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/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.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 "ref-cache.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static struct inode_operations btrfs_dir_inode_operations;
  61 static struct inode_operations btrfs_symlink_inode_operations;
  62 static struct inode_operations btrfs_dir_ro_inode_operations;
  63 static struct inode_operations btrfs_special_inode_operations;
  64 static struct inode_operations btrfs_file_inode_operations;
  65 static struct address_space_operations btrfs_aops;
  66 static struct address_space_operations btrfs_symlink_aops;
  67 static struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74
  75 #define S_SHIFT 12
  76 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  77         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  78         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  79         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  80         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  81         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  82         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  83         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  84 };
  85
  86 static void btrfs_truncate(struct inode *inode);
  87 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  88 static noinline int cow_file_range(struct inode *inode,
  89                                    struct page *locked_page,
  90                                    u64 start, u64 end, int *page_started,
  91                                    unsigned long *nr_written, int unlock);
  92
  93 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
  94 {
  95         int err;
  96
  97         err = btrfs_init_acl(inode, dir);
  98         if (!err)
  99                 err = btrfs_xattr_security_init(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         BTRFS_I(inode)->disk_i_size = inode->i_size;
 194         btrfs_update_inode(trans, root, inode);
 195         return 0;
 196 fail:
 197         btrfs_free_path(path);
 198         return err;
 199 }
 200
 201
 202 /*
 203  * conditionally insert an inline extent into the file.  This
 204  * does the checks required to make sure the data is small enough
 205  * to fit as an inline extent.
 206  */
 207 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 208                                  struct btrfs_root *root,
 209                                  struct inode *inode, u64 start, u64 end,
 210                                  size_t compressed_size,
 211                                  struct page **compressed_pages)
 212 {
 213         u64 isize = i_size_read(inode);
 214         u64 actual_end = min(end + 1, isize);
 215         u64 inline_len = actual_end - start;
 216         u64 aligned_end = (end + root->sectorsize - 1) &
 217                         ~((u64)root->sectorsize - 1);
 218         u64 hint_byte;
 219         u64 data_len = inline_len;
 220         int ret;
 221
 222         if (compressed_size)
 223                 data_len = compressed_size;
 224
 225         if (start > 0 ||
 226             actual_end >= PAGE_CACHE_SIZE ||
 227             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 228             (!compressed_size &&
 229             (actual_end & (root->sectorsize - 1)) == 0) ||
 230             end + 1 < isize ||
 231             data_len > root->fs_info->max_inline) {
 232                 return 1;
 233         }
 234
 235         ret = btrfs_drop_extents(trans, root, inode, start,
 236                                  aligned_end, aligned_end, start, &hint_byte);
 237         BUG_ON(ret);
 238
 239         if (isize > actual_end)
 240                 inline_len = min_t(u64, isize, actual_end);
 241         ret = insert_inline_extent(trans, root, inode, start,
 242                                    inline_len, compressed_size,
 243                                    compressed_pages);
 244         BUG_ON(ret);
 245         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
 246         return 0;
 247 }
 248
 249 struct async_extent {
 250         u64 start;
 251         u64 ram_size;
 252         u64 compressed_size;
 253         struct page **pages;
 254         unsigned long nr_pages;
 255         struct list_head list;
 256 };
 257
 258 struct async_cow {
 259         struct inode *inode;
 260         struct btrfs_root *root;
 261         struct page *locked_page;
 262         u64 start;
 263         u64 end;
 264         struct list_head extents;
 265         struct btrfs_work work;
 266 };
 267
 268 static noinline int add_async_extent(struct async_cow *cow,
 269                                      u64 start, u64 ram_size,
 270                                      u64 compressed_size,
 271                                      struct page **pages,
 272                                      unsigned long nr_pages)
 273 {
 274         struct async_extent *async_extent;
 275
 276         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 277         async_extent->start = start;
 278         async_extent->ram_size = ram_size;
 279         async_extent->compressed_size = compressed_size;
 280         async_extent->pages = pages;
 281         async_extent->nr_pages = nr_pages;
 282         list_add_tail(&async_extent->list, &cow->extents);
 283         return 0;
 284 }
 285
 286 /*
 287  * we create compressed extents in two phases.  The first
 288  * phase compresses a range of pages that have already been
 289  * locked (both pages and state bits are locked).
 290  *
 291  * This is done inside an ordered work queue, and the compression
 292  * is spread across many cpus.  The actual IO submission is step
 293  * two, and the ordered work queue takes care of making sure that
 294  * happens in the same order things were put onto the queue by
 295  * writepages and friends.
 296  *
 297  * If this code finds it can't get good compression, it puts an
 298  * entry onto the work queue to write the uncompressed bytes.  This
 299  * makes sure that both compressed inodes and uncompressed inodes
 300  * are written in the same order that pdflush sent them down.
 301  */
 302 static noinline int compress_file_range(struct inode *inode,
 303                                         struct page *locked_page,
 304                                         u64 start, u64 end,
 305                                         struct async_cow *async_cow,
 306                                         int *num_added)
 307 {
 308         struct btrfs_root *root = BTRFS_I(inode)->root;
 309         struct btrfs_trans_handle *trans;
 310         u64 num_bytes;
 311         u64 orig_start;
 312         u64 disk_num_bytes;
 313         u64 blocksize = root->sectorsize;
 314         u64 actual_end;
 315         u64 isize = i_size_read(inode);
 316         int ret = 0;
 317         struct page **pages = NULL;
 318         unsigned long nr_pages;
 319         unsigned long nr_pages_ret = 0;
 320         unsigned long total_compressed = 0;
 321         unsigned long total_in = 0;
 322         unsigned long max_compressed = 128 * 1024;
 323         unsigned long max_uncompressed = 128 * 1024;
 324         int i;
 325         int will_compress;
 326
 327         orig_start = start;
 328
 329         actual_end = min_t(u64, isize, end + 1);
 330 again:
 331         will_compress = 0;
 332         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 333         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 334
 335         /*
 336          * we don't want to send crud past the end of i_size through
 337          * compression, that's just a waste of CPU time.  So, if the
 338          * end of the file is before the start of our current
 339          * requested range of bytes, we bail out to the uncompressed
 340          * cleanup code that can deal with all of this.
 341          *
 342          * It isn't really the fastest way to fix things, but this is a
 343          * very uncommon corner.
 344          */
 345         if (actual_end <= start)
 346                 goto cleanup_and_bail_uncompressed;
 347
 348         total_compressed = actual_end - start;
 349
 350         /* we want to make sure that amount of ram required to uncompress
 351          * an extent is reasonable, so we limit the total size in ram
 352          * of a compressed extent to 128k.  This is a crucial number
 353          * because it also controls how easily we can spread reads across
 354          * cpus for decompression.
 355          *
 356          * We also want to make sure the amount of IO required to do
 357          * a random read is reasonably small, so we limit the size of
 358          * a compressed extent to 128k.
 359          */
 360         total_compressed = min(total_compressed, max_uncompressed);
 361         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 362         num_bytes = max(blocksize,  num_bytes);
 363         disk_num_bytes = num_bytes;
 364         total_in = 0;
 365         ret = 0;
 366
 367         /*
 368          * we do compression for mount -o compress and when the
 369          * inode has not been flagged as nocompress.  This flag can
 370          * change at any time if we discover bad compression ratios.
 371          */
 372         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
 373             btrfs_test_opt(root, COMPRESS)) {
 374                 WARN_ON(pages);
 375                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 376
 377                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 378                                                 total_compressed, pages,
 379                                                 nr_pages, &nr_pages_ret,
 380                                                 &total_in,
 381                                                 &total_compressed,
 382                                                 max_compressed);
 383
 384                 if (!ret) {
 385                         unsigned long offset = total_compressed &
 386                                 (PAGE_CACHE_SIZE - 1);
 387                         struct page *page = pages[nr_pages_ret - 1];
 388                         char *kaddr;
 389
 390                         /* zero the tail end of the last page, we might be
 391                          * sending it down to disk
 392                          */
 393                         if (offset) {
 394                                 kaddr = kmap_atomic(page, KM_USER0);
 395                                 memset(kaddr + offset, 0,
 396                                        PAGE_CACHE_SIZE - offset);
 397                                 kunmap_atomic(kaddr, KM_USER0);
 398                         }
 399                         will_compress = 1;
 400                 }
 401         }
 402         if (start == 0) {
 403                 trans = btrfs_join_transaction(root, 1);
 404                 BUG_ON(!trans);
 405                 btrfs_set_trans_block_group(trans, inode);
 406
 407                 /* lets try to make an inline extent */
 408                 if (ret || total_in < (actual_end - start)) {
 409                         /* we didn't compress the entire range, try
 410                          * to make an uncompressed inline extent.
 411                          */
 412                         ret = cow_file_range_inline(trans, root, inode,
 413                                                     start, end, 0, NULL);
 414                 } else {
 415                         /* try making a compressed inline extent */
 416                         ret = cow_file_range_inline(trans, root, inode,
 417                                                     start, end,
 418                                                     total_compressed, pages);
 419                 }
 420                 btrfs_end_transaction(trans, root);
 421                 if (ret == 0) {
 422                         /*
 423                          * inline extent creation worked, we don't need
 424                          * to create any more async work items.  Unlock
 425                          * and free up our temp pages.
 426                          */
 427                         extent_clear_unlock_delalloc(inode,
 428                                                      &BTRFS_I(inode)->io_tree,
 429                                                      start, end, NULL, 1, 0,
 430                                                      0, 1, 1, 1);
 431                         ret = 0;
 432                         goto free_pages_out;
 433                 }
 434         }
 435
 436         if (will_compress) {
 437                 /*
 438                  * we aren't doing an inline extent round the compressed size
 439                  * up to a block size boundary so the allocator does sane
 440                  * things
 441                  */
 442                 total_compressed = (total_compressed + blocksize - 1) &
 443                         ~(blocksize - 1);
 444
 445                 /*
 446                  * one last check to make sure the compression is really a
 447                  * win, compare the page count read with the blocks on disk
 448                  */
 449                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 450                         ~(PAGE_CACHE_SIZE - 1);
 451                 if (total_compressed >= total_in) {
 452                         will_compress = 0;
 453                 } else {
 454                         disk_num_bytes = total_compressed;
 455                         num_bytes = total_in;
 456                 }
 457         }
 458         if (!will_compress && pages) {
 459                 /*
 460                  * the compression code ran but failed to make things smaller,
 461                  * free any pages it allocated and our page pointer array
 462                  */
 463                 for (i = 0; i < nr_pages_ret; i++) {
 464                         WARN_ON(pages[i]->mapping);
 465                         page_cache_release(pages[i]);
 466                 }
 467                 kfree(pages);
 468                 pages = NULL;
 469                 total_compressed = 0;
 470                 nr_pages_ret = 0;
 471
 472                 /* flag the file so we don't compress in the future */
 473                 btrfs_set_flag(inode, NOCOMPRESS);
 474         }
 475         if (will_compress) {
 476                 *num_added += 1;
 477
 478                 /* the async work queues will take care of doing actual
 479                  * allocation on disk for these compressed pages,
 480                  * and will submit them to the elevator.
 481                  */
 482                 add_async_extent(async_cow, start, num_bytes,
 483                                  total_compressed, pages, nr_pages_ret);
 484
 485                 if (start + num_bytes < end && start + num_bytes < actual_end) {
 486                         start += num_bytes;
 487                         pages = NULL;
 488                         cond_resched();
 489                         goto again;
 490                 }
 491         } else {
 492 cleanup_and_bail_uncompressed:
 493                 /*
 494                  * No compression, but we still need to write the pages in
 495                  * the file we've been given so far.  redirty the locked
 496                  * page if it corresponds to our extent and set things up
 497                  * for the async work queue to run cow_file_range to do
 498                  * the normal delalloc dance
 499                  */
 500                 if (page_offset(locked_page) >= start &&
 501                     page_offset(locked_page) <= end) {
 502                         __set_page_dirty_nobuffers(locked_page);
 503                         /* unlocked later on in the async handlers */
 504                 }
 505                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 506                 *num_added += 1;
 507         }
 508
 509 out:
 510         return 0;
 511
 512 free_pages_out:
 513         for (i = 0; i < nr_pages_ret; i++) {
 514                 WARN_ON(pages[i]->mapping);
 515                 page_cache_release(pages[i]);
 516         }
 517         kfree(pages);
 518
 519         goto out;
 520 }
 521
 522 /*
 523  * phase two of compressed writeback.  This is the ordered portion
 524  * of the code, which only gets called in the order the work was
 525  * queued.  We walk all the async extents created by compress_file_range
 526  * and send them down to the disk.
 527  */
 528 static noinline int submit_compressed_extents(struct inode *inode,
 529                                               struct async_cow *async_cow)
 530 {
 531         struct async_extent *async_extent;
 532         u64 alloc_hint = 0;
 533         struct btrfs_trans_handle *trans;
 534         struct btrfs_key ins;
 535         struct extent_map *em;
 536         struct btrfs_root *root = BTRFS_I(inode)->root;
 537         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 538         struct extent_io_tree *io_tree;
 539         int ret;
 540
 541         if (list_empty(&async_cow->extents))
 542                 return 0;
 543
 544         trans = btrfs_join_transaction(root, 1);
 545
 546         while (!list_empty(&async_cow->extents)) {
 547                 async_extent = list_entry(async_cow->extents.next,
 548                                           struct async_extent, list);
 549                 list_del(&async_extent->list);
 550
 551                 io_tree = &BTRFS_I(inode)->io_tree;
 552
 553                 /* did the compression code fall back to uncompressed IO? */
 554                 if (!async_extent->pages) {
 555                         int page_started = 0;
 556                         unsigned long nr_written = 0;
 557
 558                         lock_extent(io_tree, async_extent->start,
 559                                     async_extent->start +
 560                                     async_extent->ram_size - 1, GFP_NOFS);
 561
 562                         /* allocate blocks */
 563                         cow_file_range(inode, async_cow->locked_page,
 564                                        async_extent->start,
 565                                        async_extent->start +
 566                                        async_extent->ram_size - 1,
 567                                        &page_started, &nr_written, 0);
 568
 569                         /*
 570                          * if page_started, cow_file_range inserted an
 571                          * inline extent and took care of all the unlocking
 572                          * and IO for us.  Otherwise, we need to submit
 573                          * all those pages down to the drive.
 574                          */
 575                         if (!page_started)
 576                                 extent_write_locked_range(io_tree,
 577                                                   inode, async_extent->start,
 578                                                   async_extent->start +
 579                                                   async_extent->ram_size - 1,
 580                                                   btrfs_get_extent,
 581                                                   WB_SYNC_ALL);
 582                         kfree(async_extent);
 583                         cond_resched();
 584                         continue;
 585                 }
 586
 587                 lock_extent(io_tree, async_extent->start,
 588                             async_extent->start + async_extent->ram_size - 1,
 589                             GFP_NOFS);
 590                 /*
 591                  * here we're doing allocation and writeback of the
 592                  * compressed pages
 593                  */
 594                 btrfs_drop_extent_cache(inode, async_extent->start,
 595                                         async_extent->start +
 596                                         async_extent->ram_size - 1, 0);
 597
 598                 ret = btrfs_reserve_extent(trans, root,
 599                                            async_extent->compressed_size,
 600                                            async_extent->compressed_size,
 601                                            0, alloc_hint,
 602                                            (u64)-1, &ins, 1);
 603                 BUG_ON(ret);
 604                 em = alloc_extent_map(GFP_NOFS);
 605                 em->start = async_extent->start;
 606                 em->len = async_extent->ram_size;
 607                 em->orig_start = em->start;
 608
 609                 em->block_start = ins.objectid;
 610                 em->block_len = ins.offset;
 611                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 612                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 613                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 614
 615                 while (1) {
 616                         spin_lock(&em_tree->lock);
 617                         ret = add_extent_mapping(em_tree, em);
 618                         spin_unlock(&em_tree->lock);
 619                         if (ret != -EEXIST) {
 620                                 free_extent_map(em);
 621                                 break;
 622                         }
 623                         btrfs_drop_extent_cache(inode, async_extent->start,
 624                                                 async_extent->start +
 625                                                 async_extent->ram_size - 1, 0);
 626                 }
 627
 628                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 629                                                ins.objectid,
 630                                                async_extent->ram_size,
 631                                                ins.offset,
 632                                                BTRFS_ORDERED_COMPRESSED);
 633                 BUG_ON(ret);
 634
 635                 btrfs_end_transaction(trans, root);
 636
 637                 /*
 638                  * clear dirty, set writeback and unlock the pages.
 639                  */
 640                 extent_clear_unlock_delalloc(inode,
 641                                              &BTRFS_I(inode)->io_tree,
 642                                              async_extent->start,
 643                                              async_extent->start +
 644                                              async_extent->ram_size - 1,
 645                                              NULL, 1, 1, 0, 1, 1, 0);
 646
 647                 ret = btrfs_submit_compressed_write(inode,
 648                                     async_extent->start,
 649                                     async_extent->ram_size,
 650                                     ins.objectid,
 651                                     ins.offset, async_extent->pages,
 652                                     async_extent->nr_pages);
 653
 654                 BUG_ON(ret);
 655                 trans = btrfs_join_transaction(root, 1);
 656                 alloc_hint = ins.objectid + ins.offset;
 657                 kfree(async_extent);
 658                 cond_resched();
 659         }
 660
 661         btrfs_end_transaction(trans, root);
 662         return 0;
 663 }
 664
 665 /*
 666  * when extent_io.c finds a delayed allocation range in the file,
 667  * the call backs end up in this code.  The basic idea is to
 668  * allocate extents on disk for the range, and create ordered data structs
 669  * in ram to track those extents.
 670  *
 671  * locked_page is the page that writepage had locked already.  We use
 672  * it to make sure we don't do extra locks or unlocks.
 673  *
 674  * *page_started is set to one if we unlock locked_page and do everything
 675  * required to start IO on it.  It may be clean and already done with
 676  * IO when we return.
 677  */
 678 static noinline int cow_file_range(struct inode *inode,
 679                                    struct page *locked_page,
 680                                    u64 start, u64 end, int *page_started,
 681                                    unsigned long *nr_written,
 682                                    int unlock)
 683 {
 684         struct btrfs_root *root = BTRFS_I(inode)->root;
 685         struct btrfs_trans_handle *trans;
 686         u64 alloc_hint = 0;
 687         u64 num_bytes;
 688         unsigned long ram_size;
 689         u64 disk_num_bytes;
 690         u64 cur_alloc_size;
 691         u64 blocksize = root->sectorsize;
 692         u64 actual_end;
 693         u64 isize = i_size_read(inode);
 694         struct btrfs_key ins;
 695         struct extent_map *em;
 696         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 697         int ret = 0;
 698
 699         trans = btrfs_join_transaction(root, 1);
 700         BUG_ON(!trans);
 701         btrfs_set_trans_block_group(trans, inode);
 702
 703         actual_end = min_t(u64, isize, end + 1);
 704
 705         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 706         num_bytes = max(blocksize,  num_bytes);
 707         disk_num_bytes = num_bytes;
 708         ret = 0;
 709
 710         if (start == 0) {
 711                 /* lets try to make an inline extent */
 712                 ret = cow_file_range_inline(trans, root, inode,
 713                                             start, end, 0, NULL);
 714                 if (ret == 0) {
 715                         extent_clear_unlock_delalloc(inode,
 716                                                      &BTRFS_I(inode)->io_tree,
 717                                                      start, end, NULL, 1, 1,
 718                                                      1, 1, 1, 1);
 719                         *nr_written = *nr_written +
 720                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 721                         *page_started = 1;
 722                         ret = 0;
 723                         goto out;
 724                 }
 725         }
 726
 727         BUG_ON(disk_num_bytes >
 728                btrfs_super_total_bytes(&root->fs_info->super_copy));
 729
 730         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 731
 732         while (disk_num_bytes > 0) {
 733                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 734                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 735                                            root->sectorsize, 0, alloc_hint,
 736                                            (u64)-1, &ins, 1);
 737                 BUG_ON(ret);
 738
 739                 em = alloc_extent_map(GFP_NOFS);
 740                 em->start = start;
 741                 em->orig_start = em->start;
 742
 743                 ram_size = ins.offset;
 744                 em->len = ins.offset;
 745
 746                 em->block_start = ins.objectid;
 747                 em->block_len = ins.offset;
 748                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 749                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 750
 751                 while (1) {
 752                         spin_lock(&em_tree->lock);
 753                         ret = add_extent_mapping(em_tree, em);
 754                         spin_unlock(&em_tree->lock);
 755                         if (ret != -EEXIST) {
 756                                 free_extent_map(em);
 757                                 break;
 758                         }
 759                         btrfs_drop_extent_cache(inode, start,
 760                                                 start + ram_size - 1, 0);
 761                 }
 762
 763                 cur_alloc_size = ins.offset;
 764                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 765                                                ram_size, cur_alloc_size, 0);
 766                 BUG_ON(ret);
 767
 768                 if (root->root_key.objectid ==
 769                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 770                         ret = btrfs_reloc_clone_csums(inode, start,
 771                                                       cur_alloc_size);
 772                         BUG_ON(ret);
 773                 }
 774
 775                 if (disk_num_bytes < cur_alloc_size)
 776                         break;
 777
 778                 /* we're not doing compressed IO, don't unlock the first
 779                  * page (which the caller expects to stay locked), don't
 780                  * clear any dirty bits and don't set any writeback bits
 781                  */
 782                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 783                                              start, start + ram_size - 1,
 784                                              locked_page, unlock, 1,
 785                                              1, 0, 0, 0);
 786                 disk_num_bytes -= cur_alloc_size;
 787                 num_bytes -= cur_alloc_size;
 788                 alloc_hint = ins.objectid + ins.offset;
 789                 start += cur_alloc_size;
 790         }
 791 out:
 792         ret = 0;
 793         btrfs_end_transaction(trans, root);
 794
 795         return ret;
 796 }
 797
 798 /*
 799  * work queue call back to started compression on a file and pages
 800  */
 801 static noinline void async_cow_start(struct btrfs_work *work)
 802 {
 803         struct async_cow *async_cow;
 804         int num_added = 0;
 805         async_cow = container_of(work, struct async_cow, work);
 806
 807         compress_file_range(async_cow->inode, async_cow->locked_page,
 808                             async_cow->start, async_cow->end, async_cow,
 809                             &num_added);
 810         if (num_added == 0)
 811                 async_cow->inode = NULL;
 812 }
 813
 814 /*
 815  * work queue call back to submit previously compressed pages
 816  */
 817 static noinline void async_cow_submit(struct btrfs_work *work)
 818 {
 819         struct async_cow *async_cow;
 820         struct btrfs_root *root;
 821         unsigned long nr_pages;
 822
 823         async_cow = container_of(work, struct async_cow, work);
 824
 825         root = async_cow->root;
 826         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 827                 PAGE_CACHE_SHIFT;
 828
 829         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 830
 831         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 832             5 * 1042 * 1024 &&
 833             waitqueue_active(&root->fs_info->async_submit_wait))
 834                 wake_up(&root->fs_info->async_submit_wait);
 835
 836         if (async_cow->inode)
 837                 submit_compressed_extents(async_cow->inode, async_cow);
 838 }
 839
 840 static noinline void async_cow_free(struct btrfs_work *work)
 841 {
 842         struct async_cow *async_cow;
 843         async_cow = container_of(work, struct async_cow, work);
 844         kfree(async_cow);
 845 }
 846
 847 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 848                                 u64 start, u64 end, int *page_started,
 849                                 unsigned long *nr_written)
 850 {
 851         struct async_cow *async_cow;
 852         struct btrfs_root *root = BTRFS_I(inode)->root;
 853         unsigned long nr_pages;
 854         u64 cur_end;
 855         int limit = 10 * 1024 * 1042;
 856
 857         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 858                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 859         while (start < end) {
 860                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 861                 async_cow->inode = inode;
 862                 async_cow->root = root;
 863                 async_cow->locked_page = locked_page;
 864                 async_cow->start = start;
 865
 866                 if (btrfs_test_flag(inode, NOCOMPRESS))
 867                         cur_end = end;
 868                 else
 869                         cur_end = min(end, start + 512 * 1024 - 1);
 870
 871                 async_cow->end = cur_end;
 872                 INIT_LIST_HEAD(&async_cow->extents);
 873
 874                 async_cow->work.func = async_cow_start;
 875                 async_cow->work.ordered_func = async_cow_submit;
 876                 async_cow->work.ordered_free = async_cow_free;
 877                 async_cow->work.flags = 0;
 878
 879                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 880                         PAGE_CACHE_SHIFT;
 881                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 882
 883                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 884                                    &async_cow->work);
 885
 886                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 887                         wait_event(root->fs_info->async_submit_wait,
 888                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 889                             limit));
 890                 }
 891
 892                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 893                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 894                         wait_event(root->fs_info->async_submit_wait,
 895                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 896                            0));
 897                 }
 898
 899                 *nr_written += nr_pages;
 900                 start = cur_end + 1;
 901         }
 902         *page_started = 1;
 903         return 0;
 904 }
 905
 906 static noinline int csum_exist_in_range(struct btrfs_root *root,
 907                                         u64 bytenr, u64 num_bytes)
 908 {
 909         int ret;
 910         struct btrfs_ordered_sum *sums;
 911         LIST_HEAD(list);
 912
 913         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 914                                        bytenr + num_bytes - 1, &list);
 915         if (ret == 0 && list_empty(&list))
 916                 return 0;
 917
 918         while (!list_empty(&list)) {
 919                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 920                 list_del(&sums->list);
 921                 kfree(sums);
 922         }
 923         return 1;
 924 }
 925
 926 /*
 927  * when nowcow writeback call back.  This checks for snapshots or COW copies
 928  * of the extents that exist in the file, and COWs the file as required.
 929  *
 930  * If no cow copies or snapshots exist, we write directly to the existing
 931  * blocks on disk
 932  */
 933 static noinline int run_delalloc_nocow(struct inode *inode,
 934                                        struct page *locked_page,
 935                               u64 start, u64 end, int *page_started, int force,
 936                               unsigned long *nr_written)
 937 {
 938         struct btrfs_root *root = BTRFS_I(inode)->root;
 939         struct btrfs_trans_handle *trans;
 940         struct extent_buffer *leaf;
 941         struct btrfs_path *path;
 942         struct btrfs_file_extent_item *fi;
 943         struct btrfs_key found_key;
 944         u64 cow_start;
 945         u64 cur_offset;
 946         u64 extent_end;
 947         u64 disk_bytenr;
 948         u64 num_bytes;
 949         int extent_type;
 950         int ret;
 951         int type;
 952         int nocow;
 953         int check_prev = 1;
 954
 955         path = btrfs_alloc_path();
 956         BUG_ON(!path);
 957         trans = btrfs_join_transaction(root, 1);
 958         BUG_ON(!trans);
 959
 960         cow_start = (u64)-1;
 961         cur_offset = start;
 962         while (1) {
 963                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 964                                                cur_offset, 0);
 965                 BUG_ON(ret < 0);
 966                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 967                         leaf = path->nodes[0];
 968                         btrfs_item_key_to_cpu(leaf, &found_key,
 969                                               path->slots[0] - 1);
 970                         if (found_key.objectid == inode->i_ino &&
 971                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 972                                 path->slots[0]--;
 973                 }
 974                 check_prev = 0;
 975 next_slot:
 976                 leaf = path->nodes[0];
 977                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 978                         ret = btrfs_next_leaf(root, path);
 979                         if (ret < 0)
 980                                 BUG_ON(1);
 981                         if (ret > 0)
 982                                 break;
 983                         leaf = path->nodes[0];
 984                 }
 985
 986                 nocow = 0;
 987                 disk_bytenr = 0;
 988                 num_bytes = 0;
 989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 990
 991                 if (found_key.objectid > inode->i_ino ||
 992                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
 993                     found_key.offset > end)
 994                         break;
 995
 996                 if (found_key.offset > cur_offset) {
 997                         extent_end = found_key.offset;
 998                         goto out_check;
 999                 }
1000
1001                 fi = btrfs_item_ptr(leaf, path->slots[0],
1002                                     struct btrfs_file_extent_item);
1003                 extent_type = btrfs_file_extent_type(leaf, fi);
1004
1005                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1006                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1007                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1008                         extent_end = found_key.offset +
1009                                 btrfs_file_extent_num_bytes(leaf, fi);
1010                         if (extent_end <= start) {
1011                                 path->slots[0]++;
1012                                 goto next_slot;
1013                         }
1014                         if (disk_bytenr == 0)
1015                                 goto out_check;
1016                         if (btrfs_file_extent_compression(leaf, fi) ||
1017                             btrfs_file_extent_encryption(leaf, fi) ||
1018                             btrfs_file_extent_other_encoding(leaf, fi))
1019                                 goto out_check;
1020                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1021                                 goto out_check;
1022                         if (btrfs_extent_readonly(root, disk_bytenr))
1023                                 goto out_check;
1024                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1025                                                   disk_bytenr))
1026                                 goto out_check;
1027                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1028                         disk_bytenr += cur_offset - found_key.offset;
1029                         num_bytes = min(end + 1, extent_end) - cur_offset;
1030                         /*
1031                          * force cow if csum exists in the range.
1032                          * this ensure that csum for a given extent are
1033                          * either valid or do not exist.
1034                          */
1035                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1036                                 goto out_check;
1037                         nocow = 1;
1038                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1039                         extent_end = found_key.offset +
1040                                 btrfs_file_extent_inline_len(leaf, fi);
1041                         extent_end = ALIGN(extent_end, root->sectorsize);
1042                 } else {
1043                         BUG_ON(1);
1044                 }
1045 out_check:
1046                 if (extent_end <= start) {
1047                         path->slots[0]++;
1048                         goto next_slot;
1049                 }
1050                 if (!nocow) {
1051                         if (cow_start == (u64)-1)
1052                                 cow_start = cur_offset;
1053                         cur_offset = extent_end;
1054                         if (cur_offset > end)
1055                                 break;
1056                         path->slots[0]++;
1057                         goto next_slot;
1058                 }
1059
1060                 btrfs_release_path(root, path);
1061                 if (cow_start != (u64)-1) {
1062                         ret = cow_file_range(inode, locked_page, cow_start,
1063                                         found_key.offset - 1, page_started,
1064                                         nr_written, 1);
1065                         BUG_ON(ret);
1066                         cow_start = (u64)-1;
1067                 }
1068
1069                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1070                         struct extent_map *em;
1071                         struct extent_map_tree *em_tree;
1072                         em_tree = &BTRFS_I(inode)->extent_tree;
1073                         em = alloc_extent_map(GFP_NOFS);
1074                         em->start = cur_offset;
1075                         em->orig_start = em->start;
1076                         em->len = num_bytes;
1077                         em->block_len = num_bytes;
1078                         em->block_start = disk_bytenr;
1079                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1080                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1081                         while (1) {
1082                                 spin_lock(&em_tree->lock);
1083                                 ret = add_extent_mapping(em_tree, em);
1084                                 spin_unlock(&em_tree->lock);
1085                                 if (ret != -EEXIST) {
1086                                         free_extent_map(em);
1087                                         break;
1088                                 }
1089                                 btrfs_drop_extent_cache(inode, em->start,
1090                                                 em->start + em->len - 1, 0);
1091                         }
1092                         type = BTRFS_ORDERED_PREALLOC;
1093                 } else {
1094                         type = BTRFS_ORDERED_NOCOW;
1095                 }
1096
1097                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1098                                                num_bytes, num_bytes, type);
1099                 BUG_ON(ret);
1100
1101                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1102                                         cur_offset, cur_offset + num_bytes - 1,
1103                                         locked_page, 1, 1, 1, 0, 0, 0);
1104                 cur_offset = extent_end;
1105                 if (cur_offset > end)
1106                         break;
1107         }
1108         btrfs_release_path(root, path);
1109
1110         if (cur_offset <= end && cow_start == (u64)-1)
1111                 cow_start = cur_offset;
1112         if (cow_start != (u64)-1) {
1113                 ret = cow_file_range(inode, locked_page, cow_start, end,
1114                                      page_started, nr_written, 1);
1115                 BUG_ON(ret);
1116         }
1117
1118         ret = btrfs_end_transaction(trans, root);
1119         BUG_ON(ret);
1120         btrfs_free_path(path);
1121         return 0;
1122 }
1123
1124 /*
1125  * extent_io.c call back to do delayed allocation processing
1126  */
1127 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1128                               u64 start, u64 end, int *page_started,
1129                               unsigned long *nr_written)
1130 {
1131         int ret;
1132         struct btrfs_root *root = BTRFS_I(inode)->root;
1133
1134         if (btrfs_test_flag(inode, NODATACOW))
1135                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1136                                          page_started, 1, nr_written);
1137         else if (btrfs_test_flag(inode, PREALLOC))
1138                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1139                                          page_started, 0, nr_written);
1140         else if (!btrfs_test_opt(root, COMPRESS))
1141                 ret = cow_file_range(inode, locked_page, start, end,
1142                                       page_started, nr_written, 1);
1143         else
1144                 ret = cow_file_range_async(inode, locked_page, start, end,
1145                                            page_started, nr_written);
1146         return ret;
1147 }
1148
1149 /*
1150  * extent_io.c set_bit_hook, used to track delayed allocation
1151  * bytes in this file, and to maintain the list of inodes that
1152  * have pending delalloc work to be done.
1153  */
1154 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1155                        unsigned long old, unsigned long bits)
1156 {
1157         /*
1158          * set_bit and clear bit hooks normally require _irqsave/restore
1159          * but in this case, we are only testeing for the DELALLOC
1160          * bit, which is only set or cleared with irqs on
1161          */
1162         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1163                 struct btrfs_root *root = BTRFS_I(inode)->root;
1164                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1165                 spin_lock(&root->fs_info->delalloc_lock);
1166                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1167                 root->fs_info->delalloc_bytes += end - start + 1;
1168                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1169                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1170                                       &root->fs_info->delalloc_inodes);
1171                 }
1172                 spin_unlock(&root->fs_info->delalloc_lock);
1173         }
1174         return 0;
1175 }
1176
1177 /*
1178  * extent_io.c clear_bit_hook, see set_bit_hook for why
1179  */
1180 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1181                          unsigned long old, unsigned long bits)
1182 {
1183         /*
1184          * set_bit and clear bit hooks normally require _irqsave/restore
1185          * but in this case, we are only testeing for the DELALLOC
1186          * bit, which is only set or cleared with irqs on
1187          */
1188         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1189                 struct btrfs_root *root = BTRFS_I(inode)->root;
1190
1191                 spin_lock(&root->fs_info->delalloc_lock);
1192                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1193                         printk(KERN_INFO "btrfs warning: delalloc account "
1194                                "%llu %llu\n",
1195                                (unsigned long long)end - start + 1,
1196                                (unsigned long long)
1197                                root->fs_info->delalloc_bytes);
1198                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1199                         root->fs_info->delalloc_bytes = 0;
1200                         BTRFS_I(inode)->delalloc_bytes = 0;
1201                 } else {
1202                         btrfs_delalloc_free_space(root, inode,
1203                                                   end - start + 1);
1204                         root->fs_info->delalloc_bytes -= end - start + 1;
1205                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1206                 }
1207                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1208                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1209                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1210                 }
1211                 spin_unlock(&root->fs_info->delalloc_lock);
1212         }
1213         return 0;
1214 }
1215
1216 /*
1217  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1218  * we don't create bios that span stripes or chunks
1219  */
1220 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1221                          size_t size, struct bio *bio,
1222                          unsigned long bio_flags)
1223 {
1224         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1225         struct btrfs_mapping_tree *map_tree;
1226         u64 logical = (u64)bio->bi_sector << 9;
1227         u64 length = 0;
1228         u64 map_length;
1229         int ret;
1230
1231         if (bio_flags & EXTENT_BIO_COMPRESSED)
1232                 return 0;
1233
1234         length = bio->bi_size;
1235         map_tree = &root->fs_info->mapping_tree;
1236         map_length = length;
1237         ret = btrfs_map_block(map_tree, READ, logical,
1238                               &map_length, NULL, 0);
1239
1240         if (map_length < length + size)
1241                 return 1;
1242         return 0;
1243 }
1244
1245 /*
1246  * in order to insert checksums into the metadata in large chunks,
1247  * we wait until bio submission time.   All the pages in the bio are
1248  * checksummed and sums are attached onto the ordered extent record.
1249  *
1250  * At IO completion time the cums attached on the ordered extent record
1251  * are inserted into the btree
1252  */
1253 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1254                                     struct bio *bio, int mirror_num,
1255                                     unsigned long bio_flags)
1256 {
1257         struct btrfs_root *root = BTRFS_I(inode)->root;
1258         int ret = 0;
1259
1260         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1261         BUG_ON(ret);
1262         return 0;
1263 }
1264
1265 /*
1266  * in order to insert checksums into the metadata in large chunks,
1267  * we wait until bio submission time.   All the pages in the bio are
1268  * checksummed and sums are attached onto the ordered extent record.
1269  *
1270  * At IO completion time the cums attached on the ordered extent record
1271  * are inserted into the btree
1272  */
1273 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1274                           int mirror_num, unsigned long bio_flags)
1275 {
1276         struct btrfs_root *root = BTRFS_I(inode)->root;
1277         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1278 }
1279
1280 /*
1281  * extent_io.c submission hook. This does the right thing for csum calculation
1282  * on write, or reading the csums from the tree before a read
1283  */
1284 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1285                           int mirror_num, unsigned long bio_flags)
1286 {
1287         struct btrfs_root *root = BTRFS_I(inode)->root;
1288         int ret = 0;
1289         int skip_sum;
1290
1291         skip_sum = btrfs_test_flag(inode, NODATASUM);
1292
1293         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1294         BUG_ON(ret);
1295
1296         if (!(rw & (1 << BIO_RW))) {
1297                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1298                         return btrfs_submit_compressed_read(inode, bio,
1299                                                     mirror_num, bio_flags);
1300                 } else if (!skip_sum)
1301                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1302                 goto mapit;
1303         } else if (!skip_sum) {
1304                 /* csum items have already been cloned */
1305                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1306                         goto mapit;
1307                 /* we're doing a write, do the async checksumming */
1308                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1309                                    inode, rw, bio, mirror_num,
1310                                    bio_flags, __btrfs_submit_bio_start,
1311                                    __btrfs_submit_bio_done);
1312         }
1313
1314 mapit:
1315         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1316 }
1317
1318 /*
1319  * given a list of ordered sums record them in the inode.  This happens
1320  * at IO completion time based on sums calculated at bio submission time.
1321  */
1322 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1323                              struct inode *inode, u64 file_offset,
1324                              struct list_head *list)
1325 {
1326         struct btrfs_ordered_sum *sum;
1327
1328         btrfs_set_trans_block_group(trans, inode);
1329
1330         list_for_each_entry(sum, list, list) {
1331                 btrfs_csum_file_blocks(trans,
1332                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1333         }
1334         return 0;
1335 }
1336
1337 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1338 {
1339         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1340                 WARN_ON(1);
1341         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1342                                    GFP_NOFS);
1343 }
1344
1345 /* see btrfs_writepage_start_hook for details on why this is required */
1346 struct btrfs_writepage_fixup {
1347         struct page *page;
1348         struct btrfs_work work;
1349 };
1350
1351 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1352 {
1353         struct btrfs_writepage_fixup *fixup;
1354         struct btrfs_ordered_extent *ordered;
1355         struct page *page;
1356         struct inode *inode;
1357         u64 page_start;
1358         u64 page_end;
1359
1360         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1361         page = fixup->page;
1362 again:
1363         lock_page(page);
1364         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1365                 ClearPageChecked(page);
1366                 goto out_page;
1367         }
1368
1369         inode = page->mapping->host;
1370         page_start = page_offset(page);
1371         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1372
1373         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1374
1375         /* already ordered? We're done */
1376         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1377                              EXTENT_ORDERED, 0)) {
1378                 goto out;
1379         }
1380
1381         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1382         if (ordered) {
1383                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1384                               page_end, GFP_NOFS);
1385                 unlock_page(page);
1386                 btrfs_start_ordered_extent(inode, ordered, 1);
1387                 goto again;
1388         }
1389
1390         btrfs_set_extent_delalloc(inode, page_start, page_end);
1391         ClearPageChecked(page);
1392 out:
1393         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1394 out_page:
1395         unlock_page(page);
1396         page_cache_release(page);
1397 }
1398
1399 /*
1400  * There are a few paths in the higher layers of the kernel that directly
1401  * set the page dirty bit without asking the filesystem if it is a
1402  * good idea.  This causes problems because we want to make sure COW
1403  * properly happens and the data=ordered rules are followed.
1404  *
1405  * In our case any range that doesn't have the ORDERED bit set
1406  * hasn't been properly setup for IO.  We kick off an async process
1407  * to fix it up.  The async helper will wait for ordered extents, set
1408  * the delalloc bit and make it safe to write the page.
1409  */
1410 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1411 {
1412         struct inode *inode = page->mapping->host;
1413         struct btrfs_writepage_fixup *fixup;
1414         struct btrfs_root *root = BTRFS_I(inode)->root;
1415         int ret;
1416
1417         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1418                              EXTENT_ORDERED, 0);
1419         if (ret)
1420                 return 0;
1421
1422         if (PageChecked(page))
1423                 return -EAGAIN;
1424
1425         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1426         if (!fixup)
1427                 return -EAGAIN;
1428
1429         SetPageChecked(page);
1430         page_cache_get(page);
1431         fixup->work.func = btrfs_writepage_fixup_worker;
1432         fixup->page = page;
1433         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1434         return -EAGAIN;
1435 }
1436
1437 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1438                                        struct inode *inode, u64 file_pos,
1439                                        u64 disk_bytenr, u64 disk_num_bytes,
1440                                        u64 num_bytes, u64 ram_bytes,
1441                                        u64 locked_end,
1442                                        u8 compression, u8 encryption,
1443                                        u16 other_encoding, int extent_type)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         struct btrfs_file_extent_item *fi;
1447         struct btrfs_path *path;
1448         struct extent_buffer *leaf;
1449         struct btrfs_key ins;
1450         u64 hint;
1451         int ret;
1452
1453         path = btrfs_alloc_path();
1454         BUG_ON(!path);
1455
1456         path->leave_spinning = 1;
1457         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458                                  file_pos + num_bytes, locked_end,
1459                                  file_pos, &hint);
1460         BUG_ON(ret);
1461
1462         ins.objectid = inode->i_ino;
1463         ins.offset = file_pos;
1464         ins.type = BTRFS_EXTENT_DATA_KEY;
1465         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1466         BUG_ON(ret);
1467         leaf = path->nodes[0];
1468         fi = btrfs_item_ptr(leaf, path->slots[0],
1469                             struct btrfs_file_extent_item);
1470         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1471         btrfs_set_file_extent_type(leaf, fi, extent_type);
1472         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1473         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1474         btrfs_set_file_extent_offset(leaf, fi, 0);
1475         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1476         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1477         btrfs_set_file_extent_compression(leaf, fi, compression);
1478         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1479         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1480
1481         btrfs_unlock_up_safe(path, 1);
1482         btrfs_set_lock_blocking(leaf);
1483
1484         btrfs_mark_buffer_dirty(leaf);
1485
1486         inode_add_bytes(inode, num_bytes);
1487         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1488
1489         ins.objectid = disk_bytenr;
1490         ins.offset = disk_num_bytes;
1491         ins.type = BTRFS_EXTENT_ITEM_KEY;
1492         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1493                                           root->root_key.objectid,
1494                                           trans->transid, inode->i_ino, &ins);
1495         BUG_ON(ret);
1496         btrfs_free_path(path);
1497
1498         return 0;
1499 }
1500
1501 /*
1502  * helper function for btrfs_finish_ordered_io, this
1503  * just reads in some of the csum leaves to prime them into ram
1504  * before we start the transaction.  It limits the amount of btree
1505  * reads required while inside the transaction.
1506  */
1507 static noinline void reada_csum(struct btrfs_root *root,
1508                                 struct btrfs_path *path,
1509                                 struct btrfs_ordered_extent *ordered_extent)
1510 {
1511         struct btrfs_ordered_sum *sum;
1512         u64 bytenr;
1513
1514         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1515                          list);
1516         bytenr = sum->sums[0].bytenr;
1517
1518         /*
1519          * we don't care about the results, the point of this search is
1520          * just to get the btree leaves into ram
1521          */
1522         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1523 }
1524
1525 /* as ordered data IO finishes, this gets called so we can finish
1526  * an ordered extent if the range of bytes in the file it covers are
1527  * fully written.
1528  */
1529 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1530 {
1531         struct btrfs_root *root = BTRFS_I(inode)->root;
1532         struct btrfs_trans_handle *trans;
1533         struct btrfs_ordered_extent *ordered_extent = NULL;
1534         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1535         struct btrfs_path *path;
1536         int compressed = 0;
1537         int ret;
1538
1539         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1540         if (!ret)
1541                 return 0;
1542
1543         /*
1544          * before we join the transaction, try to do some of our IO.
1545          * This will limit the amount of IO that we have to do with
1546          * the transaction running.  We're unlikely to need to do any
1547          * IO if the file extents are new, the disk_i_size checks
1548          * covers the most common case.
1549          */
1550         if (start < BTRFS_I(inode)->disk_i_size) {
1551                 path = btrfs_alloc_path();
1552                 if (path) {
1553                         ret = btrfs_lookup_file_extent(NULL, root, path,
1554                                                        inode->i_ino,
1555                                                        start, 0);
1556                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1557                                                                      start);
1558                         if (!list_empty(&ordered_extent->list)) {
1559                                 btrfs_release_path(root, path);
1560                                 reada_csum(root, path, ordered_extent);
1561                         }
1562                         btrfs_free_path(path);
1563                 }
1564         }
1565
1566         trans = btrfs_join_transaction(root, 1);
1567
1568         if (!ordered_extent)
1569                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1570         BUG_ON(!ordered_extent);
1571         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1572                 goto nocow;
1573
1574         lock_extent(io_tree, ordered_extent->file_offset,
1575                     ordered_extent->file_offset + ordered_extent->len - 1,
1576                     GFP_NOFS);
1577
1578         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1579                 compressed = 1;
1580         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1581                 BUG_ON(compressed);
1582                 ret = btrfs_mark_extent_written(trans, root, inode,
1583                                                 ordered_extent->file_offset,
1584                                                 ordered_extent->file_offset +
1585                                                 ordered_extent->len);
1586                 BUG_ON(ret);
1587         } else {
1588                 ret = insert_reserved_file_extent(trans, inode,
1589                                                 ordered_extent->file_offset,
1590                                                 ordered_extent->start,
1591                                                 ordered_extent->disk_len,
1592                                                 ordered_extent->len,
1593                                                 ordered_extent->len,
1594                                                 ordered_extent->file_offset +
1595                                                 ordered_extent->len,
1596                                                 compressed, 0, 0,
1597                                                 BTRFS_FILE_EXTENT_REG);
1598                 BUG_ON(ret);
1599         }
1600         unlock_extent(io_tree, ordered_extent->file_offset,
1601                     ordered_extent->file_offset + ordered_extent->len - 1,
1602                     GFP_NOFS);
1603 nocow:
1604         add_pending_csums(trans, inode, ordered_extent->file_offset,
1605                           &ordered_extent->list);
1606
1607         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1608         btrfs_ordered_update_i_size(inode, ordered_extent);
1609         btrfs_update_inode(trans, root, inode);
1610         btrfs_remove_ordered_extent(inode, ordered_extent);
1611         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1612
1613         /* once for us */
1614         btrfs_put_ordered_extent(ordered_extent);
1615         /* once for the tree */
1616         btrfs_put_ordered_extent(ordered_extent);
1617
1618         btrfs_end_transaction(trans, root);
1619         return 0;
1620 }
1621
1622 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1623                                 struct extent_state *state, int uptodate)
1624 {
1625         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1626 }
1627
1628 /*
1629  * When IO fails, either with EIO or csum verification fails, we
1630  * try other mirrors that might have a good copy of the data.  This
1631  * io_failure_record is used to record state as we go through all the
1632  * mirrors.  If another mirror has good data, the page is set up to date
1633  * and things continue.  If a good mirror can't be found, the original
1634  * bio end_io callback is called to indicate things have failed.
1635  */
1636 struct io_failure_record {
1637         struct page *page;
1638         u64 start;
1639         u64 len;
1640         u64 logical;
1641         unsigned long bio_flags;
1642         int last_mirror;
1643 };
1644
1645 static int btrfs_io_failed_hook(struct bio *failed_bio,
1646                          struct page *page, u64 start, u64 end,
1647                          struct extent_state *state)
1648 {
1649         struct io_failure_record *failrec = NULL;
1650         u64 private;
1651         struct extent_map *em;
1652         struct inode *inode = page->mapping->host;
1653         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1654         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1655         struct bio *bio;
1656         int num_copies;
1657         int ret;
1658         int rw;
1659         u64 logical;
1660
1661         ret = get_state_private(failure_tree, start, &private);
1662         if (ret) {
1663                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1664                 if (!failrec)
1665                         return -ENOMEM;
1666                 failrec->start = start;
1667                 failrec->len = end - start + 1;
1668                 failrec->last_mirror = 0;
1669                 failrec->bio_flags = 0;
1670
1671                 spin_lock(&em_tree->lock);
1672                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1673                 if (em->start > start || em->start + em->len < start) {
1674                         free_extent_map(em);
1675                         em = NULL;
1676                 }
1677                 spin_unlock(&em_tree->lock);
1678
1679                 if (!em || IS_ERR(em)) {
1680                         kfree(failrec);
1681                         return -EIO;
1682                 }
1683                 logical = start - em->start;
1684                 logical = em->block_start + logical;
1685                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1686                         logical = em->block_start;
1687                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1688                 }
1689                 failrec->logical = logical;
1690                 free_extent_map(em);
1691                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1692                                 EXTENT_DIRTY, GFP_NOFS);
1693                 set_state_private(failure_tree, start,
1694                                  (u64)(unsigned long)failrec);
1695         } else {
1696                 failrec = (struct io_failure_record *)(unsigned long)private;
1697         }
1698         num_copies = btrfs_num_copies(
1699                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1700                               failrec->logical, failrec->len);
1701         failrec->last_mirror++;
1702         if (!state) {
1703                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1704                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1705                                                     failrec->start,
1706                                                     EXTENT_LOCKED);
1707                 if (state && state->start != failrec->start)
1708                         state = NULL;
1709                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1710         }
1711         if (!state || failrec->last_mirror > num_copies) {
1712                 set_state_private(failure_tree, failrec->start, 0);
1713                 clear_extent_bits(failure_tree, failrec->start,
1714                                   failrec->start + failrec->len - 1,
1715                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1716                 kfree(failrec);
1717                 return -EIO;
1718         }
1719         bio = bio_alloc(GFP_NOFS, 1);
1720         bio->bi_private = state;
1721         bio->bi_end_io = failed_bio->bi_end_io;
1722         bio->bi_sector = failrec->logical >> 9;
1723         bio->bi_bdev = failed_bio->bi_bdev;
1724         bio->bi_size = 0;
1725
1726         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1727         if (failed_bio->bi_rw & (1 << BIO_RW))
1728                 rw = WRITE;
1729         else
1730                 rw = READ;
1731
1732         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1733                                                       failrec->last_mirror,
1734                                                       failrec->bio_flags);
1735         return 0;
1736 }
1737
1738 /*
1739  * each time an IO finishes, we do a fast check in the IO failure tree
1740  * to see if we need to process or clean up an io_failure_record
1741  */
1742 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1743 {
1744         u64 private;
1745         u64 private_failure;
1746         struct io_failure_record *failure;
1747         int ret;
1748
1749         private = 0;
1750         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1751                              (u64)-1, 1, EXTENT_DIRTY)) {
1752                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1753                                         start, &private_failure);
1754                 if (ret == 0) {
1755                         failure = (struct io_failure_record *)(unsigned long)
1756                                    private_failure;
1757                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1758                                           failure->start, 0);
1759                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1760                                           failure->start,
1761                                           failure->start + failure->len - 1,
1762                                           EXTENT_DIRTY | EXTENT_LOCKED,
1763                                           GFP_NOFS);
1764                         kfree(failure);
1765                 }
1766         }
1767         return 0;
1768 }
1769
1770 /*
1771  * when reads are done, we need to check csums to verify the data is correct
1772  * if there's a match, we allow the bio to finish.  If not, we go through
1773  * the io_failure_record routines to find good copies
1774  */
1775 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1776                                struct extent_state *state)
1777 {
1778         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1779         struct inode *inode = page->mapping->host;
1780         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1781         char *kaddr;
1782         u64 private = ~(u32)0;
1783         int ret;
1784         struct btrfs_root *root = BTRFS_I(inode)->root;
1785         u32 csum = ~(u32)0;
1786
1787         if (PageChecked(page)) {
1788                 ClearPageChecked(page);
1789                 goto good;
1790         }
1791         if (btrfs_test_flag(inode, NODATASUM))
1792                 return 0;
1793
1794         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1795             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1796                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1797                                   GFP_NOFS);
1798                 return 0;
1799         }
1800
1801         if (state && state->start == start) {
1802                 private = state->private;
1803                 ret = 0;
1804         } else {
1805                 ret = get_state_private(io_tree, start, &private);
1806         }
1807         kaddr = kmap_atomic(page, KM_USER0);
1808         if (ret)
1809                 goto zeroit;
1810
1811         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1812         btrfs_csum_final(csum, (char *)&csum);
1813         if (csum != private)
1814                 goto zeroit;
1815
1816         kunmap_atomic(kaddr, KM_USER0);
1817 good:
1818         /* if the io failure tree for this inode is non-empty,
1819          * check to see if we've recovered from a failed IO
1820          */
1821         btrfs_clean_io_failures(inode, start);
1822         return 0;
1823
1824 zeroit:
1825         if (printk_ratelimit()) {
1826                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1827                        "private %llu\n", page->mapping->host->i_ino,
1828                        (unsigned long long)start, csum,
1829                        (unsigned long long)private);
1830         }
1831         memset(kaddr + offset, 1, end - start + 1);
1832         flush_dcache_page(page);
1833         kunmap_atomic(kaddr, KM_USER0);
1834         if (private == 0)
1835                 return 0;
1836         return -EIO;
1837 }
1838
1839 /*
1840  * This creates an orphan entry for the given inode in case something goes
1841  * wrong in the middle of an unlink/truncate.
1842  */
1843 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1844 {
1845         struct btrfs_root *root = BTRFS_I(inode)->root;
1846         int ret = 0;
1847
1848         spin_lock(&root->list_lock);
1849
1850         /* already on the orphan list, we're good */
1851         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1852                 spin_unlock(&root->list_lock);
1853                 return 0;
1854         }
1855
1856         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1857
1858         spin_unlock(&root->list_lock);
1859
1860         /*
1861          * insert an orphan item to track this unlinked/truncated file
1862          */
1863         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1864
1865         return ret;
1866 }
1867
1868 /*
1869  * We have done the truncate/delete so we can go ahead and remove the orphan
1870  * item for this particular inode.
1871  */
1872 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1873 {
1874         struct btrfs_root *root = BTRFS_I(inode)->root;
1875         int ret = 0;
1876
1877         spin_lock(&root->list_lock);
1878
1879         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1880                 spin_unlock(&root->list_lock);
1881                 return 0;
1882         }
1883
1884         list_del_init(&BTRFS_I(inode)->i_orphan);
1885         if (!trans) {
1886                 spin_unlock(&root->list_lock);
1887                 return 0;
1888         }
1889
1890         spin_unlock(&root->list_lock);
1891
1892         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1893
1894         return ret;
1895 }
1896
1897 /*
1898  * this cleans up any orphans that may be left on the list from the last use
1899  * of this root.
1900  */
1901 void btrfs_orphan_cleanup(struct btrfs_root *root)
1902 {
1903         struct btrfs_path *path;
1904         struct extent_buffer *leaf;
1905         struct btrfs_item *item;
1906         struct btrfs_key key, found_key;
1907         struct btrfs_trans_handle *trans;
1908         struct inode *inode;
1909         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1910
1911         path = btrfs_alloc_path();
1912         if (!path)
1913                 return;
1914         path->reada = -1;
1915
1916         key.objectid = BTRFS_ORPHAN_OBJECTID;
1917         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1918         key.offset = (u64)-1;
1919
1920
1921         while (1) {
1922                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1923                 if (ret < 0) {
1924                         printk(KERN_ERR "Error searching slot for orphan: %d"
1925                                "\n", ret);
1926                         break;
1927                 }
1928
1929                 /*
1930                  * if ret == 0 means we found what we were searching for, which
1931                  * is weird, but possible, so only screw with path if we didnt
1932                  * find the key and see if we have stuff that matches
1933                  */
1934                 if (ret > 0) {
1935                         if (path->slots[0] == 0)
1936                                 break;
1937                         path->slots[0]--;
1938                 }
1939
1940                 /* pull out the item */
1941                 leaf = path->nodes[0];
1942                 item = btrfs_item_nr(leaf, path->slots[0]);
1943                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1944
1945                 /* make sure the item matches what we want */
1946                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1947                         break;
1948                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1949                         break;
1950
1951                 /* release the path since we're done with it */
1952                 btrfs_release_path(root, path);
1953
1954                 /*
1955                  * this is where we are basically btrfs_lookup, without the
1956                  * crossing root thing.  we store the inode number in the
1957                  * offset of the orphan item.
1958                  */
1959                 inode = btrfs_iget_locked(root->fs_info->sb,
1960                                           found_key.offset, root);
1961                 if (!inode)
1962                         break;
1963
1964                 if (inode->i_state & I_NEW) {
1965                         BTRFS_I(inode)->root = root;
1966
1967                         /* have to set the location manually */
1968                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1969                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1970                         BTRFS_I(inode)->location.offset = 0;
1971
1972                         btrfs_read_locked_inode(inode);
1973                         unlock_new_inode(inode);
1974                 }
1975
1976                 /*
1977                  * add this inode to the orphan list so btrfs_orphan_del does
1978                  * the proper thing when we hit it
1979                  */
1980                 spin_lock(&root->list_lock);
1981                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1982                 spin_unlock(&root->list_lock);
1983
1984                 /*
1985                  * if this is a bad inode, means we actually succeeded in
1986                  * removing the inode, but not the orphan record, which means
1987                  * we need to manually delete the orphan since iput will just
1988                  * do a destroy_inode
1989                  */
1990                 if (is_bad_inode(inode)) {
1991                         trans = btrfs_start_transaction(root, 1);
1992                         btrfs_orphan_del(trans, inode);
1993                         btrfs_end_transaction(trans, root);
1994                         iput(inode);
1995                         continue;
1996                 }
1997
1998                 /* if we have links, this was a truncate, lets do that */
1999                 if (inode->i_nlink) {
2000                         nr_truncate++;
2001                         btrfs_truncate(inode);
2002                 } else {
2003                         nr_unlink++;
2004                 }
2005
2006                 /* this will do delete_inode and everything for us */
2007                 iput(inode);
2008         }
2009
2010         if (nr_unlink)
2011                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2012         if (nr_truncate)
2013                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2014
2015         btrfs_free_path(path);
2016 }
2017
2018 /*
2019  * read an inode from the btree into the in-memory inode
2020  */
2021 void btrfs_read_locked_inode(struct inode *inode)
2022 {
2023         struct btrfs_path *path;
2024         struct extent_buffer *leaf;
2025         struct btrfs_inode_item *inode_item;
2026         struct btrfs_timespec *tspec;
2027         struct btrfs_root *root = BTRFS_I(inode)->root;
2028         struct btrfs_key location;
2029         u64 alloc_group_block;
2030         u32 rdev;
2031         int ret;
2032
2033         path = btrfs_alloc_path();
2034         BUG_ON(!path);
2035         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2036
2037         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2038         if (ret)
2039                 goto make_bad;
2040
2041         leaf = path->nodes[0];
2042         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2043                                     struct btrfs_inode_item);
2044
2045         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2046         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2047         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2048         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2049         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2050
2051         tspec = btrfs_inode_atime(inode_item);
2052         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2053         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2054
2055         tspec = btrfs_inode_mtime(inode_item);
2056         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2057         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2058
2059         tspec = btrfs_inode_ctime(inode_item);
2060         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2061         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2062
2063         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2064         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2065         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2066         inode->i_generation = BTRFS_I(inode)->generation;
2067         inode->i_rdev = 0;
2068         rdev = btrfs_inode_rdev(leaf, inode_item);
2069
2070         BTRFS_I(inode)->index_cnt = (u64)-1;
2071         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2072
2073         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2074
2075         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2076                                                 alloc_group_block, 0);
2077         btrfs_free_path(path);
2078         inode_item = NULL;
2079
2080         switch (inode->i_mode & S_IFMT) {
2081         case S_IFREG:
2082                 inode->i_mapping->a_ops = &btrfs_aops;
2083                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2084                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2085                 inode->i_fop = &btrfs_file_operations;
2086                 inode->i_op = &btrfs_file_inode_operations;
2087                 break;
2088         case S_IFDIR:
2089                 inode->i_fop = &btrfs_dir_file_operations;
2090                 if (root == root->fs_info->tree_root)
2091                         inode->i_op = &btrfs_dir_ro_inode_operations;
2092                 else
2093                         inode->i_op = &btrfs_dir_inode_operations;
2094                 break;
2095         case S_IFLNK:
2096                 inode->i_op = &btrfs_symlink_inode_operations;
2097                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2098                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2099                 break;
2100         default:
2101                 inode->i_op = &btrfs_special_inode_operations;
2102                 init_special_inode(inode, inode->i_mode, rdev);
2103                 break;
2104         }
2105         return;
2106
2107 make_bad:
2108         btrfs_free_path(path);
2109         make_bad_inode(inode);
2110 }
2111
2112 /*
2113  * given a leaf and an inode, copy the inode fields into the leaf
2114  */
2115 static void fill_inode_item(struct btrfs_trans_handle *trans,
2116                             struct extent_buffer *leaf,
2117                             struct btrfs_inode_item *item,
2118                             struct inode *inode)
2119 {
2120         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2121         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2122         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2123         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2124         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2125
2126         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2127                                inode->i_atime.tv_sec);
2128         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2129                                 inode->i_atime.tv_nsec);
2130
2131         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2132                                inode->i_mtime.tv_sec);
2133         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2134                                 inode->i_mtime.tv_nsec);
2135
2136         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2137                                inode->i_ctime.tv_sec);
2138         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2139                                 inode->i_ctime.tv_nsec);
2140
2141         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2142         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2143         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2144         btrfs_set_inode_transid(leaf, item, trans->transid);
2145         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2146         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2147         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2148 }
2149
2150 /*
2151  * copy everything in the in-memory inode into the btree.
2152  */
2153 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2154                                 struct btrfs_root *root, struct inode *inode)
2155 {
2156         struct btrfs_inode_item *inode_item;
2157         struct btrfs_path *path;
2158         struct extent_buffer *leaf;
2159         int ret;
2160
2161         path = btrfs_alloc_path();
2162         BUG_ON(!path);
2163         path->leave_spinning = 1;
2164         ret = btrfs_lookup_inode(trans, root, path,
2165                                  &BTRFS_I(inode)->location, 1);
2166         if (ret) {
2167                 if (ret > 0)
2168                         ret = -ENOENT;
2169                 goto failed;
2170         }
2171
2172         btrfs_unlock_up_safe(path, 1);
2173         leaf = path->nodes[0];
2174         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2175                                   struct btrfs_inode_item);
2176
2177         fill_inode_item(trans, leaf, inode_item, inode);
2178         btrfs_mark_buffer_dirty(leaf);
2179         btrfs_set_inode_last_trans(trans, inode);
2180         ret = 0;
2181 failed:
2182         btrfs_free_path(path);
2183         return ret;
2184 }
2185
2186
2187 /*
2188  * unlink helper that gets used here in inode.c and in the tree logging
2189  * recovery code.  It remove a link in a directory with a given name, and
2190  * also drops the back refs in the inode to the directory
2191  */
2192 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2193                        struct btrfs_root *root,
2194                        struct inode *dir, struct inode *inode,
2195                        const char *name, int name_len)
2196 {
2197         struct btrfs_path *path;
2198         int ret = 0;
2199         struct extent_buffer *leaf;
2200         struct btrfs_dir_item *di;
2201         struct btrfs_key key;
2202         u64 index;
2203
2204         path = btrfs_alloc_path();
2205         if (!path) {
2206                 ret = -ENOMEM;
2207                 goto err;
2208         }
2209
2210         path->leave_spinning = 1;
2211         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2212                                     name, name_len, -1);
2213         if (IS_ERR(di)) {
2214                 ret = PTR_ERR(di);
2215                 goto err;
2216         }
2217         if (!di) {
2218                 ret = -ENOENT;
2219                 goto err;
2220         }
2221         leaf = path->nodes[0];
2222         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2223         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2224         if (ret)
2225                 goto err;
2226         btrfs_release_path(root, path);
2227
2228         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2229                                   inode->i_ino,
2230                                   dir->i_ino, &index);
2231         if (ret) {
2232                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2233                        "inode %lu parent %lu\n", name_len, name,
2234                        inode->i_ino, dir->i_ino);
2235                 goto err;
2236         }
2237
2238         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2239                                          index, name, name_len, -1);
2240         if (IS_ERR(di)) {
2241                 ret = PTR_ERR(di);
2242                 goto err;
2243         }
2244         if (!di) {
2245                 ret = -ENOENT;
2246                 goto err;
2247         }
2248         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2249         btrfs_release_path(root, path);
2250
2251         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2252                                          inode, dir->i_ino);
2253         BUG_ON(ret != 0 && ret != -ENOENT);
2254
2255         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2256                                            dir, index);
2257         BUG_ON(ret);
2258 err:
2259         btrfs_free_path(path);
2260         if (ret)
2261                 goto out;
2262
2263         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2264         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2265         btrfs_update_inode(trans, root, dir);
2266         btrfs_drop_nlink(inode);
2267         ret = btrfs_update_inode(trans, root, inode);
2268         dir->i_sb->s_dirt = 1;
2269 out:
2270         return ret;
2271 }
2272
2273 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2274 {
2275         struct btrfs_root *root;
2276         struct btrfs_trans_handle *trans;
2277         struct inode *inode = dentry->d_inode;
2278         int ret;
2279         unsigned long nr = 0;
2280
2281         root = BTRFS_I(dir)->root;
2282
2283         trans = btrfs_start_transaction(root, 1);
2284
2285         btrfs_set_trans_block_group(trans, dir);
2286
2287         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2288
2289         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2290                                  dentry->d_name.name, dentry->d_name.len);
2291
2292         if (inode->i_nlink == 0)
2293                 ret = btrfs_orphan_add(trans, inode);
2294
2295         nr = trans->blocks_used;
2296
2297         btrfs_end_transaction_throttle(trans, root);
2298         btrfs_btree_balance_dirty(root, nr);
2299         return ret;
2300 }
2301
2302 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2303 {
2304         struct inode *inode = dentry->d_inode;
2305         int err = 0;
2306         int ret;
2307         struct btrfs_root *root = BTRFS_I(dir)->root;
2308         struct btrfs_trans_handle *trans;
2309         unsigned long nr = 0;
2310
2311         /*
2312          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2313          * the root of a subvolume or snapshot
2314          */
2315         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2316             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2317                 return -ENOTEMPTY;
2318         }
2319
2320         trans = btrfs_start_transaction(root, 1);
2321         btrfs_set_trans_block_group(trans, dir);
2322
2323         err = btrfs_orphan_add(trans, inode);
2324         if (err)
2325                 goto fail_trans;
2326
2327         /* now the directory is empty */
2328         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2329                                  dentry->d_name.name, dentry->d_name.len);
2330         if (!err)
2331                 btrfs_i_size_write(inode, 0);
2332
2333 fail_trans:
2334         nr = trans->blocks_used;
2335         ret = btrfs_end_transaction_throttle(trans, root);
2336         btrfs_btree_balance_dirty(root, nr);
2337
2338         if (ret && !err)
2339                 err = ret;
2340         return err;
2341 }
2342
2343 #if 0
2344 /*
2345  * when truncating bytes in a file, it is possible to avoid reading
2346  * the leaves that contain only checksum items.  This can be the
2347  * majority of the IO required to delete a large file, but it must
2348  * be done carefully.
2349  *
2350  * The keys in the level just above the leaves are checked to make sure
2351  * the lowest key in a given leaf is a csum key, and starts at an offset
2352  * after the new  size.
2353  *
2354  * Then the key for the next leaf is checked to make sure it also has
2355  * a checksum item for the same file.  If it does, we know our target leaf
2356  * contains only checksum items, and it can be safely freed without reading
2357  * it.
2358  *
2359  * This is just an optimization targeted at large files.  It may do
2360  * nothing.  It will return 0 unless things went badly.
2361  */
2362 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2363                                      struct btrfs_root *root,
2364                                      struct btrfs_path *path,
2365                                      struct inode *inode, u64 new_size)
2366 {
2367         struct btrfs_key key;
2368         int ret;
2369         int nritems;
2370         struct btrfs_key found_key;
2371         struct btrfs_key other_key;
2372         struct btrfs_leaf_ref *ref;
2373         u64 leaf_gen;
2374         u64 leaf_start;
2375
2376         path->lowest_level = 1;
2377         key.objectid = inode->i_ino;
2378         key.type = BTRFS_CSUM_ITEM_KEY;
2379         key.offset = new_size;
2380 again:
2381         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2382         if (ret < 0)
2383                 goto out;
2384
2385         if (path->nodes[1] == NULL) {
2386                 ret = 0;
2387                 goto out;
2388         }
2389         ret = 0;
2390         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2391         nritems = btrfs_header_nritems(path->nodes[1]);
2392
2393         if (!nritems)
2394                 goto out;
2395
2396         if (path->slots[1] >= nritems)
2397                 goto next_node;
2398
2399         /* did we find a key greater than anything we want to delete? */
2400         if (found_key.objectid > inode->i_ino ||
2401            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2402                 goto out;
2403
2404         /* we check the next key in the node to make sure the leave contains
2405          * only checksum items.  This comparison doesn't work if our
2406          * leaf is the last one in the node
2407          */
2408         if (path->slots[1] + 1 >= nritems) {
2409 next_node:
2410                 /* search forward from the last key in the node, this
2411                  * will bring us into the next node in the tree
2412                  */
2413                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2414
2415                 /* unlikely, but we inc below, so check to be safe */
2416                 if (found_key.offset == (u64)-1)
2417                         goto out;
2418
2419                 /* search_forward needs a path with locks held, do the
2420                  * search again for the original key.  It is possible
2421                  * this will race with a balance and return a path that
2422                  * we could modify, but this drop is just an optimization
2423                  * and is allowed to miss some leaves.
2424                  */
2425                 btrfs_release_path(root, path);
2426                 found_key.offset++;
2427
2428                 /* setup a max key for search_forward */
2429                 other_key.offset = (u64)-1;
2430                 other_key.type = key.type;
2431                 other_key.objectid = key.objectid;
2432
2433                 path->keep_locks = 1;
2434                 ret = btrfs_search_forward(root, &found_key, &other_key,
2435                                            path, 0, 0);
2436                 path->keep_locks = 0;
2437                 if (ret || found_key.objectid != key.objectid ||
2438                     found_key.type != key.type) {
2439                         ret = 0;
2440                         goto out;
2441                 }
2442
2443                 key.offset = found_key.offset;
2444                 btrfs_release_path(root, path);
2445                 cond_resched();
2446                 goto again;
2447         }
2448
2449         /* we know there's one more slot after us in the tree,
2450          * read that key so we can verify it is also a checksum item
2451          */
2452         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2453
2454         if (found_key.objectid < inode->i_ino)
2455                 goto next_key;
2456
2457         if (found_key.type != key.type || found_key.offset < new_size)
2458                 goto next_key;
2459
2460         /*
2461          * if the key for the next leaf isn't a csum key from this objectid,
2462          * we can't be sure there aren't good items inside this leaf.
2463          * Bail out
2464          */
2465         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2466                 goto out;
2467
2468         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2469         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2470         /*
2471          * it is safe to delete this leaf, it contains only
2472          * csum items from this inode at an offset >= new_size
2473          */
2474         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2475         BUG_ON(ret);
2476
2477         if (root->ref_cows && leaf_gen < trans->transid) {
2478                 ref = btrfs_alloc_leaf_ref(root, 0);
2479                 if (ref) {
2480                         ref->root_gen = root->root_key.offset;
2481                         ref->bytenr = leaf_start;
2482                         ref->owner = 0;
2483                         ref->generation = leaf_gen;
2484                         ref->nritems = 0;
2485
2486                         btrfs_sort_leaf_ref(ref);
2487
2488                         ret = btrfs_add_leaf_ref(root, ref, 0);
2489                         WARN_ON(ret);
2490                         btrfs_free_leaf_ref(root, ref);
2491                 } else {
2492                         WARN_ON(1);
2493                 }
2494         }
2495 next_key:
2496         btrfs_release_path(root, path);
2497
2498         if (other_key.objectid == inode->i_ino &&
2499             other_key.type == key.type && other_key.offset > key.offset) {
2500                 key.offset = other_key.offset;
2501                 cond_resched();
2502                 goto again;
2503         }
2504         ret = 0;
2505 out:
2506         /* fixup any changes we've made to the path */
2507         path->lowest_level = 0;
2508         path->keep_locks = 0;
2509         btrfs_release_path(root, path);
2510         return ret;
2511 }
2512
2513 #endif
2514
2515 /*
2516  * this can truncate away extent items, csum items and directory items.
2517  * It starts at a high offset and removes keys until it can't find
2518  * any higher than new_size
2519  *
2520  * csum items that cross the new i_size are truncated to the new size
2521  * as well.
2522  *
2523  * min_type is the minimum key type to truncate down to.  If set to 0, this
2524  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2525  */
2526 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2527                                         struct btrfs_root *root,
2528                                         struct inode *inode,
2529                                         u64 new_size, u32 min_type)
2530 {
2531         int ret;
2532         struct btrfs_path *path;
2533         struct btrfs_key key;
2534         struct btrfs_key found_key;
2535         u32 found_type = (u8)-1;
2536         struct extent_buffer *leaf;
2537         struct btrfs_file_extent_item *fi;
2538         u64 extent_start = 0;
2539         u64 extent_num_bytes = 0;
2540         u64 item_end = 0;
2541         u64 root_gen = 0;
2542         u64 root_owner = 0;
2543         int found_extent;
2544         int del_item;
2545         int pending_del_nr = 0;
2546         int pending_del_slot = 0;
2547         int extent_type = -1;
2548         int encoding;
2549         u64 mask = root->sectorsize - 1;
2550
2551         if (root->ref_cows)
2552                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2553         path = btrfs_alloc_path();
2554         path->reada = -1;
2555         BUG_ON(!path);
2556
2557         /* FIXME, add redo link to tree so we don't leak on crash */
2558         key.objectid = inode->i_ino;
2559         key.offset = (u64)-1;
2560         key.type = (u8)-1;
2561
2562 search_again:
2563         path->leave_spinning = 1;
2564         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2565         if (ret < 0)
2566                 goto error;
2567
2568         if (ret > 0) {
2569                 /* there are no items in the tree for us to truncate, we're
2570                  * done
2571                  */
2572                 if (path->slots[0] == 0) {
2573                         ret = 0;
2574                         goto error;
2575                 }
2576                 path->slots[0]--;
2577         }
2578
2579         while (1) {
2580                 fi = NULL;
2581                 leaf = path->nodes[0];
2582                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2583                 found_type = btrfs_key_type(&found_key);
2584                 encoding = 0;
2585
2586                 if (found_key.objectid != inode->i_ino)
2587                         break;
2588
2589                 if (found_type < min_type)
2590                         break;
2591
2592                 item_end = found_key.offset;
2593                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2594                         fi = btrfs_item_ptr(leaf, path->slots[0],
2595                                             struct btrfs_file_extent_item);
2596                         extent_type = btrfs_file_extent_type(leaf, fi);
2597                         encoding = btrfs_file_extent_compression(leaf, fi);
2598                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2599                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2600
2601                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2602                                 item_end +=
2603                                     btrfs_file_extent_num_bytes(leaf, fi);
2604                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2605                                 item_end += btrfs_file_extent_inline_len(leaf,
2606                                                                          fi);
2607                         }
2608                         item_end--;
2609                 }
2610                 if (item_end < new_size) {
2611                         if (found_type == BTRFS_DIR_ITEM_KEY)
2612                                 found_type = BTRFS_INODE_ITEM_KEY;
2613                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2614                                 found_type = BTRFS_EXTENT_DATA_KEY;
2615                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2616                                 found_type = BTRFS_XATTR_ITEM_KEY;
2617                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2618                                 found_type = BTRFS_INODE_REF_KEY;
2619                         else if (found_type)
2620                                 found_type--;
2621                         else
2622                                 break;
2623                         btrfs_set_key_type(&key, found_type);
2624                         goto next;
2625                 }
2626                 if (found_key.offset >= new_size)
2627                         del_item = 1;
2628                 else
2629                         del_item = 0;
2630                 found_extent = 0;
2631
2632                 /* FIXME, shrink the extent if the ref count is only 1 */
2633                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2634                         goto delete;
2635
2636                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2637                         u64 num_dec;
2638                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2639                         if (!del_item && !encoding) {
2640                                 u64 orig_num_bytes =
2641                                         btrfs_file_extent_num_bytes(leaf, fi);
2642                                 extent_num_bytes = new_size -
2643                                         found_key.offset + root->sectorsize - 1;
2644                                 extent_num_bytes = extent_num_bytes &
2645                                         ~((u64)root->sectorsize - 1);
2646                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2647                                                          extent_num_bytes);
2648                                 num_dec = (orig_num_bytes -
2649                                            extent_num_bytes);
2650                                 if (root->ref_cows && extent_start != 0)
2651                                         inode_sub_bytes(inode, num_dec);
2652                                 btrfs_mark_buffer_dirty(leaf);
2653                         } else {
2654                                 extent_num_bytes =
2655                                         btrfs_file_extent_disk_num_bytes(leaf,
2656                                                                          fi);
2657                                 /* FIXME blocksize != 4096 */
2658                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2659                                 if (extent_start != 0) {
2660                                         found_extent = 1;
2661                                         if (root->ref_cows)
2662                                                 inode_sub_bytes(inode, num_dec);
2663                                 }
2664                                 root_gen = btrfs_header_generation(leaf);
2665                                 root_owner = btrfs_header_owner(leaf);
2666                         }
2667                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2668                         /*
2669                          * we can't truncate inline items that have had
2670                          * special encodings
2671                          */
2672                         if (!del_item &&
2673                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2674                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2675                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2676                                 u32 size = new_size - found_key.offset;
2677
2678                                 if (root->ref_cows) {
2679                                         inode_sub_bytes(inode, item_end + 1 -
2680                                                         new_size);
2681                                 }
2682                                 size =
2683                                     btrfs_file_extent_calc_inline_size(size);
2684                                 ret = btrfs_truncate_item(trans, root, path,
2685                                                           size, 1);
2686                                 BUG_ON(ret);
2687                         } else if (root->ref_cows) {
2688                                 inode_sub_bytes(inode, item_end + 1 -
2689                                                 found_key.offset);
2690                         }
2691                 }
2692 delete:
2693                 if (del_item) {
2694                         if (!pending_del_nr) {
2695                                 /* no pending yet, add ourselves */
2696                                 pending_del_slot = path->slots[0];
2697                                 pending_del_nr = 1;
2698                         } else if (pending_del_nr &&
2699                                    path->slots[0] + 1 == pending_del_slot) {
2700                                 /* hop on the pending chunk */
2701                                 pending_del_nr++;
2702                                 pending_del_slot = path->slots[0];
2703                         } else {
2704                                 BUG();
2705                         }
2706                 } else {
2707                         break;
2708                 }
2709                 if (found_extent) {
2710                         btrfs_set_path_blocking(path);
2711                         ret = btrfs_free_extent(trans, root, extent_start,
2712                                                 extent_num_bytes,
2713                                                 leaf->start, root_owner,
2714                                                 root_gen, inode->i_ino, 0);
2715                         BUG_ON(ret);
2716                 }
2717 next:
2718                 if (path->slots[0] == 0) {
2719                         if (pending_del_nr)
2720                                 goto del_pending;
2721                         btrfs_release_path(root, path);
2722                         if (found_type == BTRFS_INODE_ITEM_KEY)
2723                                 break;
2724                         goto search_again;
2725                 }
2726
2727                 path->slots[0]--;
2728                 if (pending_del_nr &&
2729                     path->slots[0] + 1 != pending_del_slot) {
2730                         struct btrfs_key debug;
2731 del_pending:
2732                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2733                                               pending_del_slot);
2734                         ret = btrfs_del_items(trans, root, path,
2735                                               pending_del_slot,
2736                                               pending_del_nr);
2737                         BUG_ON(ret);
2738                         pending_del_nr = 0;
2739                         btrfs_release_path(root, path);
2740                         if (found_type == BTRFS_INODE_ITEM_KEY)
2741                                 break;
2742                         goto search_again;
2743                 }
2744         }
2745         ret = 0;
2746 error:
2747         if (pending_del_nr) {
2748                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2749                                       pending_del_nr);
2750         }
2751         btrfs_free_path(path);
2752         inode->i_sb->s_dirt = 1;
2753         return ret;
2754 }
2755
2756 /*
2757  * taken from block_truncate_page, but does cow as it zeros out
2758  * any bytes left in the last page in the file.
2759  */
2760 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2761 {
2762         struct inode *inode = mapping->host;
2763         struct btrfs_root *root = BTRFS_I(inode)->root;
2764         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2765         struct btrfs_ordered_extent *ordered;
2766         char *kaddr;
2767         u32 blocksize = root->sectorsize;
2768         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2769         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2770         struct page *page;
2771         int ret = 0;
2772         u64 page_start;
2773         u64 page_end;
2774
2775         if ((offset & (blocksize - 1)) == 0)
2776                 goto out;
2777
2778         ret = -ENOMEM;
2779 again:
2780         page = grab_cache_page(mapping, index);
2781         if (!page)
2782                 goto out;
2783
2784         page_start = page_offset(page);
2785         page_end = page_start + PAGE_CACHE_SIZE - 1;
2786
2787         if (!PageUptodate(page)) {
2788                 ret = btrfs_readpage(NULL, page);
2789                 lock_page(page);
2790                 if (page->mapping != mapping) {
2791                         unlock_page(page);
2792                         page_cache_release(page);
2793                         goto again;
2794                 }
2795                 if (!PageUptodate(page)) {
2796                         ret = -EIO;
2797                         goto out_unlock;
2798                 }
2799         }
2800         wait_on_page_writeback(page);
2801
2802         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2803         set_page_extent_mapped(page);
2804
2805         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2806         if (ordered) {
2807                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2808                 unlock_page(page);
2809                 page_cache_release(page);
2810                 btrfs_start_ordered_extent(inode, ordered, 1);
2811                 btrfs_put_ordered_extent(ordered);
2812                 goto again;
2813         }
2814
2815         btrfs_set_extent_delalloc(inode, page_start, page_end);
2816         ret = 0;
2817         if (offset != PAGE_CACHE_SIZE) {
2818                 kaddr = kmap(page);
2819                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2820                 flush_dcache_page(page);
2821                 kunmap(page);
2822         }
2823         ClearPageChecked(page);
2824         set_page_dirty(page);
2825         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2826
2827 out_unlock:
2828         unlock_page(page);
2829         page_cache_release(page);
2830 out:
2831         return ret;
2832 }
2833
2834 int btrfs_cont_expand(struct inode *inode, loff_t size)
2835 {
2836         struct btrfs_trans_handle *trans;
2837         struct btrfs_root *root = BTRFS_I(inode)->root;
2838         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2839         struct extent_map *em;
2840         u64 mask = root->sectorsize - 1;
2841         u64 hole_start = (inode->i_size + mask) & ~mask;
2842         u64 block_end = (size + mask) & ~mask;
2843         u64 last_byte;
2844         u64 cur_offset;
2845         u64 hole_size;
2846         int err;
2847
2848         if (size <= hole_start)
2849                 return 0;
2850
2851         err = btrfs_check_metadata_free_space(root);
2852         if (err)
2853                 return err;
2854
2855         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2856
2857         while (1) {
2858                 struct btrfs_ordered_extent *ordered;
2859                 btrfs_wait_ordered_range(inode, hole_start,
2860                                          block_end - hole_start);
2861                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2862                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2863                 if (!ordered)
2864                         break;
2865                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2866                 btrfs_put_ordered_extent(ordered);
2867         }
2868
2869         trans = btrfs_start_transaction(root, 1);
2870         btrfs_set_trans_block_group(trans, inode);
2871
2872         cur_offset = hole_start;
2873         while (1) {
2874                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2875                                 block_end - cur_offset, 0);
2876                 BUG_ON(IS_ERR(em) || !em);
2877                 last_byte = min(extent_map_end(em), block_end);
2878                 last_byte = (last_byte + mask) & ~mask;
2879                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2880                         u64 hint_byte = 0;
2881                         hole_size = last_byte - cur_offset;
2882                         err = btrfs_drop_extents(trans, root, inode,
2883                                                  cur_offset,
2884                                                  cur_offset + hole_size,
2885                                                  block_end,
2886                                                  cur_offset, &hint_byte);
2887                         if (err)
2888                                 break;
2889                         err = btrfs_insert_file_extent(trans, root,
2890                                         inode->i_ino, cur_offset, 0,
2891                                         0, hole_size, 0, hole_size,
2892                                         0, 0, 0);
2893                         btrfs_drop_extent_cache(inode, hole_start,
2894                                         last_byte - 1, 0);
2895                 }
2896                 free_extent_map(em);
2897                 cur_offset = last_byte;
2898                 if (err || cur_offset >= block_end)
2899                         break;
2900         }
2901
2902         btrfs_end_transaction(trans, root);
2903         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2904         return err;
2905 }
2906
2907 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2908 {
2909         struct inode *inode = dentry->d_inode;
2910         int err;
2911
2912         err = inode_change_ok(inode, attr);
2913         if (err)
2914                 return err;
2915
2916         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2917                 if (attr->ia_size > inode->i_size) {
2918                         err = btrfs_cont_expand(inode, attr->ia_size);
2919                         if (err)
2920                                 return err;
2921                 } else if (inode->i_size > 0 &&
2922                            attr->ia_size == 0) {
2923
2924                         /* we're truncating a file that used to have good
2925                          * data down to zero.  Make sure it gets into
2926                          * the ordered flush list so that any new writes
2927                          * get down to disk quickly.
2928                          */
2929                         BTRFS_I(inode)->ordered_data_close = 1;
2930                 }
2931         }
2932
2933         err = inode_setattr(inode, attr);
2934
2935         if (!err && ((attr->ia_valid & ATTR_MODE)))
2936                 err = btrfs_acl_chmod(inode);
2937         return err;
2938 }
2939
2940 void btrfs_delete_inode(struct inode *inode)
2941 {
2942         struct btrfs_trans_handle *trans;
2943         struct btrfs_root *root = BTRFS_I(inode)->root;
2944         unsigned long nr;
2945         int ret;
2946
2947         truncate_inode_pages(&inode->i_data, 0);
2948         if (is_bad_inode(inode)) {
2949                 btrfs_orphan_del(NULL, inode);
2950                 goto no_delete;
2951         }
2952         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2953
2954         btrfs_i_size_write(inode, 0);
2955         trans = btrfs_join_transaction(root, 1);
2956
2957         btrfs_set_trans_block_group(trans, inode);
2958         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2959         if (ret) {
2960                 btrfs_orphan_del(NULL, inode);
2961                 goto no_delete_lock;
2962         }
2963
2964         btrfs_orphan_del(trans, inode);
2965
2966         nr = trans->blocks_used;
2967         clear_inode(inode);
2968
2969         btrfs_end_transaction(trans, root);
2970         btrfs_btree_balance_dirty(root, nr);
2971         return;
2972
2973 no_delete_lock:
2974         nr = trans->blocks_used;
2975         btrfs_end_transaction(trans, root);
2976         btrfs_btree_balance_dirty(root, nr);
2977 no_delete:
2978         clear_inode(inode);
2979 }
2980
2981 /*
2982  * this returns the key found in the dir entry in the location pointer.
2983  * If no dir entries were found, location->objectid is 0.
2984  */
2985 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2986                                struct btrfs_key *location)
2987 {
2988         const char *name = dentry->d_name.name;
2989         int namelen = dentry->d_name.len;
2990         struct btrfs_dir_item *di;
2991         struct btrfs_path *path;
2992         struct btrfs_root *root = BTRFS_I(dir)->root;
2993         int ret = 0;
2994
2995         path = btrfs_alloc_path();
2996         BUG_ON(!path);
2997
2998         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2999                                     namelen, 0);
3000         if (IS_ERR(di))
3001                 ret = PTR_ERR(di);
3002
3003         if (!di || IS_ERR(di))
3004                 goto out_err;
3005
3006         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3007 out:
3008         btrfs_free_path(path);
3009         return ret;
3010 out_err:
3011         location->objectid = 0;
3012         goto out;
3013 }
3014
3015 /*
3016  * when we hit a tree root in a directory, the btrfs part of the inode
3017  * needs to be changed to reflect the root directory of the tree root.  This
3018  * is kind of like crossing a mount point.
3019  */
3020 static int fixup_tree_root_location(struct btrfs_root *root,
3021                              struct btrfs_key *location,
3022                              struct btrfs_root **sub_root,
3023                              struct dentry *dentry)
3024 {
3025         struct btrfs_root_item *ri;
3026
3027         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3028                 return 0;
3029         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3030                 return 0;
3031
3032         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3033                                         dentry->d_name.name,
3034                                         dentry->d_name.len);
3035         if (IS_ERR(*sub_root))
3036                 return PTR_ERR(*sub_root);
3037
3038         ri = &(*sub_root)->root_item;
3039         location->objectid = btrfs_root_dirid(ri);
3040         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3041         location->offset = 0;
3042
3043         return 0;
3044 }
3045
3046 static noinline void init_btrfs_i(struct inode *inode)
3047 {
3048         struct btrfs_inode *bi = BTRFS_I(inode);
3049
3050         bi->i_acl = BTRFS_ACL_NOT_CACHED;
3051         bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
3052
3053         bi->generation = 0;
3054         bi->sequence = 0;
3055         bi->last_trans = 0;
3056         bi->logged_trans = 0;
3057         bi->delalloc_bytes = 0;
3058         bi->reserved_bytes = 0;
3059         bi->disk_i_size = 0;
3060         bi->flags = 0;
3061         bi->index_cnt = (u64)-1;
3062         bi->last_unlink_trans = 0;
3063         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3064         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3065                              inode->i_mapping, GFP_NOFS);
3066         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3067                              inode->i_mapping, GFP_NOFS);
3068         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3069         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3070         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3071         mutex_init(&BTRFS_I(inode)->extent_mutex);
3072         mutex_init(&BTRFS_I(inode)->log_mutex);
3073 }
3074
3075 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3076 {
3077         struct btrfs_iget_args *args = p;
3078         inode->i_ino = args->ino;
3079         init_btrfs_i(inode);
3080         BTRFS_I(inode)->root = args->root;
3081         btrfs_set_inode_space_info(args->root, inode);
3082         return 0;
3083 }
3084
3085 static int btrfs_find_actor(struct inode *inode, void *opaque)
3086 {
3087         struct btrfs_iget_args *args = opaque;
3088         return args->ino == inode->i_ino &&
3089                 args->root == BTRFS_I(inode)->root;
3090 }
3091
3092 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3093                             struct btrfs_root *root, int wait)
3094 {
3095         struct inode *inode;
3096         struct btrfs_iget_args args;
3097         args.ino = objectid;
3098         args.root = root;
3099
3100         if (wait) {
3101                 inode = ilookup5(s, objectid, btrfs_find_actor,
3102                                  (void *)&args);
3103         } else {
3104                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3105                                         (void *)&args);
3106         }
3107         return inode;
3108 }
3109
3110 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3111                                 struct btrfs_root *root)
3112 {
3113         struct inode *inode;
3114         struct btrfs_iget_args args;
3115         args.ino = objectid;
3116         args.root = root;
3117
3118         inode = iget5_locked(s, objectid, btrfs_find_actor,
3119                              btrfs_init_locked_inode,
3120                              (void *)&args);
3121         return inode;
3122 }
3123
3124 /* Get an inode object given its location and corresponding root.
3125  * Returns in *is_new if the inode was read from disk
3126  */
3127 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3128                          struct btrfs_root *root, int *is_new)
3129 {
3130         struct inode *inode;
3131
3132         inode = btrfs_iget_locked(s, location->objectid, root);
3133         if (!inode)
3134                 return ERR_PTR(-EACCES);
3135
3136         if (inode->i_state & I_NEW) {
3137                 BTRFS_I(inode)->root = root;
3138                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3139                 btrfs_read_locked_inode(inode);
3140                 unlock_new_inode(inode);
3141                 if (is_new)
3142                         *is_new = 1;
3143         } else {
3144                 if (is_new)
3145                         *is_new = 0;
3146         }
3147
3148         return inode;
3149 }
3150
3151 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3152 {
3153         struct inode *inode;
3154         struct btrfs_inode *bi = BTRFS_I(dir);
3155         struct btrfs_root *root = bi->root;
3156         struct btrfs_root *sub_root = root;
3157         struct btrfs_key location;
3158         int ret, new;
3159
3160         if (dentry->d_name.len > BTRFS_NAME_LEN)
3161                 return ERR_PTR(-ENAMETOOLONG);
3162
3163         ret = btrfs_inode_by_name(dir, dentry, &location);
3164
3165         if (ret < 0)
3166                 return ERR_PTR(ret);
3167
3168         inode = NULL;
3169         if (location.objectid) {
3170                 ret = fixup_tree_root_location(root, &location, &sub_root,
3171                                                 dentry);
3172                 if (ret < 0)
3173                         return ERR_PTR(ret);
3174                 if (ret > 0)
3175                         return ERR_PTR(-ENOENT);
3176                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3177                 if (IS_ERR(inode))
3178                         return ERR_CAST(inode);
3179         }
3180         return inode;
3181 }
3182
3183 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3184                                    struct nameidata *nd)
3185 {
3186         struct inode *inode;
3187
3188         if (dentry->d_name.len > BTRFS_NAME_LEN)
3189                 return ERR_PTR(-ENAMETOOLONG);
3190
3191         inode = btrfs_lookup_dentry(dir, dentry);
3192         if (IS_ERR(inode))
3193                 return ERR_CAST(inode);
3194
3195         return d_splice_alias(inode, dentry);
3196 }
3197
3198 static unsigned char btrfs_filetype_table[] = {
3199         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3200 };
3201
3202 static int btrfs_real_readdir(struct file *filp, void *dirent,
3203                               filldir_t filldir)
3204 {
3205         struct inode *inode = filp->f_dentry->d_inode;
3206         struct btrfs_root *root = BTRFS_I(inode)->root;
3207         struct btrfs_item *item;
3208         struct btrfs_dir_item *di;
3209         struct btrfs_key key;
3210         struct btrfs_key found_key;
3211         struct btrfs_path *path;
3212         int ret;
3213         u32 nritems;
3214         struct extent_buffer *leaf;
3215         int slot;
3216         int advance;
3217         unsigned char d_type;
3218         int over = 0;
3219         u32 di_cur;
3220         u32 di_total;
3221         u32 di_len;
3222         int key_type = BTRFS_DIR_INDEX_KEY;
3223         char tmp_name[32];
3224         char *name_ptr;
3225         int name_len;
3226
3227         /* FIXME, use a real flag for deciding about the key type */
3228         if (root->fs_info->tree_root == root)
3229                 key_type = BTRFS_DIR_ITEM_KEY;
3230
3231         /* special case for "." */
3232         if (filp->f_pos == 0) {
3233                 over = filldir(dirent, ".", 1,
3234                                1, inode->i_ino,
3235                                DT_DIR);
3236                 if (over)
3237                         return 0;
3238                 filp->f_pos = 1;
3239         }
3240         /* special case for .., just use the back ref */
3241         if (filp->f_pos == 1) {
3242                 u64 pino = parent_ino(filp->f_path.dentry);
3243                 over = filldir(dirent, "..", 2,
3244                                2, pino, DT_DIR);
3245                 if (over)
3246                         return 0;
3247                 filp->f_pos = 2;
3248         }
3249         path = btrfs_alloc_path();
3250         path->reada = 2;
3251
3252         btrfs_set_key_type(&key, key_type);
3253         key.offset = filp->f_pos;
3254         key.objectid = inode->i_ino;
3255
3256         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3257         if (ret < 0)
3258                 goto err;
3259         advance = 0;
3260
3261         while (1) {
3262                 leaf = path->nodes[0];
3263                 nritems = btrfs_header_nritems(leaf);
3264                 slot = path->slots[0];
3265                 if (advance || slot >= nritems) {
3266                         if (slot >= nritems - 1) {
3267                                 ret = btrfs_next_leaf(root, path);
3268                                 if (ret)
3269                                         break;
3270                                 leaf = path->nodes[0];
3271                                 nritems = btrfs_header_nritems(leaf);
3272                                 slot = path->slots[0];
3273                         } else {
3274                                 slot++;
3275                                 path->slots[0]++;
3276                         }
3277                 }
3278
3279                 advance = 1;
3280                 item = btrfs_item_nr(leaf, slot);
3281                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3282
3283                 if (found_key.objectid != key.objectid)
3284                         break;
3285                 if (btrfs_key_type(&found_key) != key_type)
3286                         break;
3287                 if (found_key.offset < filp->f_pos)
3288                         continue;
3289
3290                 filp->f_pos = found_key.offset;
3291
3292                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3293                 di_cur = 0;
3294                 di_total = btrfs_item_size(leaf, item);
3295
3296                 while (di_cur < di_total) {
3297                         struct btrfs_key location;
3298
3299                         name_len = btrfs_dir_name_len(leaf, di);
3300                         if (name_len <= sizeof(tmp_name)) {
3301                                 name_ptr = tmp_name;
3302                         } else {
3303                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3304                                 if (!name_ptr) {
3305                                         ret = -ENOMEM;
3306                                         goto err;
3307                                 }
3308                         }
3309                         read_extent_buffer(leaf, name_ptr,
3310                                            (unsigned long)(di + 1), name_len);
3311
3312                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3313                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3314
3315                         /* is this a reference to our own snapshot? If so
3316                          * skip it
3317                          */
3318                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3319                             location.objectid == root->root_key.objectid) {
3320                                 over = 0;
3321                                 goto skip;
3322                         }
3323                         over = filldir(dirent, name_ptr, name_len,
3324                                        found_key.offset, location.objectid,
3325                                        d_type);
3326
3327 skip:
3328                         if (name_ptr != tmp_name)
3329                                 kfree(name_ptr);
3330
3331                         if (over)
3332                                 goto nopos;
3333                         di_len = btrfs_dir_name_len(leaf, di) +
3334                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3335                         di_cur += di_len;
3336                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3337                 }
3338         }
3339
3340         /* Reached end of directory/root. Bump pos past the last item. */
3341         if (key_type == BTRFS_DIR_INDEX_KEY)
3342                 filp->f_pos = INT_LIMIT(off_t);
3343         else
3344                 filp->f_pos++;
3345 nopos:
3346         ret = 0;
3347 err:
3348         btrfs_free_path(path);
3349         return ret;
3350 }
3351
3352 int btrfs_write_inode(struct inode *inode, int wait)
3353 {
3354         struct btrfs_root *root = BTRFS_I(inode)->root;
3355         struct btrfs_trans_handle *trans;
3356         int ret = 0;
3357
3358         if (root->fs_info->btree_inode == inode)
3359                 return 0;
3360
3361         if (wait) {
3362                 trans = btrfs_join_transaction(root, 1);
3363                 btrfs_set_trans_block_group(trans, inode);
3364                 ret = btrfs_commit_transaction(trans, root);
3365         }
3366         return ret;
3367 }
3368
3369 /*
3370  * This is somewhat expensive, updating the tree every time the
3371  * inode changes.  But, it is most likely to find the inode in cache.
3372  * FIXME, needs more benchmarking...there are no reasons other than performance
3373  * to keep or drop this code.
3374  */
3375 void btrfs_dirty_inode(struct inode *inode)
3376 {
3377         struct btrfs_root *root = BTRFS_I(inode)->root;
3378         struct btrfs_trans_handle *trans;
3379
3380         trans = btrfs_join_transaction(root, 1);
3381         btrfs_set_trans_block_group(trans, inode);
3382         btrfs_update_inode(trans, root, inode);
3383         btrfs_end_transaction(trans, root);
3384 }
3385
3386 /*
3387  * find the highest existing sequence number in a directory
3388  * and then set the in-memory index_cnt variable to reflect
3389  * free sequence numbers
3390  */
3391 static int btrfs_set_inode_index_count(struct inode *inode)
3392 {
3393         struct btrfs_root *root = BTRFS_I(inode)->root;
3394         struct btrfs_key key, found_key;
3395         struct btrfs_path *path;
3396         struct extent_buffer *leaf;
3397         int ret;
3398
3399         key.objectid = inode->i_ino;
3400         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3401         key.offset = (u64)-1;
3402
3403         path = btrfs_alloc_path();
3404         if (!path)
3405                 return -ENOMEM;
3406
3407         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3408         if (ret < 0)
3409                 goto out;
3410         /* FIXME: we should be able to handle this */
3411         if (ret == 0)
3412                 goto out;
3413         ret = 0;
3414
3415         /*
3416          * MAGIC NUMBER EXPLANATION:
3417          * since we search a directory based on f_pos we have to start at 2
3418          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3419          * else has to start at 2
3420          */
3421         if (path->slots[0] == 0) {
3422                 BTRFS_I(inode)->index_cnt = 2;
3423                 goto out;
3424         }
3425
3426         path->slots[0]--;
3427
3428         leaf = path->nodes[0];
3429         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3430
3431         if (found_key.objectid != inode->i_ino ||
3432             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3433                 BTRFS_I(inode)->index_cnt = 2;
3434                 goto out;
3435         }
3436
3437         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3438 out:
3439         btrfs_free_path(path);
3440         return ret;
3441 }
3442
3443 /*
3444  * helper to find a free sequence number in a given directory.  This current
3445  * code is very simple, later versions will do smarter things in the btree
3446  */
3447 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3448 {
3449         int ret = 0;
3450
3451         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3452                 ret = btrfs_set_inode_index_count(dir);
3453                 if (ret)
3454                         return ret;
3455         }
3456
3457         *index = BTRFS_I(dir)->index_cnt;
3458         BTRFS_I(dir)->index_cnt++;
3459
3460         return ret;
3461 }
3462
3463 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3464                                      struct btrfs_root *root,
3465                                      struct inode *dir,
3466                                      const char *name, int name_len,
3467                                      u64 ref_objectid, u64 objectid,
3468                                      u64 alloc_hint, int mode, u64 *index)
3469 {
3470         struct inode *inode;
3471         struct btrfs_inode_item *inode_item;
3472         struct btrfs_key *location;
3473         struct btrfs_path *path;
3474         struct btrfs_inode_ref *ref;
3475         struct btrfs_key key[2];
3476         u32 sizes[2];
3477         unsigned long ptr;
3478         int ret;
3479         int owner;
3480
3481         path = btrfs_alloc_path();
3482         BUG_ON(!path);
3483
3484         inode = new_inode(root->fs_info->sb);
3485         if (!inode)
3486                 return ERR_PTR(-ENOMEM);
3487
3488         if (dir) {
3489                 ret = btrfs_set_inode_index(dir, index);
3490                 if (ret) {
3491                         iput(inode);
3492                         return ERR_PTR(ret);
3493                 }
3494         }
3495         /*
3496          * index_cnt is ignored for everything but a dir,
3497          * btrfs_get_inode_index_count has an explanation for the magic
3498          * number
3499          */
3500         init_btrfs_i(inode);
3501         BTRFS_I(inode)->index_cnt = 2;
3502         BTRFS_I(inode)->root = root;
3503         BTRFS_I(inode)->generation = trans->transid;
3504         btrfs_set_inode_space_info(root, inode);
3505
3506         if (mode & S_IFDIR)
3507                 owner = 0;
3508         else
3509                 owner = 1;
3510         BTRFS_I(inode)->block_group =
3511                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3512         if ((mode & S_IFREG)) {
3513                 if (btrfs_test_opt(root, NODATASUM))
3514                         btrfs_set_flag(inode, NODATASUM);
3515                 if (btrfs_test_opt(root, NODATACOW))
3516                         btrfs_set_flag(inode, NODATACOW);
3517         }
3518
3519         key[0].objectid = objectid;
3520         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3521         key[0].offset = 0;
3522
3523         key[1].objectid = objectid;
3524         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3525         key[1].offset = ref_objectid;
3526
3527         sizes[0] = sizeof(struct btrfs_inode_item);
3528         sizes[1] = name_len + sizeof(*ref);
3529
3530         path->leave_spinning = 1;
3531         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3532         if (ret != 0)
3533                 goto fail;
3534
3535         if (objectid > root->highest_inode)
3536                 root->highest_inode = objectid;
3537
3538         inode->i_uid = current_fsuid();
3539
3540         if (dir && (dir->i_mode & S_ISGID)) {
3541                 inode->i_gid = dir->i_gid;
3542                 if (S_ISDIR(mode))
3543                         mode |= S_ISGID;
3544         } else
3545                 inode->i_gid = current_fsgid();
3546
3547         inode->i_mode = mode;
3548         inode->i_ino = objectid;
3549         inode_set_bytes(inode, 0);
3550         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3551         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3552                                   struct btrfs_inode_item);
3553         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3554
3555         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3556                              struct btrfs_inode_ref);
3557         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3558         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3559         ptr = (unsigned long)(ref + 1);
3560         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3561
3562         btrfs_mark_buffer_dirty(path->nodes[0]);
3563         btrfs_free_path(path);
3564
3565         location = &BTRFS_I(inode)->location;
3566         location->objectid = objectid;
3567         location->offset = 0;
3568         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3569
3570         insert_inode_hash(inode);
3571         return inode;
3572 fail:
3573         if (dir)
3574                 BTRFS_I(dir)->index_cnt--;
3575         btrfs_free_path(path);
3576         iput(inode);
3577         return ERR_PTR(ret);
3578 }
3579
3580 static inline u8 btrfs_inode_type(struct inode *inode)
3581 {
3582         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3583 }
3584
3585 /*
3586  * utility function to add 'inode' into 'parent_inode' with
3587  * a give name and a given sequence number.
3588  * if 'add_backref' is true, also insert a backref from the
3589  * inode to the parent directory.
3590  */
3591 int btrfs_add_link(struct btrfs_trans_handle *trans,
3592                    struct inode *parent_inode, struct inode *inode,
3593                    const char *name, int name_len, int add_backref, u64 index)
3594 {
3595         int ret;
3596         struct btrfs_key key;
3597         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3598
3599         key.objectid = inode->i_ino;
3600         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3601         key.offset = 0;
3602
3603         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3604                                     parent_inode->i_ino,
3605                                     &key, btrfs_inode_type(inode),
3606                                     index);
3607         if (ret == 0) {
3608                 if (add_backref) {
3609                         ret = btrfs_insert_inode_ref(trans, root,
3610                                                      name, name_len,
3611                                                      inode->i_ino,
3612                                                      parent_inode->i_ino,
3613                                                      index);
3614                 }
3615                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3616                                    name_len * 2);
3617                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3618                 ret = btrfs_update_inode(trans, root, parent_inode);
3619         }
3620         return ret;
3621 }
3622
3623 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3624                             struct dentry *dentry, struct inode *inode,
3625                             int backref, u64 index)
3626 {
3627         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3628                                  inode, dentry->d_name.name,
3629                                  dentry->d_name.len, backref, index);
3630         if (!err) {
3631                 d_instantiate(dentry, inode);
3632                 return 0;
3633         }
3634         if (err > 0)
3635                 err = -EEXIST;
3636         return err;
3637 }
3638
3639 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3640                         int mode, dev_t rdev)
3641 {
3642         struct btrfs_trans_handle *trans;
3643         struct btrfs_root *root = BTRFS_I(dir)->root;
3644         struct inode *inode = NULL;
3645         int err;
3646         int drop_inode = 0;
3647         u64 objectid;
3648         unsigned long nr = 0;
3649         u64 index = 0;
3650
3651         if (!new_valid_dev(rdev))
3652                 return -EINVAL;
3653
3654         err = btrfs_check_metadata_free_space(root);
3655         if (err)
3656                 goto fail;
3657
3658         trans = btrfs_start_transaction(root, 1);
3659         btrfs_set_trans_block_group(trans, dir);
3660
3661         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3662         if (err) {
3663                 err = -ENOSPC;
3664                 goto out_unlock;
3665         }
3666
3667         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3668                                 dentry->d_name.len,
3669                                 dentry->d_parent->d_inode->i_ino, objectid,
3670                                 BTRFS_I(dir)->block_group, mode, &index);
3671         err = PTR_ERR(inode);
3672         if (IS_ERR(inode))
3673                 goto out_unlock;
3674
3675         err = btrfs_init_inode_security(inode, dir);
3676         if (err) {
3677                 drop_inode = 1;
3678                 goto out_unlock;
3679         }
3680
3681         btrfs_set_trans_block_group(trans, inode);
3682         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3683         if (err)
3684                 drop_inode = 1;
3685         else {
3686                 inode->i_op = &btrfs_special_inode_operations;
3687                 init_special_inode(inode, inode->i_mode, rdev);
3688                 btrfs_update_inode(trans, root, inode);
3689         }
3690         dir->i_sb->s_dirt = 1;
3691         btrfs_update_inode_block_group(trans, inode);
3692         btrfs_update_inode_block_group(trans, dir);
3693 out_unlock:
3694         nr = trans->blocks_used;
3695         btrfs_end_transaction_throttle(trans, root);
3696 fail:
3697         if (drop_inode) {
3698                 inode_dec_link_count(inode);
3699                 iput(inode);
3700         }
3701         btrfs_btree_balance_dirty(root, nr);
3702         return err;
3703 }
3704
3705 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3706                         int mode, struct nameidata *nd)
3707 {
3708         struct btrfs_trans_handle *trans;
3709         struct btrfs_root *root = BTRFS_I(dir)->root;
3710         struct inode *inode = NULL;
3711         int err;
3712         int drop_inode = 0;
3713         unsigned long nr = 0;
3714         u64 objectid;
3715         u64 index = 0;
3716
3717         err = btrfs_check_metadata_free_space(root);
3718         if (err)
3719                 goto fail;
3720         trans = btrfs_start_transaction(root, 1);
3721         btrfs_set_trans_block_group(trans, dir);
3722
3723         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3724         if (err) {
3725                 err = -ENOSPC;
3726                 goto out_unlock;
3727         }
3728
3729         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3730                                 dentry->d_name.len,
3731                                 dentry->d_parent->d_inode->i_ino,
3732                                 objectid, BTRFS_I(dir)->block_group, mode,
3733                                 &index);
3734         err = PTR_ERR(inode);
3735         if (IS_ERR(inode))
3736                 goto out_unlock;
3737
3738         err = btrfs_init_inode_security(inode, dir);
3739         if (err) {
3740                 drop_inode = 1;
3741                 goto out_unlock;
3742         }
3743
3744         btrfs_set_trans_block_group(trans, inode);
3745         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3746         if (err)
3747                 drop_inode = 1;
3748         else {
3749                 inode->i_mapping->a_ops = &btrfs_aops;
3750                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3751                 inode->i_fop = &btrfs_file_operations;
3752                 inode->i_op = &btrfs_file_inode_operations;
3753                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3754         }
3755         dir->i_sb->s_dirt = 1;
3756         btrfs_update_inode_block_group(trans, inode);
3757         btrfs_update_inode_block_group(trans, dir);
3758 out_unlock:
3759         nr = trans->blocks_used;
3760         btrfs_end_transaction_throttle(trans, root);
3761 fail:
3762         if (drop_inode) {
3763                 inode_dec_link_count(inode);
3764                 iput(inode);
3765         }
3766         btrfs_btree_balance_dirty(root, nr);
3767         return err;
3768 }
3769
3770 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3771                       struct dentry *dentry)
3772 {
3773         struct btrfs_trans_handle *trans;
3774         struct btrfs_root *root = BTRFS_I(dir)->root;
3775         struct inode *inode = old_dentry->d_inode;
3776         u64 index;
3777         unsigned long nr = 0;
3778         int err;
3779         int drop_inode = 0;
3780
3781         if (inode->i_nlink == 0)
3782                 return -ENOENT;
3783
3784         btrfs_inc_nlink(inode);
3785         err = btrfs_check_metadata_free_space(root);
3786         if (err)
3787                 goto fail;
3788         err = btrfs_set_inode_index(dir, &index);
3789         if (err)
3790                 goto fail;
3791
3792         trans = btrfs_start_transaction(root, 1);
3793
3794         btrfs_set_trans_block_group(trans, dir);
3795         atomic_inc(&inode->i_count);
3796
3797         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3798
3799         if (err)
3800                 drop_inode = 1;
3801
3802         dir->i_sb->s_dirt = 1;
3803         btrfs_update_inode_block_group(trans, dir);
3804         err = btrfs_update_inode(trans, root, inode);
3805
3806         if (err)
3807                 drop_inode = 1;
3808
3809         nr = trans->blocks_used;
3810
3811         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3812         btrfs_end_transaction_throttle(trans, root);
3813 fail:
3814         if (drop_inode) {
3815                 inode_dec_link_count(inode);
3816                 iput(inode);
3817         }
3818         btrfs_btree_balance_dirty(root, nr);
3819         return err;
3820 }
3821
3822 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3823 {
3824         struct inode *inode = NULL;
3825         struct btrfs_trans_handle *trans;
3826         struct btrfs_root *root = BTRFS_I(dir)->root;
3827         int err = 0;
3828         int drop_on_err = 0;
3829         u64 objectid = 0;
3830         u64 index = 0;
3831         unsigned long nr = 1;
3832
3833         err = btrfs_check_metadata_free_space(root);
3834         if (err)
3835                 goto out_unlock;
3836
3837         trans = btrfs_start_transaction(root, 1);
3838         btrfs_set_trans_block_group(trans, dir);
3839
3840         if (IS_ERR(trans)) {
3841                 err = PTR_ERR(trans);
3842                 goto out_unlock;
3843         }
3844
3845         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3846         if (err) {
3847                 err = -ENOSPC;
3848                 goto out_unlock;
3849         }
3850
3851         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3852                                 dentry->d_name.len,
3853                                 dentry->d_parent->d_inode->i_ino, objectid,
3854                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3855                                 &index);
3856         if (IS_ERR(inode)) {
3857                 err = PTR_ERR(inode);
3858                 goto out_fail;
3859         }
3860
3861         drop_on_err = 1;
3862
3863         err = btrfs_init_inode_security(inode, dir);
3864         if (err)
3865                 goto out_fail;
3866
3867         inode->i_op = &btrfs_dir_inode_operations;
3868         inode->i_fop = &btrfs_dir_file_operations;
3869         btrfs_set_trans_block_group(trans, inode);
3870
3871         btrfs_i_size_write(inode, 0);
3872         err = btrfs_update_inode(trans, root, inode);
3873         if (err)
3874                 goto out_fail;
3875
3876         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3877                                  inode, dentry->d_name.name,
3878                                  dentry->d_name.len, 0, index);
3879         if (err)
3880                 goto out_fail;
3881
3882         d_instantiate(dentry, inode);
3883         drop_on_err = 0;
3884         dir->i_sb->s_dirt = 1;
3885         btrfs_update_inode_block_group(trans, inode);
3886         btrfs_update_inode_block_group(trans, dir);
3887
3888 out_fail:
3889         nr = trans->blocks_used;
3890         btrfs_end_transaction_throttle(trans, root);
3891
3892 out_unlock:
3893         if (drop_on_err)
3894                 iput(inode);
3895         btrfs_btree_balance_dirty(root, nr);
3896         return err;
3897 }
3898
3899 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3900  * and an extent that you want to insert, deal with overlap and insert
3901  * the new extent into the tree.
3902  */
3903 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3904                                 struct extent_map *existing,
3905                                 struct extent_map *em,
3906                                 u64 map_start, u64 map_len)
3907 {
3908         u64 start_diff;
3909
3910         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3911         start_diff = map_start - em->start;
3912         em->start = map_start;
3913         em->len = map_len;
3914         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3915             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3916                 em->block_start += start_diff;
3917                 em->block_len -= start_diff;
3918         }
3919         return add_extent_mapping(em_tree, em);
3920 }
3921
3922 static noinline int uncompress_inline(struct btrfs_path *path,
3923                                       struct inode *inode, struct page *page,
3924                                       size_t pg_offset, u64 extent_offset,
3925                                       struct btrfs_file_extent_item *item)
3926 {
3927         int ret;
3928         struct extent_buffer *leaf = path->nodes[0];
3929         char *tmp;
3930         size_t max_size;
3931         unsigned long inline_size;
3932         unsigned long ptr;
3933
3934         WARN_ON(pg_offset != 0);
3935         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3936         inline_size = btrfs_file_extent_inline_item_len(leaf,
3937                                         btrfs_item_nr(leaf, path->slots[0]));
3938         tmp = kmalloc(inline_size, GFP_NOFS);
3939         ptr = btrfs_file_extent_inline_start(item);
3940
3941         read_extent_buffer(leaf, tmp, ptr, inline_size);
3942
3943         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3944         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3945                                     inline_size, max_size);
3946         if (ret) {
3947                 char *kaddr = kmap_atomic(page, KM_USER0);
3948                 unsigned long copy_size = min_t(u64,
3949                                   PAGE_CACHE_SIZE - pg_offset,
3950                                   max_size - extent_offset);
3951                 memset(kaddr + pg_offset, 0, copy_size);
3952                 kunmap_atomic(kaddr, KM_USER0);
3953         }
3954         kfree(tmp);
3955         return 0;
3956 }
3957
3958 /*
3959  * a bit scary, this does extent mapping from logical file offset to the disk.
3960  * the ugly parts come from merging extents from the disk with the in-ram
3961  * representation.  This gets more complex because of the data=ordered code,
3962  * where the in-ram extents might be locked pending data=ordered completion.
3963  *
3964  * This also copies inline extents directly into the page.
3965  */
3966
3967 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3968                                     size_t pg_offset, u64 start, u64 len,
3969                                     int create)
3970 {
3971         int ret;
3972         int err = 0;
3973         u64 bytenr;
3974         u64 extent_start = 0;
3975         u64 extent_end = 0;
3976         u64 objectid = inode->i_ino;
3977         u32 found_type;
3978         struct btrfs_path *path = NULL;
3979         struct btrfs_root *root = BTRFS_I(inode)->root;
3980         struct btrfs_file_extent_item *item;
3981         struct extent_buffer *leaf;
3982         struct btrfs_key found_key;
3983         struct extent_map *em = NULL;
3984         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3985         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3986         struct btrfs_trans_handle *trans = NULL;
3987         int compressed;
3988
3989 again:
3990         spin_lock(&em_tree->lock);
3991         em = lookup_extent_mapping(em_tree, start, len);
3992         if (em)
3993                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3994         spin_unlock(&em_tree->lock);
3995
3996         if (em) {
3997                 if (em->start > start || em->start + em->len <= start)
3998                         free_extent_map(em);
3999                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4000                         free_extent_map(em);
4001                 else
4002                         goto out;
4003         }
4004         em = alloc_extent_map(GFP_NOFS);
4005         if (!em) {
4006                 err = -ENOMEM;
4007                 goto out;
4008         }
4009         em->bdev = root->fs_info->fs_devices->latest_bdev;
4010         em->start = EXTENT_MAP_HOLE;
4011         em->orig_start = EXTENT_MAP_HOLE;
4012         em->len = (u64)-1;
4013         em->block_len = (u64)-1;
4014
4015         if (!path) {
4016                 path = btrfs_alloc_path();
4017                 BUG_ON(!path);
4018         }
4019
4020         ret = btrfs_lookup_file_extent(trans, root, path,
4021                                        objectid, start, trans != NULL);
4022         if (ret < 0) {
4023                 err = ret;
4024                 goto out;
4025         }
4026
4027         if (ret != 0) {
4028                 if (path->slots[0] == 0)
4029                         goto not_found;
4030                 path->slots[0]--;
4031         }
4032
4033         leaf = path->nodes[0];
4034         item = btrfs_item_ptr(leaf, path->slots[0],
4035                               struct btrfs_file_extent_item);
4036         /* are we inside the extent that was found? */
4037         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4038         found_type = btrfs_key_type(&found_key);
4039         if (found_key.objectid != objectid ||
4040             found_type != BTRFS_EXTENT_DATA_KEY) {
4041                 goto not_found;
4042         }
4043
4044         found_type = btrfs_file_extent_type(leaf, item);
4045         extent_start = found_key.offset;
4046         compressed = btrfs_file_extent_compression(leaf, item);
4047         if (found_type == BTRFS_FILE_EXTENT_REG ||
4048             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4049                 extent_end = extent_start +
4050                        btrfs_file_extent_num_bytes(leaf, item);
4051         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4052                 size_t size;
4053                 size = btrfs_file_extent_inline_len(leaf, item);
4054                 extent_end = (extent_start + size + root->sectorsize - 1) &
4055                         ~((u64)root->sectorsize - 1);
4056         }
4057
4058         if (start >= extent_end) {
4059                 path->slots[0]++;
4060                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4061                         ret = btrfs_next_leaf(root, path);
4062                         if (ret < 0) {
4063                                 err = ret;
4064                                 goto out;
4065                         }
4066                         if (ret > 0)
4067                                 goto not_found;
4068                         leaf = path->nodes[0];
4069                 }
4070                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4071                 if (found_key.objectid != objectid ||
4072                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4073                         goto not_found;
4074                 if (start + len <= found_key.offset)
4075                         goto not_found;
4076                 em->start = start;
4077                 em->len = found_key.offset - start;
4078                 goto not_found_em;
4079         }
4080
4081         if (found_type == BTRFS_FILE_EXTENT_REG ||
4082             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4083                 em->start = extent_start;
4084                 em->len = extent_end - extent_start;
4085                 em->orig_start = extent_start -
4086                                  btrfs_file_extent_offset(leaf, item);
4087                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4088                 if (bytenr == 0) {
4089                         em->block_start = EXTENT_MAP_HOLE;
4090                         goto insert;
4091                 }
4092                 if (compressed) {
4093                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4094                         em->block_start = bytenr;
4095                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4096                                                                          item);
4097                 } else {
4098                         bytenr += btrfs_file_extent_offset(leaf, item);
4099                         em->block_start = bytenr;
4100                         em->block_len = em->len;
4101                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4102                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4103                 }
4104                 goto insert;
4105         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4106                 unsigned long ptr;
4107                 char *map;
4108                 size_t size;
4109                 size_t extent_offset;
4110                 size_t copy_size;
4111
4112                 em->block_start = EXTENT_MAP_INLINE;
4113                 if (!page || create) {
4114                         em->start = extent_start;
4115                         em->len = extent_end - extent_start;
4116                         goto out;
4117                 }
4118
4119                 size = btrfs_file_extent_inline_len(leaf, item);
4120                 extent_offset = page_offset(page) + pg_offset - extent_start;
4121                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4122                                 size - extent_offset);
4123                 em->start = extent_start + extent_offset;
4124                 em->len = (copy_size + root->sectorsize - 1) &
4125                         ~((u64)root->sectorsize - 1);
4126                 em->orig_start = EXTENT_MAP_INLINE;
4127                 if (compressed)
4128                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4129                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4130                 if (create == 0 && !PageUptodate(page)) {
4131                         if (btrfs_file_extent_compression(leaf, item) ==
4132                             BTRFS_COMPRESS_ZLIB) {
4133                                 ret = uncompress_inline(path, inode, page,
4134                                                         pg_offset,
4135                                                         extent_offset, item);
4136                                 BUG_ON(ret);
4137                         } else {
4138                                 map = kmap(page);
4139                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4140                                                    copy_size);
4141                                 kunmap(page);
4142                         }
4143                         flush_dcache_page(page);
4144                 } else if (create && PageUptodate(page)) {
4145                         if (!trans) {
4146                                 kunmap(page);
4147                                 free_extent_map(em);
4148                                 em = NULL;
4149                                 btrfs_release_path(root, path);
4150                                 trans = btrfs_join_transaction(root, 1);
4151                                 goto again;
4152                         }
4153                         map = kmap(page);
4154                         write_extent_buffer(leaf, map + pg_offset, ptr,
4155                                             copy_size);
4156                         kunmap(page);
4157                         btrfs_mark_buffer_dirty(leaf);
4158                 }
4159                 set_extent_uptodate(io_tree, em->start,
4160                                     extent_map_end(em) - 1, GFP_NOFS);
4161                 goto insert;
4162         } else {
4163                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4164                 WARN_ON(1);
4165         }
4166 not_found:
4167         em->start = start;
4168         em->len = len;
4169 not_found_em:
4170         em->block_start = EXTENT_MAP_HOLE;
4171         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4172 insert:
4173         btrfs_release_path(root, path);
4174         if (em->start > start || extent_map_end(em) <= start) {
4175                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4176                        "[%llu %llu]\n", (unsigned long long)em->start,
4177                        (unsigned long long)em->len,
4178                        (unsigned long long)start,
4179                        (unsigned long long)len);
4180                 err = -EIO;
4181                 goto out;
4182         }
4183
4184         err = 0;
4185         spin_lock(&em_tree->lock);
4186         ret = add_extent_mapping(em_tree, em);
4187         /* it is possible that someone inserted the extent into the tree
4188          * while we had the lock dropped.  It is also possible that
4189          * an overlapping map exists in the tree
4190          */
4191         if (ret == -EEXIST) {
4192                 struct extent_map *existing;
4193
4194                 ret = 0;
4195
4196                 existing = lookup_extent_mapping(em_tree, start, len);
4197                 if (existing && (existing->start > start ||
4198                     existing->start + existing->len <= start)) {
4199                         free_extent_map(existing);
4200                         existing = NULL;
4201                 }
4202                 if (!existing) {
4203                         existing = lookup_extent_mapping(em_tree, em->start,
4204                                                          em->len);
4205                         if (existing) {
4206                                 err = merge_extent_mapping(em_tree, existing,
4207                                                            em, start,
4208                                                            root->sectorsize);
4209                                 free_extent_map(existing);
4210                                 if (err) {
4211                                         free_extent_map(em);
4212                                         em = NULL;
4213                                 }
4214                         } else {
4215                                 err = -EIO;
4216                                 free_extent_map(em);
4217                                 em = NULL;
4218                         }
4219                 } else {
4220                         free_extent_map(em);
4221                         em = existing;
4222                         err = 0;
4223                 }
4224         }
4225         spin_unlock(&em_tree->lock);
4226 out:
4227         if (path)
4228                 btrfs_free_path(path);
4229         if (trans) {
4230                 ret = btrfs_end_transaction(trans, root);
4231                 if (!err)
4232                         err = ret;
4233         }
4234         if (err) {
4235                 free_extent_map(em);
4236                 WARN_ON(1);
4237                 return ERR_PTR(err);
4238         }
4239         return em;
4240 }
4241
4242 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4243                         const struct iovec *iov, loff_t offset,
4244                         unsigned long nr_segs)
4245 {
4246         return -EINVAL;
4247 }
4248
4249 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4250                 __u64 start, __u64 len)
4251 {
4252         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4253 }
4254
4255 int btrfs_readpage(struct file *file, struct page *page)
4256 {
4257         struct extent_io_tree *tree;
4258         tree = &BTRFS_I(page->mapping->host)->io_tree;
4259         return extent_read_full_page(tree, page, btrfs_get_extent);
4260 }
4261
4262 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4263 {
4264         struct extent_io_tree *tree;
4265
4266
4267         if (current->flags & PF_MEMALLOC) {
4268                 redirty_page_for_writepage(wbc, page);
4269                 unlock_page(page);
4270                 return 0;
4271         }
4272         tree = &BTRFS_I(page->mapping->host)->io_tree;
4273         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4274 }
4275
4276 int btrfs_writepages(struct address_space *mapping,
4277                      struct writeback_control *wbc)
4278 {
4279         struct extent_io_tree *tree;
4280
4281         tree = &BTRFS_I(mapping->host)->io_tree;
4282         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4283 }
4284
4285 static int
4286 btrfs_readpages(struct file *file, struct address_space *mapping,
4287                 struct list_head *pages, unsigned nr_pages)
4288 {
4289         struct extent_io_tree *tree;
4290         tree = &BTRFS_I(mapping->host)->io_tree;
4291         return extent_readpages(tree, mapping, pages, nr_pages,
4292                                 btrfs_get_extent);
4293 }
4294 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4295 {
4296         struct extent_io_tree *tree;
4297         struct extent_map_tree *map;
4298         int ret;
4299
4300         tree = &BTRFS_I(page->mapping->host)->io_tree;
4301         map = &BTRFS_I(page->mapping->host)->extent_tree;
4302         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4303         if (ret == 1) {
4304                 ClearPagePrivate(page);
4305                 set_page_private(page, 0);
4306                 page_cache_release(page);
4307         }
4308         return ret;
4309 }
4310
4311 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4312 {
4313         if (PageWriteback(page) || PageDirty(page))
4314                 return 0;
4315         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4316 }
4317
4318 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4319 {
4320         struct extent_io_tree *tree;
4321         struct btrfs_ordered_extent *ordered;
4322         u64 page_start = page_offset(page);
4323         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4324
4325         wait_on_page_writeback(page);
4326         tree = &BTRFS_I(page->mapping->host)->io_tree;
4327         if (offset) {
4328                 btrfs_releasepage(page, GFP_NOFS);
4329                 return;
4330         }
4331
4332         lock_extent(tree, page_start, page_end, GFP_NOFS);
4333         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4334                                            page_offset(page));
4335         if (ordered) {
4336                 /*
4337                  * IO on this page will never be started, so we need
4338                  * to account for any ordered extents now
4339                  */
4340                 clear_extent_bit(tree, page_start, page_end,
4341                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4342                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4343                 btrfs_finish_ordered_io(page->mapping->host,
4344                                         page_start, page_end);
4345                 btrfs_put_ordered_extent(ordered);
4346                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4347         }
4348         clear_extent_bit(tree, page_start, page_end,
4349                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4350                  EXTENT_ORDERED,
4351                  1, 1, GFP_NOFS);
4352         __btrfs_releasepage(page, GFP_NOFS);
4353
4354         ClearPageChecked(page);
4355         if (PagePrivate(page)) {
4356                 ClearPagePrivate(page);
4357                 set_page_private(page, 0);
4358                 page_cache_release(page);
4359         }
4360 }
4361
4362 /*
4363  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4364  * called from a page fault handler when a page is first dirtied. Hence we must
4365  * be careful to check for EOF conditions here. We set the page up correctly
4366  * for a written page which means we get ENOSPC checking when writing into
4367  * holes and correct delalloc and unwritten extent mapping on filesystems that
4368  * support these features.
4369  *
4370  * We are not allowed to take the i_mutex here so we have to play games to
4371  * protect against truncate races as the page could now be beyond EOF.  Because
4372  * vmtruncate() writes the inode size before removing pages, once we have the
4373  * page lock we can determine safely if the page is beyond EOF. If it is not
4374  * beyond EOF, then the page is guaranteed safe against truncation until we
4375  * unlock the page.
4376  */
4377 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4378 {
4379         struct page *page = vmf->page;
4380         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4381         struct btrfs_root *root = BTRFS_I(inode)->root;
4382         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4383         struct btrfs_ordered_extent *ordered;
4384         char *kaddr;
4385         unsigned long zero_start;
4386         loff_t size;
4387         int ret;
4388         u64 page_start;
4389         u64 page_end;
4390
4391         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4392         if (ret) {
4393                 if (ret == -ENOMEM)
4394                         ret = VM_FAULT_OOM;
4395                 else /* -ENOSPC, -EIO, etc */
4396                         ret = VM_FAULT_SIGBUS;
4397                 goto out;
4398         }
4399
4400         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4401 again:
4402         lock_page(page);
4403         size = i_size_read(inode);
4404         page_start = page_offset(page);
4405         page_end = page_start + PAGE_CACHE_SIZE - 1;
4406
4407         if ((page->mapping != inode->i_mapping) ||
4408             (page_start >= size)) {
4409                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4410                 /* page got truncated out from underneath us */
4411                 goto out_unlock;
4412         }
4413         wait_on_page_writeback(page);
4414
4415         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4416         set_page_extent_mapped(page);
4417
4418         /*
4419          * we can't set the delalloc bits if there are pending ordered
4420          * extents.  Drop our locks and wait for them to finish
4421          */
4422         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4423         if (ordered) {
4424                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4425                 unlock_page(page);
4426                 btrfs_start_ordered_extent(inode, ordered, 1);
4427                 btrfs_put_ordered_extent(ordered);
4428                 goto again;
4429         }
4430
4431         btrfs_set_extent_delalloc(inode, page_start, page_end);
4432         ret = 0;
4433
4434         /* page is wholly or partially inside EOF */
4435         if (page_start + PAGE_CACHE_SIZE > size)
4436                 zero_start = size & ~PAGE_CACHE_MASK;
4437         else
4438                 zero_start = PAGE_CACHE_SIZE;
4439
4440         if (zero_start != PAGE_CACHE_SIZE) {
4441                 kaddr = kmap(page);
4442                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4443                 flush_dcache_page(page);
4444                 kunmap(page);
4445         }
4446         ClearPageChecked(page);
4447         set_page_dirty(page);
4448
4449         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4450         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4451
4452 out_unlock:
4453         unlock_page(page);
4454 out:
4455         return ret;
4456 }
4457
4458 static void btrfs_truncate(struct inode *inode)
4459 {
4460         struct btrfs_root *root = BTRFS_I(inode)->root;
4461         int ret;
4462         struct btrfs_trans_handle *trans;
4463         unsigned long nr;
4464         u64 mask = root->sectorsize - 1;
4465
4466         if (!S_ISREG(inode->i_mode))
4467                 return;
4468         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4469                 return;
4470
4471         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4472         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4473
4474         trans = btrfs_start_transaction(root, 1);
4475
4476         /*
4477          * setattr is responsible for setting the ordered_data_close flag,
4478          * but that is only tested during the last file release.  That
4479          * could happen well after the next commit, leaving a great big
4480          * window where new writes may get lost if someone chooses to write
4481          * to this file after truncating to zero
4482          *
4483          * The inode doesn't have any dirty data here, and so if we commit
4484          * this is a noop.  If someone immediately starts writing to the inode
4485          * it is very likely we'll catch some of their writes in this
4486          * transaction, and the commit will find this file on the ordered
4487          * data list with good things to send down.
4488          *
4489          * This is a best effort solution, there is still a window where
4490          * using truncate to replace the contents of the file will
4491          * end up with a zero length file after a crash.
4492          */
4493         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4494                 btrfs_add_ordered_operation(trans, root, inode);
4495
4496         btrfs_set_trans_block_group(trans, inode);
4497         btrfs_i_size_write(inode, inode->i_size);
4498
4499         ret = btrfs_orphan_add(trans, inode);
4500         if (ret)
4501                 goto out;
4502         /* FIXME, add redo link to tree so we don't leak on crash */
4503         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4504                                       BTRFS_EXTENT_DATA_KEY);
4505         btrfs_update_inode(trans, root, inode);
4506
4507         ret = btrfs_orphan_del(trans, inode);
4508         BUG_ON(ret);
4509
4510 out:
4511         nr = trans->blocks_used;
4512         ret = btrfs_end_transaction_throttle(trans, root);
4513         BUG_ON(ret);
4514         btrfs_btree_balance_dirty(root, nr);
4515 }
4516
4517 /*
4518  * create a new subvolume directory/inode (helper for the ioctl).
4519  */
4520 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4521                              struct btrfs_root *new_root, struct dentry *dentry,
4522                              u64 new_dirid, u64 alloc_hint)
4523 {
4524         struct inode *inode;
4525         int error;
4526         u64 index = 0;
4527
4528         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4529                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4530         if (IS_ERR(inode))
4531                 return PTR_ERR(inode);
4532         inode->i_op = &btrfs_dir_inode_operations;
4533         inode->i_fop = &btrfs_dir_file_operations;
4534
4535         inode->i_nlink = 1;
4536         btrfs_i_size_write(inode, 0);
4537
4538         error = btrfs_update_inode(trans, new_root, inode);
4539         if (error)
4540                 return error;
4541
4542         d_instantiate(dentry, inode);
4543         return 0;
4544 }
4545
4546 /* helper function for file defrag and space balancing.  This
4547  * forces readahead on a given range of bytes in an inode
4548  */
4549 unsigned long btrfs_force_ra(struct address_space *mapping,
4550                               struct file_ra_state *ra, struct file *file,
4551                               pgoff_t offset, pgoff_t last_index)
4552 {
4553         pgoff_t req_size = last_index - offset + 1;
4554
4555         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4556         return offset + req_size;
4557 }
4558
4559 struct inode *btrfs_alloc_inode(struct super_block *sb)
4560 {
4561         struct btrfs_inode *ei;
4562
4563         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4564         if (!ei)
4565                 return NULL;
4566         ei->last_trans = 0;
4567         ei->logged_trans = 0;
4568         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4569         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4570         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4571         INIT_LIST_HEAD(&ei->i_orphan);
4572         INIT_LIST_HEAD(&ei->ordered_operations);
4573         return &ei->vfs_inode;
4574 }
4575
4576 void btrfs_destroy_inode(struct inode *inode)
4577 {
4578         struct btrfs_ordered_extent *ordered;
4579         struct btrfs_root *root = BTRFS_I(inode)->root;
4580
4581         WARN_ON(!list_empty(&inode->i_dentry));
4582         WARN_ON(inode->i_data.nrpages);
4583
4584         if (BTRFS_I(inode)->i_acl &&
4585             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4586                 posix_acl_release(BTRFS_I(inode)->i_acl);
4587         if (BTRFS_I(inode)->i_default_acl &&
4588             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4589                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4590
4591         /*
4592          * Make sure we're properly removed from the ordered operation
4593          * lists.
4594          */
4595         smp_mb();
4596         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4597                 spin_lock(&root->fs_info->ordered_extent_lock);
4598                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4599                 spin_unlock(&root->fs_info->ordered_extent_lock);
4600         }
4601
4602         spin_lock(&root->list_lock);
4603         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4604                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4605                        " list\n", inode->i_ino);
4606                 dump_stack();
4607         }
4608         spin_unlock(&root->list_lock);
4609
4610         while (1) {
4611                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4612                 if (!ordered)
4613                         break;
4614                 else {
4615                         printk(KERN_ERR "btrfs found ordered "
4616                                "extent %llu %llu on inode cleanup\n",
4617                                (unsigned long long)ordered->file_offset,
4618                                (unsigned long long)ordered->len);
4619                         btrfs_remove_ordered_extent(inode, ordered);
4620                         btrfs_put_ordered_extent(ordered);
4621                         btrfs_put_ordered_extent(ordered);
4622                 }
4623         }
4624         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4625         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4626 }
4627
4628 static void init_once(void *foo)
4629 {
4630         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4631
4632         inode_init_once(&ei->vfs_inode);
4633 }
4634
4635 void btrfs_destroy_cachep(void)
4636 {
4637         if (btrfs_inode_cachep)
4638                 kmem_cache_destroy(btrfs_inode_cachep);
4639         if (btrfs_trans_handle_cachep)
4640                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4641         if (btrfs_transaction_cachep)
4642                 kmem_cache_destroy(btrfs_transaction_cachep);
4643         if (btrfs_path_cachep)
4644                 kmem_cache_destroy(btrfs_path_cachep);
4645 }
4646
4647 int btrfs_init_cachep(void)
4648 {
4649         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4650                         sizeof(struct btrfs_inode), 0,
4651                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4652         if (!btrfs_inode_cachep)
4653                 goto fail;
4654
4655         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4656                         sizeof(struct btrfs_trans_handle), 0,
4657                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4658         if (!btrfs_trans_handle_cachep)
4659                 goto fail;
4660
4661         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4662                         sizeof(struct btrfs_transaction), 0,
4663                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4664         if (!btrfs_transaction_cachep)
4665                 goto fail;
4666
4667         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4668                         sizeof(struct btrfs_path), 0,
4669                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4670         if (!btrfs_path_cachep)
4671                 goto fail;
4672
4673         return 0;
4674 fail:
4675         btrfs_destroy_cachep();
4676         return -ENOMEM;
4677 }
4678
4679 static int btrfs_getattr(struct vfsmount *mnt,
4680                          struct dentry *dentry, struct kstat *stat)
4681 {
4682         struct inode *inode = dentry->d_inode;
4683         generic_fillattr(inode, stat);
4684         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4685         stat->blksize = PAGE_CACHE_SIZE;
4686         stat->blocks = (inode_get_bytes(inode) +
4687                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4688         return 0;
4689 }
4690
4691 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4692                            struct inode *new_dir, struct dentry *new_dentry)
4693 {
4694         struct btrfs_trans_handle *trans;
4695         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4696         struct inode *new_inode = new_dentry->d_inode;
4697         struct inode *old_inode = old_dentry->d_inode;
4698         struct timespec ctime = CURRENT_TIME;
4699         u64 index = 0;
4700         int ret;
4701
4702         /* we're not allowed to rename between subvolumes */
4703         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4704             BTRFS_I(new_dir)->root->root_key.objectid)
4705                 return -EXDEV;
4706
4707         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4708             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4709                 return -ENOTEMPTY;
4710         }
4711
4712         /* to rename a snapshot or subvolume, we need to juggle the
4713          * backrefs.  This isn't coded yet
4714          */
4715         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4716                 return -EXDEV;
4717
4718         ret = btrfs_check_metadata_free_space(root);
4719         if (ret)
4720                 goto out_unlock;
4721
4722         /*
4723          * we're using rename to replace one file with another.
4724          * and the replacement file is large.  Start IO on it now so
4725          * we don't add too much work to the end of the transaction
4726          */
4727         if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4728             new_inode->i_size &&
4729             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4730                 filemap_flush(old_inode->i_mapping);
4731
4732         trans = btrfs_start_transaction(root, 1);
4733
4734         /*
4735          * make sure the inode gets flushed if it is replacing
4736          * something.
4737          */
4738         if (new_inode && new_inode->i_size &&
4739             old_inode && S_ISREG(old_inode->i_mode)) {
4740                 btrfs_add_ordered_operation(trans, root, old_inode);
4741         }
4742
4743         /*
4744          * this is an ugly little race, but the rename is required to make
4745          * sure that if we crash, the inode is either at the old name
4746          * or the new one.  pinning the log transaction lets us make sure
4747          * we don't allow a log commit to come in after we unlink the
4748          * name but before we add the new name back in.
4749          */
4750         btrfs_pin_log_trans(root);
4751
4752         btrfs_set_trans_block_group(trans, new_dir);
4753
4754         btrfs_inc_nlink(old_dentry->d_inode);
4755         old_dir->i_ctime = old_dir->i_mtime = ctime;
4756         new_dir->i_ctime = new_dir->i_mtime = ctime;
4757         old_inode->i_ctime = ctime;
4758
4759         if (old_dentry->d_parent != new_dentry->d_parent)
4760                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4761
4762         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4763                                  old_dentry->d_name.name,
4764                                  old_dentry->d_name.len);
4765         if (ret)
4766                 goto out_fail;
4767
4768         if (new_inode) {
4769                 new_inode->i_ctime = CURRENT_TIME;
4770                 ret = btrfs_unlink_inode(trans, root, new_dir,
4771                                          new_dentry->d_inode,
4772                                          new_dentry->d_name.name,
4773                                          new_dentry->d_name.len);
4774                 if (ret)
4775                         goto out_fail;
4776                 if (new_inode->i_nlink == 0) {
4777                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4778                         if (ret)
4779                                 goto out_fail;
4780                 }
4781
4782         }
4783         ret = btrfs_set_inode_index(new_dir, &index);
4784         if (ret)
4785                 goto out_fail;
4786
4787         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4788                              old_inode, new_dentry->d_name.name,
4789                              new_dentry->d_name.len, 1, index);
4790         if (ret)
4791                 goto out_fail;
4792
4793         btrfs_log_new_name(trans, old_inode, old_dir,
4794                                        new_dentry->d_parent);
4795 out_fail:
4796
4797         /* this btrfs_end_log_trans just allows the current
4798          * log-sub transaction to complete
4799          */
4800         btrfs_end_log_trans(root);
4801         btrfs_end_transaction_throttle(trans, root);
4802 out_unlock:
4803         return ret;
4804 }
4805
4806 /*
4807  * some fairly slow code that needs optimization. This walks the list
4808  * of all the inodes with pending delalloc and forces them to disk.
4809  */
4810 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4811 {
4812         struct list_head *head = &root->fs_info->delalloc_inodes;
4813         struct btrfs_inode *binode;
4814         struct inode *inode;
4815
4816         if (root->fs_info->sb->s_flags & MS_RDONLY)
4817                 return -EROFS;
4818
4819         spin_lock(&root->fs_info->delalloc_lock);
4820         while (!list_empty(head)) {
4821                 binode = list_entry(head->next, struct btrfs_inode,
4822                                     delalloc_inodes);
4823                 inode = igrab(&binode->vfs_inode);
4824                 if (!inode)
4825                         list_del_init(&binode->delalloc_inodes);
4826                 spin_unlock(&root->fs_info->delalloc_lock);
4827                 if (inode) {
4828                         filemap_flush(inode->i_mapping);
4829                         iput(inode);
4830                 }
4831                 cond_resched();
4832                 spin_lock(&root->fs_info->delalloc_lock);
4833         }
4834         spin_unlock(&root->fs_info->delalloc_lock);
4835
4836         /* the filemap_flush will queue IO into the worker threads, but
4837          * we have to make sure the IO is actually started and that
4838          * ordered extents get created before we return
4839          */
4840         atomic_inc(&root->fs_info->async_submit_draining);
4841         while (atomic_read(&root->fs_info->nr_async_submits) ||
4842               atomic_read(&root->fs_info->async_delalloc_pages)) {
4843                 wait_event(root->fs_info->async_submit_wait,
4844                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4845                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4846         }
4847         atomic_dec(&root->fs_info->async_submit_draining);
4848         return 0;
4849 }
4850
4851 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4852                          const char *symname)
4853 {
4854         struct btrfs_trans_handle *trans;
4855         struct btrfs_root *root = BTRFS_I(dir)->root;
4856         struct btrfs_path *path;
4857         struct btrfs_key key;
4858         struct inode *inode = NULL;
4859         int err;
4860         int drop_inode = 0;
4861         u64 objectid;
4862         u64 index = 0 ;
4863         int name_len;
4864         int datasize;
4865         unsigned long ptr;
4866         struct btrfs_file_extent_item *ei;
4867         struct extent_buffer *leaf;
4868         unsigned long nr = 0;
4869
4870         name_len = strlen(symname) + 1;
4871         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4872                 return -ENAMETOOLONG;
4873
4874         err = btrfs_check_metadata_free_space(root);
4875         if (err)
4876                 goto out_fail;
4877
4878         trans = btrfs_start_transaction(root, 1);
4879         btrfs_set_trans_block_group(trans, dir);
4880
4881         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4882         if (err) {
4883                 err = -ENOSPC;
4884                 goto out_unlock;
4885         }
4886
4887         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4888                                 dentry->d_name.len,
4889                                 dentry->d_parent->d_inode->i_ino, objectid,
4890                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4891                                 &index);
4892         err = PTR_ERR(inode);
4893         if (IS_ERR(inode))
4894                 goto out_unlock;
4895
4896         err = btrfs_init_inode_security(inode, dir);
4897         if (err) {
4898                 drop_inode = 1;
4899                 goto out_unlock;
4900         }
4901
4902         btrfs_set_trans_block_group(trans, inode);
4903         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4904         if (err)
4905                 drop_inode = 1;
4906         else {
4907                 inode->i_mapping->a_ops = &btrfs_aops;
4908                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4909                 inode->i_fop = &btrfs_file_operations;
4910                 inode->i_op = &btrfs_file_inode_operations;
4911                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4912         }
4913         dir->i_sb->s_dirt = 1;
4914         btrfs_update_inode_block_group(trans, inode);
4915         btrfs_update_inode_block_group(trans, dir);
4916         if (drop_inode)
4917                 goto out_unlock;
4918
4919         path = btrfs_alloc_path();
4920         BUG_ON(!path);
4921         key.objectid = inode->i_ino;
4922         key.offset = 0;
4923         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4924         datasize = btrfs_file_extent_calc_inline_size(name_len);
4925         err = btrfs_insert_empty_item(trans, root, path, &key,
4926                                       datasize);
4927         if (err) {
4928                 drop_inode = 1;
4929                 goto out_unlock;
4930         }
4931         leaf = path->nodes[0];
4932         ei = btrfs_item_ptr(leaf, path->slots[0],
4933                             struct btrfs_file_extent_item);
4934         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4935         btrfs_set_file_extent_type(leaf, ei,
4936                                    BTRFS_FILE_EXTENT_INLINE);
4937         btrfs_set_file_extent_encryption(leaf, ei, 0);
4938         btrfs_set_file_extent_compression(leaf, ei, 0);
4939         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4940         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4941
4942         ptr = btrfs_file_extent_inline_start(ei);
4943         write_extent_buffer(leaf, symname, ptr, name_len);
4944         btrfs_mark_buffer_dirty(leaf);
4945         btrfs_free_path(path);
4946
4947         inode->i_op = &btrfs_symlink_inode_operations;
4948         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4949         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4950         inode_set_bytes(inode, name_len);
4951         btrfs_i_size_write(inode, name_len - 1);
4952         err = btrfs_update_inode(trans, root, inode);
4953         if (err)
4954                 drop_inode = 1;
4955
4956 out_unlock:
4957         nr = trans->blocks_used;
4958         btrfs_end_transaction_throttle(trans, root);
4959 out_fail:
4960         if (drop_inode) {
4961                 inode_dec_link_count(inode);
4962                 iput(inode);
4963         }
4964         btrfs_btree_balance_dirty(root, nr);
4965         return err;
4966 }
4967
4968 static int prealloc_file_range(struct btrfs_trans_handle *trans,
4969                                struct inode *inode, u64 start, u64 end,
4970                                u64 locked_end, u64 alloc_hint, int mode)
4971 {
4972         struct btrfs_root *root = BTRFS_I(inode)->root;
4973         struct btrfs_key ins;
4974         u64 alloc_size;
4975         u64 cur_offset = start;
4976         u64 num_bytes = end - start;
4977         int ret = 0;
4978
4979         while (num_bytes > 0) {
4980                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4981                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4982                                            root->sectorsize, 0, alloc_hint,
4983                                            (u64)-1, &ins, 1);
4984                 if (ret) {
4985                         WARN_ON(1);
4986                         goto out;
4987                 }
4988                 ret = insert_reserved_file_extent(trans, inode,
4989                                                   cur_offset, ins.objectid,
4990                                                   ins.offset, ins.offset,
4991                                                   ins.offset, locked_end,
4992                                                   0, 0, 0,
4993                                                   BTRFS_FILE_EXTENT_PREALLOC);
4994                 BUG_ON(ret);
4995                 num_bytes -= ins.offset;
4996                 cur_offset += ins.offset;
4997                 alloc_hint = ins.objectid + ins.offset;
4998         }
4999 out:
5000         if (cur_offset > start) {
5001                 inode->i_ctime = CURRENT_TIME;
5002                 btrfs_set_flag(inode, PREALLOC);
5003                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5004                     cur_offset > i_size_read(inode))
5005                         btrfs_i_size_write(inode, cur_offset);
5006                 ret = btrfs_update_inode(trans, root, inode);
5007                 BUG_ON(ret);
5008         }
5009
5010         return ret;
5011 }
5012
5013 static long btrfs_fallocate(struct inode *inode, int mode,
5014                             loff_t offset, loff_t len)
5015 {
5016         u64 cur_offset;
5017         u64 last_byte;
5018         u64 alloc_start;
5019         u64 alloc_end;
5020         u64 alloc_hint = 0;
5021         u64 locked_end;
5022         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5023         struct extent_map *em;
5024         struct btrfs_trans_handle *trans;
5025         int ret;
5026
5027         alloc_start = offset & ~mask;
5028         alloc_end =  (offset + len + mask) & ~mask;
5029
5030         /*
5031          * wait for ordered IO before we have any locks.  We'll loop again
5032          * below with the locks held.
5033          */
5034         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5035
5036         mutex_lock(&inode->i_mutex);
5037         if (alloc_start > inode->i_size) {
5038                 ret = btrfs_cont_expand(inode, alloc_start);
5039                 if (ret)
5040                         goto out;
5041         }
5042
5043         locked_end = alloc_end - 1;
5044         while (1) {
5045                 struct btrfs_ordered_extent *ordered;
5046
5047                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5048                 if (!trans) {
5049                         ret = -EIO;
5050                         goto out;
5051                 }
5052
5053                 /* the extent lock is ordered inside the running
5054                  * transaction
5055                  */
5056                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5057                             GFP_NOFS);
5058                 ordered = btrfs_lookup_first_ordered_extent(inode,
5059                                                             alloc_end - 1);
5060                 if (ordered &&
5061                     ordered->file_offset + ordered->len > alloc_start &&
5062                     ordered->file_offset < alloc_end) {
5063                         btrfs_put_ordered_extent(ordered);
5064                         unlock_extent(&BTRFS_I(inode)->io_tree,
5065                                       alloc_start, locked_end, GFP_NOFS);
5066                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5067
5068                         /*
5069                          * we can't wait on the range with the transaction
5070                          * running or with the extent lock held
5071                          */
5072                         btrfs_wait_ordered_range(inode, alloc_start,
5073                                                  alloc_end - alloc_start);
5074                 } else {
5075                         if (ordered)
5076                                 btrfs_put_ordered_extent(ordered);
5077                         break;
5078                 }
5079         }
5080
5081         cur_offset = alloc_start;
5082         while (1) {
5083                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5084                                       alloc_end - cur_offset, 0);
5085                 BUG_ON(IS_ERR(em) || !em);
5086                 last_byte = min(extent_map_end(em), alloc_end);
5087                 last_byte = (last_byte + mask) & ~mask;
5088                 if (em->block_start == EXTENT_MAP_HOLE) {
5089                         ret = prealloc_file_range(trans, inode, cur_offset,
5090                                         last_byte, locked_end + 1,
5091                                         alloc_hint, mode);
5092                         if (ret < 0) {
5093                                 free_extent_map(em);
5094                                 break;
5095                         }
5096                 }
5097                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5098                         alloc_hint = em->block_start;
5099                 free_extent_map(em);
5100
5101                 cur_offset = last_byte;
5102                 if (cur_offset >= alloc_end) {
5103                         ret = 0;
5104                         break;
5105                 }
5106         }
5107         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5108                       GFP_NOFS);
5109
5110         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5111 out:
5112         mutex_unlock(&inode->i_mutex);
5113         return ret;
5114 }
5115
5116 static int btrfs_set_page_dirty(struct page *page)
5117 {
5118         return __set_page_dirty_nobuffers(page);
5119 }
5120
5121 static int btrfs_permission(struct inode *inode, int mask)
5122 {
5123         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
5124                 return -EACCES;
5125         return generic_permission(inode, mask, btrfs_check_acl);
5126 }
5127
5128 static struct inode_operations btrfs_dir_inode_operations = {
5129         .getattr        = btrfs_getattr,
5130         .lookup         = btrfs_lookup,
5131         .create         = btrfs_create,
5132         .unlink         = btrfs_unlink,
5133         .link           = btrfs_link,
5134         .mkdir          = btrfs_mkdir,
5135         .rmdir          = btrfs_rmdir,
5136         .rename         = btrfs_rename,
5137         .symlink        = btrfs_symlink,
5138         .setattr        = btrfs_setattr,
5139         .mknod          = btrfs_mknod,
5140         .setxattr       = btrfs_setxattr,
5141         .getxattr       = btrfs_getxattr,
5142         .listxattr      = btrfs_listxattr,
5143         .removexattr    = btrfs_removexattr,
5144         .permission     = btrfs_permission,
5145 };
5146 static struct inode_operations btrfs_dir_ro_inode_operations = {
5147         .lookup         = btrfs_lookup,
5148         .permission     = btrfs_permission,
5149 };
5150 static struct file_operations btrfs_dir_file_operations = {
5151         .llseek         = generic_file_llseek,
5152         .read           = generic_read_dir,
5153         .readdir        = btrfs_real_readdir,
5154         .unlocked_ioctl = btrfs_ioctl,
5155 #ifdef CONFIG_COMPAT
5156         .compat_ioctl   = btrfs_ioctl,
5157 #endif
5158         .release        = btrfs_release_file,
5159         .fsync          = btrfs_sync_file,
5160 };
5161
5162 static struct extent_io_ops btrfs_extent_io_ops = {
5163         .fill_delalloc = run_delalloc_range,
5164         .submit_bio_hook = btrfs_submit_bio_hook,
5165         .merge_bio_hook = btrfs_merge_bio_hook,
5166         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5167         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5168         .writepage_start_hook = btrfs_writepage_start_hook,
5169         .readpage_io_failed_hook = btrfs_io_failed_hook,
5170         .set_bit_hook = btrfs_set_bit_hook,
5171         .clear_bit_hook = btrfs_clear_bit_hook,
5172 };
5173
5174 /*
5175  * btrfs doesn't support the bmap operation because swapfiles
5176  * use bmap to make a mapping of extents in the file.  They assume
5177  * these extents won't change over the life of the file and they
5178  * use the bmap result to do IO directly to the drive.
5179  *
5180  * the btrfs bmap call would return logical addresses that aren't
5181  * suitable for IO and they also will change frequently as COW
5182  * operations happen.  So, swapfile + btrfs == corruption.
5183  *
5184  * For now we're avoiding this by dropping bmap.
5185  */
5186 static struct address_space_operations btrfs_aops = {
5187         .readpage       = btrfs_readpage,
5188         .writepage      = btrfs_writepage,
5189         .writepages     = btrfs_writepages,
5190         .readpages      = btrfs_readpages,
5191         .sync_page      = block_sync_page,
5192         .direct_IO      = btrfs_direct_IO,
5193         .invalidatepage = btrfs_invalidatepage,
5194         .releasepage    = btrfs_releasepage,
5195         .set_page_dirty = btrfs_set_page_dirty,
5196 };
5197
5198 static struct address_space_operations btrfs_symlink_aops = {
5199         .readpage       = btrfs_readpage,
5200         .writepage      = btrfs_writepage,
5201         .invalidatepage = btrfs_invalidatepage,
5202         .releasepage    = btrfs_releasepage,
5203 };
5204
5205 static struct inode_operations btrfs_file_inode_operations = {
5206         .truncate       = btrfs_truncate,
5207         .getattr        = btrfs_getattr,
5208         .setattr        = btrfs_setattr,
5209         .setxattr       = btrfs_setxattr,
5210         .getxattr       = btrfs_getxattr,
5211         .listxattr      = btrfs_listxattr,
5212         .removexattr    = btrfs_removexattr,
5213         .permission     = btrfs_permission,
5214         .fallocate      = btrfs_fallocate,
5215         .fiemap         = btrfs_fiemap,
5216 };
5217 static struct inode_operations btrfs_special_inode_operations = {
5218         .getattr        = btrfs_getattr,
5219         .setattr        = btrfs_setattr,
5220         .permission     = btrfs_permission,
5221         .setxattr       = btrfs_setxattr,
5222         .getxattr       = btrfs_getxattr,
5223         .listxattr      = btrfs_listxattr,
5224         .removexattr    = btrfs_removexattr,
5225 };
5226 static struct inode_operations btrfs_symlink_inode_operations = {
5227         .readlink       = generic_readlink,
5228         .follow_link    = page_follow_link_light,
5229         .put_link       = page_put_link,
5230         .permission     = btrfs_permission,
5231         .setxattr       = btrfs_setxattr,
5232         .getxattr       = btrfs_getxattr,
5233         .listxattr      = btrfs_listxattr,
5234         .removexattr    = btrfs_removexattr,
5235 };