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  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2020  * don't find any xattrs, we know there can't be any acls.
2021  *
2022  * slot is the slot the inode is in, objectid is the objectid of the inode
2023  */
2024 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2025                                           int slot, u64 objectid)
2026 {
2027         u32 nritems = btrfs_header_nritems(leaf);
2028         struct btrfs_key found_key;
2029         int scanned = 0;
2030
2031         slot++;
2032         while (slot < nritems) {
2033                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2034
2035                 /* we found a different objectid, there must not be acls */
2036                 if (found_key.objectid != objectid)
2037                         return 0;
2038
2039                 /* we found an xattr, assume we've got an acl */
2040                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2041                         return 1;
2042
2043                 /*
2044                  * we found a key greater than an xattr key, there can't
2045                  * be any acls later on
2046                  */
2047                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2048                         return 0;
2049
2050                 slot++;
2051                 scanned++;
2052
2053                 /*
2054                  * it goes inode, inode backrefs, xattrs, extents,
2055                  * so if there are a ton of hard links to an inode there can
2056                  * be a lot of backrefs.  Don't waste time searching too hard,
2057                  * this is just an optimization
2058                  */
2059                 if (scanned >= 8)
2060                         break;
2061         }
2062         /* we hit the end of the leaf before we found an xattr or
2063          * something larger than an xattr.  We have to assume the inode
2064          * has acls
2065          */
2066         return 1;
2067 }
2068
2069 /*
2070  * read an inode from the btree into the in-memory inode
2071  */
2072 void btrfs_read_locked_inode(struct inode *inode)
2073 {
2074         struct btrfs_path *path;
2075         struct extent_buffer *leaf;
2076         struct btrfs_inode_item *inode_item;
2077         struct btrfs_timespec *tspec;
2078         struct btrfs_root *root = BTRFS_I(inode)->root;
2079         struct btrfs_key location;
2080         int maybe_acls;
2081         u64 alloc_group_block;
2082         u32 rdev;
2083         int ret;
2084
2085         path = btrfs_alloc_path();
2086         BUG_ON(!path);
2087         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2088
2089         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2090         if (ret)
2091                 goto make_bad;
2092
2093         leaf = path->nodes[0];
2094         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2095                                     struct btrfs_inode_item);
2096
2097         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2098         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2099         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2100         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2101         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2102
2103         tspec = btrfs_inode_atime(inode_item);
2104         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2105         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2106
2107         tspec = btrfs_inode_mtime(inode_item);
2108         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2109         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2110
2111         tspec = btrfs_inode_ctime(inode_item);
2112         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2113         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2114
2115         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2116         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2117         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2118         inode->i_generation = BTRFS_I(inode)->generation;
2119         inode->i_rdev = 0;
2120         rdev = btrfs_inode_rdev(leaf, inode_item);
2121
2122         BTRFS_I(inode)->index_cnt = (u64)-1;
2123         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2124
2125         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2126
2127         /*
2128          * try to precache a NULL acl entry for files that don't have
2129          * any xattrs or acls
2130          */
2131         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2132         if (!maybe_acls) {
2133                 BTRFS_I(inode)->i_acl = NULL;
2134                 BTRFS_I(inode)->i_default_acl = NULL;
2135         }
2136
2137         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2138                                                 alloc_group_block, 0);
2139         btrfs_free_path(path);
2140         inode_item = NULL;
2141
2142         switch (inode->i_mode & S_IFMT) {
2143         case S_IFREG:
2144                 inode->i_mapping->a_ops = &btrfs_aops;
2145                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2146                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2147                 inode->i_fop = &btrfs_file_operations;
2148                 inode->i_op = &btrfs_file_inode_operations;
2149                 break;
2150         case S_IFDIR:
2151                 inode->i_fop = &btrfs_dir_file_operations;
2152                 if (root == root->fs_info->tree_root)
2153                         inode->i_op = &btrfs_dir_ro_inode_operations;
2154                 else
2155                         inode->i_op = &btrfs_dir_inode_operations;
2156                 break;
2157         case S_IFLNK:
2158                 inode->i_op = &btrfs_symlink_inode_operations;
2159                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2160                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2161                 break;
2162         default:
2163                 inode->i_op = &btrfs_special_inode_operations;
2164                 init_special_inode(inode, inode->i_mode, rdev);
2165                 break;
2166         }
2167         return;
2168
2169 make_bad:
2170         btrfs_free_path(path);
2171         make_bad_inode(inode);
2172 }
2173
2174 /*
2175  * given a leaf and an inode, copy the inode fields into the leaf
2176  */
2177 static void fill_inode_item(struct btrfs_trans_handle *trans,
2178                             struct extent_buffer *leaf,
2179                             struct btrfs_inode_item *item,
2180                             struct inode *inode)
2181 {
2182         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2183         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2184         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2185         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2186         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2187
2188         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2189                                inode->i_atime.tv_sec);
2190         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2191                                 inode->i_atime.tv_nsec);
2192
2193         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2194                                inode->i_mtime.tv_sec);
2195         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2196                                 inode->i_mtime.tv_nsec);
2197
2198         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2199                                inode->i_ctime.tv_sec);
2200         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2201                                 inode->i_ctime.tv_nsec);
2202
2203         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2204         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2205         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2206         btrfs_set_inode_transid(leaf, item, trans->transid);
2207         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2208         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2209         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2210 }
2211
2212 /*
2213  * copy everything in the in-memory inode into the btree.
2214  */
2215 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2216                                 struct btrfs_root *root, struct inode *inode)
2217 {
2218         struct btrfs_inode_item *inode_item;
2219         struct btrfs_path *path;
2220         struct extent_buffer *leaf;
2221         int ret;
2222
2223         path = btrfs_alloc_path();
2224         BUG_ON(!path);
2225         path->leave_spinning = 1;
2226         ret = btrfs_lookup_inode(trans, root, path,
2227                                  &BTRFS_I(inode)->location, 1);
2228         if (ret) {
2229                 if (ret > 0)
2230                         ret = -ENOENT;
2231                 goto failed;
2232         }
2233
2234         btrfs_unlock_up_safe(path, 1);
2235         leaf = path->nodes[0];
2236         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2237                                   struct btrfs_inode_item);
2238
2239         fill_inode_item(trans, leaf, inode_item, inode);
2240         btrfs_mark_buffer_dirty(leaf);
2241         btrfs_set_inode_last_trans(trans, inode);
2242         ret = 0;
2243 failed:
2244         btrfs_free_path(path);
2245         return ret;
2246 }
2247
2248
2249 /*
2250  * unlink helper that gets used here in inode.c and in the tree logging
2251  * recovery code.  It remove a link in a directory with a given name, and
2252  * also drops the back refs in the inode to the directory
2253  */
2254 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2255                        struct btrfs_root *root,
2256                        struct inode *dir, struct inode *inode,
2257                        const char *name, int name_len)
2258 {
2259         struct btrfs_path *path;
2260         int ret = 0;
2261         struct extent_buffer *leaf;
2262         struct btrfs_dir_item *di;
2263         struct btrfs_key key;
2264         u64 index;
2265
2266         path = btrfs_alloc_path();
2267         if (!path) {
2268                 ret = -ENOMEM;
2269                 goto err;
2270         }
2271
2272         path->leave_spinning = 1;
2273         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2274                                     name, name_len, -1);
2275         if (IS_ERR(di)) {
2276                 ret = PTR_ERR(di);
2277                 goto err;
2278         }
2279         if (!di) {
2280                 ret = -ENOENT;
2281                 goto err;
2282         }
2283         leaf = path->nodes[0];
2284         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2285         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2286         if (ret)
2287                 goto err;
2288         btrfs_release_path(root, path);
2289
2290         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2291                                   inode->i_ino,
2292                                   dir->i_ino, &index);
2293         if (ret) {
2294                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2295                        "inode %lu parent %lu\n", name_len, name,
2296                        inode->i_ino, dir->i_ino);
2297                 goto err;
2298         }
2299
2300         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2301                                          index, name, name_len, -1);
2302         if (IS_ERR(di)) {
2303                 ret = PTR_ERR(di);
2304                 goto err;
2305         }
2306         if (!di) {
2307                 ret = -ENOENT;
2308                 goto err;
2309         }
2310         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2311         btrfs_release_path(root, path);
2312
2313         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2314                                          inode, dir->i_ino);
2315         BUG_ON(ret != 0 && ret != -ENOENT);
2316
2317         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2318                                            dir, index);
2319         BUG_ON(ret);
2320 err:
2321         btrfs_free_path(path);
2322         if (ret)
2323                 goto out;
2324
2325         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2326         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2327         btrfs_update_inode(trans, root, dir);
2328         btrfs_drop_nlink(inode);
2329         ret = btrfs_update_inode(trans, root, inode);
2330         dir->i_sb->s_dirt = 1;
2331 out:
2332         return ret;
2333 }
2334
2335 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2336 {
2337         struct btrfs_root *root;
2338         struct btrfs_trans_handle *trans;
2339         struct inode *inode = dentry->d_inode;
2340         int ret;
2341         unsigned long nr = 0;
2342
2343         root = BTRFS_I(dir)->root;
2344
2345         trans = btrfs_start_transaction(root, 1);
2346
2347         btrfs_set_trans_block_group(trans, dir);
2348
2349         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2350
2351         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2352                                  dentry->d_name.name, dentry->d_name.len);
2353
2354         if (inode->i_nlink == 0)
2355                 ret = btrfs_orphan_add(trans, inode);
2356
2357         nr = trans->blocks_used;
2358
2359         btrfs_end_transaction_throttle(trans, root);
2360         btrfs_btree_balance_dirty(root, nr);
2361         return ret;
2362 }
2363
2364 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2365 {
2366         struct inode *inode = dentry->d_inode;
2367         int err = 0;
2368         int ret;
2369         struct btrfs_root *root = BTRFS_I(dir)->root;
2370         struct btrfs_trans_handle *trans;
2371         unsigned long nr = 0;
2372
2373         /*
2374          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2375          * the root of a subvolume or snapshot
2376          */
2377         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2378             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2379                 return -ENOTEMPTY;
2380         }
2381
2382         trans = btrfs_start_transaction(root, 1);
2383         btrfs_set_trans_block_group(trans, dir);
2384
2385         err = btrfs_orphan_add(trans, inode);
2386         if (err)
2387                 goto fail_trans;
2388
2389         /* now the directory is empty */
2390         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2391                                  dentry->d_name.name, dentry->d_name.len);
2392         if (!err)
2393                 btrfs_i_size_write(inode, 0);
2394
2395 fail_trans:
2396         nr = trans->blocks_used;
2397         ret = btrfs_end_transaction_throttle(trans, root);
2398         btrfs_btree_balance_dirty(root, nr);
2399
2400         if (ret && !err)
2401                 err = ret;
2402         return err;
2403 }
2404
2405 #if 0
2406 /*
2407  * when truncating bytes in a file, it is possible to avoid reading
2408  * the leaves that contain only checksum items.  This can be the
2409  * majority of the IO required to delete a large file, but it must
2410  * be done carefully.
2411  *
2412  * The keys in the level just above the leaves are checked to make sure
2413  * the lowest key in a given leaf is a csum key, and starts at an offset
2414  * after the new  size.
2415  *
2416  * Then the key for the next leaf is checked to make sure it also has
2417  * a checksum item for the same file.  If it does, we know our target leaf
2418  * contains only checksum items, and it can be safely freed without reading
2419  * it.
2420  *
2421  * This is just an optimization targeted at large files.  It may do
2422  * nothing.  It will return 0 unless things went badly.
2423  */
2424 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2425                                      struct btrfs_root *root,
2426                                      struct btrfs_path *path,
2427                                      struct inode *inode, u64 new_size)
2428 {
2429         struct btrfs_key key;
2430         int ret;
2431         int nritems;
2432         struct btrfs_key found_key;
2433         struct btrfs_key other_key;
2434         struct btrfs_leaf_ref *ref;
2435         u64 leaf_gen;
2436         u64 leaf_start;
2437
2438         path->lowest_level = 1;
2439         key.objectid = inode->i_ino;
2440         key.type = BTRFS_CSUM_ITEM_KEY;
2441         key.offset = new_size;
2442 again:
2443         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2444         if (ret < 0)
2445                 goto out;
2446
2447         if (path->nodes[1] == NULL) {
2448                 ret = 0;
2449                 goto out;
2450         }
2451         ret = 0;
2452         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2453         nritems = btrfs_header_nritems(path->nodes[1]);
2454
2455         if (!nritems)
2456                 goto out;
2457
2458         if (path->slots[1] >= nritems)
2459                 goto next_node;
2460
2461         /* did we find a key greater than anything we want to delete? */
2462         if (found_key.objectid > inode->i_ino ||
2463            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2464                 goto out;
2465
2466         /* we check the next key in the node to make sure the leave contains
2467          * only checksum items.  This comparison doesn't work if our
2468          * leaf is the last one in the node
2469          */
2470         if (path->slots[1] + 1 >= nritems) {
2471 next_node:
2472                 /* search forward from the last key in the node, this
2473                  * will bring us into the next node in the tree
2474                  */
2475                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2476
2477                 /* unlikely, but we inc below, so check to be safe */
2478                 if (found_key.offset == (u64)-1)
2479                         goto out;
2480
2481                 /* search_forward needs a path with locks held, do the
2482                  * search again for the original key.  It is possible
2483                  * this will race with a balance and return a path that
2484                  * we could modify, but this drop is just an optimization
2485                  * and is allowed to miss some leaves.
2486                  */
2487                 btrfs_release_path(root, path);
2488                 found_key.offset++;
2489
2490                 /* setup a max key for search_forward */
2491                 other_key.offset = (u64)-1;
2492                 other_key.type = key.type;
2493                 other_key.objectid = key.objectid;
2494
2495                 path->keep_locks = 1;
2496                 ret = btrfs_search_forward(root, &found_key, &other_key,
2497                                            path, 0, 0);
2498                 path->keep_locks = 0;
2499                 if (ret || found_key.objectid != key.objectid ||
2500                     found_key.type != key.type) {
2501                         ret = 0;
2502                         goto out;
2503                 }
2504
2505                 key.offset = found_key.offset;
2506                 btrfs_release_path(root, path);
2507                 cond_resched();
2508                 goto again;
2509         }
2510
2511         /* we know there's one more slot after us in the tree,
2512          * read that key so we can verify it is also a checksum item
2513          */
2514         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2515
2516         if (found_key.objectid < inode->i_ino)
2517                 goto next_key;
2518
2519         if (found_key.type != key.type || found_key.offset < new_size)
2520                 goto next_key;
2521
2522         /*
2523          * if the key for the next leaf isn't a csum key from this objectid,
2524          * we can't be sure there aren't good items inside this leaf.
2525          * Bail out
2526          */
2527         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2528                 goto out;
2529
2530         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2531         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2532         /*
2533          * it is safe to delete this leaf, it contains only
2534          * csum items from this inode at an offset >= new_size
2535          */
2536         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2537         BUG_ON(ret);
2538
2539         if (root->ref_cows && leaf_gen < trans->transid) {
2540                 ref = btrfs_alloc_leaf_ref(root, 0);
2541                 if (ref) {
2542                         ref->root_gen = root->root_key.offset;
2543                         ref->bytenr = leaf_start;
2544                         ref->owner = 0;
2545                         ref->generation = leaf_gen;
2546                         ref->nritems = 0;
2547
2548                         btrfs_sort_leaf_ref(ref);
2549
2550                         ret = btrfs_add_leaf_ref(root, ref, 0);
2551                         WARN_ON(ret);
2552                         btrfs_free_leaf_ref(root, ref);
2553                 } else {
2554                         WARN_ON(1);
2555                 }
2556         }
2557 next_key:
2558         btrfs_release_path(root, path);
2559
2560         if (other_key.objectid == inode->i_ino &&
2561             other_key.type == key.type && other_key.offset > key.offset) {
2562                 key.offset = other_key.offset;
2563                 cond_resched();
2564                 goto again;
2565         }
2566         ret = 0;
2567 out:
2568         /* fixup any changes we've made to the path */
2569         path->lowest_level = 0;
2570         path->keep_locks = 0;
2571         btrfs_release_path(root, path);
2572         return ret;
2573 }
2574
2575 #endif
2576
2577 /*
2578  * this can truncate away extent items, csum items and directory items.
2579  * It starts at a high offset and removes keys until it can't find
2580  * any higher than new_size
2581  *
2582  * csum items that cross the new i_size are truncated to the new size
2583  * as well.
2584  *
2585  * min_type is the minimum key type to truncate down to.  If set to 0, this
2586  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2587  */
2588 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2589                                         struct btrfs_root *root,
2590                                         struct inode *inode,
2591                                         u64 new_size, u32 min_type)
2592 {
2593         int ret;
2594         struct btrfs_path *path;
2595         struct btrfs_key key;
2596         struct btrfs_key found_key;
2597         u32 found_type = (u8)-1;
2598         struct extent_buffer *leaf;
2599         struct btrfs_file_extent_item *fi;
2600         u64 extent_start = 0;
2601         u64 extent_num_bytes = 0;
2602         u64 item_end = 0;
2603         u64 root_gen = 0;
2604         u64 root_owner = 0;
2605         int found_extent;
2606         int del_item;
2607         int pending_del_nr = 0;
2608         int pending_del_slot = 0;
2609         int extent_type = -1;
2610         int encoding;
2611         u64 mask = root->sectorsize - 1;
2612
2613         if (root->ref_cows)
2614                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2615         path = btrfs_alloc_path();
2616         path->reada = -1;
2617         BUG_ON(!path);
2618
2619         /* FIXME, add redo link to tree so we don't leak on crash */
2620         key.objectid = inode->i_ino;
2621         key.offset = (u64)-1;
2622         key.type = (u8)-1;
2623
2624 search_again:
2625         path->leave_spinning = 1;
2626         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2627         if (ret < 0)
2628                 goto error;
2629
2630         if (ret > 0) {
2631                 /* there are no items in the tree for us to truncate, we're
2632                  * done
2633                  */
2634                 if (path->slots[0] == 0) {
2635                         ret = 0;
2636                         goto error;
2637                 }
2638                 path->slots[0]--;
2639         }
2640
2641         while (1) {
2642                 fi = NULL;
2643                 leaf = path->nodes[0];
2644                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2645                 found_type = btrfs_key_type(&found_key);
2646                 encoding = 0;
2647
2648                 if (found_key.objectid != inode->i_ino)
2649                         break;
2650
2651                 if (found_type < min_type)
2652                         break;
2653
2654                 item_end = found_key.offset;
2655                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2656                         fi = btrfs_item_ptr(leaf, path->slots[0],
2657                                             struct btrfs_file_extent_item);
2658                         extent_type = btrfs_file_extent_type(leaf, fi);
2659                         encoding = btrfs_file_extent_compression(leaf, fi);
2660                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2661                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2662
2663                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2664                                 item_end +=
2665                                     btrfs_file_extent_num_bytes(leaf, fi);
2666                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2667                                 item_end += btrfs_file_extent_inline_len(leaf,
2668                                                                          fi);
2669                         }
2670                         item_end--;
2671                 }
2672                 if (item_end < new_size) {
2673                         if (found_type == BTRFS_DIR_ITEM_KEY)
2674                                 found_type = BTRFS_INODE_ITEM_KEY;
2675                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2676                                 found_type = BTRFS_EXTENT_DATA_KEY;
2677                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2678                                 found_type = BTRFS_XATTR_ITEM_KEY;
2679                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2680                                 found_type = BTRFS_INODE_REF_KEY;
2681                         else if (found_type)
2682                                 found_type--;
2683                         else
2684                                 break;
2685                         btrfs_set_key_type(&key, found_type);
2686                         goto next;
2687                 }
2688                 if (found_key.offset >= new_size)
2689                         del_item = 1;
2690                 else
2691                         del_item = 0;
2692                 found_extent = 0;
2693
2694                 /* FIXME, shrink the extent if the ref count is only 1 */
2695                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2696                         goto delete;
2697
2698                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2699                         u64 num_dec;
2700                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2701                         if (!del_item && !encoding) {
2702                                 u64 orig_num_bytes =
2703                                         btrfs_file_extent_num_bytes(leaf, fi);
2704                                 extent_num_bytes = new_size -
2705                                         found_key.offset + root->sectorsize - 1;
2706                                 extent_num_bytes = extent_num_bytes &
2707                                         ~((u64)root->sectorsize - 1);
2708                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2709                                                          extent_num_bytes);
2710                                 num_dec = (orig_num_bytes -
2711                                            extent_num_bytes);
2712                                 if (root->ref_cows && extent_start != 0)
2713                                         inode_sub_bytes(inode, num_dec);
2714                                 btrfs_mark_buffer_dirty(leaf);
2715                         } else {
2716                                 extent_num_bytes =
2717                                         btrfs_file_extent_disk_num_bytes(leaf,
2718                                                                          fi);
2719                                 /* FIXME blocksize != 4096 */
2720                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2721                                 if (extent_start != 0) {
2722                                         found_extent = 1;
2723                                         if (root->ref_cows)
2724                                                 inode_sub_bytes(inode, num_dec);
2725                                 }
2726                                 root_gen = btrfs_header_generation(leaf);
2727                                 root_owner = btrfs_header_owner(leaf);
2728                         }
2729                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2730                         /*
2731                          * we can't truncate inline items that have had
2732                          * special encodings
2733                          */
2734                         if (!del_item &&
2735                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2736                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2737                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2738                                 u32 size = new_size - found_key.offset;
2739
2740                                 if (root->ref_cows) {
2741                                         inode_sub_bytes(inode, item_end + 1 -
2742                                                         new_size);
2743                                 }
2744                                 size =
2745                                     btrfs_file_extent_calc_inline_size(size);
2746                                 ret = btrfs_truncate_item(trans, root, path,
2747                                                           size, 1);
2748                                 BUG_ON(ret);
2749                         } else if (root->ref_cows) {
2750                                 inode_sub_bytes(inode, item_end + 1 -
2751                                                 found_key.offset);
2752                         }
2753                 }
2754 delete:
2755                 if (del_item) {
2756                         if (!pending_del_nr) {
2757                                 /* no pending yet, add ourselves */
2758                                 pending_del_slot = path->slots[0];
2759                                 pending_del_nr = 1;
2760                         } else if (pending_del_nr &&
2761                                    path->slots[0] + 1 == pending_del_slot) {
2762                                 /* hop on the pending chunk */
2763                                 pending_del_nr++;
2764                                 pending_del_slot = path->slots[0];
2765                         } else {
2766                                 BUG();
2767                         }
2768                 } else {
2769                         break;
2770                 }
2771                 if (found_extent) {
2772                         btrfs_set_path_blocking(path);
2773                         ret = btrfs_free_extent(trans, root, extent_start,
2774                                                 extent_num_bytes,
2775                                                 leaf->start, root_owner,
2776                                                 root_gen, inode->i_ino, 0);
2777                         BUG_ON(ret);
2778                 }
2779 next:
2780                 if (path->slots[0] == 0) {
2781                         if (pending_del_nr)
2782                                 goto del_pending;
2783                         btrfs_release_path(root, path);
2784                         if (found_type == BTRFS_INODE_ITEM_KEY)
2785                                 break;
2786                         goto search_again;
2787                 }
2788
2789                 path->slots[0]--;
2790                 if (pending_del_nr &&
2791                     path->slots[0] + 1 != pending_del_slot) {
2792                         struct btrfs_key debug;
2793 del_pending:
2794                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2795                                               pending_del_slot);
2796                         ret = btrfs_del_items(trans, root, path,
2797                                               pending_del_slot,
2798                                               pending_del_nr);
2799                         BUG_ON(ret);
2800                         pending_del_nr = 0;
2801                         btrfs_release_path(root, path);
2802                         if (found_type == BTRFS_INODE_ITEM_KEY)
2803                                 break;
2804                         goto search_again;
2805                 }
2806         }
2807         ret = 0;
2808 error:
2809         if (pending_del_nr) {
2810                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2811                                       pending_del_nr);
2812         }
2813         btrfs_free_path(path);
2814         inode->i_sb->s_dirt = 1;
2815         return ret;
2816 }
2817
2818 /*
2819  * taken from block_truncate_page, but does cow as it zeros out
2820  * any bytes left in the last page in the file.
2821  */
2822 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2823 {
2824         struct inode *inode = mapping->host;
2825         struct btrfs_root *root = BTRFS_I(inode)->root;
2826         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2827         struct btrfs_ordered_extent *ordered;
2828         char *kaddr;
2829         u32 blocksize = root->sectorsize;
2830         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2831         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2832         struct page *page;
2833         int ret = 0;
2834         u64 page_start;
2835         u64 page_end;
2836
2837         if ((offset & (blocksize - 1)) == 0)
2838                 goto out;
2839
2840         ret = -ENOMEM;
2841 again:
2842         page = grab_cache_page(mapping, index);
2843         if (!page)
2844                 goto out;
2845
2846         page_start = page_offset(page);
2847         page_end = page_start + PAGE_CACHE_SIZE - 1;
2848
2849         if (!PageUptodate(page)) {
2850                 ret = btrfs_readpage(NULL, page);
2851                 lock_page(page);
2852                 if (page->mapping != mapping) {
2853                         unlock_page(page);
2854                         page_cache_release(page);
2855                         goto again;
2856                 }
2857                 if (!PageUptodate(page)) {
2858                         ret = -EIO;
2859                         goto out_unlock;
2860                 }
2861         }
2862         wait_on_page_writeback(page);
2863
2864         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2865         set_page_extent_mapped(page);
2866
2867         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2868         if (ordered) {
2869                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2870                 unlock_page(page);
2871                 page_cache_release(page);
2872                 btrfs_start_ordered_extent(inode, ordered, 1);
2873                 btrfs_put_ordered_extent(ordered);
2874                 goto again;
2875         }
2876
2877         btrfs_set_extent_delalloc(inode, page_start, page_end);
2878         ret = 0;
2879         if (offset != PAGE_CACHE_SIZE) {
2880                 kaddr = kmap(page);
2881                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2882                 flush_dcache_page(page);
2883                 kunmap(page);
2884         }
2885         ClearPageChecked(page);
2886         set_page_dirty(page);
2887         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2888
2889 out_unlock:
2890         unlock_page(page);
2891         page_cache_release(page);
2892 out:
2893         return ret;
2894 }
2895
2896 int btrfs_cont_expand(struct inode *inode, loff_t size)
2897 {
2898         struct btrfs_trans_handle *trans;
2899         struct btrfs_root *root = BTRFS_I(inode)->root;
2900         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2901         struct extent_map *em;
2902         u64 mask = root->sectorsize - 1;
2903         u64 hole_start = (inode->i_size + mask) & ~mask;
2904         u64 block_end = (size + mask) & ~mask;
2905         u64 last_byte;
2906         u64 cur_offset;
2907         u64 hole_size;
2908         int err;
2909
2910         if (size <= hole_start)
2911                 return 0;
2912
2913         err = btrfs_check_metadata_free_space(root);
2914         if (err)
2915                 return err;
2916
2917         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2918
2919         while (1) {
2920                 struct btrfs_ordered_extent *ordered;
2921                 btrfs_wait_ordered_range(inode, hole_start,
2922                                          block_end - hole_start);
2923                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2924                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2925                 if (!ordered)
2926                         break;
2927                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2928                 btrfs_put_ordered_extent(ordered);
2929         }
2930
2931         trans = btrfs_start_transaction(root, 1);
2932         btrfs_set_trans_block_group(trans, inode);
2933
2934         cur_offset = hole_start;
2935         while (1) {
2936                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2937                                 block_end - cur_offset, 0);
2938                 BUG_ON(IS_ERR(em) || !em);
2939                 last_byte = min(extent_map_end(em), block_end);
2940                 last_byte = (last_byte + mask) & ~mask;
2941                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2942                         u64 hint_byte = 0;
2943                         hole_size = last_byte - cur_offset;
2944                         err = btrfs_drop_extents(trans, root, inode,
2945                                                  cur_offset,
2946                                                  cur_offset + hole_size,
2947                                                  block_end,
2948                                                  cur_offset, &hint_byte);
2949                         if (err)
2950                                 break;
2951                         err = btrfs_insert_file_extent(trans, root,
2952                                         inode->i_ino, cur_offset, 0,
2953                                         0, hole_size, 0, hole_size,
2954                                         0, 0, 0);
2955                         btrfs_drop_extent_cache(inode, hole_start,
2956                                         last_byte - 1, 0);
2957                 }
2958                 free_extent_map(em);
2959                 cur_offset = last_byte;
2960                 if (err || cur_offset >= block_end)
2961                         break;
2962         }
2963
2964         btrfs_end_transaction(trans, root);
2965         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2966         return err;
2967 }
2968
2969 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2970 {
2971         struct inode *inode = dentry->d_inode;
2972         int err;
2973
2974         err = inode_change_ok(inode, attr);
2975         if (err)
2976                 return err;
2977
2978         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2979                 if (attr->ia_size > inode->i_size) {
2980                         err = btrfs_cont_expand(inode, attr->ia_size);
2981                         if (err)
2982                                 return err;
2983                 } else if (inode->i_size > 0 &&
2984                            attr->ia_size == 0) {
2985
2986                         /* we're truncating a file that used to have good
2987                          * data down to zero.  Make sure it gets into
2988                          * the ordered flush list so that any new writes
2989                          * get down to disk quickly.
2990                          */
2991                         BTRFS_I(inode)->ordered_data_close = 1;
2992                 }
2993         }
2994
2995         err = inode_setattr(inode, attr);
2996
2997         if (!err && ((attr->ia_valid & ATTR_MODE)))
2998                 err = btrfs_acl_chmod(inode);
2999         return err;
3000 }
3001
3002 void btrfs_delete_inode(struct inode *inode)
3003 {
3004         struct btrfs_trans_handle *trans;
3005         struct btrfs_root *root = BTRFS_I(inode)->root;
3006         unsigned long nr;
3007         int ret;
3008
3009         truncate_inode_pages(&inode->i_data, 0);
3010         if (is_bad_inode(inode)) {
3011                 btrfs_orphan_del(NULL, inode);
3012                 goto no_delete;
3013         }
3014         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3015
3016         btrfs_i_size_write(inode, 0);
3017         trans = btrfs_join_transaction(root, 1);
3018
3019         btrfs_set_trans_block_group(trans, inode);
3020         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3021         if (ret) {
3022                 btrfs_orphan_del(NULL, inode);
3023                 goto no_delete_lock;
3024         }
3025
3026         btrfs_orphan_del(trans, inode);
3027
3028         nr = trans->blocks_used;
3029         clear_inode(inode);
3030
3031         btrfs_end_transaction(trans, root);
3032         btrfs_btree_balance_dirty(root, nr);
3033         return;
3034
3035 no_delete_lock:
3036         nr = trans->blocks_used;
3037         btrfs_end_transaction(trans, root);
3038         btrfs_btree_balance_dirty(root, nr);
3039 no_delete:
3040         clear_inode(inode);
3041 }
3042
3043 /*
3044  * this returns the key found in the dir entry in the location pointer.
3045  * If no dir entries were found, location->objectid is 0.
3046  */
3047 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3048                                struct btrfs_key *location)
3049 {
3050         const char *name = dentry->d_name.name;
3051         int namelen = dentry->d_name.len;
3052         struct btrfs_dir_item *di;
3053         struct btrfs_path *path;
3054         struct btrfs_root *root = BTRFS_I(dir)->root;
3055         int ret = 0;
3056
3057         path = btrfs_alloc_path();
3058         BUG_ON(!path);
3059
3060         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3061                                     namelen, 0);
3062         if (IS_ERR(di))
3063                 ret = PTR_ERR(di);
3064
3065         if (!di || IS_ERR(di))
3066                 goto out_err;
3067
3068         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3069 out:
3070         btrfs_free_path(path);
3071         return ret;
3072 out_err:
3073         location->objectid = 0;
3074         goto out;
3075 }
3076
3077 /*
3078  * when we hit a tree root in a directory, the btrfs part of the inode
3079  * needs to be changed to reflect the root directory of the tree root.  This
3080  * is kind of like crossing a mount point.
3081  */
3082 static int fixup_tree_root_location(struct btrfs_root *root,
3083                              struct btrfs_key *location,
3084                              struct btrfs_root **sub_root,
3085                              struct dentry *dentry)
3086 {
3087         struct btrfs_root_item *ri;
3088
3089         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3090                 return 0;
3091         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3092                 return 0;
3093
3094         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3095                                         dentry->d_name.name,
3096                                         dentry->d_name.len);
3097         if (IS_ERR(*sub_root))
3098                 return PTR_ERR(*sub_root);
3099
3100         ri = &(*sub_root)->root_item;
3101         location->objectid = btrfs_root_dirid(ri);
3102         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3103         location->offset = 0;
3104
3105         return 0;
3106 }
3107
3108 static noinline void init_btrfs_i(struct inode *inode)
3109 {
3110         struct btrfs_inode *bi = BTRFS_I(inode);
3111
3112         bi->i_acl = BTRFS_ACL_NOT_CACHED;
3113         bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
3114
3115         bi->generation = 0;
3116         bi->sequence = 0;
3117         bi->last_trans = 0;
3118         bi->logged_trans = 0;
3119         bi->delalloc_bytes = 0;
3120         bi->reserved_bytes = 0;
3121         bi->disk_i_size = 0;
3122         bi->flags = 0;
3123         bi->index_cnt = (u64)-1;
3124         bi->last_unlink_trans = 0;
3125         bi->ordered_data_close = 0;
3126         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3127         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3128                              inode->i_mapping, GFP_NOFS);
3129         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3130                              inode->i_mapping, GFP_NOFS);
3131         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3132         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3133         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3134         mutex_init(&BTRFS_I(inode)->extent_mutex);
3135         mutex_init(&BTRFS_I(inode)->log_mutex);
3136 }
3137
3138 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3139 {
3140         struct btrfs_iget_args *args = p;
3141         inode->i_ino = args->ino;
3142         init_btrfs_i(inode);
3143         BTRFS_I(inode)->root = args->root;
3144         btrfs_set_inode_space_info(args->root, inode);
3145         return 0;
3146 }
3147
3148 static int btrfs_find_actor(struct inode *inode, void *opaque)
3149 {
3150         struct btrfs_iget_args *args = opaque;
3151         return args->ino == inode->i_ino &&
3152                 args->root == BTRFS_I(inode)->root;
3153 }
3154
3155 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3156                             struct btrfs_root *root, int wait)
3157 {
3158         struct inode *inode;
3159         struct btrfs_iget_args args;
3160         args.ino = objectid;
3161         args.root = root;
3162
3163         if (wait) {
3164                 inode = ilookup5(s, objectid, btrfs_find_actor,
3165                                  (void *)&args);
3166         } else {
3167                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3168                                         (void *)&args);
3169         }
3170         return inode;
3171 }
3172
3173 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3174                                 struct btrfs_root *root)
3175 {
3176         struct inode *inode;
3177         struct btrfs_iget_args args;
3178         args.ino = objectid;
3179         args.root = root;
3180
3181         inode = iget5_locked(s, objectid, btrfs_find_actor,
3182                              btrfs_init_locked_inode,
3183                              (void *)&args);
3184         return inode;
3185 }
3186
3187 /* Get an inode object given its location and corresponding root.
3188  * Returns in *is_new if the inode was read from disk
3189  */
3190 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3191                          struct btrfs_root *root, int *is_new)
3192 {
3193         struct inode *inode;
3194
3195         inode = btrfs_iget_locked(s, location->objectid, root);
3196         if (!inode)
3197                 return ERR_PTR(-EACCES);
3198
3199         if (inode->i_state & I_NEW) {
3200                 BTRFS_I(inode)->root = root;
3201                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3202                 btrfs_read_locked_inode(inode);
3203                 unlock_new_inode(inode);
3204                 if (is_new)
3205                         *is_new = 1;
3206         } else {
3207                 if (is_new)
3208                         *is_new = 0;
3209         }
3210
3211         return inode;
3212 }
3213
3214 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3215 {
3216         struct inode *inode;
3217         struct btrfs_inode *bi = BTRFS_I(dir);
3218         struct btrfs_root *root = bi->root;
3219         struct btrfs_root *sub_root = root;
3220         struct btrfs_key location;
3221         int ret, new;
3222
3223         if (dentry->d_name.len > BTRFS_NAME_LEN)
3224                 return ERR_PTR(-ENAMETOOLONG);
3225
3226         ret = btrfs_inode_by_name(dir, dentry, &location);
3227
3228         if (ret < 0)
3229                 return ERR_PTR(ret);
3230
3231         inode = NULL;
3232         if (location.objectid) {
3233                 ret = fixup_tree_root_location(root, &location, &sub_root,
3234                                                 dentry);
3235                 if (ret < 0)
3236                         return ERR_PTR(ret);
3237                 if (ret > 0)
3238                         return ERR_PTR(-ENOENT);
3239                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3240                 if (IS_ERR(inode))
3241                         return ERR_CAST(inode);
3242         }
3243         return inode;
3244 }
3245
3246 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3247                                    struct nameidata *nd)
3248 {
3249         struct inode *inode;
3250
3251         if (dentry->d_name.len > BTRFS_NAME_LEN)
3252                 return ERR_PTR(-ENAMETOOLONG);
3253
3254         inode = btrfs_lookup_dentry(dir, dentry);
3255         if (IS_ERR(inode))
3256                 return ERR_CAST(inode);
3257
3258         return d_splice_alias(inode, dentry);
3259 }
3260
3261 static unsigned char btrfs_filetype_table[] = {
3262         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3263 };
3264
3265 static int btrfs_real_readdir(struct file *filp, void *dirent,
3266                               filldir_t filldir)
3267 {
3268         struct inode *inode = filp->f_dentry->d_inode;
3269         struct btrfs_root *root = BTRFS_I(inode)->root;
3270         struct btrfs_item *item;
3271         struct btrfs_dir_item *di;
3272         struct btrfs_key key;
3273         struct btrfs_key found_key;
3274         struct btrfs_path *path;
3275         int ret;
3276         u32 nritems;
3277         struct extent_buffer *leaf;
3278         int slot;
3279         int advance;
3280         unsigned char d_type;
3281         int over = 0;
3282         u32 di_cur;
3283         u32 di_total;
3284         u32 di_len;
3285         int key_type = BTRFS_DIR_INDEX_KEY;
3286         char tmp_name[32];
3287         char *name_ptr;
3288         int name_len;
3289
3290         /* FIXME, use a real flag for deciding about the key type */
3291         if (root->fs_info->tree_root == root)
3292                 key_type = BTRFS_DIR_ITEM_KEY;
3293
3294         /* special case for "." */
3295         if (filp->f_pos == 0) {
3296                 over = filldir(dirent, ".", 1,
3297                                1, inode->i_ino,
3298                                DT_DIR);
3299                 if (over)
3300                         return 0;
3301                 filp->f_pos = 1;
3302         }
3303         /* special case for .., just use the back ref */
3304         if (filp->f_pos == 1) {
3305                 u64 pino = parent_ino(filp->f_path.dentry);
3306                 over = filldir(dirent, "..", 2,
3307                                2, pino, DT_DIR);
3308                 if (over)
3309                         return 0;
3310                 filp->f_pos = 2;
3311         }
3312         path = btrfs_alloc_path();
3313         path->reada = 2;
3314
3315         btrfs_set_key_type(&key, key_type);
3316         key.offset = filp->f_pos;
3317         key.objectid = inode->i_ino;
3318
3319         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3320         if (ret < 0)
3321                 goto err;
3322         advance = 0;
3323
3324         while (1) {
3325                 leaf = path->nodes[0];
3326                 nritems = btrfs_header_nritems(leaf);
3327                 slot = path->slots[0];
3328                 if (advance || slot >= nritems) {
3329                         if (slot >= nritems - 1) {
3330                                 ret = btrfs_next_leaf(root, path);
3331                                 if (ret)
3332                                         break;
3333                                 leaf = path->nodes[0];
3334                                 nritems = btrfs_header_nritems(leaf);
3335                                 slot = path->slots[0];
3336                         } else {
3337                                 slot++;
3338                                 path->slots[0]++;
3339                         }
3340                 }
3341
3342                 advance = 1;
3343                 item = btrfs_item_nr(leaf, slot);
3344                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3345
3346                 if (found_key.objectid != key.objectid)
3347                         break;
3348                 if (btrfs_key_type(&found_key) != key_type)
3349                         break;
3350                 if (found_key.offset < filp->f_pos)
3351                         continue;
3352
3353                 filp->f_pos = found_key.offset;
3354
3355                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3356                 di_cur = 0;
3357                 di_total = btrfs_item_size(leaf, item);
3358
3359                 while (di_cur < di_total) {
3360                         struct btrfs_key location;
3361
3362                         name_len = btrfs_dir_name_len(leaf, di);
3363                         if (name_len <= sizeof(tmp_name)) {
3364                                 name_ptr = tmp_name;
3365                         } else {
3366                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3367                                 if (!name_ptr) {
3368                                         ret = -ENOMEM;
3369                                         goto err;
3370                                 }
3371                         }
3372                         read_extent_buffer(leaf, name_ptr,
3373                                            (unsigned long)(di + 1), name_len);
3374
3375                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3376                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3377
3378                         /* is this a reference to our own snapshot? If so
3379                          * skip it
3380                          */
3381                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3382                             location.objectid == root->root_key.objectid) {
3383                                 over = 0;
3384                                 goto skip;
3385                         }
3386                         over = filldir(dirent, name_ptr, name_len,
3387                                        found_key.offset, location.objectid,
3388                                        d_type);
3389
3390 skip:
3391                         if (name_ptr != tmp_name)
3392                                 kfree(name_ptr);
3393
3394                         if (over)
3395                                 goto nopos;
3396                         di_len = btrfs_dir_name_len(leaf, di) +
3397                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3398                         di_cur += di_len;
3399                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3400                 }
3401         }
3402
3403         /* Reached end of directory/root. Bump pos past the last item. */
3404         if (key_type == BTRFS_DIR_INDEX_KEY)
3405                 filp->f_pos = INT_LIMIT(off_t);
3406         else
3407                 filp->f_pos++;
3408 nopos:
3409         ret = 0;
3410 err:
3411         btrfs_free_path(path);
3412         return ret;
3413 }
3414
3415 int btrfs_write_inode(struct inode *inode, int wait)
3416 {
3417         struct btrfs_root *root = BTRFS_I(inode)->root;
3418         struct btrfs_trans_handle *trans;
3419         int ret = 0;
3420
3421         if (root->fs_info->btree_inode == inode)
3422                 return 0;
3423
3424         if (wait) {
3425                 trans = btrfs_join_transaction(root, 1);
3426                 btrfs_set_trans_block_group(trans, inode);
3427                 ret = btrfs_commit_transaction(trans, root);
3428         }
3429         return ret;
3430 }
3431
3432 /*
3433  * This is somewhat expensive, updating the tree every time the
3434  * inode changes.  But, it is most likely to find the inode in cache.
3435  * FIXME, needs more benchmarking...there are no reasons other than performance
3436  * to keep or drop this code.
3437  */
3438 void btrfs_dirty_inode(struct inode *inode)
3439 {
3440         struct btrfs_root *root = BTRFS_I(inode)->root;
3441         struct btrfs_trans_handle *trans;
3442
3443         trans = btrfs_join_transaction(root, 1);
3444         btrfs_set_trans_block_group(trans, inode);
3445         btrfs_update_inode(trans, root, inode);
3446         btrfs_end_transaction(trans, root);
3447 }
3448
3449 /*
3450  * find the highest existing sequence number in a directory
3451  * and then set the in-memory index_cnt variable to reflect
3452  * free sequence numbers
3453  */
3454 static int btrfs_set_inode_index_count(struct inode *inode)
3455 {
3456         struct btrfs_root *root = BTRFS_I(inode)->root;
3457         struct btrfs_key key, found_key;
3458         struct btrfs_path *path;
3459         struct extent_buffer *leaf;
3460         int ret;
3461
3462         key.objectid = inode->i_ino;
3463         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3464         key.offset = (u64)-1;
3465
3466         path = btrfs_alloc_path();
3467         if (!path)
3468                 return -ENOMEM;
3469
3470         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3471         if (ret < 0)
3472                 goto out;
3473         /* FIXME: we should be able to handle this */
3474         if (ret == 0)
3475                 goto out;
3476         ret = 0;
3477
3478         /*
3479          * MAGIC NUMBER EXPLANATION:
3480          * since we search a directory based on f_pos we have to start at 2
3481          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3482          * else has to start at 2
3483          */
3484         if (path->slots[0] == 0) {
3485                 BTRFS_I(inode)->index_cnt = 2;
3486                 goto out;
3487         }
3488
3489         path->slots[0]--;
3490
3491         leaf = path->nodes[0];
3492         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3493
3494         if (found_key.objectid != inode->i_ino ||
3495             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3496                 BTRFS_I(inode)->index_cnt = 2;
3497                 goto out;
3498         }
3499
3500         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3501 out:
3502         btrfs_free_path(path);
3503         return ret;
3504 }
3505
3506 /*
3507  * helper to find a free sequence number in a given directory.  This current
3508  * code is very simple, later versions will do smarter things in the btree
3509  */
3510 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3511 {
3512         int ret = 0;
3513
3514         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3515                 ret = btrfs_set_inode_index_count(dir);
3516                 if (ret)
3517                         return ret;
3518         }
3519
3520         *index = BTRFS_I(dir)->index_cnt;
3521         BTRFS_I(dir)->index_cnt++;
3522
3523         return ret;
3524 }
3525
3526 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3527                                      struct btrfs_root *root,
3528                                      struct inode *dir,
3529                                      const char *name, int name_len,
3530                                      u64 ref_objectid, u64 objectid,
3531                                      u64 alloc_hint, int mode, u64 *index)
3532 {
3533         struct inode *inode;
3534         struct btrfs_inode_item *inode_item;
3535         struct btrfs_key *location;
3536         struct btrfs_path *path;
3537         struct btrfs_inode_ref *ref;
3538         struct btrfs_key key[2];
3539         u32 sizes[2];
3540         unsigned long ptr;
3541         int ret;
3542         int owner;
3543
3544         path = btrfs_alloc_path();
3545         BUG_ON(!path);
3546
3547         inode = new_inode(root->fs_info->sb);
3548         if (!inode)
3549                 return ERR_PTR(-ENOMEM);
3550
3551         if (dir) {
3552                 ret = btrfs_set_inode_index(dir, index);
3553                 if (ret) {
3554                         iput(inode);
3555                         return ERR_PTR(ret);
3556                 }
3557         }
3558         /*
3559          * index_cnt is ignored for everything but a dir,
3560          * btrfs_get_inode_index_count has an explanation for the magic
3561          * number
3562          */
3563         init_btrfs_i(inode);
3564         BTRFS_I(inode)->index_cnt = 2;
3565         BTRFS_I(inode)->root = root;
3566         BTRFS_I(inode)->generation = trans->transid;
3567         btrfs_set_inode_space_info(root, inode);
3568
3569         if (mode & S_IFDIR)
3570                 owner = 0;
3571         else
3572                 owner = 1;
3573         BTRFS_I(inode)->block_group =
3574                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3575         if ((mode & S_IFREG)) {
3576                 if (btrfs_test_opt(root, NODATASUM))
3577                         btrfs_set_flag(inode, NODATASUM);
3578                 if (btrfs_test_opt(root, NODATACOW))
3579                         btrfs_set_flag(inode, NODATACOW);
3580         }
3581
3582         key[0].objectid = objectid;
3583         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3584         key[0].offset = 0;
3585
3586         key[1].objectid = objectid;
3587         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3588         key[1].offset = ref_objectid;
3589
3590         sizes[0] = sizeof(struct btrfs_inode_item);
3591         sizes[1] = name_len + sizeof(*ref);
3592
3593         path->leave_spinning = 1;
3594         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3595         if (ret != 0)
3596                 goto fail;
3597
3598         if (objectid > root->highest_inode)
3599                 root->highest_inode = objectid;
3600
3601         inode->i_uid = current_fsuid();
3602
3603         if (dir && (dir->i_mode & S_ISGID)) {
3604                 inode->i_gid = dir->i_gid;
3605                 if (S_ISDIR(mode))
3606                         mode |= S_ISGID;
3607         } else
3608                 inode->i_gid = current_fsgid();
3609
3610         inode->i_mode = mode;
3611         inode->i_ino = objectid;
3612         inode_set_bytes(inode, 0);
3613         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3614         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3615                                   struct btrfs_inode_item);
3616         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3617
3618         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3619                              struct btrfs_inode_ref);
3620         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3621         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3622         ptr = (unsigned long)(ref + 1);
3623         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3624
3625         btrfs_mark_buffer_dirty(path->nodes[0]);
3626         btrfs_free_path(path);
3627
3628         location = &BTRFS_I(inode)->location;
3629         location->objectid = objectid;
3630         location->offset = 0;
3631         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3632
3633         insert_inode_hash(inode);
3634         return inode;
3635 fail:
3636         if (dir)
3637                 BTRFS_I(dir)->index_cnt--;
3638         btrfs_free_path(path);
3639         iput(inode);
3640         return ERR_PTR(ret);
3641 }
3642
3643 static inline u8 btrfs_inode_type(struct inode *inode)
3644 {
3645         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3646 }
3647
3648 /*
3649  * utility function to add 'inode' into 'parent_inode' with
3650  * a give name and a given sequence number.
3651  * if 'add_backref' is true, also insert a backref from the
3652  * inode to the parent directory.
3653  */
3654 int btrfs_add_link(struct btrfs_trans_handle *trans,
3655                    struct inode *parent_inode, struct inode *inode,
3656                    const char *name, int name_len, int add_backref, u64 index)
3657 {
3658         int ret;
3659         struct btrfs_key key;
3660         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3661
3662         key.objectid = inode->i_ino;
3663         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3664         key.offset = 0;
3665
3666         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3667                                     parent_inode->i_ino,
3668                                     &key, btrfs_inode_type(inode),
3669                                     index);
3670         if (ret == 0) {
3671                 if (add_backref) {
3672                         ret = btrfs_insert_inode_ref(trans, root,
3673                                                      name, name_len,
3674                                                      inode->i_ino,
3675                                                      parent_inode->i_ino,
3676                                                      index);
3677                 }
3678                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3679                                    name_len * 2);
3680                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3681                 ret = btrfs_update_inode(trans, root, parent_inode);
3682         }
3683         return ret;
3684 }
3685
3686 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3687                             struct dentry *dentry, struct inode *inode,
3688                             int backref, u64 index)
3689 {
3690         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3691                                  inode, dentry->d_name.name,
3692                                  dentry->d_name.len, backref, index);
3693         if (!err) {
3694                 d_instantiate(dentry, inode);
3695                 return 0;
3696         }
3697         if (err > 0)
3698                 err = -EEXIST;
3699         return err;
3700 }
3701
3702 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3703                         int mode, dev_t rdev)
3704 {
3705         struct btrfs_trans_handle *trans;
3706         struct btrfs_root *root = BTRFS_I(dir)->root;
3707         struct inode *inode = NULL;
3708         int err;
3709         int drop_inode = 0;
3710         u64 objectid;
3711         unsigned long nr = 0;
3712         u64 index = 0;
3713
3714         if (!new_valid_dev(rdev))
3715                 return -EINVAL;
3716
3717         err = btrfs_check_metadata_free_space(root);
3718         if (err)
3719                 goto fail;
3720
3721         trans = btrfs_start_transaction(root, 1);
3722         btrfs_set_trans_block_group(trans, dir);
3723
3724         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3725         if (err) {
3726                 err = -ENOSPC;
3727                 goto out_unlock;
3728         }
3729
3730         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3731                                 dentry->d_name.len,
3732                                 dentry->d_parent->d_inode->i_ino, objectid,
3733                                 BTRFS_I(dir)->block_group, mode, &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_op = &btrfs_special_inode_operations;
3750                 init_special_inode(inode, inode->i_mode, rdev);
3751                 btrfs_update_inode(trans, root, inode);
3752         }
3753         dir->i_sb->s_dirt = 1;
3754         btrfs_update_inode_block_group(trans, inode);
3755         btrfs_update_inode_block_group(trans, dir);
3756 out_unlock:
3757         nr = trans->blocks_used;
3758         btrfs_end_transaction_throttle(trans, root);
3759 fail:
3760         if (drop_inode) {
3761                 inode_dec_link_count(inode);
3762                 iput(inode);
3763         }
3764         btrfs_btree_balance_dirty(root, nr);
3765         return err;
3766 }
3767
3768 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3769                         int mode, struct nameidata *nd)
3770 {
3771         struct btrfs_trans_handle *trans;
3772         struct btrfs_root *root = BTRFS_I(dir)->root;
3773         struct inode *inode = NULL;
3774         int err;
3775         int drop_inode = 0;
3776         unsigned long nr = 0;
3777         u64 objectid;
3778         u64 index = 0;
3779
3780         err = btrfs_check_metadata_free_space(root);
3781         if (err)
3782                 goto fail;
3783         trans = btrfs_start_transaction(root, 1);
3784         btrfs_set_trans_block_group(trans, dir);
3785
3786         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3787         if (err) {
3788                 err = -ENOSPC;
3789                 goto out_unlock;
3790         }
3791
3792         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3793                                 dentry->d_name.len,
3794                                 dentry->d_parent->d_inode->i_ino,
3795                                 objectid, BTRFS_I(dir)->block_group, mode,
3796                                 &index);
3797         err = PTR_ERR(inode);
3798         if (IS_ERR(inode))
3799                 goto out_unlock;
3800
3801         err = btrfs_init_inode_security(inode, dir);
3802         if (err) {
3803                 drop_inode = 1;
3804                 goto out_unlock;
3805         }
3806
3807         btrfs_set_trans_block_group(trans, inode);
3808         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3809         if (err)
3810                 drop_inode = 1;
3811         else {
3812                 inode->i_mapping->a_ops = &btrfs_aops;
3813                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3814                 inode->i_fop = &btrfs_file_operations;
3815                 inode->i_op = &btrfs_file_inode_operations;
3816                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3817         }
3818         dir->i_sb->s_dirt = 1;
3819         btrfs_update_inode_block_group(trans, inode);
3820         btrfs_update_inode_block_group(trans, dir);
3821 out_unlock:
3822         nr = trans->blocks_used;
3823         btrfs_end_transaction_throttle(trans, root);
3824 fail:
3825         if (drop_inode) {
3826                 inode_dec_link_count(inode);
3827                 iput(inode);
3828         }
3829         btrfs_btree_balance_dirty(root, nr);
3830         return err;
3831 }
3832
3833 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3834                       struct dentry *dentry)
3835 {
3836         struct btrfs_trans_handle *trans;
3837         struct btrfs_root *root = BTRFS_I(dir)->root;
3838         struct inode *inode = old_dentry->d_inode;
3839         u64 index;
3840         unsigned long nr = 0;
3841         int err;
3842         int drop_inode = 0;
3843
3844         if (inode->i_nlink == 0)
3845                 return -ENOENT;
3846
3847         btrfs_inc_nlink(inode);
3848         err = btrfs_check_metadata_free_space(root);
3849         if (err)
3850                 goto fail;
3851         err = btrfs_set_inode_index(dir, &index);
3852         if (err)
3853                 goto fail;
3854
3855         trans = btrfs_start_transaction(root, 1);
3856
3857         btrfs_set_trans_block_group(trans, dir);
3858         atomic_inc(&inode->i_count);
3859
3860         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3861
3862         if (err)
3863                 drop_inode = 1;
3864
3865         dir->i_sb->s_dirt = 1;
3866         btrfs_update_inode_block_group(trans, dir);
3867         err = btrfs_update_inode(trans, root, inode);
3868
3869         if (err)
3870                 drop_inode = 1;
3871
3872         nr = trans->blocks_used;
3873
3874         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3875         btrfs_end_transaction_throttle(trans, root);
3876 fail:
3877         if (drop_inode) {
3878                 inode_dec_link_count(inode);
3879                 iput(inode);
3880         }
3881         btrfs_btree_balance_dirty(root, nr);
3882         return err;
3883 }
3884
3885 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3886 {
3887         struct inode *inode = NULL;
3888         struct btrfs_trans_handle *trans;
3889         struct btrfs_root *root = BTRFS_I(dir)->root;
3890         int err = 0;
3891         int drop_on_err = 0;
3892         u64 objectid = 0;
3893         u64 index = 0;
3894         unsigned long nr = 1;
3895
3896         err = btrfs_check_metadata_free_space(root);
3897         if (err)
3898                 goto out_unlock;
3899
3900         trans = btrfs_start_transaction(root, 1);
3901         btrfs_set_trans_block_group(trans, dir);
3902
3903         if (IS_ERR(trans)) {
3904                 err = PTR_ERR(trans);
3905                 goto out_unlock;
3906         }
3907
3908         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3909         if (err) {
3910                 err = -ENOSPC;
3911                 goto out_unlock;
3912         }
3913
3914         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3915                                 dentry->d_name.len,
3916                                 dentry->d_parent->d_inode->i_ino, objectid,
3917                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3918                                 &index);
3919         if (IS_ERR(inode)) {
3920                 err = PTR_ERR(inode);
3921                 goto out_fail;
3922         }
3923
3924         drop_on_err = 1;
3925
3926         err = btrfs_init_inode_security(inode, dir);
3927         if (err)
3928                 goto out_fail;
3929
3930         inode->i_op = &btrfs_dir_inode_operations;
3931         inode->i_fop = &btrfs_dir_file_operations;
3932         btrfs_set_trans_block_group(trans, inode);
3933
3934         btrfs_i_size_write(inode, 0);
3935         err = btrfs_update_inode(trans, root, inode);
3936         if (err)
3937                 goto out_fail;
3938
3939         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3940                                  inode, dentry->d_name.name,
3941                                  dentry->d_name.len, 0, index);
3942         if (err)
3943                 goto out_fail;
3944
3945         d_instantiate(dentry, inode);
3946         drop_on_err = 0;
3947         dir->i_sb->s_dirt = 1;
3948         btrfs_update_inode_block_group(trans, inode);
3949         btrfs_update_inode_block_group(trans, dir);
3950
3951 out_fail:
3952         nr = trans->blocks_used;
3953         btrfs_end_transaction_throttle(trans, root);
3954
3955 out_unlock:
3956         if (drop_on_err)
3957                 iput(inode);
3958         btrfs_btree_balance_dirty(root, nr);
3959         return err;
3960 }
3961
3962 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3963  * and an extent that you want to insert, deal with overlap and insert
3964  * the new extent into the tree.
3965  */
3966 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3967                                 struct extent_map *existing,
3968                                 struct extent_map *em,
3969                                 u64 map_start, u64 map_len)
3970 {
3971         u64 start_diff;
3972
3973         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3974         start_diff = map_start - em->start;
3975         em->start = map_start;
3976         em->len = map_len;
3977         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3978             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3979                 em->block_start += start_diff;
3980                 em->block_len -= start_diff;
3981         }
3982         return add_extent_mapping(em_tree, em);
3983 }
3984
3985 static noinline int uncompress_inline(struct btrfs_path *path,
3986                                       struct inode *inode, struct page *page,
3987                                       size_t pg_offset, u64 extent_offset,
3988                                       struct btrfs_file_extent_item *item)
3989 {
3990         int ret;
3991         struct extent_buffer *leaf = path->nodes[0];
3992         char *tmp;
3993         size_t max_size;
3994         unsigned long inline_size;
3995         unsigned long ptr;
3996
3997         WARN_ON(pg_offset != 0);
3998         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3999         inline_size = btrfs_file_extent_inline_item_len(leaf,
4000                                         btrfs_item_nr(leaf, path->slots[0]));
4001         tmp = kmalloc(inline_size, GFP_NOFS);
4002         ptr = btrfs_file_extent_inline_start(item);
4003
4004         read_extent_buffer(leaf, tmp, ptr, inline_size);
4005
4006         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4007         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4008                                     inline_size, max_size);
4009         if (ret) {
4010                 char *kaddr = kmap_atomic(page, KM_USER0);
4011                 unsigned long copy_size = min_t(u64,
4012                                   PAGE_CACHE_SIZE - pg_offset,
4013                                   max_size - extent_offset);
4014                 memset(kaddr + pg_offset, 0, copy_size);
4015                 kunmap_atomic(kaddr, KM_USER0);
4016         }
4017         kfree(tmp);
4018         return 0;
4019 }
4020
4021 /*
4022  * a bit scary, this does extent mapping from logical file offset to the disk.
4023  * the ugly parts come from merging extents from the disk with the in-ram
4024  * representation.  This gets more complex because of the data=ordered code,
4025  * where the in-ram extents might be locked pending data=ordered completion.
4026  *
4027  * This also copies inline extents directly into the page.
4028  */
4029
4030 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4031                                     size_t pg_offset, u64 start, u64 len,
4032                                     int create)
4033 {
4034         int ret;
4035         int err = 0;
4036         u64 bytenr;
4037         u64 extent_start = 0;
4038         u64 extent_end = 0;
4039         u64 objectid = inode->i_ino;
4040         u32 found_type;
4041         struct btrfs_path *path = NULL;
4042         struct btrfs_root *root = BTRFS_I(inode)->root;
4043         struct btrfs_file_extent_item *item;
4044         struct extent_buffer *leaf;
4045         struct btrfs_key found_key;
4046         struct extent_map *em = NULL;
4047         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4048         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4049         struct btrfs_trans_handle *trans = NULL;
4050         int compressed;
4051
4052 again:
4053         spin_lock(&em_tree->lock);
4054         em = lookup_extent_mapping(em_tree, start, len);
4055         if (em)
4056                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4057         spin_unlock(&em_tree->lock);
4058
4059         if (em) {
4060                 if (em->start > start || em->start + em->len <= start)
4061                         free_extent_map(em);
4062                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4063                         free_extent_map(em);
4064                 else
4065                         goto out;
4066         }
4067         em = alloc_extent_map(GFP_NOFS);
4068         if (!em) {
4069                 err = -ENOMEM;
4070                 goto out;
4071         }
4072         em->bdev = root->fs_info->fs_devices->latest_bdev;
4073         em->start = EXTENT_MAP_HOLE;
4074         em->orig_start = EXTENT_MAP_HOLE;
4075         em->len = (u64)-1;
4076         em->block_len = (u64)-1;
4077
4078         if (!path) {
4079                 path = btrfs_alloc_path();
4080                 BUG_ON(!path);
4081         }
4082
4083         ret = btrfs_lookup_file_extent(trans, root, path,
4084                                        objectid, start, trans != NULL);
4085         if (ret < 0) {
4086                 err = ret;
4087                 goto out;
4088         }
4089
4090         if (ret != 0) {
4091                 if (path->slots[0] == 0)
4092                         goto not_found;
4093                 path->slots[0]--;
4094         }
4095
4096         leaf = path->nodes[0];
4097         item = btrfs_item_ptr(leaf, path->slots[0],
4098                               struct btrfs_file_extent_item);
4099         /* are we inside the extent that was found? */
4100         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4101         found_type = btrfs_key_type(&found_key);
4102         if (found_key.objectid != objectid ||
4103             found_type != BTRFS_EXTENT_DATA_KEY) {
4104                 goto not_found;
4105         }
4106
4107         found_type = btrfs_file_extent_type(leaf, item);
4108         extent_start = found_key.offset;
4109         compressed = btrfs_file_extent_compression(leaf, item);
4110         if (found_type == BTRFS_FILE_EXTENT_REG ||
4111             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4112                 extent_end = extent_start +
4113                        btrfs_file_extent_num_bytes(leaf, item);
4114         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4115                 size_t size;
4116                 size = btrfs_file_extent_inline_len(leaf, item);
4117                 extent_end = (extent_start + size + root->sectorsize - 1) &
4118                         ~((u64)root->sectorsize - 1);
4119         }
4120
4121         if (start >= extent_end) {
4122                 path->slots[0]++;
4123                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4124                         ret = btrfs_next_leaf(root, path);
4125                         if (ret < 0) {
4126                                 err = ret;
4127                                 goto out;
4128                         }
4129                         if (ret > 0)
4130                                 goto not_found;
4131                         leaf = path->nodes[0];
4132                 }
4133                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4134                 if (found_key.objectid != objectid ||
4135                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4136                         goto not_found;
4137                 if (start + len <= found_key.offset)
4138                         goto not_found;
4139                 em->start = start;
4140                 em->len = found_key.offset - start;
4141                 goto not_found_em;
4142         }
4143
4144         if (found_type == BTRFS_FILE_EXTENT_REG ||
4145             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4146                 em->start = extent_start;
4147                 em->len = extent_end - extent_start;
4148                 em->orig_start = extent_start -
4149                                  btrfs_file_extent_offset(leaf, item);
4150                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4151                 if (bytenr == 0) {
4152                         em->block_start = EXTENT_MAP_HOLE;
4153                         goto insert;
4154                 }
4155                 if (compressed) {
4156                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4157                         em->block_start = bytenr;
4158                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4159                                                                          item);
4160                 } else {
4161                         bytenr += btrfs_file_extent_offset(leaf, item);
4162                         em->block_start = bytenr;
4163                         em->block_len = em->len;
4164                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4165                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4166                 }
4167                 goto insert;
4168         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4169                 unsigned long ptr;
4170                 char *map;
4171                 size_t size;
4172                 size_t extent_offset;
4173                 size_t copy_size;
4174
4175                 em->block_start = EXTENT_MAP_INLINE;
4176                 if (!page || create) {
4177                         em->start = extent_start;
4178                         em->len = extent_end - extent_start;
4179                         goto out;
4180                 }
4181
4182                 size = btrfs_file_extent_inline_len(leaf, item);
4183                 extent_offset = page_offset(page) + pg_offset - extent_start;
4184                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4185                                 size - extent_offset);
4186                 em->start = extent_start + extent_offset;
4187                 em->len = (copy_size + root->sectorsize - 1) &
4188                         ~((u64)root->sectorsize - 1);
4189                 em->orig_start = EXTENT_MAP_INLINE;
4190                 if (compressed)
4191                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4192                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4193                 if (create == 0 && !PageUptodate(page)) {
4194                         if (btrfs_file_extent_compression(leaf, item) ==
4195                             BTRFS_COMPRESS_ZLIB) {
4196                                 ret = uncompress_inline(path, inode, page,
4197                                                         pg_offset,
4198                                                         extent_offset, item);
4199                                 BUG_ON(ret);
4200                         } else {
4201                                 map = kmap(page);
4202                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4203                                                    copy_size);
4204                                 kunmap(page);
4205                         }
4206                         flush_dcache_page(page);
4207                 } else if (create && PageUptodate(page)) {
4208                         if (!trans) {
4209                                 kunmap(page);
4210                                 free_extent_map(em);
4211                                 em = NULL;
4212                                 btrfs_release_path(root, path);
4213                                 trans = btrfs_join_transaction(root, 1);
4214                                 goto again;
4215                         }
4216                         map = kmap(page);
4217                         write_extent_buffer(leaf, map + pg_offset, ptr,
4218                                             copy_size);
4219                         kunmap(page);
4220                         btrfs_mark_buffer_dirty(leaf);
4221                 }
4222                 set_extent_uptodate(io_tree, em->start,
4223                                     extent_map_end(em) - 1, GFP_NOFS);
4224                 goto insert;
4225         } else {
4226                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4227                 WARN_ON(1);
4228         }
4229 not_found:
4230         em->start = start;
4231         em->len = len;
4232 not_found_em:
4233         em->block_start = EXTENT_MAP_HOLE;
4234         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4235 insert:
4236         btrfs_release_path(root, path);
4237         if (em->start > start || extent_map_end(em) <= start) {
4238                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4239                        "[%llu %llu]\n", (unsigned long long)em->start,
4240                        (unsigned long long)em->len,
4241                        (unsigned long long)start,
4242                        (unsigned long long)len);
4243                 err = -EIO;
4244                 goto out;
4245         }
4246
4247         err = 0;
4248         spin_lock(&em_tree->lock);
4249         ret = add_extent_mapping(em_tree, em);
4250         /* it is possible that someone inserted the extent into the tree
4251          * while we had the lock dropped.  It is also possible that
4252          * an overlapping map exists in the tree
4253          */
4254         if (ret == -EEXIST) {
4255                 struct extent_map *existing;
4256
4257                 ret = 0;
4258
4259                 existing = lookup_extent_mapping(em_tree, start, len);
4260                 if (existing && (existing->start > start ||
4261                     existing->start + existing->len <= start)) {
4262                         free_extent_map(existing);
4263                         existing = NULL;
4264                 }
4265                 if (!existing) {
4266                         existing = lookup_extent_mapping(em_tree, em->start,
4267                                                          em->len);
4268                         if (existing) {
4269                                 err = merge_extent_mapping(em_tree, existing,
4270                                                            em, start,
4271                                                            root->sectorsize);
4272                                 free_extent_map(existing);
4273                                 if (err) {
4274                                         free_extent_map(em);
4275                                         em = NULL;
4276                                 }
4277                         } else {
4278                                 err = -EIO;
4279                                 free_extent_map(em);
4280                                 em = NULL;
4281                         }
4282                 } else {
4283                         free_extent_map(em);
4284                         em = existing;
4285                         err = 0;
4286                 }
4287         }
4288         spin_unlock(&em_tree->lock);
4289 out:
4290         if (path)
4291                 btrfs_free_path(path);
4292         if (trans) {
4293                 ret = btrfs_end_transaction(trans, root);
4294                 if (!err)
4295                         err = ret;
4296         }
4297         if (err) {
4298                 free_extent_map(em);
4299                 return ERR_PTR(err);
4300         }
4301         return em;
4302 }
4303
4304 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4305                         const struct iovec *iov, loff_t offset,
4306                         unsigned long nr_segs)
4307 {
4308         return -EINVAL;
4309 }
4310
4311 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4312                 __u64 start, __u64 len)
4313 {
4314         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4315 }
4316
4317 int btrfs_readpage(struct file *file, struct page *page)
4318 {
4319         struct extent_io_tree *tree;
4320         tree = &BTRFS_I(page->mapping->host)->io_tree;
4321         return extent_read_full_page(tree, page, btrfs_get_extent);
4322 }
4323
4324 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4325 {
4326         struct extent_io_tree *tree;
4327
4328
4329         if (current->flags & PF_MEMALLOC) {
4330                 redirty_page_for_writepage(wbc, page);
4331                 unlock_page(page);
4332                 return 0;
4333         }
4334         tree = &BTRFS_I(page->mapping->host)->io_tree;
4335         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4336 }
4337
4338 int btrfs_writepages(struct address_space *mapping,
4339                      struct writeback_control *wbc)
4340 {
4341         struct extent_io_tree *tree;
4342
4343         tree = &BTRFS_I(mapping->host)->io_tree;
4344         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4345 }
4346
4347 static int
4348 btrfs_readpages(struct file *file, struct address_space *mapping,
4349                 struct list_head *pages, unsigned nr_pages)
4350 {
4351         struct extent_io_tree *tree;
4352         tree = &BTRFS_I(mapping->host)->io_tree;
4353         return extent_readpages(tree, mapping, pages, nr_pages,
4354                                 btrfs_get_extent);
4355 }
4356 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4357 {
4358         struct extent_io_tree *tree;
4359         struct extent_map_tree *map;
4360         int ret;
4361
4362         tree = &BTRFS_I(page->mapping->host)->io_tree;
4363         map = &BTRFS_I(page->mapping->host)->extent_tree;
4364         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4365         if (ret == 1) {
4366                 ClearPagePrivate(page);
4367                 set_page_private(page, 0);
4368                 page_cache_release(page);
4369         }
4370         return ret;
4371 }
4372
4373 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4374 {
4375         if (PageWriteback(page) || PageDirty(page))
4376                 return 0;
4377         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4378 }
4379
4380 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4381 {
4382         struct extent_io_tree *tree;
4383         struct btrfs_ordered_extent *ordered;
4384         u64 page_start = page_offset(page);
4385         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4386
4387         wait_on_page_writeback(page);
4388         tree = &BTRFS_I(page->mapping->host)->io_tree;
4389         if (offset) {
4390                 btrfs_releasepage(page, GFP_NOFS);
4391                 return;
4392         }
4393
4394         lock_extent(tree, page_start, page_end, GFP_NOFS);
4395         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4396                                            page_offset(page));
4397         if (ordered) {
4398                 /*
4399                  * IO on this page will never be started, so we need
4400                  * to account for any ordered extents now
4401                  */
4402                 clear_extent_bit(tree, page_start, page_end,
4403                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4404                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4405                 btrfs_finish_ordered_io(page->mapping->host,
4406                                         page_start, page_end);
4407                 btrfs_put_ordered_extent(ordered);
4408                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4409         }
4410         clear_extent_bit(tree, page_start, page_end,
4411                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4412                  EXTENT_ORDERED,
4413                  1, 1, GFP_NOFS);
4414         __btrfs_releasepage(page, GFP_NOFS);
4415
4416         ClearPageChecked(page);
4417         if (PagePrivate(page)) {
4418                 ClearPagePrivate(page);
4419                 set_page_private(page, 0);
4420                 page_cache_release(page);
4421         }
4422 }
4423
4424 /*
4425  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4426  * called from a page fault handler when a page is first dirtied. Hence we must
4427  * be careful to check for EOF conditions here. We set the page up correctly
4428  * for a written page which means we get ENOSPC checking when writing into
4429  * holes and correct delalloc and unwritten extent mapping on filesystems that
4430  * support these features.
4431  *
4432  * We are not allowed to take the i_mutex here so we have to play games to
4433  * protect against truncate races as the page could now be beyond EOF.  Because
4434  * vmtruncate() writes the inode size before removing pages, once we have the
4435  * page lock we can determine safely if the page is beyond EOF. If it is not
4436  * beyond EOF, then the page is guaranteed safe against truncation until we
4437  * unlock the page.
4438  */
4439 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4440 {
4441         struct page *page = vmf->page;
4442         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4443         struct btrfs_root *root = BTRFS_I(inode)->root;
4444         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4445         struct btrfs_ordered_extent *ordered;
4446         char *kaddr;
4447         unsigned long zero_start;
4448         loff_t size;
4449         int ret;
4450         u64 page_start;
4451         u64 page_end;
4452
4453         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4454         if (ret) {
4455                 if (ret == -ENOMEM)
4456                         ret = VM_FAULT_OOM;
4457                 else /* -ENOSPC, -EIO, etc */
4458                         ret = VM_FAULT_SIGBUS;
4459                 goto out;
4460         }
4461
4462         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4463 again:
4464         lock_page(page);
4465         size = i_size_read(inode);
4466         page_start = page_offset(page);
4467         page_end = page_start + PAGE_CACHE_SIZE - 1;
4468
4469         if ((page->mapping != inode->i_mapping) ||
4470             (page_start >= size)) {
4471                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4472                 /* page got truncated out from underneath us */
4473                 goto out_unlock;
4474         }
4475         wait_on_page_writeback(page);
4476
4477         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4478         set_page_extent_mapped(page);
4479
4480         /*
4481          * we can't set the delalloc bits if there are pending ordered
4482          * extents.  Drop our locks and wait for them to finish
4483          */
4484         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4485         if (ordered) {
4486                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4487                 unlock_page(page);
4488                 btrfs_start_ordered_extent(inode, ordered, 1);
4489                 btrfs_put_ordered_extent(ordered);
4490                 goto again;
4491         }
4492
4493         btrfs_set_extent_delalloc(inode, page_start, page_end);
4494         ret = 0;
4495
4496         /* page is wholly or partially inside EOF */
4497         if (page_start + PAGE_CACHE_SIZE > size)
4498                 zero_start = size & ~PAGE_CACHE_MASK;
4499         else
4500                 zero_start = PAGE_CACHE_SIZE;
4501
4502         if (zero_start != PAGE_CACHE_SIZE) {
4503                 kaddr = kmap(page);
4504                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4505                 flush_dcache_page(page);
4506                 kunmap(page);
4507         }
4508         ClearPageChecked(page);
4509         set_page_dirty(page);
4510
4511         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4512         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4513
4514 out_unlock:
4515         unlock_page(page);
4516 out:
4517         return ret;
4518 }
4519
4520 static void btrfs_truncate(struct inode *inode)
4521 {
4522         struct btrfs_root *root = BTRFS_I(inode)->root;
4523         int ret;
4524         struct btrfs_trans_handle *trans;
4525         unsigned long nr;
4526         u64 mask = root->sectorsize - 1;
4527
4528         if (!S_ISREG(inode->i_mode))
4529                 return;
4530         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4531                 return;
4532
4533         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4534         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4535
4536         trans = btrfs_start_transaction(root, 1);
4537
4538         /*
4539          * setattr is responsible for setting the ordered_data_close flag,
4540          * but that is only tested during the last file release.  That
4541          * could happen well after the next commit, leaving a great big
4542          * window where new writes may get lost if someone chooses to write
4543          * to this file after truncating to zero
4544          *
4545          * The inode doesn't have any dirty data here, and so if we commit
4546          * this is a noop.  If someone immediately starts writing to the inode
4547          * it is very likely we'll catch some of their writes in this
4548          * transaction, and the commit will find this file on the ordered
4549          * data list with good things to send down.
4550          *
4551          * This is a best effort solution, there is still a window where
4552          * using truncate to replace the contents of the file will
4553          * end up with a zero length file after a crash.
4554          */
4555         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4556                 btrfs_add_ordered_operation(trans, root, inode);
4557
4558         btrfs_set_trans_block_group(trans, inode);
4559         btrfs_i_size_write(inode, inode->i_size);
4560
4561         ret = btrfs_orphan_add(trans, inode);
4562         if (ret)
4563                 goto out;
4564         /* FIXME, add redo link to tree so we don't leak on crash */
4565         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4566                                       BTRFS_EXTENT_DATA_KEY);
4567         btrfs_update_inode(trans, root, inode);
4568
4569         ret = btrfs_orphan_del(trans, inode);
4570         BUG_ON(ret);
4571
4572 out:
4573         nr = trans->blocks_used;
4574         ret = btrfs_end_transaction_throttle(trans, root);
4575         BUG_ON(ret);
4576         btrfs_btree_balance_dirty(root, nr);
4577 }
4578
4579 /*
4580  * create a new subvolume directory/inode (helper for the ioctl).
4581  */
4582 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4583                              struct btrfs_root *new_root, struct dentry *dentry,
4584                              u64 new_dirid, u64 alloc_hint)
4585 {
4586         struct inode *inode;
4587         int error;
4588         u64 index = 0;
4589
4590         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4591                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4592         if (IS_ERR(inode))
4593                 return PTR_ERR(inode);
4594         inode->i_op = &btrfs_dir_inode_operations;
4595         inode->i_fop = &btrfs_dir_file_operations;
4596
4597         inode->i_nlink = 1;
4598         btrfs_i_size_write(inode, 0);
4599
4600         error = btrfs_update_inode(trans, new_root, inode);
4601         if (error)
4602                 return error;
4603
4604         d_instantiate(dentry, inode);
4605         return 0;
4606 }
4607
4608 /* helper function for file defrag and space balancing.  This
4609  * forces readahead on a given range of bytes in an inode
4610  */
4611 unsigned long btrfs_force_ra(struct address_space *mapping,
4612                               struct file_ra_state *ra, struct file *file,
4613                               pgoff_t offset, pgoff_t last_index)
4614 {
4615         pgoff_t req_size = last_index - offset + 1;
4616
4617         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4618         return offset + req_size;
4619 }
4620
4621 struct inode *btrfs_alloc_inode(struct super_block *sb)
4622 {
4623         struct btrfs_inode *ei;
4624
4625         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4626         if (!ei)
4627                 return NULL;
4628         ei->last_trans = 0;
4629         ei->logged_trans = 0;
4630         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4631         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4632         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4633         INIT_LIST_HEAD(&ei->i_orphan);
4634         INIT_LIST_HEAD(&ei->ordered_operations);
4635         return &ei->vfs_inode;
4636 }
4637
4638 void btrfs_destroy_inode(struct inode *inode)
4639 {
4640         struct btrfs_ordered_extent *ordered;
4641         struct btrfs_root *root = BTRFS_I(inode)->root;
4642
4643         WARN_ON(!list_empty(&inode->i_dentry));
4644         WARN_ON(inode->i_data.nrpages);
4645
4646         if (BTRFS_I(inode)->i_acl &&
4647             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4648                 posix_acl_release(BTRFS_I(inode)->i_acl);
4649         if (BTRFS_I(inode)->i_default_acl &&
4650             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4651                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4652
4653         /*
4654          * Make sure we're properly removed from the ordered operation
4655          * lists.
4656          */
4657         smp_mb();
4658         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4659                 spin_lock(&root->fs_info->ordered_extent_lock);
4660                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4661                 spin_unlock(&root->fs_info->ordered_extent_lock);
4662         }
4663
4664         spin_lock(&root->list_lock);
4665         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4666                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4667                        " list\n", inode->i_ino);
4668                 dump_stack();
4669         }
4670         spin_unlock(&root->list_lock);
4671
4672         while (1) {
4673                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4674                 if (!ordered)
4675                         break;
4676                 else {
4677                         printk(KERN_ERR "btrfs found ordered "
4678                                "extent %llu %llu on inode cleanup\n",
4679                                (unsigned long long)ordered->file_offset,
4680                                (unsigned long long)ordered->len);
4681                         btrfs_remove_ordered_extent(inode, ordered);
4682                         btrfs_put_ordered_extent(ordered);
4683                         btrfs_put_ordered_extent(ordered);
4684                 }
4685         }
4686         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4687         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4688 }
4689
4690 static void init_once(void *foo)
4691 {
4692         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4693
4694         inode_init_once(&ei->vfs_inode);
4695 }
4696
4697 void btrfs_destroy_cachep(void)
4698 {
4699         if (btrfs_inode_cachep)
4700                 kmem_cache_destroy(btrfs_inode_cachep);
4701         if (btrfs_trans_handle_cachep)
4702                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4703         if (btrfs_transaction_cachep)
4704                 kmem_cache_destroy(btrfs_transaction_cachep);
4705         if (btrfs_path_cachep)
4706                 kmem_cache_destroy(btrfs_path_cachep);
4707 }
4708
4709 int btrfs_init_cachep(void)
4710 {
4711         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4712                         sizeof(struct btrfs_inode), 0,
4713                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4714         if (!btrfs_inode_cachep)
4715                 goto fail;
4716
4717         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4718                         sizeof(struct btrfs_trans_handle), 0,
4719                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4720         if (!btrfs_trans_handle_cachep)
4721                 goto fail;
4722
4723         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4724                         sizeof(struct btrfs_transaction), 0,
4725                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4726         if (!btrfs_transaction_cachep)
4727                 goto fail;
4728
4729         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4730                         sizeof(struct btrfs_path), 0,
4731                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4732         if (!btrfs_path_cachep)
4733                 goto fail;
4734
4735         return 0;
4736 fail:
4737         btrfs_destroy_cachep();
4738         return -ENOMEM;
4739 }
4740
4741 static int btrfs_getattr(struct vfsmount *mnt,
4742                          struct dentry *dentry, struct kstat *stat)
4743 {
4744         struct inode *inode = dentry->d_inode;
4745         generic_fillattr(inode, stat);
4746         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4747         stat->blksize = PAGE_CACHE_SIZE;
4748         stat->blocks = (inode_get_bytes(inode) +
4749                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4750         return 0;
4751 }
4752
4753 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4754                            struct inode *new_dir, struct dentry *new_dentry)
4755 {
4756         struct btrfs_trans_handle *trans;
4757         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4758         struct inode *new_inode = new_dentry->d_inode;
4759         struct inode *old_inode = old_dentry->d_inode;
4760         struct timespec ctime = CURRENT_TIME;
4761         u64 index = 0;
4762         int ret;
4763
4764         /* we're not allowed to rename between subvolumes */
4765         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4766             BTRFS_I(new_dir)->root->root_key.objectid)
4767                 return -EXDEV;
4768
4769         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4770             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4771                 return -ENOTEMPTY;
4772         }
4773
4774         /* to rename a snapshot or subvolume, we need to juggle the
4775          * backrefs.  This isn't coded yet
4776          */
4777         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4778                 return -EXDEV;
4779
4780         ret = btrfs_check_metadata_free_space(root);
4781         if (ret)
4782                 goto out_unlock;
4783
4784         /*
4785          * we're using rename to replace one file with another.
4786          * and the replacement file is large.  Start IO on it now so
4787          * we don't add too much work to the end of the transaction
4788          */
4789         if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4790             new_inode->i_size &&
4791             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4792                 filemap_flush(old_inode->i_mapping);
4793
4794         trans = btrfs_start_transaction(root, 1);
4795
4796         /*
4797          * make sure the inode gets flushed if it is replacing
4798          * something.
4799          */
4800         if (new_inode && new_inode->i_size &&
4801             old_inode && S_ISREG(old_inode->i_mode)) {
4802                 btrfs_add_ordered_operation(trans, root, old_inode);
4803         }
4804
4805         /*
4806          * this is an ugly little race, but the rename is required to make
4807          * sure that if we crash, the inode is either at the old name
4808          * or the new one.  pinning the log transaction lets us make sure
4809          * we don't allow a log commit to come in after we unlink the
4810          * name but before we add the new name back in.
4811          */
4812         btrfs_pin_log_trans(root);
4813
4814         btrfs_set_trans_block_group(trans, new_dir);
4815
4816         btrfs_inc_nlink(old_dentry->d_inode);
4817         old_dir->i_ctime = old_dir->i_mtime = ctime;
4818         new_dir->i_ctime = new_dir->i_mtime = ctime;
4819         old_inode->i_ctime = ctime;
4820
4821         if (old_dentry->d_parent != new_dentry->d_parent)
4822                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4823
4824         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4825                                  old_dentry->d_name.name,
4826                                  old_dentry->d_name.len);
4827         if (ret)
4828                 goto out_fail;
4829
4830         if (new_inode) {
4831                 new_inode->i_ctime = CURRENT_TIME;
4832                 ret = btrfs_unlink_inode(trans, root, new_dir,
4833                                          new_dentry->d_inode,
4834                                          new_dentry->d_name.name,
4835                                          new_dentry->d_name.len);
4836                 if (ret)
4837                         goto out_fail;
4838                 if (new_inode->i_nlink == 0) {
4839                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4840                         if (ret)
4841                                 goto out_fail;
4842                 }
4843
4844         }
4845         ret = btrfs_set_inode_index(new_dir, &index);
4846         if (ret)
4847                 goto out_fail;
4848
4849         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4850                              old_inode, new_dentry->d_name.name,
4851                              new_dentry->d_name.len, 1, index);
4852         if (ret)
4853                 goto out_fail;
4854
4855         btrfs_log_new_name(trans, old_inode, old_dir,
4856                                        new_dentry->d_parent);
4857 out_fail:
4858
4859         /* this btrfs_end_log_trans just allows the current
4860          * log-sub transaction to complete
4861          */
4862         btrfs_end_log_trans(root);
4863         btrfs_end_transaction_throttle(trans, root);
4864 out_unlock:
4865         return ret;
4866 }
4867
4868 /*
4869  * some fairly slow code that needs optimization. This walks the list
4870  * of all the inodes with pending delalloc and forces them to disk.
4871  */
4872 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4873 {
4874         struct list_head *head = &root->fs_info->delalloc_inodes;
4875         struct btrfs_inode *binode;
4876         struct inode *inode;
4877
4878         if (root->fs_info->sb->s_flags & MS_RDONLY)
4879                 return -EROFS;
4880
4881         spin_lock(&root->fs_info->delalloc_lock);
4882         while (!list_empty(head)) {
4883                 binode = list_entry(head->next, struct btrfs_inode,
4884                                     delalloc_inodes);
4885                 inode = igrab(&binode->vfs_inode);
4886                 if (!inode)
4887                         list_del_init(&binode->delalloc_inodes);
4888                 spin_unlock(&root->fs_info->delalloc_lock);
4889                 if (inode) {
4890                         filemap_flush(inode->i_mapping);
4891                         iput(inode);
4892                 }
4893                 cond_resched();
4894                 spin_lock(&root->fs_info->delalloc_lock);
4895         }
4896         spin_unlock(&root->fs_info->delalloc_lock);
4897
4898         /* the filemap_flush will queue IO into the worker threads, but
4899          * we have to make sure the IO is actually started and that
4900          * ordered extents get created before we return
4901          */
4902         atomic_inc(&root->fs_info->async_submit_draining);
4903         while (atomic_read(&root->fs_info->nr_async_submits) ||
4904               atomic_read(&root->fs_info->async_delalloc_pages)) {
4905                 wait_event(root->fs_info->async_submit_wait,
4906                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4907                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4908         }
4909         atomic_dec(&root->fs_info->async_submit_draining);
4910         return 0;
4911 }
4912
4913 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4914                          const char *symname)
4915 {
4916         struct btrfs_trans_handle *trans;
4917         struct btrfs_root *root = BTRFS_I(dir)->root;
4918         struct btrfs_path *path;
4919         struct btrfs_key key;
4920         struct inode *inode = NULL;
4921         int err;
4922         int drop_inode = 0;
4923         u64 objectid;
4924         u64 index = 0 ;
4925         int name_len;
4926         int datasize;
4927         unsigned long ptr;
4928         struct btrfs_file_extent_item *ei;
4929         struct extent_buffer *leaf;
4930         unsigned long nr = 0;
4931
4932         name_len = strlen(symname) + 1;
4933         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4934                 return -ENAMETOOLONG;
4935
4936         err = btrfs_check_metadata_free_space(root);
4937         if (err)
4938                 goto out_fail;
4939
4940         trans = btrfs_start_transaction(root, 1);
4941         btrfs_set_trans_block_group(trans, dir);
4942
4943         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4944         if (err) {
4945                 err = -ENOSPC;
4946                 goto out_unlock;
4947         }
4948
4949         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4950                                 dentry->d_name.len,
4951                                 dentry->d_parent->d_inode->i_ino, objectid,
4952                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4953                                 &index);
4954         err = PTR_ERR(inode);
4955         if (IS_ERR(inode))
4956                 goto out_unlock;
4957
4958         err = btrfs_init_inode_security(inode, dir);
4959         if (err) {
4960                 drop_inode = 1;
4961                 goto out_unlock;
4962         }
4963
4964         btrfs_set_trans_block_group(trans, inode);
4965         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4966         if (err)
4967                 drop_inode = 1;
4968         else {
4969                 inode->i_mapping->a_ops = &btrfs_aops;
4970                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4971                 inode->i_fop = &btrfs_file_operations;
4972                 inode->i_op = &btrfs_file_inode_operations;
4973                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4974         }
4975         dir->i_sb->s_dirt = 1;
4976         btrfs_update_inode_block_group(trans, inode);
4977         btrfs_update_inode_block_group(trans, dir);
4978         if (drop_inode)
4979                 goto out_unlock;
4980
4981         path = btrfs_alloc_path();
4982         BUG_ON(!path);
4983         key.objectid = inode->i_ino;
4984         key.offset = 0;
4985         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4986         datasize = btrfs_file_extent_calc_inline_size(name_len);
4987         err = btrfs_insert_empty_item(trans, root, path, &key,
4988                                       datasize);
4989         if (err) {
4990                 drop_inode = 1;
4991                 goto out_unlock;
4992         }
4993         leaf = path->nodes[0];
4994         ei = btrfs_item_ptr(leaf, path->slots[0],
4995                             struct btrfs_file_extent_item);
4996         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4997         btrfs_set_file_extent_type(leaf, ei,
4998                                    BTRFS_FILE_EXTENT_INLINE);
4999         btrfs_set_file_extent_encryption(leaf, ei, 0);
5000         btrfs_set_file_extent_compression(leaf, ei, 0);
5001         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5002         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5003
5004         ptr = btrfs_file_extent_inline_start(ei);
5005         write_extent_buffer(leaf, symname, ptr, name_len);
5006         btrfs_mark_buffer_dirty(leaf);
5007         btrfs_free_path(path);
5008
5009         inode->i_op = &btrfs_symlink_inode_operations;
5010         inode->i_mapping->a_ops = &btrfs_symlink_aops;
5011         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5012         inode_set_bytes(inode, name_len);
5013         btrfs_i_size_write(inode, name_len - 1);
5014         err = btrfs_update_inode(trans, root, inode);
5015         if (err)
5016                 drop_inode = 1;
5017
5018 out_unlock:
5019         nr = trans->blocks_used;
5020         btrfs_end_transaction_throttle(trans, root);
5021 out_fail:
5022         if (drop_inode) {
5023                 inode_dec_link_count(inode);
5024                 iput(inode);
5025         }
5026         btrfs_btree_balance_dirty(root, nr);
5027         return err;
5028 }
5029
5030 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5031                                struct inode *inode, u64 start, u64 end,
5032                                u64 locked_end, u64 alloc_hint, int mode)
5033 {
5034         struct btrfs_root *root = BTRFS_I(inode)->root;
5035         struct btrfs_key ins;
5036         u64 alloc_size;
5037         u64 cur_offset = start;
5038         u64 num_bytes = end - start;
5039         int ret = 0;
5040
5041         while (num_bytes > 0) {
5042                 alloc_size = min(num_bytes, root->fs_info->max_extent);
5043                 ret = btrfs_reserve_extent(trans, root, alloc_size,
5044                                            root->sectorsize, 0, alloc_hint,
5045                                            (u64)-1, &ins, 1);
5046                 if (ret) {
5047                         WARN_ON(1);
5048                         goto out;
5049                 }
5050                 ret = insert_reserved_file_extent(trans, inode,
5051                                                   cur_offset, ins.objectid,
5052                                                   ins.offset, ins.offset,
5053                                                   ins.offset, locked_end,
5054                                                   0, 0, 0,
5055                                                   BTRFS_FILE_EXTENT_PREALLOC);
5056                 BUG_ON(ret);
5057                 num_bytes -= ins.offset;
5058                 cur_offset += ins.offset;
5059                 alloc_hint = ins.objectid + ins.offset;
5060         }
5061 out:
5062         if (cur_offset > start) {
5063                 inode->i_ctime = CURRENT_TIME;
5064                 btrfs_set_flag(inode, PREALLOC);
5065                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5066                     cur_offset > i_size_read(inode))
5067                         btrfs_i_size_write(inode, cur_offset);
5068                 ret = btrfs_update_inode(trans, root, inode);
5069                 BUG_ON(ret);
5070         }
5071
5072         return ret;
5073 }
5074
5075 static long btrfs_fallocate(struct inode *inode, int mode,
5076                             loff_t offset, loff_t len)
5077 {
5078         u64 cur_offset;
5079         u64 last_byte;
5080         u64 alloc_start;
5081         u64 alloc_end;
5082         u64 alloc_hint = 0;
5083         u64 locked_end;
5084         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5085         struct extent_map *em;
5086         struct btrfs_trans_handle *trans;
5087         int ret;
5088
5089         alloc_start = offset & ~mask;
5090         alloc_end =  (offset + len + mask) & ~mask;
5091
5092         /*
5093          * wait for ordered IO before we have any locks.  We'll loop again
5094          * below with the locks held.
5095          */
5096         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5097
5098         mutex_lock(&inode->i_mutex);
5099         if (alloc_start > inode->i_size) {
5100                 ret = btrfs_cont_expand(inode, alloc_start);
5101                 if (ret)
5102                         goto out;
5103         }
5104
5105         locked_end = alloc_end - 1;
5106         while (1) {
5107                 struct btrfs_ordered_extent *ordered;
5108
5109                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5110                 if (!trans) {
5111                         ret = -EIO;
5112                         goto out;
5113                 }
5114
5115                 /* the extent lock is ordered inside the running
5116                  * transaction
5117                  */
5118                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5119                             GFP_NOFS);
5120                 ordered = btrfs_lookup_first_ordered_extent(inode,
5121                                                             alloc_end - 1);
5122                 if (ordered &&
5123                     ordered->file_offset + ordered->len > alloc_start &&
5124                     ordered->file_offset < alloc_end) {
5125                         btrfs_put_ordered_extent(ordered);
5126                         unlock_extent(&BTRFS_I(inode)->io_tree,
5127                                       alloc_start, locked_end, GFP_NOFS);
5128                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5129
5130                         /*
5131                          * we can't wait on the range with the transaction
5132                          * running or with the extent lock held
5133                          */
5134                         btrfs_wait_ordered_range(inode, alloc_start,
5135                                                  alloc_end - alloc_start);
5136                 } else {
5137                         if (ordered)
5138                                 btrfs_put_ordered_extent(ordered);
5139                         break;
5140                 }
5141         }
5142
5143         cur_offset = alloc_start;
5144         while (1) {
5145                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5146                                       alloc_end - cur_offset, 0);
5147                 BUG_ON(IS_ERR(em) || !em);
5148                 last_byte = min(extent_map_end(em), alloc_end);
5149                 last_byte = (last_byte + mask) & ~mask;
5150                 if (em->block_start == EXTENT_MAP_HOLE) {
5151                         ret = prealloc_file_range(trans, inode, cur_offset,
5152                                         last_byte, locked_end + 1,
5153                                         alloc_hint, mode);
5154                         if (ret < 0) {
5155                                 free_extent_map(em);
5156                                 break;
5157                         }
5158                 }
5159                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5160                         alloc_hint = em->block_start;
5161                 free_extent_map(em);
5162
5163                 cur_offset = last_byte;
5164                 if (cur_offset >= alloc_end) {
5165                         ret = 0;
5166                         break;
5167                 }
5168         }
5169         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5170                       GFP_NOFS);
5171
5172         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5173 out:
5174         mutex_unlock(&inode->i_mutex);
5175         return ret;
5176 }
5177
5178 static int btrfs_set_page_dirty(struct page *page)
5179 {
5180         return __set_page_dirty_nobuffers(page);
5181 }
5182
5183 static int btrfs_permission(struct inode *inode, int mask)
5184 {
5185         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
5186                 return -EACCES;
5187         return generic_permission(inode, mask, btrfs_check_acl);
5188 }
5189
5190 static struct inode_operations btrfs_dir_inode_operations = {
5191         .getattr        = btrfs_getattr,
5192         .lookup         = btrfs_lookup,
5193         .create         = btrfs_create,
5194         .unlink         = btrfs_unlink,
5195         .link           = btrfs_link,
5196         .mkdir          = btrfs_mkdir,
5197         .rmdir          = btrfs_rmdir,
5198         .rename         = btrfs_rename,
5199         .symlink        = btrfs_symlink,
5200         .setattr        = btrfs_setattr,
5201         .mknod          = btrfs_mknod,
5202         .setxattr       = btrfs_setxattr,
5203         .getxattr       = btrfs_getxattr,
5204         .listxattr      = btrfs_listxattr,
5205         .removexattr    = btrfs_removexattr,
5206         .permission     = btrfs_permission,
5207 };
5208 static struct inode_operations btrfs_dir_ro_inode_operations = {
5209         .lookup         = btrfs_lookup,
5210         .permission     = btrfs_permission,
5211 };
5212 static struct file_operations btrfs_dir_file_operations = {
5213         .llseek         = generic_file_llseek,
5214         .read           = generic_read_dir,
5215         .readdir        = btrfs_real_readdir,
5216         .unlocked_ioctl = btrfs_ioctl,
5217 #ifdef CONFIG_COMPAT
5218         .compat_ioctl   = btrfs_ioctl,
5219 #endif
5220         .release        = btrfs_release_file,
5221         .fsync          = btrfs_sync_file,
5222 };
5223
5224 static struct extent_io_ops btrfs_extent_io_ops = {
5225         .fill_delalloc = run_delalloc_range,
5226         .submit_bio_hook = btrfs_submit_bio_hook,
5227         .merge_bio_hook = btrfs_merge_bio_hook,
5228         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5229         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5230         .writepage_start_hook = btrfs_writepage_start_hook,
5231         .readpage_io_failed_hook = btrfs_io_failed_hook,
5232         .set_bit_hook = btrfs_set_bit_hook,
5233         .clear_bit_hook = btrfs_clear_bit_hook,
5234 };
5235
5236 /*
5237  * btrfs doesn't support the bmap operation because swapfiles
5238  * use bmap to make a mapping of extents in the file.  They assume
5239  * these extents won't change over the life of the file and they
5240  * use the bmap result to do IO directly to the drive.
5241  *
5242  * the btrfs bmap call would return logical addresses that aren't
5243  * suitable for IO and they also will change frequently as COW
5244  * operations happen.  So, swapfile + btrfs == corruption.
5245  *
5246  * For now we're avoiding this by dropping bmap.
5247  */
5248 static struct address_space_operations btrfs_aops = {
5249         .readpage       = btrfs_readpage,
5250         .writepage      = btrfs_writepage,
5251         .writepages     = btrfs_writepages,
5252         .readpages      = btrfs_readpages,
5253         .sync_page      = block_sync_page,
5254         .direct_IO      = btrfs_direct_IO,
5255         .invalidatepage = btrfs_invalidatepage,
5256         .releasepage    = btrfs_releasepage,
5257         .set_page_dirty = btrfs_set_page_dirty,
5258 };
5259
5260 static struct address_space_operations btrfs_symlink_aops = {
5261         .readpage       = btrfs_readpage,
5262         .writepage      = btrfs_writepage,
5263         .invalidatepage = btrfs_invalidatepage,
5264         .releasepage    = btrfs_releasepage,
5265 };
5266
5267 static struct inode_operations btrfs_file_inode_operations = {
5268         .truncate       = btrfs_truncate,
5269         .getattr        = btrfs_getattr,
5270         .setattr        = btrfs_setattr,
5271         .setxattr       = btrfs_setxattr,
5272         .getxattr       = btrfs_getxattr,
5273         .listxattr      = btrfs_listxattr,
5274         .removexattr    = btrfs_removexattr,
5275         .permission     = btrfs_permission,
5276         .fallocate      = btrfs_fallocate,
5277         .fiemap         = btrfs_fiemap,
5278 };
5279 static struct inode_operations btrfs_special_inode_operations = {
5280         .getattr        = btrfs_getattr,
5281         .setattr        = btrfs_setattr,
5282         .permission     = btrfs_permission,
5283         .setxattr       = btrfs_setxattr,
5284         .getxattr       = btrfs_getxattr,
5285         .listxattr      = btrfs_listxattr,
5286         .removexattr    = btrfs_removexattr,
5287 };
5288 static struct inode_operations btrfs_symlink_inode_operations = {
5289         .readlink       = generic_readlink,
5290         .follow_link    = page_follow_link_light,
5291         .put_link       = page_put_link,
5292         .permission     = btrfs_permission,
5293         .setxattr       = btrfs_setxattr,
5294         .getxattr       = btrfs_getxattr,
5295         .listxattr      = btrfs_listxattr,
5296         .removexattr    = btrfs_removexattr,
5297 };