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