(inode);
6370         page_start = page_offset(page);
6371         page_end = page_start + PAGE_CACHE_SIZE - 1;
6372
6373         if ((page->mapping != inode->i_mapping) ||
6374             (page_start >= size)) {
6375                 /* page got truncated out from underneath us */
6376                 goto out_unlock;
6377         }
6378         wait_on_page_writeback(page);
6379
6380         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6381                          GFP_NOFS);
6382         set_page_extent_mapped(page);
6383
6384         /*
6385          * we can't set the delalloc bits if there are pending ordered
6386          * extents.  Drop our locks and wait for them to finish
6387          */
6388         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6389         if (ordered) {
6390                 unlock_extent_cached(io_tree, page_start, page_end,
6391                                      &cached_state, GFP_NOFS);
6392                 unlock_page(page);
6393                 btrfs_start_ordered_extent(inode, ordered, 1);
6394                 btrfs_put_ordered_extent(ordered);
6395                 goto again;
6396         }
6397
6398         /*
6399          * XXX - page_mkwrite gets called every time the page is dirtied, even
6400          * if it was already dirty, so for space accounting reasons we need to
6401          * clear any delalloc bits for the range we are fixing to save.  There
6402          * is probably a better way to do this, but for now keep consistent with
6403          * prepare_pages in the normal write path.
6404          */
6405         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6406                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6407                           0, 0, &cached_state, GFP_NOFS);
6408
6409         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6410                                         &cached_state);
6411         if (ret) {
6412                 unlock_extent_cached(io_tree, page_start, page_end,
6413                                      &cached_state, GFP_NOFS);
6414                 ret = VM_FAULT_SIGBUS;
6415                 goto out_unlock;
6416         }
6417         ret = 0;
6418
6419         /* page is wholly or partially inside EOF */
6420         if (page_start + PAGE_CACHE_SIZE > size)
6421                 zero_start = size & ~PAGE_CACHE_MASK;
6422         else
6423                 zero_start = PAGE_CACHE_SIZE;
6424
6425         if (zero_start != PAGE_CACHE_SIZE) {
6426                 kaddr = kmap(page);
6427                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6428                 flush_dcache_page(page);
6429                 kunmap(page);
6430         }
6431         ClearPageChecked(page);
6432         set_page_dirty(page);
6433         SetPageUptodate(page);
6434
6435         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6436         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6437
6438         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6439
6440 out_unlock:
6441         if (!ret)
6442                 return VM_FAULT_LOCKED;
6443         unlock_page(page);
6444         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6445 out:
6446         return ret;
6447 }
6448
6449 static int btrfs_truncate(struct inode *inode)
6450 {
6451         struct btrfs_root *root = BTRFS_I(inode)->root;
6452         struct btrfs_block_rsv *rsv;
6453         int ret;
6454         int err = 0;
6455         struct btrfs_trans_handle *trans;
6456         unsigned long nr;
6457         u64 mask = root->sectorsize - 1;
6458         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6459
6460         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6461         if (ret)
6462                 return ret;
6463
6464         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6465         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6466
6467         /*
6468          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6469          * 3 things going on here
6470          *
6471          * 1) We need to reserve space for our orphan item and the space to
6472          * delete our orphan item.  Lord knows we don't want to have a dangling
6473          * orphan item because we didn't reserve space to remove it.
6474          *
6475          * 2) We need to reserve space to update our inode.
6476          *
6477          * 3) We need to have something to cache all the space that is going to
6478          * be free'd up by the truncate operation, but also have some slack
6479          * space reserved in case it uses space during the truncate (thank you
6480          * very much snapshotting).
6481          *
6482          * And we need these to all be seperate.  The fact is we can use alot of
6483          * space doing the truncate, and we have no earthly idea how much space
6484          * we will use, so we need the truncate reservation to be seperate so it
6485          * doesn't end up using space reserved for updating the inode or
6486          * removing the orphan item.  We also need to be able to stop the
6487          * transaction and start a new one, which means we need to be able to
6488          * update the inode several times, and we have no idea of knowing how
6489          * many times that will be, so we can't just reserve 1 item for the
6490          * entirety of the opration, so that has to be done seperately as well.
6491          * Then there is the orphan item, which does indeed need to be held on
6492          * to for the whole operation, and we need nobody to touch this reserved
6493          * space except the orphan code.
6494          *
6495          * So that leaves us with
6496          *
6497          * 1) root->orphan_block_rsv - for the orphan deletion.
6498          * 2) rsv - for the truncate reservation, which we will steal from the
6499          * transaction reservation.
6500          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6501          * updating the inode.
6502          */
6503         rsv = btrfs_alloc_block_rsv(root);
6504         if (!rsv)
6505                 return -ENOMEM;
6506         rsv->size = min_size;
6507
6508         /*
6509          * 1 for the truncate slack space
6510          * 1 for the orphan item we're going to add
6511          * 1 for the orphan item deletion
6512          * 1 for updating the inode.
6513          */
6514         trans = btrfs_start_transaction(root, 4);
6515         if (IS_ERR(trans)) {
6516                 err = PTR_ERR(trans);
6517                 goto out;
6518         }
6519
6520         /* Migrate the slack space for the truncate to our reserve */
6521         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6522                                       min_size);
6523         BUG_ON(ret);
6524
6525         ret = btrfs_orphan_add(trans, inode);
6526         if (ret) {
6527                 btrfs_end_transaction(trans, root);
6528                 goto out;
6529         }
6530
6531         /*
6532          * setattr is responsible for setting the ordered_data_close flag,
6533          * but that is only tested during the last file release.  That
6534          * could happen well after the next commit, leaving a great big
6535          * window where new writes may get lost if someone chooses to write
6536          * to this file after truncating to zero
6537          *
6538          * The inode doesn't have any dirty data here, and so if we commit
6539          * this is a noop.  If someone immediately starts writing to the inode
6540          * it is very likely we'll catch some of their writes in this
6541          * transaction, and the commit will find this file on the ordered
6542          * data list with good things to send down.
6543          *
6544          * This is a best effort solution, there is still a window where
6545          * using truncate to replace the contents of the file will
6546          * end up with a zero length file after a crash.
6547          */
6548         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6549                 btrfs_add_ordered_operation(trans, root, inode);
6550
6551         while (1) {
6552                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6553                 if (ret) {
6554                         /*
6555                          * This can only happen with the original transaction we
6556                          * started above, every other time we shouldn't have a
6557                          * transaction started yet.
6558                          */
6559                         if (ret == -EAGAIN)
6560                                 goto end_trans;
6561                         err = ret;
6562                         break;
6563                 }
6564
6565                 if (!trans) {
6566                         /* Just need the 1 for updating the inode */
6567                         trans = btrfs_start_transaction(root, 1);
6568                         if (IS_ERR(trans)) {
6569                                 err = PTR_ERR(trans);
6570                                 goto out;
6571                         }
6572                 }
6573
6574                 trans->block_rsv = rsv;
6575
6576                 ret = btrfs_truncate_inode_items(trans, root, inode,
6577                                                  inode->i_size,
6578                                                  BTRFS_EXTENT_DATA_KEY);
6579                 if (ret != -EAGAIN) {
6580                         err = ret;
6581                         break;
6582                 }
6583
6584                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6585                 ret = btrfs_update_inode(trans, root, inode);
6586                 if (ret) {
6587                         err = ret;
6588                         break;
6589                 }
6590 end_trans:
6591                 nr = trans->blocks_used;
6592                 btrfs_end_transaction(trans, root);
6593                 trans = NULL;
6594                 btrfs_btree_balance_dirty(root, nr);
6595         }
6596
6597         if (ret == 0 && inode->i_nlink > 0) {
6598                 trans->block_rsv = root->orphan_block_rsv;
6599                 ret = btrfs_orphan_del(trans, inode);
6600                 if (ret)
6601                         err = ret;
6602         } else if (ret && inode->i_nlink > 0) {
6603                 /*
6604                  * Failed to do the truncate, remove us from the in memory
6605                  * orphan list.
6606                  */
6607                 ret = btrfs_orphan_del(NULL, inode);
6608         }
6609
6610         if (trans) {
6611                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6612                 ret = btrfs_update_inode(trans, root, inode);
6613                 if (ret && !err)
6614                         err = ret;
6615
6616                 nr = trans->blocks_used;
6617                 ret = btrfs_end_transaction_throttle(trans, root);
6618                 btrfs_btree_balance_dirty(root, nr);
6619         }
6620
6621 out:
6622         btrfs_free_block_rsv(root, rsv);
6623
6624         if (ret && !err)
6625                 err = ret;
6626
6627         return err;
6628 }
6629
6630 /*
6631  * create a new subvolume directory/inode (helper for the ioctl).
6632  */
6633 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6634                              struct btrfs_root *new_root, u64 new_dirid)
6635 {
6636         struct inode *inode;
6637         int err;
6638         u64 index = 0;
6639
6640         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6641                                 new_dirid, S_IFDIR | 0700, &index);
6642         if (IS_ERR(inode))
6643                 return PTR_ERR(inode);
6644         inode->i_op = &btrfs_dir_inode_operations;
6645         inode->i_fop = &btrfs_dir_file_operations;
6646
6647         inode->i_nlink = 1;
6648         btrfs_i_size_write(inode, 0);
6649
6650         err = btrfs_update_inode(trans, new_root, inode);
6651         BUG_ON(err);
6652
6653         iput(inode);
6654         return 0;
6655 }
6656
6657 struct inode *btrfs_alloc_inode(struct super_block *sb)
6658 {
6659         struct btrfs_inode *ei;
6660         struct inode *inode;
6661
6662         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6663         if (!ei)
6664                 return NULL;
6665
6666         ei->root = NULL;
6667         ei->space_info = NULL;
6668         ei->generation = 0;
6669         ei->sequence = 0;
6670         ei->last_trans = 0;
6671         ei->last_sub_trans = 0;
6672         ei->logged_trans = 0;
6673         ei->delalloc_bytes = 0;
6674         ei->disk_i_size = 0;
6675         ei->flags = 0;
6676         ei->csum_bytes = 0;
6677         ei->index_cnt = (u64)-1;
6678         ei->last_unlink_trans = 0;
6679
6680         spin_lock_init(&ei->lock);
6681         ei->outstanding_extents = 0;
6682         ei->reserved_extents = 0;
6683
6684         ei->ordered_data_close = 0;
6685         ei->orphan_meta_reserved = 0;
6686         ei->dummy_inode = 0;
6687         ei->in_defrag = 0;
6688         ei->delalloc_meta_reserved = 0;
6689         ei->force_compress = BTRFS_COMPRESS_NONE;
6690
6691         ei->delayed_node = NULL;
6692
6693         inode = &ei->vfs_inode;
6694         extent_map_tree_init(&ei->extent_tree);
6695         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6696         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6697         mutex_init(&ei->log_mutex);
6698         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6699         INIT_LIST_HEAD(&ei->i_orphan);
6700         INIT_LIST_HEAD(&ei->delalloc_inodes);
6701         INIT_LIST_HEAD(&ei->ordered_operations);
6702         RB_CLEAR_NODE(&ei->rb_node);
6703
6704         return inode;
6705 }
6706
6707 static void btrfs_i_callback(struct rcu_head *head)
6708 {
6709         struct inode *inode = container_of(head, struct inode, i_rcu);
6710         INIT_LIST_HEAD(&inode->i_dentry);
6711         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6712 }
6713
6714 void btrfs_destroy_inode(struct inode *inode)
6715 {
6716         struct btrfs_ordered_extent *ordered;
6717         struct btrfs_root *root = BTRFS_I(inode)->root;
6718
6719         WARN_ON(!list_empty(&inode->i_dentry));
6720         WARN_ON(inode->i_data.nrpages);
6721         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6722         WARN_ON(BTRFS_I(inode)->reserved_extents);
6723         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6724         WARN_ON(BTRFS_I(inode)->csum_bytes);
6725
6726         /*
6727          * This can happen where we create an inode, but somebody else also
6728          * created the same inode and we need to destroy the one we already
6729          * created.
6730          */
6731         if (!root)
6732                 goto free;
6733
6734         /*
6735          * Make sure we're properly removed from the ordered operation
6736          * lists.
6737          */
6738         smp_mb();
6739         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6740                 spin_lock(&root->fs_info->ordered_extent_lock);
6741                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6742                 spin_unlock(&root->fs_info->ordered_extent_lock);
6743         }
6744
6745         spin_lock(&root->orphan_lock);
6746         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6747                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6748                        (unsigned long long)btrfs_ino(inode));
6749                 list_del_init(&BTRFS_I(inode)->i_orphan);
6750         }
6751         spin_unlock(&root->orphan_lock);
6752
6753         while (1) {
6754                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6755                 if (!ordered)
6756                         break;
6757                 else {
6758                         printk(KERN_ERR "btrfs found ordered "
6759                                "extent %llu %llu on inode cleanup\n",
6760                                (unsigned long long)ordered->file_offset,
6761                                (unsigned long long)ordered->len);
6762                         btrfs_remove_ordered_extent(inode, ordered);
6763                         btrfs_put_ordered_extent(ordered);
6764                         btrfs_put_ordered_extent(ordered);
6765                 }
6766         }
6767         inode_tree_del(inode);
6768         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6769 free:
6770         btrfs_remove_delayed_node(inode);
6771         call_rcu(&inode->i_rcu, btrfs_i_callback);
6772 }
6773
6774 int btrfs_drop_inode(struct inode *inode)
6775 {
6776         struct btrfs_root *root = BTRFS_I(inode)->root;
6777
6778         if (btrfs_root_refs(&root->root_item) == 0 &&
6779             !btrfs_is_free_space_inode(root, inode))
6780                 return 1;
6781         else
6782                 return generic_drop_inode(inode);
6783 }
6784
6785 static void init_once(void *foo)
6786 {
6787         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6788
6789         inode_init_once(&ei->vfs_inode);
6790 }
6791
6792 void btrfs_destroy_cachep(void)
6793 {
6794         if (btrfs_inode_cachep)
6795                 kmem_cache_destroy(btrfs_inode_cachep);
6796         if (btrfs_trans_handle_cachep)
6797                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6798         if (btrfs_transaction_cachep)
6799                 kmem_cache_destroy(btrfs_transaction_cachep);
6800         if (btrfs_path_cachep)
6801                 kmem_cache_destroy(btrfs_path_cachep);
6802         if (btrfs_free_space_cachep)
6803                 kmem_cache_destroy(btrfs_free_space_cachep);
6804 }
6805
6806 int btrfs_init_cachep(void)
6807 {
6808         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6809                         sizeof(struct btrfs_inode), 0,
6810                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6811         if (!btrfs_inode_cachep)
6812                 goto fail;
6813
6814         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6815                         sizeof(struct btrfs_trans_handle), 0,
6816                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6817         if (!btrfs_trans_handle_cachep)
6818                 goto fail;
6819
6820         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6821                         sizeof(struct btrfs_transaction), 0,
6822                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6823         if (!btrfs_transaction_cachep)
6824                 goto fail;
6825
6826         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6827                         sizeof(struct btrfs_path), 0,
6828                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6829         if (!btrfs_path_cachep)
6830                 goto fail;
6831
6832         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6833                         sizeof(struct btrfs_free_space), 0,
6834                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6835         if (!btrfs_free_space_cachep)
6836                 goto fail;
6837
6838         return 0;
6839 fail:
6840         btrfs_destroy_cachep();
6841         return -ENOMEM;
6842 }
6843
6844 static int btrfs_getattr(struct vfsmount *mnt,
6845                          struct dentry *dentry, struct kstat *stat)
6846 {
6847         struct inode *inode = dentry->d_inode;
6848         u32 blocksize = inode->i_sb->s_blocksize;
6849
6850         generic_fillattr(inode, stat);
6851         stat->dev = BTRFS_I(inode)->root->anon_dev;
6852         stat->blksize = PAGE_CACHE_SIZE;
6853         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6854                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6855         return 0;
6856 }
6857
6858 /*
6859  * If a file is moved, it will inherit the cow and compression flags of the new
6860  * directory.
6861  */
6862 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6863 {
6864         struct btrfs_inode *b_dir = BTRFS_I(dir);
6865         struct btrfs_inode *b_inode = BTRFS_I(inode);
6866
6867         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6868                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6869         else
6870                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6871
6872         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6873                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6874         else
6875                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6876 }
6877
6878 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6879                            struct inode *new_dir, struct dentry *new_dentry)
6880 {
6881         struct btrfs_trans_handle *trans;
6882         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6883         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6884         struct inode *new_inode = new_dentry->d_inode;
6885         struct inode *old_inode = old_dentry->d_inode;
6886         struct timespec ctime = CURRENT_TIME;
6887         u64 index = 0;
6888         u64 root_objectid;
6889         int ret;
6890         u64 old_ino = btrfs_ino(old_inode);
6891
6892         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6893                 return -EPERM;
6894
6895         /* we only allow rename subvolume link between subvolumes */
6896         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6897                 return -EXDEV;
6898
6899         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6900             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6901                 return -ENOTEMPTY;
6902
6903         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6904             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6905                 return -ENOTEMPTY;
6906         /*
6907          * we're using rename to replace one file with another.
6908          * and the replacement file is large.  Start IO on it now so
6909          * we don't add too much work to the end of the transaction
6910          */
6911         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6912             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6913                 filemap_flush(old_inode->i_mapping);
6914
6915         /* close the racy window with snapshot create/destroy ioctl */
6916         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6917                 down_read(&root->fs_info->subvol_sem);
6918         /*
6919          * We want to reserve the absolute worst case amount of items.  So if
6920          * both inodes are subvols and we need to unlink them then that would
6921          * require 4 item modifications, but if they are both normal inodes it
6922          * would require 5 item modifications, so we'll assume their normal
6923          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6924          * should cover the worst case number of items we'll modify.
6925          */
6926         trans = btrfs_start_transaction(root, 20);
6927         if (IS_ERR(trans)) {
6928                 ret = PTR_ERR(trans);
6929                 goto out_notrans;
6930         }
6931
6932         if (dest != root)
6933                 btrfs_record_root_in_trans(trans, dest);
6934
6935         ret = btrfs_set_inode_index(new_dir, &index);
6936         if (ret)
6937                 goto out_fail;
6938
6939         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6940                 /* force full log commit if subvolume involved. */
6941                 root->fs_info->last_trans_log_full_commit = trans->transid;
6942         } else {
6943                 ret = btrfs_insert_inode_ref(trans, dest,
6944                                              new_dentry->d_name.name,
6945                                              new_dentry->d_name.len,
6946                                              old_ino,
6947                                              btrfs_ino(new_dir), index);
6948                 if (ret)
6949                         goto out_fail;
6950                 /*
6951                  * this is an ugly little race, but the rename is required
6952                  * to make sure that if we crash, the inode is either at the
6953                  * old name or the new one.  pinning the log transaction lets
6954                  * us make sure we don't allow a log commit to come in after
6955                  * we unlink the name but before we add the new name back in.
6956                  */
6957                 btrfs_pin_log_trans(root);
6958         }
6959         /*
6960          * make sure the inode gets flushed if it is replacing
6961          * something.
6962          */
6963         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6964                 btrfs_add_ordered_operation(trans, root, old_inode);
6965
6966         old_dir->i_ctime = old_dir->i_mtime = ctime;
6967         new_dir->i_ctime = new_dir->i_mtime = ctime;
6968         old_inode->i_ctime = ctime;
6969
6970         if (old_dentry->d_parent != new_dentry->d_parent)
6971                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6972
6973         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6974                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6975                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6976                                         old_dentry->d_name.name,
6977                                         old_dentry->d_name.len);
6978         } else {
6979                 ret = __btrfs_unlink_inode(trans, root, old_dir,
6980                                         old_dentry->d_inode,
6981                                         old_dentry->d_name.name,
6982                                         old_dentry->d_name.len);
6983                 if (!ret)
6984                         ret = btrfs_update_inode(trans, root, old_inode);
6985         }
6986         BUG_ON(ret);
6987
6988         if (new_inode) {
6989                 new_inode->i_ctime = CURRENT_TIME;
6990                 if (unlikely(btrfs_ino(new_inode) ==
6991                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6992                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6993                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6994                                                 root_objectid,
6995                                                 new_dentry->d_name.name,
6996                                                 new_dentry->d_name.len);
6997                         BUG_ON(new_inode->i_nlink == 0);
6998                 } else {
6999                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7000                                                  new_dentry->d_inode,
7001                                                  new_dentry->d_name.name,
7002                                                  new_dentry->d_name.len);
7003                 }
7004                 BUG_ON(ret);
7005                 if (new_inode->i_nlink == 0) {
7006                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7007                         BUG_ON(ret);
7008                 }
7009         }
7010
7011         fixup_inode_flags(new_dir, old_inode);
7012
7013         ret = btrfs_add_link(trans, new_dir, old_inode,
7014                              new_dentry->d_name.name,
7015                              new_dentry->d_name.len, 0, index);
7016         BUG_ON(ret);
7017
7018         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7019                 struct dentry *parent = new_dentry->d_parent;
7020                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7021                 btrfs_end_log_trans(root);
7022         }
7023 out_fail:
7024         btrfs_end_transaction_throttle(trans, root);
7025 out_notrans:
7026         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7027                 up_read(&root->fs_info->subvol_sem);
7028
7029         return ret;
7030 }
7031
7032 /*
7033  * some fairly slow code that needs optimization. This walks the list
7034  * of all the inodes with pending delalloc and forces them to disk.
7035  */
7036 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7037 {
7038         struct list_head *head = &root->fs_info->delalloc_inodes;
7039         struct btrfs_inode *binode;
7040         struct inode *inode;
7041
7042         if (root->fs_info->sb->s_flags & MS_RDONLY)
7043                 return -EROFS;
7044
7045         spin_lock(&root->fs_info->delalloc_lock);
7046         while (!list_empty(head)) {
7047                 binode = list_entry(head->next, struct btrfs_inode,
7048                                     delalloc_inodes);
7049                 inode = igrab(&binode->vfs_inode);
7050                 if (!inode)
7051                         list_del_init(&binode->delalloc_inodes);
7052                 spin_unlock(&root->fs_info->delalloc_lock);
7053                 if (inode) {
7054                         filemap_flush(inode->i_mapping);
7055                         if (delay_iput)
7056                                 btrfs_add_delayed_iput(inode);
7057                         else
7058                                 iput(inode);
7059                 }
7060                 cond_resched();
7061                 spin_lock(&root->fs_info->delalloc_lock);
7062         }
7063         spin_unlock(&root->fs_info->delalloc_lock);
7064
7065         /* the filemap_flush will queue IO into the worker threads, but
7066          * we have to make sure the IO is actually started and that
7067          * ordered extents get created before we return
7068          */
7069         atomic_inc(&root->fs_info->async_submit_draining);
7070         while (atomic_read(&root->fs_info->nr_async_submits) ||
7071               atomic_read(&root->fs_info->async_delalloc_pages)) {
7072                 wait_event(root->fs_info->async_submit_wait,
7073                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7074                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7075         }
7076         atomic_dec(&root->fs_info->async_submit_draining);
7077         return 0;
7078 }
7079
7080 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7081                          const char *symname)
7082 {
7083         struct btrfs_trans_handle *trans;
7084         struct btrfs_root *root = BTRFS_I(dir)->root;
7085         struct btrfs_path *path;
7086         struct btrfs_key key;
7087         struct inode *inode = NULL;
7088         int err;
7089         int drop_inode = 0;
7090         u64 objectid;
7091         u64 index = 0 ;
7092         int name_len;
7093         int datasize;
7094         unsigned long ptr;
7095         struct btrfs_file_extent_item *ei;
7096         struct extent_buffer *leaf;
7097         unsigned long nr = 0;
7098
7099         name_len = strlen(symname) + 1;
7100         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7101                 return -ENAMETOOLONG;
7102
7103         /*
7104          * 2 items for inode item and ref
7105          * 2 items for dir items
7106          * 1 item for xattr if selinux is on
7107          */
7108         trans = btrfs_start_transaction(root, 5);
7109         if (IS_ERR(trans))
7110                 return PTR_ERR(trans);
7111
7112         err = btrfs_find_free_ino(root, &objectid);
7113         if (err)
7114                 goto out_unlock;
7115
7116         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7117                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7118                                 S_IFLNK|S_IRWXUGO, &index);
7119         if (IS_ERR(inode)) {
7120                 err = PTR_ERR(inode);
7121                 goto out_unlock;
7122         }
7123
7124         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7125         if (err) {
7126                 drop_inode = 1;
7127                 goto out_unlock;
7128         }
7129
7130         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7131         if (err)
7132                 drop_inode = 1;
7133         else {
7134                 inode->i_mapping->a_ops = &btrfs_aops;
7135                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7136                 inode->i_fop = &btrfs_file_operations;
7137                 inode->i_op = &btrfs_file_inode_operations;
7138                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7139         }
7140         if (drop_inode)
7141                 goto out_unlock;
7142
7143         path = btrfs_alloc_path();
7144         if (!path) {
7145                 err = -ENOMEM;
7146                 drop_inode = 1;
7147                 goto out_unlock;
7148         }
7149         key.objectid = btrfs_ino(inode);
7150         key.offset = 0;
7151         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7152         datasize = btrfs_file_extent_calc_inline_size(name_len);
7153         err = btrfs_insert_empty_item(trans, root, path, &key,
7154                                       datasize);
7155         if (err) {
7156                 drop_inode = 1;
7157                 btrfs_free_path(path);
7158                 goto out_unlock;
7159         }
7160         leaf = path->nodes[0];
7161         ei = btrfs_item_ptr(leaf, path->slots[0],
7162                             struct btrfs_file_extent_item);
7163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7164         btrfs_set_file_extent_type(leaf, ei,
7165                                    BTRFS_FILE_EXTENT_INLINE);
7166         btrfs_set_file_extent_encryption(leaf, ei, 0);
7167         btrfs_set_file_extent_compression(leaf, ei, 0);
7168         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7169         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7170
7171         ptr = btrfs_file_extent_inline_start(ei);
7172         write_extent_buffer(leaf, symname, ptr, name_len);
7173         btrfs_mark_buffer_dirty(leaf);
7174         btrfs_free_path(path);
7175
7176         inode->i_op = &btrfs_symlink_inode_operations;
7177         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7178         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7179         inode_set_bytes(inode, name_len);
7180         btrfs_i_size_write(inode, name_len - 1);
7181         err = btrfs_update_inode(trans, root, inode);
7182         if (err)
7183                 drop_inode = 1;
7184
7185 out_unlock:
7186         nr = trans->blocks_used;
7187         btrfs_end_transaction_throttle(trans, root);
7188         if (drop_inode) {
7189                 inode_dec_link_count(inode);
7190                 iput(inode);
7191         }
7192         btrfs_btree_balance_dirty(root, nr);
7193         return err;
7194 }
7195
7196 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7197                                        u64 start, u64 num_bytes, u64 min_size,
7198                                        loff_t actual_len, u64 *alloc_hint,
7199                                        struct btrfs_trans_handle *trans)
7200 {
7201         struct btrfs_root *root = BTRFS_I(inode)->root;
7202         struct btrfs_key ins;
7203         u64 cur_offset = start;
7204         u64 i_size;
7205         int ret = 0;
7206         bool own_trans = true;
7207
7208         if (trans)
7209                 own_trans = false;
7210         while (num_bytes > 0) {
7211                 if (own_trans) {
7212                         trans = btrfs_start_transaction(root, 3);
7213                         if (IS_ERR(trans)) {
7214                                 ret = PTR_ERR(trans);
7215                                 break;
7216                         }
7217                 }
7218
7219                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7220                                            0, *alloc_hint, (u64)-1, &ins, 1);
7221                 if (ret) {
7222                         if (own_trans)
7223                                 btrfs_end_transaction(trans, root);
7224                         break;
7225                 }
7226
7227                 ret = insert_reserved_file_extent(trans, inode,
7228                                                   cur_offset, ins.objectid,
7229                                                   ins.offset, ins.offset,
7230                                                   ins.offset, 0, 0, 0,
7231                                                   BTRFS_FILE_EXTENT_PREALLOC);
7232                 BUG_ON(ret);
7233                 btrfs_drop_extent_cache(inode, cur_offset,
7234                                         cur_offset + ins.offset -1, 0);
7235
7236                 num_bytes -= ins.offset;
7237                 cur_offset += ins.offset;
7238                 *alloc_hint = ins.objectid + ins.offset;
7239
7240                 inode->i_ctime = CURRENT_TIME;
7241                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7242                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7243                     (actual_len > inode->i_size) &&
7244                     (cur_offset > inode->i_size)) {
7245                         if (cur_offset > actual_len)
7246                                 i_size = actual_len;
7247                         else
7248                                 i_size = cur_offset;
7249                         i_size_write(inode, i_size);
7250                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7251                 }
7252
7253                 ret = btrfs_update_inode(trans, root, inode);
7254                 BUG_ON(ret);
7255
7256                 if (own_trans)
7257                         btrfs_end_transaction(trans, root);
7258         }
7259         return ret;
7260 }
7261
7262 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7263                               u64 start, u64 num_bytes, u64 min_size,
7264                               loff_t actual_len, u64 *alloc_hint)
7265 {
7266         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7267                                            min_size, actual_len, alloc_hint,
7268                                            NULL);
7269 }
7270
7271 int btrfs_prealloc_file_range_trans(struct inode *inode,
7272                                     struct btrfs_trans_handle *trans, int mode,
7273                                     u64 start, u64 num_bytes, u64 min_size,
7274                                     loff_t actual_len, u64 *alloc_hint)
7275 {
7276         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7277                                            min_size, actual_len, alloc_hint, trans);
7278 }
7279
7280 static int btrfs_set_page_dirty(struct page *page)
7281 {
7282         return __set_page_dirty_nobuffers(page);
7283 }
7284
7285 static int btrfs_permission(struct inode *inode, int mask)
7286 {
7287         struct btrfs_root *root = BTRFS_I(inode)->root;
7288         umode_t mode = inode->i_mode;
7289
7290         if (mask & MAY_WRITE &&
7291             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7292                 if (btrfs_root_readonly(root))
7293                         return -EROFS;
7294                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7295                         return -EACCES;
7296         }
7297         return generic_permission(inode, mask);
7298 }
7299
7300 static const struct inode_operations btrfs_dir_inode_operations = {
7301         .getattr        = btrfs_getattr,
7302         .lookup         = btrfs_lookup,
7303         .create         = btrfs_create,
7304         .unlink         = btrfs_unlink,
7305         .link           = btrfs_link,
7306         .mkdir          = btrfs_mkdir,
7307         .rmdir          = btrfs_rmdir,
7308         .rename         = btrfs_rename,
7309         .symlink        = btrfs_symlink,
7310         .setattr        = btrfs_setattr,
7311         .mknod          = btrfs_mknod,
7312         .setxattr       = btrfs_setxattr,
7313         .getxattr       = btrfs_getxattr,
7314         .listxattr      = btrfs_listxattr,
7315         .removexattr    = btrfs_removexattr,
7316         .permission     = btrfs_permission,
7317         .get_acl        = btrfs_get_acl,
7318 };
7319 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7320         .lookup         = btrfs_lookup,
7321         .permission     = btrfs_permission,
7322         .get_acl        = btrfs_get_acl,
7323 };
7324
7325 static const struct file_operations btrfs_dir_file_operations = {
7326         .llseek         = generic_file_llseek,
7327         .read           = generic_read_dir,
7328         .readdir        = btrfs_real_readdir,
7329         .unlocked_ioctl = btrfs_ioctl,
7330 #ifdef CONFIG_COMPAT
7331         .compat_ioctl   = btrfs_ioctl,
7332 #endif
7333         .release        = btrfs_release_file,
7334         .fsync          = btrfs_sync_file,
7335 };
7336
7337 static struct extent_io_ops btrfs_extent_io_ops = {
7338         .fill_delalloc = run_delalloc_range,
7339         .submit_bio_hook = btrfs_submit_bio_hook,
7340         .merge_bio_hook = btrfs_merge_bio_hook,
7341         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7342         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7343         .writepage_start_hook = btrfs_writepage_start_hook,
7344         .set_bit_hook = btrfs_set_bit_hook,
7345         .clear_bit_hook = btrfs_clear_bit_hook,
7346         .merge_extent_hook = btrfs_merge_extent_hook,
7347         .split_extent_hook = btrfs_split_extent_hook,
7348 };
7349
7350 /*
7351  * btrfs doesn't support the bmap operation because swapfiles
7352  * use bmap to make a mapping of extents in the file.  They assume
7353  * these extents won't change over the life of the file and they
7354  * use the bmap result to do IO directly to the drive.
7355  *
7356  * the btrfs bmap call would return logical addresses that aren't
7357  * suitable for IO and they also will change frequently as COW
7358  * operations happen.  So, swapfile + btrfs == corruption.
7359  *
7360  * For now we're avoiding this by dropping bmap.
7361  */
7362 static const struct address_space_operations btrfs_aops = {
7363         .readpage       = btrfs_readpage,
7364         .writepage      = btrfs_writepage,
7365         .writepages     = btrfs_writepages,
7366         .readpages      = btrfs_readpages,
7367         .direct_IO      = btrfs_direct_IO,
7368         .invalidatepage = btrfs_invalidatepage,
7369         .releasepage    = btrfs_releasepage,
7370         .set_page_dirty = btrfs_set_page_dirty,
7371         .error_remove_page = generic_error_remove_page,
7372 };
7373
7374 static const struct address_space_operations btrfs_symlink_aops = {
7375         .readpage       = btrfs_readpage,
7376         .writepage      = btrfs_writepage,
7377         .invalidatepage = btrfs_invalidatepage,
7378         .releasepage    = btrfs_releasepage,
7379 };
7380
7381 static const struct inode_operations btrfs_file_inode_operations = {
7382         .getattr        = btrfs_getattr,
7383         .setattr        = btrfs_setattr,
7384         .setxattr       = btrfs_setxattr,
7385         .getxattr       = btrfs_getxattr,
7386         .listxattr      = btrfs_listxattr,
7387         .removexattr    = btrfs_removexattr,
7388         .permission     = btrfs_permission,
7389         .fiemap         = btrfs_fiemap,
7390         .get_acl        = btrfs_get_acl,
7391 };
7392 static const struct inode_operations btrfs_special_inode_operations = {
7393         .getattr        = btrfs_getattr,
7394         .setattr        = btrfs_setattr,
7395         .permission     = btrfs_permission,
7396         .setxattr       = btrfs_setxattr,
7397         .getxattr       = btrfs_getxattr,
7398         .listxattr      = btrfs_listxattr,
7399         .removexattr    = btrfs_removexattr,
7400         .get_acl        = btrfs_get_acl,
7401 };
7402 static const struct inode_operations btrfs_symlink_inode_operations = {
7403         .readlink       = generic_readlink,
7404         .follow_link    = page_follow_link_light,
7405         .put_link       = page_put_link,
7406         .getattr        = btrfs_getattr,
7407         .setattr        = btrfs_setattr,
7408         .permission     = btrfs_permission,
7409         .setxattr       = btrfs_setxattr,
7410         .getxattr       = btrfs_getxattr,
7411         .listxattr      = btrfs_listxattr,
7412         .removexattr    = btrfs_removexattr,
7413         .get_acl        = btrfs_get_acl,
7414 };
7415
7416 const struct dentry_operations btrfs_dentry_operations = {
7417         .d_delete       = btrfs_dentry_delete,
7418         .d_release      = btrfs_dentry_release,
7419 };