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