fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir,
  97                                      const struct qstr *qstr)
  98 {
  99         int err;
 100
 101         err = btrfs_init_acl(trans, inode, dir);
 102         if (!err)
 103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 104         return err;
 105 }
 106
 107 /*
 108  * this does all the hard work for inserting an inline extent into
 109  * the btree.  The caller should have done a btrfs_drop_extents so that
 110  * no overlapping inline items exist in the btree
 111  */
 112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 113                                 struct btrfs_root *root, struct inode *inode,
 114                                 u64 start, size_t size, size_t compressed_size,
 115                                 int compress_type,
 116                                 struct page **compressed_pages)
 117 {
 118         struct btrfs_key key;
 119         struct btrfs_path *path;
 120         struct extent_buffer *leaf;
 121         struct page *page = NULL;
 122         char *kaddr;
 123         unsigned long ptr;
 124         struct btrfs_file_extent_item *ei;
 125         int err = 0;
 126         int ret;
 127         size_t cur_size = size;
 128         size_t datasize;
 129         unsigned long offset;
 130
 131         if (compressed_size && compressed_pages)
 132                 cur_size = compressed_size;
 133
 134         path = btrfs_alloc_path();
 135         if (!path)
 136                 return -ENOMEM;
 137
 138         path->leave_spinning = 1;
 139         btrfs_set_trans_block_group(trans, inode);
 140
 141         key.objectid = inode->i_ino;
 142         key.offset = start;
 143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 145
 146         inode_add_bytes(inode, size);
 147         ret = btrfs_insert_empty_item(trans, root, path, &key,
 148                                       datasize);
 149         BUG_ON(ret);
 150         if (ret) {
 151                 err = ret;
 152                 goto fail;
 153         }
 154         leaf = path->nodes[0];
 155         ei = btrfs_item_ptr(leaf, path->slots[0],
 156                             struct btrfs_file_extent_item);
 157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 159         btrfs_set_file_extent_encryption(leaf, ei, 0);
 160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 162         ptr = btrfs_file_extent_inline_start(ei);
 163
 164         if (compress_type != BTRFS_COMPRESS_NONE) {
 165                 struct page *cpage;
 166                 int i = 0;
 167                 while (compressed_size > 0) {
 168                         cpage = compressed_pages[i];
 169                         cur_size = min_t(unsigned long, compressed_size,
 170                                        PAGE_CACHE_SIZE);
 171
 172                         kaddr = kmap_atomic(cpage, KM_USER0);
 173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 174                         kunmap_atomic(kaddr, KM_USER0);
 175
 176                         i++;
 177                         ptr += cur_size;
 178                         compressed_size -= cur_size;
 179                 }
 180                 btrfs_set_file_extent_compression(leaf, ei,
 181                                                   compress_type);
 182         } else {
 183                 page = find_get_page(inode->i_mapping,
 184                                      start >> PAGE_CACHE_SHIFT);
 185                 btrfs_set_file_extent_compression(leaf, ei, 0);
 186                 kaddr = kmap_atomic(page, KM_USER0);
 187                 offset = start & (PAGE_CACHE_SIZE - 1);
 188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 189                 kunmap_atomic(kaddr, KM_USER0);
 190                 page_cache_release(page);
 191         }
 192         btrfs_mark_buffer_dirty(leaf);
 193         btrfs_free_path(path);
 194
 195         /*
 196          * we're an inline extent, so nobody can
 197          * extend the file past i_size without locking
 198          * a page we already have locked.
 199          *
 200          * We must do any isize and inode updates
 201          * before we unlock the pages.  Otherwise we
 202          * could end up racing with unlink.
 203          */
 204         BTRFS_I(inode)->disk_i_size = inode->i_size;
 205         btrfs_update_inode(trans, root, inode);
 206
 207         return 0;
 208 fail:
 209         btrfs_free_path(path);
 210         return err;
 211 }
 212
 213
 214 /*
 215  * conditionally insert an inline extent into the file.  This
 216  * does the checks required to make sure the data is small enough
 217  * to fit as an inline extent.
 218  */
 219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 220                                  struct btrfs_root *root,
 221                                  struct inode *inode, u64 start, u64 end,
 222                                  size_t compressed_size, int compress_type,
 223                                  struct page **compressed_pages)
 224 {
 225         u64 isize = i_size_read(inode);
 226         u64 actual_end = min(end + 1, isize);
 227         u64 inline_len = actual_end - start;
 228         u64 aligned_end = (end + root->sectorsize - 1) &
 229                         ~((u64)root->sectorsize - 1);
 230         u64 hint_byte;
 231         u64 data_len = inline_len;
 232         int ret;
 233
 234         if (compressed_size)
 235                 data_len = compressed_size;
 236
 237         if (start > 0 ||
 238             actual_end >= PAGE_CACHE_SIZE ||
 239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 240             (!compressed_size &&
 241             (actual_end & (root->sectorsize - 1)) == 0) ||
 242             end + 1 < isize ||
 243             data_len > root->fs_info->max_inline) {
 244                 return 1;
 245         }
 246
 247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 248                                  &hint_byte, 1);
 249         BUG_ON(ret);
 250
 251         if (isize > actual_end)
 252                 inline_len = min_t(u64, isize, actual_end);
 253         ret = insert_inline_extent(trans, root, inode, start,
 254                                    inline_len, compressed_size,
 255                                    compress_type, compressed_pages);
 256         BUG_ON(ret);
 257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 259         return 0;
 260 }
 261
 262 struct async_extent {
 263         u64 start;
 264         u64 ram_size;
 265         u64 compressed_size;
 266         struct page **pages;
 267         unsigned long nr_pages;
 268         int compress_type;
 269         struct list_head list;
 270 };
 271
 272 struct async_cow {
 273         struct inode *inode;
 274         struct btrfs_root *root;
 275         struct page *locked_page;
 276         u64 start;
 277         u64 end;
 278         struct list_head extents;
 279         struct btrfs_work work;
 280 };
 281
 282 static noinline int add_async_extent(struct async_cow *cow,
 283                                      u64 start, u64 ram_size,
 284                                      u64 compressed_size,
 285                                      struct page **pages,
 286                                      unsigned long nr_pages,
 287                                      int compress_type)
 288 {
 289         struct async_extent *async_extent;
 290
 291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 292         BUG_ON(!async_extent);
 293         async_extent->start = start;
 294         async_extent->ram_size = ram_size;
 295         async_extent->compressed_size = compressed_size;
 296         async_extent->pages = pages;
 297         async_extent->nr_pages = nr_pages;
 298         async_extent->compress_type = compress_type;
 299         list_add_tail(&async_extent->list, &cow->extents);
 300         return 0;
 301 }
 302
 303 /*
 304  * we create compressed extents in two phases.  The first
 305  * phase compresses a range of pages that have already been
 306  * locked (both pages and state bits are locked).
 307  *
 308  * This is done inside an ordered work queue, and the compression
 309  * is spread across many cpus.  The actual IO submission is step
 310  * two, and the ordered work queue takes care of making sure that
 311  * happens in the same order things were put onto the queue by
 312  * writepages and friends.
 313  *
 314  * If this code finds it can't get good compression, it puts an
 315  * entry onto the work queue to write the uncompressed bytes.  This
 316  * makes sure that both compressed inodes and uncompressed inodes
 317  * are written in the same order that pdflush sent them down.
 318  */
 319 static noinline int compress_file_range(struct inode *inode,
 320                                         struct page *locked_page,
 321                                         u64 start, u64 end,
 322                                         struct async_cow *async_cow,
 323                                         int *num_added)
 324 {
 325         struct btrfs_root *root = BTRFS_I(inode)->root;
 326         struct btrfs_trans_handle *trans;
 327         u64 num_bytes;
 328         u64 blocksize = root->sectorsize;
 329         u64 actual_end;
 330         u64 isize = i_size_read(inode);
 331         int ret = 0;
 332         struct page **pages = NULL;
 333         unsigned long nr_pages;
 334         unsigned long nr_pages_ret = 0;
 335         unsigned long total_compressed = 0;
 336         unsigned long total_in = 0;
 337         unsigned long max_compressed = 128 * 1024;
 338         unsigned long max_uncompressed = 128 * 1024;
 339         int i;
 340         int will_compress;
 341         int compress_type = root->fs_info->compress_type;
 342
 343         actual_end = min_t(u64, isize, end + 1);
 344 again:
 345         will_compress = 0;
 346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 348
 349         /*
 350          * we don't want to send crud past the end of i_size through
 351          * compression, that's just a waste of CPU time.  So, if the
 352          * end of the file is before the start of our current
 353          * requested range of bytes, we bail out to the uncompressed
 354          * cleanup code that can deal with all of this.
 355          *
 356          * It isn't really the fastest way to fix things, but this is a
 357          * very uncommon corner.
 358          */
 359         if (actual_end <= start)
 360                 goto cleanup_and_bail_uncompressed;
 361
 362         total_compressed = actual_end - start;
 363
 364         /* we want to make sure that amount of ram required to uncompress
 365          * an extent is reasonable, so we limit the total size in ram
 366          * of a compressed extent to 128k.  This is a crucial number
 367          * because it also controls how easily we can spread reads across
 368          * cpus for decompression.
 369          *
 370          * We also want to make sure the amount of IO required to do
 371          * a random read is reasonably small, so we limit the size of
 372          * a compressed extent to 128k.
 373          */
 374         total_compressed = min(total_compressed, max_uncompressed);
 375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 376         num_bytes = max(blocksize,  num_bytes);
 377         total_in = 0;
 378         ret = 0;
 379
 380         /*
 381          * we do compression for mount -o compress and when the
 382          * inode has not been flagged as nocompress.  This flag can
 383          * change at any time if we discover bad compression ratios.
 384          */
 385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 386             (btrfs_test_opt(root, COMPRESS) ||
 387              (BTRFS_I(inode)->force_compress) ||
 388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 389                 WARN_ON(pages);
 390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 391                 BUG_ON(!pages);
 392
 393                 if (BTRFS_I(inode)->force_compress)
 394                         compress_type = BTRFS_I(inode)->force_compress;
 395
 396                 ret = btrfs_compress_pages(compress_type,
 397                                            inode->i_mapping, start,
 398                                            total_compressed, pages,
 399                                            nr_pages, &nr_pages_ret,
 400                                            &total_in,
 401                                            &total_compressed,
 402                                            max_compressed);
 403
 404                 if (!ret) {
 405                         unsigned long offset = total_compressed &
 406                                 (PAGE_CACHE_SIZE - 1);
 407                         struct page *page = pages[nr_pages_ret - 1];
 408                         char *kaddr;
 409
 410                         /* zero the tail end of the last page, we might be
 411                          * sending it down to disk
 412                          */
 413                         if (offset) {
 414                                 kaddr = kmap_atomic(page, KM_USER0);
 415                                 memset(kaddr + offset, 0,
 416                                        PAGE_CACHE_SIZE - offset);
 417                                 kunmap_atomic(kaddr, KM_USER0);
 418                         }
 419                         will_compress = 1;
 420                 }
 421         }
 422         if (start == 0) {
 423                 trans = btrfs_join_transaction(root, 1);
 424                 BUG_ON(IS_ERR(trans));
 425                 btrfs_set_trans_block_group(trans, inode);
 426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 427
 428                 /* lets try to make an inline extent */
 429                 if (ret || total_in < (actual_end - start)) {
 430                         /* we didn't compress the entire range, try
 431                          * to make an uncompressed inline extent.
 432                          */
 433                         ret = cow_file_range_inline(trans, root, inode,
 434                                                     start, end, 0, 0, NULL);
 435                 } else {
 436                         /* try making a compressed inline extent */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end,
 439                                                     total_compressed,
 440                                                     compress_type, pages);
 441                 }
 442                 if (ret == 0) {
 443                         /*
 444                          * inline extent creation worked, we don't need
 445                          * to create any more async work items.  Unlock
 446                          * and free up our temp pages.
 447                          */
 448                         extent_clear_unlock_delalloc(inode,
 449                              &BTRFS_I(inode)->io_tree,
 450                              start, end, NULL,
 451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 452                              EXTENT_CLEAR_DELALLOC |
 453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 454
 455                         btrfs_end_transaction(trans, root);
 456                         goto free_pages_out;
 457                 }
 458                 btrfs_end_transaction(trans, root);
 459         }
 460
 461         if (will_compress) {
 462                 /*
 463                  * we aren't doing an inline extent round the compressed size
 464                  * up to a block size boundary so the allocator does sane
 465                  * things
 466                  */
 467                 total_compressed = (total_compressed + blocksize - 1) &
 468                         ~(blocksize - 1);
 469
 470                 /*
 471                  * one last check to make sure the compression is really a
 472                  * win, compare the page count read with the blocks on disk
 473                  */
 474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 475                         ~(PAGE_CACHE_SIZE - 1);
 476                 if (total_compressed >= total_in) {
 477                         will_compress = 0;
 478                 } else {
 479                         num_bytes = total_in;
 480                 }
 481         }
 482         if (!will_compress && pages) {
 483                 /*
 484                  * the compression code ran but failed to make things smaller,
 485                  * free any pages it allocated and our page pointer array
 486                  */
 487                 for (i = 0; i < nr_pages_ret; i++) {
 488                         WARN_ON(pages[i]->mapping);
 489                         page_cache_release(pages[i]);
 490                 }
 491                 kfree(pages);
 492                 pages = NULL;
 493                 total_compressed = 0;
 494                 nr_pages_ret = 0;
 495
 496                 /* flag the file so we don't compress in the future */
 497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 498                     !(BTRFS_I(inode)->force_compress)) {
 499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 500                 }
 501         }
 502         if (will_compress) {
 503                 *num_added += 1;
 504
 505                 /* the async work queues will take care of doing actual
 506                  * allocation on disk for these compressed pages,
 507                  * and will submit them to the elevator.
 508                  */
 509                 add_async_extent(async_cow, start, num_bytes,
 510                                  total_compressed, pages, nr_pages_ret,
 511                                  compress_type);
 512
 513                 if (start + num_bytes < end) {
 514                         start += num_bytes;
 515                         pages = NULL;
 516                         cond_resched();
 517                         goto again;
 518                 }
 519         } else {
 520 cleanup_and_bail_uncompressed:
 521                 /*
 522                  * No compression, but we still need to write the pages in
 523                  * the file we've been given so far.  redirty the locked
 524                  * page if it corresponds to our extent and set things up
 525                  * for the async work queue to run cow_file_range to do
 526                  * the normal delalloc dance
 527                  */
 528                 if (page_offset(locked_page) >= start &&
 529                     page_offset(locked_page) <= end) {
 530                         __set_page_dirty_nobuffers(locked_page);
 531                         /* unlocked later on in the async handlers */
 532                 }
 533                 add_async_extent(async_cow, start, end - start + 1,
 534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 535                 *num_added += 1;
 536         }
 537
 538 out:
 539         return 0;
 540
 541 free_pages_out:
 542         for (i = 0; i < nr_pages_ret; i++) {
 543                 WARN_ON(pages[i]->mapping);
 544                 page_cache_release(pages[i]);
 545         }
 546         kfree(pages);
 547
 548         goto out;
 549 }
 550
 551 /*
 552  * phase two of compressed writeback.  This is the ordered portion
 553  * of the code, which only gets called in the order the work was
 554  * queued.  We walk all the async extents created by compress_file_range
 555  * and send them down to the disk.
 556  */
 557 static noinline int submit_compressed_extents(struct inode *inode,
 558                                               struct async_cow *async_cow)
 559 {
 560         struct async_extent *async_extent;
 561         u64 alloc_hint = 0;
 562         struct btrfs_trans_handle *trans;
 563         struct btrfs_key ins;
 564         struct extent_map *em;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 567         struct extent_io_tree *io_tree;
 568         int ret = 0;
 569
 570         if (list_empty(&async_cow->extents))
 571                 return 0;
 572
 573
 574         while (!list_empty(&async_cow->extents)) {
 575                 async_extent = list_entry(async_cow->extents.next,
 576                                           struct async_extent, list);
 577                 list_del(&async_extent->list);
 578
 579                 io_tree = &BTRFS_I(inode)->io_tree;
 580
 581 retry:
 582                 /* did the compression code fall back to uncompressed IO? */
 583                 if (!async_extent->pages) {
 584                         int page_started = 0;
 585                         unsigned long nr_written = 0;
 586
 587                         lock_extent(io_tree, async_extent->start,
 588                                          async_extent->start +
 589                                          async_extent->ram_size - 1, GFP_NOFS);
 590
 591                         /* allocate blocks */
 592                         ret = cow_file_range(inode, async_cow->locked_page,
 593                                              async_extent->start,
 594                                              async_extent->start +
 595                                              async_extent->ram_size - 1,
 596                                              &page_started, &nr_written, 0);
 597
 598                         /*
 599                          * if page_started, cow_file_range inserted an
 600                          * inline extent and took care of all the unlocking
 601                          * and IO for us.  Otherwise, we need to submit
 602                          * all those pages down to the drive.
 603                          */
 604                         if (!page_started && !ret)
 605                                 extent_write_locked_range(io_tree,
 606                                                   inode, async_extent->start,
 607                                                   async_extent->start +
 608                                                   async_extent->ram_size - 1,
 609                                                   btrfs_get_extent,
 610                                                   WB_SYNC_ALL);
 611                         kfree(async_extent);
 612                         cond_resched();
 613                         continue;
 614                 }
 615
 616                 lock_extent(io_tree, async_extent->start,
 617                             async_extent->start + async_extent->ram_size - 1,
 618                             GFP_NOFS);
 619
 620                 trans = btrfs_join_transaction(root, 1);
 621                 BUG_ON(IS_ERR(trans));
 622                 ret = btrfs_reserve_extent(trans, root,
 623                                            async_extent->compressed_size,
 624                                            async_extent->compressed_size,
 625                                            0, alloc_hint,
 626                                            (u64)-1, &ins, 1);
 627                 btrfs_end_transaction(trans, root);
 628
 629                 if (ret) {
 630                         int i;
 631                         for (i = 0; i < async_extent->nr_pages; i++) {
 632                                 WARN_ON(async_extent->pages[i]->mapping);
 633                                 page_cache_release(async_extent->pages[i]);
 634                         }
 635                         kfree(async_extent->pages);
 636                         async_extent->nr_pages = 0;
 637                         async_extent->pages = NULL;
 638                         unlock_extent(io_tree, async_extent->start,
 639                                       async_extent->start +
 640                                       async_extent->ram_size - 1, GFP_NOFS);
 641                         goto retry;
 642                 }
 643
 644                 /*
 645                  * here we're doing allocation and writeback of the
 646                  * compressed pages
 647                  */
 648                 btrfs_drop_extent_cache(inode, async_extent->start,
 649                                         async_extent->start +
 650                                         async_extent->ram_size - 1, 0);
 651
 652                 em = alloc_extent_map(GFP_NOFS);
 653                 BUG_ON(!em);
 654                 em->start = async_extent->start;
 655                 em->len = async_extent->ram_size;
 656                 em->orig_start = em->start;
 657
 658                 em->block_start = ins.objectid;
 659                 em->block_len = ins.offset;
 660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 661                 em->compress_type = async_extent->compress_type;
 662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 664
 665                 while (1) {
 666                         write_lock(&em_tree->lock);
 667                         ret = add_extent_mapping(em_tree, em);
 668                         write_unlock(&em_tree->lock);
 669                         if (ret != -EEXIST) {
 670                                 free_extent_map(em);
 671                                 break;
 672                         }
 673                         btrfs_drop_extent_cache(inode, async_extent->start,
 674                                                 async_extent->start +
 675                                                 async_extent->ram_size - 1, 0);
 676                 }
 677
 678                 ret = btrfs_add_ordered_extent_compress(inode,
 679                                                 async_extent->start,
 680                                                 ins.objectid,
 681                                                 async_extent->ram_size,
 682                                                 ins.offset,
 683                                                 BTRFS_ORDERED_COMPRESSED,
 684                                                 async_extent->compress_type);
 685                 BUG_ON(ret);
 686
 687                 /*
 688                  * clear dirty, set writeback and unlock the pages.
 689                  */
 690                 extent_clear_unlock_delalloc(inode,
 691                                 &BTRFS_I(inode)->io_tree,
 692                                 async_extent->start,
 693                                 async_extent->start +
 694                                 async_extent->ram_size - 1,
 695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 696                                 EXTENT_CLEAR_UNLOCK |
 697                                 EXTENT_CLEAR_DELALLOC |
 698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 699
 700                 ret = btrfs_submit_compressed_write(inode,
 701                                     async_extent->start,
 702                                     async_extent->ram_size,
 703                                     ins.objectid,
 704                                     ins.offset, async_extent->pages,
 705                                     async_extent->nr_pages);
 706
 707                 BUG_ON(ret);
 708                 alloc_hint = ins.objectid + ins.offset;
 709                 kfree(async_extent);
 710                 cond_resched();
 711         }
 712
 713         return 0;
 714 }
 715
 716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 717                                       u64 num_bytes)
 718 {
 719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 720         struct extent_map *em;
 721         u64 alloc_hint = 0;
 722
 723         read_lock(&em_tree->lock);
 724         em = search_extent_mapping(em_tree, start, num_bytes);
 725         if (em) {
 726                 /*
 727                  * if block start isn't an actual block number then find the
 728                  * first block in this inode and use that as a hint.  If that
 729                  * block is also bogus then just don't worry about it.
 730                  */
 731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 732                         free_extent_map(em);
 733                         em = search_extent_mapping(em_tree, 0, 0);
 734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 735                                 alloc_hint = em->block_start;
 736                         if (em)
 737                                 free_extent_map(em);
 738                 } else {
 739                         alloc_hint = em->block_start;
 740                         free_extent_map(em);
 741                 }
 742         }
 743         read_unlock(&em_tree->lock);
 744
 745         return alloc_hint;
 746 }
 747
 748 /*
 749  * when extent_io.c finds a delayed allocation range in the file,
 750  * the call backs end up in this code.  The basic idea is to
 751  * allocate extents on disk for the range, and create ordered data structs
 752  * in ram to track those extents.
 753  *
 754  * locked_page is the page that writepage had locked already.  We use
 755  * it to make sure we don't do extra locks or unlocks.
 756  *
 757  * *page_started is set to one if we unlock locked_page and do everything
 758  * required to start IO on it.  It may be clean and already done with
 759  * IO when we return.
 760  */
 761 static noinline int cow_file_range(struct inode *inode,
 762                                    struct page *locked_page,
 763                                    u64 start, u64 end, int *page_started,
 764                                    unsigned long *nr_written,
 765                                    int unlock)
 766 {
 767         struct btrfs_root *root = BTRFS_I(inode)->root;
 768         struct btrfs_trans_handle *trans;
 769         u64 alloc_hint = 0;
 770         u64 num_bytes;
 771         unsigned long ram_size;
 772         u64 disk_num_bytes;
 773         u64 cur_alloc_size;
 774         u64 blocksize = root->sectorsize;
 775         struct btrfs_key ins;
 776         struct extent_map *em;
 777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 778         int ret = 0;
 779
 780         BUG_ON(root == root->fs_info->tree_root);
 781         trans = btrfs_join_transaction(root, 1);
 782         BUG_ON(IS_ERR(trans));
 783         btrfs_set_trans_block_group(trans, inode);
 784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 785
 786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 787         num_bytes = max(blocksize,  num_bytes);
 788         disk_num_bytes = num_bytes;
 789         ret = 0;
 790
 791         if (start == 0) {
 792                 /* lets try to make an inline extent */
 793                 ret = cow_file_range_inline(trans, root, inode,
 794                                             start, end, 0, 0, NULL);
 795                 if (ret == 0) {
 796                         extent_clear_unlock_delalloc(inode,
 797                                      &BTRFS_I(inode)->io_tree,
 798                                      start, end, NULL,
 799                                      EXTENT_CLEAR_UNLOCK_PAGE |
 800                                      EXTENT_CLEAR_UNLOCK |
 801                                      EXTENT_CLEAR_DELALLOC |
 802                                      EXTENT_CLEAR_DIRTY |
 803                                      EXTENT_SET_WRITEBACK |
 804                                      EXTENT_END_WRITEBACK);
 805
 806                         *nr_written = *nr_written +
 807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 808                         *page_started = 1;
 809                         ret = 0;
 810                         goto out;
 811                 }
 812         }
 813
 814         BUG_ON(disk_num_bytes >
 815                btrfs_super_total_bytes(&root->fs_info->super_copy));
 816
 817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 819
 820         while (disk_num_bytes > 0) {
 821                 unsigned long op;
 822
 823                 cur_alloc_size = disk_num_bytes;
 824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 825                                            root->sectorsize, 0, alloc_hint,
 826                                            (u64)-1, &ins, 1);
 827                 BUG_ON(ret);
 828
 829                 em = alloc_extent_map(GFP_NOFS);
 830                 BUG_ON(!em);
 831                 em->start = start;
 832                 em->orig_start = em->start;
 833                 ram_size = ins.offset;
 834                 em->len = ins.offset;
 835
 836                 em->block_start = ins.objectid;
 837                 em->block_len = ins.offset;
 838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 840
 841                 while (1) {
 842                         write_lock(&em_tree->lock);
 843                         ret = add_extent_mapping(em_tree, em);
 844                         write_unlock(&em_tree->lock);
 845                         if (ret != -EEXIST) {
 846                                 free_extent_map(em);
 847                                 break;
 848                         }
 849                         btrfs_drop_extent_cache(inode, start,
 850                                                 start + ram_size - 1, 0);
 851                 }
 852
 853                 cur_alloc_size = ins.offset;
 854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 855                                                ram_size, cur_alloc_size, 0);
 856                 BUG_ON(ret);
 857
 858                 if (root->root_key.objectid ==
 859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 860                         ret = btrfs_reloc_clone_csums(inode, start,
 861                                                       cur_alloc_size);
 862                         BUG_ON(ret);
 863                 }
 864
 865                 if (disk_num_bytes < cur_alloc_size)
 866                         break;
 867
 868                 /* we're not doing compressed IO, don't unlock the first
 869                  * page (which the caller expects to stay locked), don't
 870                  * clear any dirty bits and don't set any writeback bits
 871                  *
 872                  * Do set the Private2 bit so we know this page was properly
 873                  * setup for writepage
 874                  */
 875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 877                         EXTENT_SET_PRIVATE2;
 878
 879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 880                                              start, start + ram_size - 1,
 881                                              locked_page, op);
 882                 disk_num_bytes -= cur_alloc_size;
 883                 num_bytes -= cur_alloc_size;
 884                 alloc_hint = ins.objectid + ins.offset;
 885                 start += cur_alloc_size;
 886         }
 887 out:
 888         ret = 0;
 889         btrfs_end_transaction(trans, root);
 890
 891         return ret;
 892 }
 893
 894 /*
 895  * work queue call back to started compression on a file and pages
 896  */
 897 static noinline void async_cow_start(struct btrfs_work *work)
 898 {
 899         struct async_cow *async_cow;
 900         int num_added = 0;
 901         async_cow = container_of(work, struct async_cow, work);
 902
 903         compress_file_range(async_cow->inode, async_cow->locked_page,
 904                             async_cow->start, async_cow->end, async_cow,
 905                             &num_added);
 906         if (num_added == 0)
 907                 async_cow->inode = NULL;
 908 }
 909
 910 /*
 911  * work queue call back to submit previously compressed pages
 912  */
 913 static noinline void async_cow_submit(struct btrfs_work *work)
 914 {
 915         struct async_cow *async_cow;
 916         struct btrfs_root *root;
 917         unsigned long nr_pages;
 918
 919         async_cow = container_of(work, struct async_cow, work);
 920
 921         root = async_cow->root;
 922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 923                 PAGE_CACHE_SHIFT;
 924
 925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 926
 927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 928             5 * 1042 * 1024 &&
 929             waitqueue_active(&root->fs_info->async_submit_wait))
 930                 wake_up(&root->fs_info->async_submit_wait);
 931
 932         if (async_cow->inode)
 933                 submit_compressed_extents(async_cow->inode, async_cow);
 934 }
 935
 936 static noinline void async_cow_free(struct btrfs_work *work)
 937 {
 938         struct async_cow *async_cow;
 939         async_cow = container_of(work, struct async_cow, work);
 940         kfree(async_cow);
 941 }
 942
 943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 944                                 u64 start, u64 end, int *page_started,
 945                                 unsigned long *nr_written)
 946 {
 947         struct async_cow *async_cow;
 948         struct btrfs_root *root = BTRFS_I(inode)->root;
 949         unsigned long nr_pages;
 950         u64 cur_end;
 951         int limit = 10 * 1024 * 1042;
 952
 953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 954                          1, 0, NULL, GFP_NOFS);
 955         while (start < end) {
 956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 957                 async_cow->inode = inode;
 958                 async_cow->root = root;
 959                 async_cow->locked_page = locked_page;
 960                 async_cow->start = start;
 961
 962                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 963                         cur_end = end;
 964                 else
 965                         cur_end = min(end, start + 512 * 1024 - 1);
 966
 967                 async_cow->end = cur_end;
 968                 INIT_LIST_HEAD(&async_cow->extents);
 969
 970                 async_cow->work.func = async_cow_start;
 971                 async_cow->work.ordered_func = async_cow_submit;
 972                 async_cow->work.ordered_free = async_cow_free;
 973                 async_cow->work.flags = 0;
 974
 975                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 976                         PAGE_CACHE_SHIFT;
 977                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 978
 979                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 980                                    &async_cow->work);
 981
 982                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 983                         wait_event(root->fs_info->async_submit_wait,
 984                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 985                             limit));
 986                 }
 987
 988                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 989                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 990                         wait_event(root->fs_info->async_submit_wait,
 991                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 992                            0));
 993                 }
 994
 995                 *nr_written += nr_pages;
 996                 start = cur_end + 1;
 997         }
 998         *page_started = 1;
 999         return 0;
1000 }
1001
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003                                         u64 bytenr, u64 num_bytes)
1004 {
1005         int ret;
1006         struct btrfs_ordered_sum *sums;
1007         LIST_HEAD(list);
1008
1009         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010                                        bytenr + num_bytes - 1, &list);
1011         if (ret == 0 && list_empty(&list))
1012                 return 0;
1013
1014         while (!list_empty(&list)) {
1015                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016                 list_del(&sums->list);
1017                 kfree(sums);
1018         }
1019         return 1;
1020 }
1021
1022 /*
1023  * when nowcow writeback call back.  This checks for snapshots or COW copies
1024  * of the extents that exist in the file, and COWs the file as required.
1025  *
1026  * If no cow copies or snapshots exist, we write directly to the existing
1027  * blocks on disk
1028  */
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030                                        struct page *locked_page,
1031                               u64 start, u64 end, int *page_started, int force,
1032                               unsigned long *nr_written)
1033 {
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         struct btrfs_trans_handle *trans;
1036         struct extent_buffer *leaf;
1037         struct btrfs_path *path;
1038         struct btrfs_file_extent_item *fi;
1039         struct btrfs_key found_key;
1040         u64 cow_start;
1041         u64 cur_offset;
1042         u64 extent_end;
1043         u64 extent_offset;
1044         u64 disk_bytenr;
1045         u64 num_bytes;
1046         int extent_type;
1047         int ret;
1048         int type;
1049         int nocow;
1050         int check_prev = 1;
1051         bool nolock = false;
1052
1053         path = btrfs_alloc_path();
1054         BUG_ON(!path);
1055         if (root == root->fs_info->tree_root) {
1056                 nolock = true;
1057                 trans = btrfs_join_transaction_nolock(root, 1);
1058         } else {
1059                 trans = btrfs_join_transaction(root, 1);
1060         }
1061         BUG_ON(IS_ERR(trans));
1062
1063         cow_start = (u64)-1;
1064         cur_offset = start;
1065         while (1) {
1066                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1067                                                cur_offset, 0);
1068                 BUG_ON(ret < 0);
1069                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070                         leaf = path->nodes[0];
1071                         btrfs_item_key_to_cpu(leaf, &found_key,
1072                                               path->slots[0] - 1);
1073                         if (found_key.objectid == inode->i_ino &&
1074                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1075                                 path->slots[0]--;
1076                 }
1077                 check_prev = 0;
1078 next_slot:
1079                 leaf = path->nodes[0];
1080                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081                         ret = btrfs_next_leaf(root, path);
1082                         if (ret < 0)
1083                                 BUG_ON(1);
1084                         if (ret > 0)
1085                                 break;
1086                         leaf = path->nodes[0];
1087                 }
1088
1089                 nocow = 0;
1090                 disk_bytenr = 0;
1091                 num_bytes = 0;
1092                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093
1094                 if (found_key.objectid > inode->i_ino ||
1095                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096                     found_key.offset > end)
1097                         break;
1098
1099                 if (found_key.offset > cur_offset) {
1100                         extent_end = found_key.offset;
1101                         extent_type = 0;
1102                         goto out_check;
1103                 }
1104
1105                 fi = btrfs_item_ptr(leaf, path->slots[0],
1106                                     struct btrfs_file_extent_item);
1107                 extent_type = btrfs_file_extent_type(leaf, fi);
1108
1109                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1113                         extent_end = found_key.offset +
1114                                 btrfs_file_extent_num_bytes(leaf, fi);
1115                         if (extent_end <= start) {
1116                                 path->slots[0]++;
1117                                 goto next_slot;
1118                         }
1119                         if (disk_bytenr == 0)
1120                                 goto out_check;
1121                         if (btrfs_file_extent_compression(leaf, fi) ||
1122                             btrfs_file_extent_encryption(leaf, fi) ||
1123                             btrfs_file_extent_other_encoding(leaf, fi))
1124                                 goto out_check;
1125                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126                                 goto out_check;
1127                         if (btrfs_extent_readonly(root, disk_bytenr))
1128                                 goto out_check;
1129                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130                                                   found_key.offset -
1131                                                   extent_offset, disk_bytenr))
1132                                 goto out_check;
1133                         disk_bytenr += extent_offset;
1134                         disk_bytenr += cur_offset - found_key.offset;
1135                         num_bytes = min(end + 1, extent_end) - cur_offset;
1136                         /*
1137                          * force cow if csum exists in the range.
1138                          * this ensure that csum for a given extent are
1139                          * either valid or do not exist.
1140                          */
1141                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1142                                 goto out_check;
1143                         nocow = 1;
1144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145                         extent_end = found_key.offset +
1146                                 btrfs_file_extent_inline_len(leaf, fi);
1147                         extent_end = ALIGN(extent_end, root->sectorsize);
1148                 } else {
1149                         BUG_ON(1);
1150                 }
1151 out_check:
1152                 if (extent_end <= start) {
1153                         path->slots[0]++;
1154                         goto next_slot;
1155                 }
1156                 if (!nocow) {
1157                         if (cow_start == (u64)-1)
1158                                 cow_start = cur_offset;
1159                         cur_offset = extent_end;
1160                         if (cur_offset > end)
1161                                 break;
1162                         path->slots[0]++;
1163                         goto next_slot;
1164                 }
1165
1166                 btrfs_release_path(root, path);
1167                 if (cow_start != (u64)-1) {
1168                         ret = cow_file_range(inode, locked_page, cow_start,
1169                                         found_key.offset - 1, page_started,
1170                                         nr_written, 1);
1171                         BUG_ON(ret);
1172                         cow_start = (u64)-1;
1173                 }
1174
1175                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176                         struct extent_map *em;
1177                         struct extent_map_tree *em_tree;
1178                         em_tree = &BTRFS_I(inode)->extent_tree;
1179                         em = alloc_extent_map(GFP_NOFS);
1180                         BUG_ON(!em);
1181                         em->start = cur_offset;
1182                         em->orig_start = em->start;
1183                         em->len = num_bytes;
1184                         em->block_len = num_bytes;
1185                         em->block_start = disk_bytenr;
1186                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1187                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188                         while (1) {
1189                                 write_lock(&em_tree->lock);
1190                                 ret = add_extent_mapping(em_tree, em);
1191                                 write_unlock(&em_tree->lock);
1192                                 if (ret != -EEXIST) {
1193                                         free_extent_map(em);
1194                                         break;
1195                                 }
1196                                 btrfs_drop_extent_cache(inode, em->start,
1197                                                 em->start + em->len - 1, 0);
1198                         }
1199                         type = BTRFS_ORDERED_PREALLOC;
1200                 } else {
1201                         type = BTRFS_ORDERED_NOCOW;
1202                 }
1203
1204                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205                                                num_bytes, num_bytes, type);
1206                 BUG_ON(ret);
1207
1208                 if (root->root_key.objectid ==
1209                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1211                                                       num_bytes);
1212                         BUG_ON(ret);
1213                 }
1214
1215                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216                                 cur_offset, cur_offset + num_bytes - 1,
1217                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219                                 EXTENT_SET_PRIVATE2);
1220                 cur_offset = extent_end;
1221                 if (cur_offset > end)
1222                         break;
1223         }
1224         btrfs_release_path(root, path);
1225
1226         if (cur_offset <= end && cow_start == (u64)-1)
1227                 cow_start = cur_offset;
1228         if (cow_start != (u64)-1) {
1229                 ret = cow_file_range(inode, locked_page, cow_start, end,
1230                                      page_started, nr_written, 1);
1231                 BUG_ON(ret);
1232         }
1233
1234         if (nolock) {
1235                 ret = btrfs_end_transaction_nolock(trans, root);
1236                 BUG_ON(ret);
1237         } else {
1238                 ret = btrfs_end_transaction(trans, root);
1239                 BUG_ON(ret);
1240         }
1241         btrfs_free_path(path);
1242         return 0;
1243 }
1244
1245 /*
1246  * extent_io.c call back to do delayed allocation processing
1247  */
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249                               u64 start, u64 end, int *page_started,
1250                               unsigned long *nr_written)
1251 {
1252         int ret;
1253         struct btrfs_root *root = BTRFS_I(inode)->root;
1254
1255         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257                                          page_started, 1, nr_written);
1258         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260                                          page_started, 0, nr_written);
1261         else if (!btrfs_test_opt(root, COMPRESS) &&
1262                  !(BTRFS_I(inode)->force_compress) &&
1263                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264                 ret = cow_file_range(inode, locked_page, start, end,
1265                                       page_started, nr_written, 1);
1266         else
1267                 ret = cow_file_range_async(inode, locked_page, start, end,
1268                                            page_started, nr_written);
1269         return ret;
1270 }
1271
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273                                    struct extent_state *orig, u64 split)
1274 {
1275         /* not delalloc, ignore it */
1276         if (!(orig->state & EXTENT_DELALLOC))
1277                 return 0;
1278
1279         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1280         return 0;
1281 }
1282
1283 /*
1284  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285  * extents so we can keep track of new extents that are just merged onto old
1286  * extents, such as when we are doing sequential writes, so we can properly
1287  * account for the metadata space we'll need.
1288  */
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290                                    struct extent_state *new,
1291                                    struct extent_state *other)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(other->state & EXTENT_DELALLOC))
1295                 return 0;
1296
1297         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1298         return 0;
1299 }
1300
1301 /*
1302  * extent_io.c set_bit_hook, used to track delayed allocation
1303  * bytes in this file, and to maintain the list of inodes that
1304  * have pending delalloc work to be done.
1305  */
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307                               struct extent_state *state, int *bits)
1308 {
1309
1310         /*
1311          * set_bit and clear bit hooks normally require _irqsave/restore
1312          * but in this case, we are only testeing for the DELALLOC
1313          * bit, which is only set or cleared with irqs on
1314          */
1315         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316                 struct btrfs_root *root = BTRFS_I(inode)->root;
1317                 u64 len = state->end + 1 - state->start;
1318                 int do_list = (root->root_key.objectid !=
1319                                BTRFS_ROOT_TREE_OBJECTID);
1320
1321                 if (*bits & EXTENT_FIRST_DELALLOC)
1322                         *bits &= ~EXTENT_FIRST_DELALLOC;
1323                 else
1324                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325
1326                 spin_lock(&root->fs_info->delalloc_lock);
1327                 BTRFS_I(inode)->delalloc_bytes += len;
1328                 root->fs_info->delalloc_bytes += len;
1329                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331                                       &root->fs_info->delalloc_inodes);
1332                 }
1333                 spin_unlock(&root->fs_info->delalloc_lock);
1334         }
1335         return 0;
1336 }
1337
1338 /*
1339  * extent_io.c clear_bit_hook, see set_bit_hook for why
1340  */
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342                                 struct extent_state *state, int *bits)
1343 {
1344         /*
1345          * set_bit and clear bit hooks normally require _irqsave/restore
1346          * but in this case, we are only testeing for the DELALLOC
1347          * bit, which is only set or cleared with irqs on
1348          */
1349         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350                 struct btrfs_root *root = BTRFS_I(inode)->root;
1351                 u64 len = state->end + 1 - state->start;
1352                 int do_list = (root->root_key.objectid !=
1353                                BTRFS_ROOT_TREE_OBJECTID);
1354
1355                 if (*bits & EXTENT_FIRST_DELALLOC)
1356                         *bits &= ~EXTENT_FIRST_DELALLOC;
1357                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359
1360                 if (*bits & EXTENT_DO_ACCOUNTING)
1361                         btrfs_delalloc_release_metadata(inode, len);
1362
1363                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364                     && do_list)
1365                         btrfs_free_reserved_data_space(inode, len);
1366
1367                 spin_lock(&root->fs_info->delalloc_lock);
1368                 root->fs_info->delalloc_bytes -= len;
1369                 BTRFS_I(inode)->delalloc_bytes -= len;
1370
1371                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374                 }
1375                 spin_unlock(&root->fs_info->delalloc_lock);
1376         }
1377         return 0;
1378 }
1379
1380 /*
1381  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382  * we don't create bios that span stripes or chunks
1383  */
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385                          size_t size, struct bio *bio,
1386                          unsigned long bio_flags)
1387 {
1388         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389         struct btrfs_mapping_tree *map_tree;
1390         u64 logical = (u64)bio->bi_sector << 9;
1391         u64 length = 0;
1392         u64 map_length;
1393         int ret;
1394
1395         if (bio_flags & EXTENT_BIO_COMPRESSED)
1396                 return 0;
1397
1398         length = bio->bi_size;
1399         map_tree = &root->fs_info->mapping_tree;
1400         map_length = length;
1401         ret = btrfs_map_block(map_tree, READ, logical,
1402                               &map_length, NULL, 0);
1403
1404         if (map_length < length + size)
1405                 return 1;
1406         return ret;
1407 }
1408
1409 /*
1410  * in order to insert checksums into the metadata in large chunks,
1411  * we wait until bio submission time.   All the pages in the bio are
1412  * checksummed and sums are attached onto the ordered extent record.
1413  *
1414  * At IO completion time the cums attached on the ordered extent record
1415  * are inserted into the btree
1416  */
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418                                     struct bio *bio, int mirror_num,
1419                                     unsigned long bio_flags,
1420                                     u64 bio_offset)
1421 {
1422         struct btrfs_root *root = BTRFS_I(inode)->root;
1423         int ret = 0;
1424
1425         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1426         BUG_ON(ret);
1427         return 0;
1428 }
1429
1430 /*
1431  * in order to insert checksums into the metadata in large chunks,
1432  * we wait until bio submission time.   All the pages in the bio are
1433  * checksummed and sums are attached onto the ordered extent record.
1434  *
1435  * At IO completion time the cums attached on the ordered extent record
1436  * are inserted into the btree
1437  */
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439                           int mirror_num, unsigned long bio_flags,
1440                           u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1444 }
1445
1446 /*
1447  * extent_io.c submission hook. This does the right thing for csum calculation
1448  * on write, or reading the csums from the tree before a read
1449  */
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451                           int mirror_num, unsigned long bio_flags,
1452                           u64 bio_offset)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         int ret = 0;
1456         int skip_sum;
1457
1458         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459
1460         if (root == root->fs_info->tree_root)
1461                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462         else
1463                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1464         BUG_ON(ret);
1465
1466         if (!(rw & REQ_WRITE)) {
1467                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468                         return btrfs_submit_compressed_read(inode, bio,
1469                                                     mirror_num, bio_flags);
1470                 } else if (!skip_sum) {
1471                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1472                         if (ret)
1473                                 return ret;
1474                 }
1475                 goto mapit;
1476         } else if (!skip_sum) {
1477                 /* csum items have already been cloned */
1478                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479                         goto mapit;
1480                 /* we're doing a write, do the async checksumming */
1481                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482                                    inode, rw, bio, mirror_num,
1483                                    bio_flags, bio_offset,
1484                                    __btrfs_submit_bio_start,
1485                                    __btrfs_submit_bio_done);
1486         }
1487
1488 mapit:
1489         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1490 }
1491
1492 /*
1493  * given a list of ordered sums record them in the inode.  This happens
1494  * at IO completion time based on sums calculated at bio submission time.
1495  */
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497                              struct inode *inode, u64 file_offset,
1498                              struct list_head *list)
1499 {
1500         struct btrfs_ordered_sum *sum;
1501
1502         btrfs_set_trans_block_group(trans, inode);
1503
1504         list_for_each_entry(sum, list, list) {
1505                 btrfs_csum_file_blocks(trans,
1506                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1507         }
1508         return 0;
1509 }
1510
1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1512                               struct extent_state **cached_state)
1513 {
1514         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515                 WARN_ON(1);
1516         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1517                                    cached_state, GFP_NOFS);
1518 }
1519
1520 /* see btrfs_writepage_start_hook for details on why this is required */
1521 struct btrfs_writepage_fixup {
1522         struct page *page;
1523         struct btrfs_work work;
1524 };
1525
1526 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 {
1528         struct btrfs_writepage_fixup *fixup;
1529         struct btrfs_ordered_extent *ordered;
1530         struct extent_state *cached_state = NULL;
1531         struct page *page;
1532         struct inode *inode;
1533         u64 page_start;
1534         u64 page_end;
1535
1536         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1537         page = fixup->page;
1538 again:
1539         lock_page(page);
1540         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1541                 ClearPageChecked(page);
1542                 goto out_page;
1543         }
1544
1545         inode = page->mapping->host;
1546         page_start = page_offset(page);
1547         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548
1549         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1550                          &cached_state, GFP_NOFS);
1551
1552         /* already ordered? We're done */
1553         if (PagePrivate2(page))
1554                 goto out;
1555
1556         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557         if (ordered) {
1558                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1559                                      page_end, &cached_state, GFP_NOFS);
1560                 unlock_page(page);
1561                 btrfs_start_ordered_extent(inode, ordered, 1);
1562                 goto again;
1563         }
1564
1565         BUG();
1566         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1567         ClearPageChecked(page);
1568 out:
1569         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1570                              &cached_state, GFP_NOFS);
1571 out_page:
1572         unlock_page(page);
1573         page_cache_release(page);
1574         kfree(fixup);
1575 }
1576
1577 /*
1578  * There are a few paths in the higher layers of the kernel that directly
1579  * set the page dirty bit without asking the filesystem if it is a
1580  * good idea.  This causes problems because we want to make sure COW
1581  * properly happens and the data=ordered rules are followed.
1582  *
1583  * In our case any range that doesn't have the ORDERED bit set
1584  * hasn't been properly setup for IO.  We kick off an async process
1585  * to fix it up.  The async helper will wait for ordered extents, set
1586  * the delalloc bit and make it safe to write the page.
1587  */
1588 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 {
1590         struct inode *inode = page->mapping->host;
1591         struct btrfs_writepage_fixup *fixup;
1592         struct btrfs_root *root = BTRFS_I(inode)->root;
1593
1594         /* this page is properly in the ordered list */
1595         if (TestClearPagePrivate2(page))
1596                 return 0;
1597
1598         if (PageChecked(page))
1599                 return -EAGAIN;
1600
1601         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1602         if (!fixup)
1603                 return -EAGAIN;
1604
1605         SetPageChecked(page);
1606         page_cache_get(page);
1607         fixup->work.func = btrfs_writepage_fixup_worker;
1608         fixup->page = page;
1609         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1610         return -EAGAIN;
1611 }
1612
1613 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1614                                        struct inode *inode, u64 file_pos,
1615                                        u64 disk_bytenr, u64 disk_num_bytes,
1616                                        u64 num_bytes, u64 ram_bytes,
1617                                        u8 compression, u8 encryption,
1618                                        u16 other_encoding, int extent_type)
1619 {
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621         struct btrfs_file_extent_item *fi;
1622         struct btrfs_path *path;
1623         struct extent_buffer *leaf;
1624         struct btrfs_key ins;
1625         u64 hint;
1626         int ret;
1627
1628         path = btrfs_alloc_path();
1629         BUG_ON(!path);
1630
1631         path->leave_spinning = 1;
1632
1633         /*
1634          * we may be replacing one extent in the tree with another.
1635          * The new extent is pinned in the extent map, and we don't want
1636          * to drop it from the cache until it is completely in the btree.
1637          *
1638          * So, tell btrfs_drop_extents to leave this extent in the cache.
1639          * the caller is expected to unpin it and allow it to be merged
1640          * with the others.
1641          */
1642         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1643                                  &hint, 0);
1644         BUG_ON(ret);
1645
1646         ins.objectid = inode->i_ino;
1647         ins.offset = file_pos;
1648         ins.type = BTRFS_EXTENT_DATA_KEY;
1649         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650         BUG_ON(ret);
1651         leaf = path->nodes[0];
1652         fi = btrfs_item_ptr(leaf, path->slots[0],
1653                             struct btrfs_file_extent_item);
1654         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1655         btrfs_set_file_extent_type(leaf, fi, extent_type);
1656         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1657         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1658         btrfs_set_file_extent_offset(leaf, fi, 0);
1659         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1660         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1661         btrfs_set_file_extent_compression(leaf, fi, compression);
1662         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1663         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664
1665         btrfs_unlock_up_safe(path, 1);
1666         btrfs_set_lock_blocking(leaf);
1667
1668         btrfs_mark_buffer_dirty(leaf);
1669
1670         inode_add_bytes(inode, num_bytes);
1671
1672         ins.objectid = disk_bytenr;
1673         ins.offset = disk_num_bytes;
1674         ins.type = BTRFS_EXTENT_ITEM_KEY;
1675         ret = btrfs_alloc_reserved_file_extent(trans, root,
1676                                         root->root_key.objectid,
1677                                         inode->i_ino, file_pos, &ins);
1678         BUG_ON(ret);
1679         btrfs_free_path(path);
1680
1681         return 0;
1682 }
1683
1684 /*
1685  * helper function for btrfs_finish_ordered_io, this
1686  * just reads in some of the csum leaves to prime them into ram
1687  * before we start the transaction.  It limits the amount of btree
1688  * reads required while inside the transaction.
1689  */
1690 /* as ordered data IO finishes, this gets called so we can finish
1691  * an ordered extent if the range of bytes in the file it covers are
1692  * fully written.
1693  */
1694 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 {
1696         struct btrfs_root *root = BTRFS_I(inode)->root;
1697         struct btrfs_trans_handle *trans = NULL;
1698         struct btrfs_ordered_extent *ordered_extent = NULL;
1699         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1700         struct extent_state *cached_state = NULL;
1701         int compress_type = 0;
1702         int ret;
1703         bool nolock = false;
1704
1705         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1706                                              end - start + 1);
1707         if (!ret)
1708                 return 0;
1709         BUG_ON(!ordered_extent);
1710
1711         nolock = (root == root->fs_info->tree_root);
1712
1713         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1714                 BUG_ON(!list_empty(&ordered_extent->list));
1715                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716                 if (!ret) {
1717                         if (nolock)
1718                                 trans = btrfs_join_transaction_nolock(root, 1);
1719                         else
1720                                 trans = btrfs_join_transaction(root, 1);
1721                         BUG_ON(IS_ERR(trans));
1722                         btrfs_set_trans_block_group(trans, inode);
1723                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1724                         ret = btrfs_update_inode(trans, root, inode);
1725                         BUG_ON(ret);
1726                 }
1727                 goto out;
1728         }
1729
1730         lock_extent_bits(io_tree, ordered_extent->file_offset,
1731                          ordered_extent->file_offset + ordered_extent->len - 1,
1732                          0, &cached_state, GFP_NOFS);
1733
1734         if (nolock)
1735                 trans = btrfs_join_transaction_nolock(root, 1);
1736         else
1737                 trans = btrfs_join_transaction(root, 1);
1738         BUG_ON(IS_ERR(trans));
1739         btrfs_set_trans_block_group(trans, inode);
1740         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741
1742         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1743                 compress_type = ordered_extent->compress_type;
1744         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1745                 BUG_ON(compress_type);
1746                 ret = btrfs_mark_extent_written(trans, inode,
1747                                                 ordered_extent->file_offset,
1748                                                 ordered_extent->file_offset +
1749                                                 ordered_extent->len);
1750                 BUG_ON(ret);
1751         } else {
1752                 BUG_ON(root == root->fs_info->tree_root);
1753                 ret = insert_reserved_file_extent(trans, inode,
1754                                                 ordered_extent->file_offset,
1755                                                 ordered_extent->start,
1756                                                 ordered_extent->disk_len,
1757                                                 ordered_extent->len,
1758                                                 ordered_extent->len,
1759                                                 compress_type, 0, 0,
1760                                                 BTRFS_FILE_EXTENT_REG);
1761                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1762                                    ordered_extent->file_offset,
1763                                    ordered_extent->len);
1764                 BUG_ON(ret);
1765         }
1766         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1767                              ordered_extent->file_offset +
1768                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769
1770         add_pending_csums(trans, inode, ordered_extent->file_offset,
1771                           &ordered_extent->list);
1772
1773         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1774         if (!ret) {
1775                 ret = btrfs_update_inode(trans, root, inode);
1776                 BUG_ON(ret);
1777         }
1778         ret = 0;
1779 out:
1780         if (nolock) {
1781                 if (trans)
1782                         btrfs_end_transaction_nolock(trans, root);
1783         } else {
1784                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1785                 if (trans)
1786                         btrfs_end_transaction(trans, root);
1787         }
1788
1789         /* once for us */
1790         btrfs_put_ordered_extent(ordered_extent);
1791         /* once for the tree */
1792         btrfs_put_ordered_extent(ordered_extent);
1793
1794         return 0;
1795 }
1796
1797 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1798                                 struct extent_state *state, int uptodate)
1799 {
1800         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1801
1802         ClearPagePrivate2(page);
1803         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1804 }
1805
1806 /*
1807  * When IO fails, either with EIO or csum verification fails, we
1808  * try other mirrors that might have a good copy of the data.  This
1809  * io_failure_record is used to record state as we go through all the
1810  * mirrors.  If another mirror has good data, the page is set up to date
1811  * and things continue.  If a good mirror can't be found, the original
1812  * bio end_io callback is called to indicate things have failed.
1813  */
1814 struct io_failure_record {
1815         struct page *page;
1816         u64 start;
1817         u64 len;
1818         u64 logical;
1819         unsigned long bio_flags;
1820         int last_mirror;
1821 };
1822
1823 static int btrfs_io_failed_hook(struct bio *failed_bio,
1824                          struct page *page, u64 start, u64 end,
1825                          struct extent_state *state)
1826 {
1827         struct io_failure_record *failrec = NULL;
1828         u64 private;
1829         struct extent_map *em;
1830         struct inode *inode = page->mapping->host;
1831         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1832         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1833         struct bio *bio;
1834         int num_copies;
1835         int ret;
1836         int rw;
1837         u64 logical;
1838
1839         ret = get_state_private(failure_tree, start, &private);
1840         if (ret) {
1841                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1842                 if (!failrec)
1843                         return -ENOMEM;
1844                 failrec->start = start;
1845                 failrec->len = end - start + 1;
1846                 failrec->last_mirror = 0;
1847                 failrec->bio_flags = 0;
1848
1849                 read_lock(&em_tree->lock);
1850                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1851                 if (em->start > start || em->start + em->len < start) {
1852                         free_extent_map(em);
1853                         em = NULL;
1854                 }
1855                 read_unlock(&em_tree->lock);
1856
1857                 if (!em || IS_ERR(em)) {
1858                         kfree(failrec);
1859                         return -EIO;
1860                 }
1861                 logical = start - em->start;
1862                 logical = em->block_start + logical;
1863                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1864                         logical = em->block_start;
1865                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1866                         extent_set_compress_type(&failrec->bio_flags,
1867                                                  em->compress_type);
1868                 }
1869                 failrec->logical = logical;
1870                 free_extent_map(em);
1871                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1872                                 EXTENT_DIRTY, GFP_NOFS);
1873                 set_state_private(failure_tree, start,
1874                                  (u64)(unsigned long)failrec);
1875         } else {
1876                 failrec = (struct io_failure_record *)(unsigned long)private;
1877         }
1878         num_copies = btrfs_num_copies(
1879                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1880                               failrec->logical, failrec->len);
1881         failrec->last_mirror++;
1882         if (!state) {
1883                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1884                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1885                                                     failrec->start,
1886                                                     EXTENT_LOCKED);
1887                 if (state && state->start != failrec->start)
1888                         state = NULL;
1889                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1890         }
1891         if (!state || failrec->last_mirror > num_copies) {
1892                 set_state_private(failure_tree, failrec->start, 0);
1893                 clear_extent_bits(failure_tree, failrec->start,
1894                                   failrec->start + failrec->len - 1,
1895                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1896                 kfree(failrec);
1897                 return -EIO;
1898         }
1899         bio = bio_alloc(GFP_NOFS, 1);
1900         bio->bi_private = state;
1901         bio->bi_end_io = failed_bio->bi_end_io;
1902         bio->bi_sector = failrec->logical >> 9;
1903         bio->bi_bdev = failed_bio->bi_bdev;
1904         bio->bi_size = 0;
1905
1906         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1907         if (failed_bio->bi_rw & REQ_WRITE)
1908                 rw = WRITE;
1909         else
1910                 rw = READ;
1911
1912         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1913                                                       failrec->last_mirror,
1914                                                       failrec->bio_flags, 0);
1915         return ret;
1916 }
1917
1918 /*
1919  * each time an IO finishes, we do a fast check in the IO failure tree
1920  * to see if we need to process or clean up an io_failure_record
1921  */
1922 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1923 {
1924         u64 private;
1925         u64 private_failure;
1926         struct io_failure_record *failure;
1927         int ret;
1928
1929         private = 0;
1930         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1931                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1932                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1933                                         start, &private_failure);
1934                 if (ret == 0) {
1935                         failure = (struct io_failure_record *)(unsigned long)
1936                                    private_failure;
1937                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1938                                           failure->start, 0);
1939                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1940                                           failure->start,
1941                                           failure->start + failure->len - 1,
1942                                           EXTENT_DIRTY | EXTENT_LOCKED,
1943                                           GFP_NOFS);
1944                         kfree(failure);
1945                 }
1946         }
1947         return 0;
1948 }
1949
1950 /*
1951  * when reads are done, we need to check csums to verify the data is correct
1952  * if there's a match, we allow the bio to finish.  If not, we go through
1953  * the io_failure_record routines to find good copies
1954  */
1955 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1956                                struct extent_state *state)
1957 {
1958         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1959         struct inode *inode = page->mapping->host;
1960         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1961         char *kaddr;
1962         u64 private = ~(u32)0;
1963         int ret;
1964         struct btrfs_root *root = BTRFS_I(inode)->root;
1965         u32 csum = ~(u32)0;
1966
1967         if (PageChecked(page)) {
1968                 ClearPageChecked(page);
1969                 goto good;
1970         }
1971
1972         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1973                 return 0;
1974
1975         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1976             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1977                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1978                                   GFP_NOFS);
1979                 return 0;
1980         }
1981
1982         if (state && state->start == start) {
1983                 private = state->private;
1984                 ret = 0;
1985         } else {
1986                 ret = get_state_private(io_tree, start, &private);
1987         }
1988         kaddr = kmap_atomic(page, KM_USER0);
1989         if (ret)
1990                 goto zeroit;
1991
1992         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1993         btrfs_csum_final(csum, (char *)&csum);
1994         if (csum != private)
1995                 goto zeroit;
1996
1997         kunmap_atomic(kaddr, KM_USER0);
1998 good:
1999         /* if the io failure tree for this inode is non-empty,
2000          * check to see if we've recovered from a failed IO
2001          */
2002         btrfs_clean_io_failures(inode, start);
2003         return 0;
2004
2005 zeroit:
2006         if (printk_ratelimit()) {
2007                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2008                        "private %llu\n", page->mapping->host->i_ino,
2009                        (unsigned long long)start, csum,
2010                        (unsigned long long)private);
2011         }
2012         memset(kaddr + offset, 1, end - start + 1);
2013         flush_dcache_page(page);
2014         kunmap_atomic(kaddr, KM_USER0);
2015         if (private == 0)
2016                 return 0;
2017         return -EIO;
2018 }
2019
2020 struct delayed_iput {
2021         struct list_head list;
2022         struct inode *inode;
2023 };
2024
2025 void btrfs_add_delayed_iput(struct inode *inode)
2026 {
2027         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2028         struct delayed_iput *delayed;
2029
2030         if (atomic_add_unless(&inode->i_count, -1, 1))
2031                 return;
2032
2033         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2034         delayed->inode = inode;
2035
2036         spin_lock(&fs_info->delayed_iput_lock);
2037         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2038         spin_unlock(&fs_info->delayed_iput_lock);
2039 }
2040
2041 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2042 {
2043         LIST_HEAD(list);
2044         struct btrfs_fs_info *fs_info = root->fs_info;
2045         struct delayed_iput *delayed;
2046         int empty;
2047
2048         spin_lock(&fs_info->delayed_iput_lock);
2049         empty = list_empty(&fs_info->delayed_iputs);
2050         spin_unlock(&fs_info->delayed_iput_lock);
2051         if (empty)
2052                 return;
2053
2054         down_read(&root->fs_info->cleanup_work_sem);
2055         spin_lock(&fs_info->delayed_iput_lock);
2056         list_splice_init(&fs_info->delayed_iputs, &list);
2057         spin_unlock(&fs_info->delayed_iput_lock);
2058
2059         while (!list_empty(&list)) {
2060                 delayed = list_entry(list.next, struct delayed_iput, list);
2061                 list_del(&delayed->list);
2062                 iput(delayed->inode);
2063                 kfree(delayed);
2064         }
2065         up_read(&root->fs_info->cleanup_work_sem);
2066 }
2067
2068 /*
2069  * calculate extra metadata reservation when snapshotting a subvolume
2070  * contains orphan files.
2071  */
2072 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2073                                 struct btrfs_pending_snapshot *pending,
2074                                 u64 *bytes_to_reserve)
2075 {
2076         struct btrfs_root *root;
2077         struct btrfs_block_rsv *block_rsv;
2078         u64 num_bytes;
2079         int index;
2080
2081         root = pending->root;
2082         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2083                 return;
2084
2085         block_rsv = root->orphan_block_rsv;
2086
2087         /* orphan block reservation for the snapshot */
2088         num_bytes = block_rsv->size;
2089
2090         /*
2091          * after the snapshot is created, COWing tree blocks may use more
2092          * space than it frees. So we should make sure there is enough
2093          * reserved space.
2094          */
2095         index = trans->transid & 0x1;
2096         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2097                 num_bytes += block_rsv->size -
2098                              (block_rsv->reserved + block_rsv->freed[index]);
2099         }
2100
2101         *bytes_to_reserve += num_bytes;
2102 }
2103
2104 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2105                                 struct btrfs_pending_snapshot *pending)
2106 {
2107         struct btrfs_root *root = pending->root;
2108         struct btrfs_root *snap = pending->snap;
2109         struct btrfs_block_rsv *block_rsv;
2110         u64 num_bytes;
2111         int index;
2112         int ret;
2113
2114         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2115                 return;
2116
2117         /* refill source subvolume's orphan block reservation */
2118         block_rsv = root->orphan_block_rsv;
2119         index = trans->transid & 0x1;
2120         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2121                 num_bytes = block_rsv->size -
2122                             (block_rsv->reserved + block_rsv->freed[index]);
2123                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2124                                               root->orphan_block_rsv,
2125                                               num_bytes);
2126                 BUG_ON(ret);
2127         }
2128
2129         /* setup orphan block reservation for the snapshot */
2130         block_rsv = btrfs_alloc_block_rsv(snap);
2131         BUG_ON(!block_rsv);
2132
2133         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2134         snap->orphan_block_rsv = block_rsv;
2135
2136         num_bytes = root->orphan_block_rsv->size;
2137         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138                                       block_rsv, num_bytes);
2139         BUG_ON(ret);
2140
2141 #if 0
2142         /* insert orphan item for the snapshot */
2143         WARN_ON(!root->orphan_item_inserted);
2144         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2145                                        snap->root_key.objectid);
2146         BUG_ON(ret);
2147         snap->orphan_item_inserted = 1;
2148 #endif
2149 }
2150
2151 enum btrfs_orphan_cleanup_state {
2152         ORPHAN_CLEANUP_STARTED  = 1,
2153         ORPHAN_CLEANUP_DONE     = 2,
2154 };
2155
2156 /*
2157  * This is called in transaction commmit time. If there are no orphan
2158  * files in the subvolume, it removes orphan item and frees block_rsv
2159  * structure.
2160  */
2161 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2162                               struct btrfs_root *root)
2163 {
2164         int ret;
2165
2166         if (!list_empty(&root->orphan_list) ||
2167             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2168                 return;
2169
2170         if (root->orphan_item_inserted &&
2171             btrfs_root_refs(&root->root_item) > 0) {
2172                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2173                                             root->root_key.objectid);
2174                 BUG_ON(ret);
2175                 root->orphan_item_inserted = 0;
2176         }
2177
2178         if (root->orphan_block_rsv) {
2179                 WARN_ON(root->orphan_block_rsv->size > 0);
2180                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2181                 root->orphan_block_rsv = NULL;
2182         }
2183 }
2184
2185 /*
2186  * This creates an orphan entry for the given inode in case something goes
2187  * wrong in the middle of an unlink/truncate.
2188  *
2189  * NOTE: caller of this function should reserve 5 units of metadata for
2190  *       this function.
2191  */
2192 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         struct btrfs_block_rsv *block_rsv = NULL;
2196         int reserve = 0;
2197         int insert = 0;
2198         int ret;
2199
2200         if (!root->orphan_block_rsv) {
2201                 block_rsv = btrfs_alloc_block_rsv(root);
2202                 BUG_ON(!block_rsv);
2203         }
2204
2205         spin_lock(&root->orphan_lock);
2206         if (!root->orphan_block_rsv) {
2207                 root->orphan_block_rsv = block_rsv;
2208         } else if (block_rsv) {
2209                 btrfs_free_block_rsv(root, block_rsv);
2210                 block_rsv = NULL;
2211         }
2212
2213         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2214                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2215 #if 0
2216                 /*
2217                  * For proper ENOSPC handling, we should do orphan
2218                  * cleanup when mounting. But this introduces backward
2219                  * compatibility issue.
2220                  */
2221                 if (!xchg(&root->orphan_item_inserted, 1))
2222                         insert = 2;
2223                 else
2224                         insert = 1;
2225 #endif
2226                 insert = 1;
2227         }
2228
2229         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2230                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2231                 reserve = 1;
2232         }
2233         spin_unlock(&root->orphan_lock);
2234
2235         if (block_rsv)
2236                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2237
2238         /* grab metadata reservation from transaction handle */
2239         if (reserve) {
2240                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2241                 BUG_ON(ret);
2242         }
2243
2244         /* insert an orphan item to track this unlinked/truncated file */
2245         if (insert >= 1) {
2246                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2247                 BUG_ON(ret);
2248         }
2249
2250         /* insert an orphan item to track subvolume contains orphan files */
2251         if (insert >= 2) {
2252                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2253                                                root->root_key.objectid);
2254                 BUG_ON(ret);
2255         }
2256         return 0;
2257 }
2258
2259 /*
2260  * We have done the truncate/delete so we can go ahead and remove the orphan
2261  * item for this particular inode.
2262  */
2263 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2264 {
2265         struct btrfs_root *root = BTRFS_I(inode)->root;
2266         int delete_item = 0;
2267         int release_rsv = 0;
2268         int ret = 0;
2269
2270         spin_lock(&root->orphan_lock);
2271         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2272                 list_del_init(&BTRFS_I(inode)->i_orphan);
2273                 delete_item = 1;
2274         }
2275
2276         if (BTRFS_I(inode)->orphan_meta_reserved) {
2277                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2278                 release_rsv = 1;
2279         }
2280         spin_unlock(&root->orphan_lock);
2281
2282         if (trans && delete_item) {
2283                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2284                 BUG_ON(ret);
2285         }
2286
2287         if (release_rsv)
2288                 btrfs_orphan_release_metadata(inode);
2289
2290         return 0;
2291 }
2292
2293 /*
2294  * this cleans up any orphans that may be left on the list from the last use
2295  * of this root.
2296  */
2297 int btrfs_orphan_cleanup(struct btrfs_root *root)
2298 {
2299         struct btrfs_path *path;
2300         struct extent_buffer *leaf;
2301         struct btrfs_key key, found_key;
2302         struct btrfs_trans_handle *trans;
2303         struct inode *inode;
2304         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2305
2306         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2307                 return 0;
2308
2309         path = btrfs_alloc_path();
2310         if (!path) {
2311                 ret = -ENOMEM;
2312                 goto out;
2313         }
2314         path->reada = -1;
2315
2316         key.objectid = BTRFS_ORPHAN_OBJECTID;
2317         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2318         key.offset = (u64)-1;
2319
2320         while (1) {
2321                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2322                 if (ret < 0)
2323                         goto out;
2324
2325                 /*
2326                  * if ret == 0 means we found what we were searching for, which
2327                  * is weird, but possible, so only screw with path if we didn't
2328                  * find the key and see if we have stuff that matches
2329                  */
2330                 if (ret > 0) {
2331                         ret = 0;
2332                         if (path->slots[0] == 0)
2333                                 break;
2334                         path->slots[0]--;
2335                 }
2336
2337                 /* pull out the item */
2338                 leaf = path->nodes[0];
2339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2340
2341                 /* make sure the item matches what we want */
2342                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2343                         break;
2344                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2345                         break;
2346
2347                 /* release the path since we're done with it */
2348                 btrfs_release_path(root, path);
2349
2350                 /*
2351                  * this is where we are basically btrfs_lookup, without the
2352                  * crossing root thing.  we store the inode number in the
2353                  * offset of the orphan item.
2354                  */
2355                 found_key.objectid = found_key.offset;
2356                 found_key.type = BTRFS_INODE_ITEM_KEY;
2357                 found_key.offset = 0;
2358                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2359                 if (IS_ERR(inode)) {
2360                         ret = PTR_ERR(inode);
2361                         goto out;
2362                 }
2363
2364                 /*
2365                  * add this inode to the orphan list so btrfs_orphan_del does
2366                  * the proper thing when we hit it
2367                  */
2368                 spin_lock(&root->orphan_lock);
2369                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2370                 spin_unlock(&root->orphan_lock);
2371
2372                 /*
2373                  * if this is a bad inode, means we actually succeeded in
2374                  * removing the inode, but not the orphan record, which means
2375                  * we need to manually delete the orphan since iput will just
2376                  * do a destroy_inode
2377                  */
2378                 if (is_bad_inode(inode)) {
2379                         trans = btrfs_start_transaction(root, 0);
2380                         if (IS_ERR(trans)) {
2381                                 ret = PTR_ERR(trans);
2382                                 goto out;
2383                         }
2384                         btrfs_orphan_del(trans, inode);
2385                         btrfs_end_transaction(trans, root);
2386                         iput(inode);
2387                         continue;
2388                 }
2389
2390                 /* if we have links, this was a truncate, lets do that */
2391                 if (inode->i_nlink) {
2392                         if (!S_ISREG(inode->i_mode)) {
2393                                 WARN_ON(1);
2394                                 iput(inode);
2395                                 continue;
2396                         }
2397                         nr_truncate++;
2398                         ret = btrfs_truncate(inode);
2399                 } else {
2400                         nr_unlink++;
2401                 }
2402
2403                 /* this will do delete_inode and everything for us */
2404                 iput(inode);
2405                 if (ret)
2406                         goto out;
2407         }
2408         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2409
2410         if (root->orphan_block_rsv)
2411                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2412                                         (u64)-1);
2413
2414         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2415                 trans = btrfs_join_transaction(root, 1);
2416                 if (!IS_ERR(trans))
2417                         btrfs_end_transaction(trans, root);
2418         }
2419
2420         if (nr_unlink)
2421                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2422         if (nr_truncate)
2423                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2424
2425 out:
2426         if (ret)
2427                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2428         btrfs_free_path(path);
2429         return ret;
2430 }
2431
2432 /*
2433  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2434  * don't find any xattrs, we know there can't be any acls.
2435  *
2436  * slot is the slot the inode is in, objectid is the objectid of the inode
2437  */
2438 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2439                                           int slot, u64 objectid)
2440 {
2441         u32 nritems = btrfs_header_nritems(leaf);
2442         struct btrfs_key found_key;
2443         int scanned = 0;
2444
2445         slot++;
2446         while (slot < nritems) {
2447                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2448
2449                 /* we found a different objectid, there must not be acls */
2450                 if (found_key.objectid != objectid)
2451                         return 0;
2452
2453                 /* we found an xattr, assume we've got an acl */
2454                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2455                         return 1;
2456
2457                 /*
2458                  * we found a key greater than an xattr key, there can't
2459                  * be any acls later on
2460                  */
2461                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2462                         return 0;
2463
2464                 slot++;
2465                 scanned++;
2466
2467                 /*
2468                  * it goes inode, inode backrefs, xattrs, extents,
2469                  * so if there are a ton of hard links to an inode there can
2470                  * be a lot of backrefs.  Don't waste time searching too hard,
2471                  * this is just an optimization
2472                  */
2473                 if (scanned >= 8)
2474                         break;
2475         }
2476         /* we hit the end of the leaf before we found an xattr or
2477          * something larger than an xattr.  We have to assume the inode
2478          * has acls
2479          */
2480         return 1;
2481 }
2482
2483 /*
2484  * read an inode from the btree into the in-memory inode
2485  */
2486 static void btrfs_read_locked_inode(struct inode *inode)
2487 {
2488         struct btrfs_path *path;
2489         struct extent_buffer *leaf;
2490         struct btrfs_inode_item *inode_item;
2491         struct btrfs_timespec *tspec;
2492         struct btrfs_root *root = BTRFS_I(inode)->root;
2493         struct btrfs_key location;
2494         int maybe_acls;
2495         u64 alloc_group_block;
2496         u32 rdev;
2497         int ret;
2498
2499         path = btrfs_alloc_path();
2500         BUG_ON(!path);
2501         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2502
2503         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2504         if (ret)
2505                 goto make_bad;
2506
2507         leaf = path->nodes[0];
2508         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2509                                     struct btrfs_inode_item);
2510
2511         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2512         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2513         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2514         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2515         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2516
2517         tspec = btrfs_inode_atime(inode_item);
2518         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520
2521         tspec = btrfs_inode_mtime(inode_item);
2522         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524
2525         tspec = btrfs_inode_ctime(inode_item);
2526         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2527         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2528
2529         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2530         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2531         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2532         inode->i_generation = BTRFS_I(inode)->generation;
2533         inode->i_rdev = 0;
2534         rdev = btrfs_inode_rdev(leaf, inode_item);
2535
2536         BTRFS_I(inode)->index_cnt = (u64)-1;
2537         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2538
2539         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2540
2541         /*
2542          * try to precache a NULL acl entry for files that don't have
2543          * any xattrs or acls
2544          */
2545         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2546         if (!maybe_acls)
2547                 cache_no_acl(inode);
2548
2549         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2550                                                 alloc_group_block, 0);
2551         btrfs_free_path(path);
2552         inode_item = NULL;
2553
2554         switch (inode->i_mode & S_IFMT) {
2555         case S_IFREG:
2556                 inode->i_mapping->a_ops = &btrfs_aops;
2557                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2558                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2559                 inode->i_fop = &btrfs_file_operations;
2560                 inode->i_op = &btrfs_file_inode_operations;
2561                 break;
2562         case S_IFDIR:
2563                 inode->i_fop = &btrfs_dir_file_operations;
2564                 if (root == root->fs_info->tree_root)
2565                         inode->i_op = &btrfs_dir_ro_inode_operations;
2566                 else
2567                         inode->i_op = &btrfs_dir_inode_operations;
2568                 break;
2569         case S_IFLNK:
2570                 inode->i_op = &btrfs_symlink_inode_operations;
2571                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2572                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2573                 break;
2574         default:
2575                 inode->i_op = &btrfs_special_inode_operations;
2576                 init_special_inode(inode, inode->i_mode, rdev);
2577                 break;
2578         }
2579
2580         btrfs_update_iflags(inode);
2581         return;
2582
2583 make_bad:
2584         btrfs_free_path(path);
2585         make_bad_inode(inode);
2586 }
2587
2588 /*
2589  * given a leaf and an inode, copy the inode fields into the leaf
2590  */
2591 static void fill_inode_item(struct btrfs_trans_handle *trans,
2592                             struct extent_buffer *leaf,
2593                             struct btrfs_inode_item *item,
2594                             struct inode *inode)
2595 {
2596         if (!leaf->map_token)
2597                 map_private_extent_buffer(leaf, (unsigned long)item,
2598                                           sizeof(struct btrfs_inode_item),
2599                                           &leaf->map_token, &leaf->kaddr,
2600                                           &leaf->map_start, &leaf->map_len,
2601                                           KM_USER1);
2602
2603         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2604         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2605         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2606         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2607         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2608
2609         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2610                                inode->i_atime.tv_sec);
2611         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2612                                 inode->i_atime.tv_nsec);
2613
2614         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2615                                inode->i_mtime.tv_sec);
2616         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2617                                 inode->i_mtime.tv_nsec);
2618
2619         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2620                                inode->i_ctime.tv_sec);
2621         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2622                                 inode->i_ctime.tv_nsec);
2623
2624         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2625         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2626         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2627         btrfs_set_inode_transid(leaf, item, trans->transid);
2628         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2629         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2630         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2631
2632         if (leaf->map_token) {
2633                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2634                 leaf->map_token = NULL;
2635         }
2636 }
2637
2638 /*
2639  * copy everything in the in-memory inode into the btree.
2640  */
2641 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2642                                 struct btrfs_root *root, struct inode *inode)
2643 {
2644         struct btrfs_inode_item *inode_item;
2645         struct btrfs_path *path;
2646         struct extent_buffer *leaf;
2647         int ret;
2648
2649         path = btrfs_alloc_path();
2650         BUG_ON(!path);
2651         path->leave_spinning = 1;
2652         ret = btrfs_lookup_inode(trans, root, path,
2653                                  &BTRFS_I(inode)->location, 1);
2654         if (ret) {
2655                 if (ret > 0)
2656                         ret = -ENOENT;
2657                 goto failed;
2658         }
2659
2660         btrfs_unlock_up_safe(path, 1);
2661         leaf = path->nodes[0];
2662         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2663                                   struct btrfs_inode_item);
2664
2665         fill_inode_item(trans, leaf, inode_item, inode);
2666         btrfs_mark_buffer_dirty(leaf);
2667         btrfs_set_inode_last_trans(trans, inode);
2668         ret = 0;
2669 failed:
2670         btrfs_free_path(path);
2671         return ret;
2672 }
2673
2674
2675 /*
2676  * unlink helper that gets used here in inode.c and in the tree logging
2677  * recovery code.  It remove a link in a directory with a given name, and
2678  * also drops the back refs in the inode to the directory
2679  */
2680 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2681                                 struct btrfs_root *root,
2682                                 struct inode *dir, struct inode *inode,
2683                                 const char *name, int name_len)
2684 {
2685         struct btrfs_path *path;
2686         int ret = 0;
2687         struct extent_buffer *leaf;
2688         struct btrfs_dir_item *di;
2689         struct btrfs_key key;
2690         u64 index;
2691
2692         path = btrfs_alloc_path();
2693         if (!path) {
2694                 ret = -ENOMEM;
2695                 goto out;
2696         }
2697
2698         path->leave_spinning = 1;
2699         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2700                                     name, name_len, -1);
2701         if (IS_ERR(di)) {
2702                 ret = PTR_ERR(di);
2703                 goto err;
2704         }
2705         if (!di) {
2706                 ret = -ENOENT;
2707                 goto err;
2708         }
2709         leaf = path->nodes[0];
2710         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2711         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2712         if (ret)
2713                 goto err;
2714         btrfs_release_path(root, path);
2715
2716         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2717                                   inode->i_ino,
2718                                   dir->i_ino, &index);
2719         if (ret) {
2720                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2721                        "inode %lu parent %lu\n", name_len, name,
2722                        inode->i_ino, dir->i_ino);
2723                 goto err;
2724         }
2725
2726         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2727                                          index, name, name_len, -1);
2728         if (IS_ERR(di)) {
2729                 ret = PTR_ERR(di);
2730                 goto err;
2731         }
2732         if (!di) {
2733                 ret = -ENOENT;
2734                 goto err;
2735         }
2736         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2737         btrfs_release_path(root, path);
2738
2739         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2740                                          inode, dir->i_ino);
2741         BUG_ON(ret != 0 && ret != -ENOENT);
2742
2743         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2744                                            dir, index);
2745         if (ret == -ENOENT)
2746                 ret = 0;
2747 err:
2748         btrfs_free_path(path);
2749         if (ret)
2750                 goto out;
2751
2752         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2753         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2754         btrfs_update_inode(trans, root, dir);
2755 out:
2756         return ret;
2757 }
2758
2759 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2760                        struct btrfs_root *root,
2761                        struct inode *dir, struct inode *inode,
2762                        const char *name, int name_len)
2763 {
2764         int ret;
2765         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2766         if (!ret) {
2767                 btrfs_drop_nlink(inode);
2768                 ret = btrfs_update_inode(trans, root, inode);
2769         }
2770         return ret;
2771 }
2772
2773
2774 /* helper to check if there is any shared block in the path */
2775 static int check_path_shared(struct btrfs_root *root,
2776                              struct btrfs_path *path)
2777 {
2778         struct extent_buffer *eb;
2779         int level;
2780         u64 refs = 1;
2781
2782         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2783                 int ret;
2784
2785                 if (!path->nodes[level])
2786                         break;
2787                 eb = path->nodes[level];
2788                 if (!btrfs_block_can_be_shared(root, eb))
2789                         continue;
2790                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2791                                                &refs, NULL);
2792                 if (refs > 1)
2793                         return 1;
2794         }
2795         return 0;
2796 }
2797
2798 /*
2799  * helper to start transaction for unlink and rmdir.
2800  *
2801  * unlink and rmdir are special in btrfs, they do not always free space.
2802  * so in enospc case, we should make sure they will free space before
2803  * allowing them to use the global metadata reservation.
2804  */
2805 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2806                                                        struct dentry *dentry)
2807 {
2808         struct btrfs_trans_handle *trans;
2809         struct btrfs_root *root = BTRFS_I(dir)->root;
2810         struct btrfs_path *path;
2811         struct btrfs_inode_ref *ref;
2812         struct btrfs_dir_item *di;
2813         struct inode *inode = dentry->d_inode;
2814         u64 index;
2815         int check_link = 1;
2816         int err = -ENOSPC;
2817         int ret;
2818
2819         trans = btrfs_start_transaction(root, 10);
2820         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2821                 return trans;
2822
2823         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2824                 return ERR_PTR(-ENOSPC);
2825
2826         /* check if there is someone else holds reference */
2827         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2828                 return ERR_PTR(-ENOSPC);
2829
2830         if (atomic_read(&inode->i_count) > 2)
2831                 return ERR_PTR(-ENOSPC);
2832
2833         if (xchg(&root->fs_info->enospc_unlink, 1))
2834                 return ERR_PTR(-ENOSPC);
2835
2836         path = btrfs_alloc_path();
2837         if (!path) {
2838                 root->fs_info->enospc_unlink = 0;
2839                 return ERR_PTR(-ENOMEM);
2840         }
2841
2842         trans = btrfs_start_transaction(root, 0);
2843         if (IS_ERR(trans)) {
2844                 btrfs_free_path(path);
2845                 root->fs_info->enospc_unlink = 0;
2846                 return trans;
2847         }
2848
2849         path->skip_locking = 1;
2850         path->search_commit_root = 1;
2851
2852         ret = btrfs_lookup_inode(trans, root, path,
2853                                 &BTRFS_I(dir)->location, 0);
2854         if (ret < 0) {
2855                 err = ret;
2856                 goto out;
2857         }
2858         if (ret == 0) {
2859                 if (check_path_shared(root, path))
2860                         goto out;
2861         } else {
2862                 check_link = 0;
2863         }
2864         btrfs_release_path(root, path);
2865
2866         ret = btrfs_lookup_inode(trans, root, path,
2867                                 &BTRFS_I(inode)->location, 0);
2868         if (ret < 0) {
2869                 err = ret;
2870                 goto out;
2871         }
2872         if (ret == 0) {
2873                 if (check_path_shared(root, path))
2874                         goto out;
2875         } else {
2876                 check_link = 0;
2877         }
2878         btrfs_release_path(root, path);
2879
2880         if (ret == 0 && S_ISREG(inode->i_mode)) {
2881                 ret = btrfs_lookup_file_extent(trans, root, path,
2882                                                inode->i_ino, (u64)-1, 0);
2883                 if (ret < 0) {
2884                         err = ret;
2885                         goto out;
2886                 }
2887                 BUG_ON(ret == 0);
2888                 if (check_path_shared(root, path))
2889                         goto out;
2890                 btrfs_release_path(root, path);
2891         }
2892
2893         if (!check_link) {
2894                 err = 0;
2895                 goto out;
2896         }
2897
2898         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2899                                 dentry->d_name.name, dentry->d_name.len, 0);
2900         if (IS_ERR(di)) {
2901                 err = PTR_ERR(di);
2902                 goto out;
2903         }
2904         if (di) {
2905                 if (check_path_shared(root, path))
2906                         goto out;
2907         } else {
2908                 err = 0;
2909                 goto out;
2910         }
2911         btrfs_release_path(root, path);
2912
2913         ref = btrfs_lookup_inode_ref(trans, root, path,
2914                                 dentry->d_name.name, dentry->d_name.len,
2915                                 inode->i_ino, dir->i_ino, 0);
2916         if (IS_ERR(ref)) {
2917                 err = PTR_ERR(ref);
2918                 goto out;
2919         }
2920         BUG_ON(!ref);
2921         if (check_path_shared(root, path))
2922                 goto out;
2923         index = btrfs_inode_ref_index(path->nodes[0], ref);
2924         btrfs_release_path(root, path);
2925
2926         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2927                                 dentry->d_name.name, dentry->d_name.len, 0);
2928         if (IS_ERR(di)) {
2929                 err = PTR_ERR(di);
2930                 goto out;
2931         }
2932         BUG_ON(ret == -ENOENT);
2933         if (check_path_shared(root, path))
2934                 goto out;
2935
2936         err = 0;
2937 out:
2938         btrfs_free_path(path);
2939         if (err) {
2940                 btrfs_end_transaction(trans, root);
2941                 root->fs_info->enospc_unlink = 0;
2942                 return ERR_PTR(err);
2943         }
2944
2945         trans->block_rsv = &root->fs_info->global_block_rsv;
2946         return trans;
2947 }
2948
2949 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2950                                struct btrfs_root *root)
2951 {
2952         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2953                 BUG_ON(!root->fs_info->enospc_unlink);
2954                 root->fs_info->enospc_unlink = 0;
2955         }
2956         btrfs_end_transaction_throttle(trans, root);
2957 }
2958
2959 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2960 {
2961         struct btrfs_root *root = BTRFS_I(dir)->root;
2962         struct btrfs_trans_handle *trans;
2963         struct inode *inode = dentry->d_inode;
2964         int ret;
2965         unsigned long nr = 0;
2966
2967         trans = __unlink_start_trans(dir, dentry);
2968         if (IS_ERR(trans))
2969                 return PTR_ERR(trans);
2970
2971         btrfs_set_trans_block_group(trans, dir);
2972
2973         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2974
2975         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2976                                  dentry->d_name.name, dentry->d_name.len);
2977         BUG_ON(ret);
2978
2979         if (inode->i_nlink == 0) {
2980                 ret = btrfs_orphan_add(trans, inode);
2981                 BUG_ON(ret);
2982         }
2983
2984         nr = trans->blocks_used;
2985         __unlink_end_trans(trans, root);
2986         btrfs_btree_balance_dirty(root, nr);
2987         return ret;
2988 }
2989
2990 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2991                         struct btrfs_root *root,
2992                         struct inode *dir, u64 objectid,
2993                         const char *name, int name_len)
2994 {
2995         struct btrfs_path *path;
2996         struct extent_buffer *leaf;
2997         struct btrfs_dir_item *di;
2998         struct btrfs_key key;
2999         u64 index;
3000         int ret;
3001
3002         path = btrfs_alloc_path();
3003         if (!path)
3004                 return -ENOMEM;
3005
3006         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3007                                    name, name_len, -1);
3008         BUG_ON(!di || IS_ERR(di));
3009
3010         leaf = path->nodes[0];
3011         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3012         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3013         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3014         BUG_ON(ret);
3015         btrfs_release_path(root, path);
3016
3017         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3018                                  objectid, root->root_key.objectid,
3019                                  dir->i_ino, &index, name, name_len);
3020         if (ret < 0) {
3021                 BUG_ON(ret != -ENOENT);
3022                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3023                                                  name, name_len);
3024                 BUG_ON(!di || IS_ERR(di));
3025
3026                 leaf = path->nodes[0];
3027                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3028                 btrfs_release_path(root, path);
3029                 index = key.offset;
3030         }
3031
3032         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3033                                          index, name, name_len, -1);
3034         BUG_ON(!di || IS_ERR(di));
3035
3036         leaf = path->nodes[0];
3037         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3038         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3039         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3040         BUG_ON(ret);
3041         btrfs_release_path(root, path);
3042
3043         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3044         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3045         ret = btrfs_update_inode(trans, root, dir);
3046         BUG_ON(ret);
3047
3048         btrfs_free_path(path);
3049         return 0;
3050 }
3051
3052 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3053 {
3054         struct inode *inode = dentry->d_inode;
3055         int err = 0;
3056         struct btrfs_root *root = BTRFS_I(dir)->root;
3057         struct btrfs_trans_handle *trans;
3058         unsigned long nr = 0;
3059
3060         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3061             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3062                 return -ENOTEMPTY;
3063
3064         trans = __unlink_start_trans(dir, dentry);
3065         if (IS_ERR(trans))
3066                 return PTR_ERR(trans);
3067
3068         btrfs_set_trans_block_group(trans, dir);
3069
3070         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3071                 err = btrfs_unlink_subvol(trans, root, dir,
3072                                           BTRFS_I(inode)->location.objectid,
3073                                           dentry->d_name.name,
3074                                           dentry->d_name.len);
3075                 goto out;
3076         }
3077
3078         err = btrfs_orphan_add(trans, inode);
3079         if (err)
3080                 goto out;
3081
3082         /* now the directory is empty */
3083         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3084                                  dentry->d_name.name, dentry->d_name.len);
3085         if (!err)
3086                 btrfs_i_size_write(inode, 0);
3087 out:
3088         nr = trans->blocks_used;
3089         __unlink_end_trans(trans, root);
3090         btrfs_btree_balance_dirty(root, nr);
3091
3092         return err;
3093 }
3094
3095 #if 0
3096 /*
3097  * when truncating bytes in a file, it is possible to avoid reading
3098  * the leaves that contain only checksum items.  This can be the
3099  * majority of the IO required to delete a large file, but it must
3100  * be done carefully.
3101  *
3102  * The keys in the level just above the leaves are checked to make sure
3103  * the lowest key in a given leaf is a csum key, and starts at an offset
3104  * after the new  size.
3105  *
3106  * Then the key for the next leaf is checked to make sure it also has
3107  * a checksum item for the same file.  If it does, we know our target leaf
3108  * contains only checksum items, and it can be safely freed without reading
3109  * it.
3110  *
3111  * This is just an optimization targeted at large files.  It may do
3112  * nothing.  It will return 0 unless things went badly.
3113  */
3114 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3115                                      struct btrfs_root *root,
3116                                      struct btrfs_path *path,
3117                                      struct inode *inode, u64 new_size)
3118 {
3119         struct btrfs_key key;
3120         int ret;
3121         int nritems;
3122         struct btrfs_key found_key;
3123         struct btrfs_key other_key;
3124         struct btrfs_leaf_ref *ref;
3125         u64 leaf_gen;
3126         u64 leaf_start;
3127
3128         path->lowest_level = 1;
3129         key.objectid = inode->i_ino;
3130         key.type = BTRFS_CSUM_ITEM_KEY;
3131         key.offset = new_size;
3132 again:
3133         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3134         if (ret < 0)
3135                 goto out;
3136
3137         if (path->nodes[1] == NULL) {
3138                 ret = 0;
3139                 goto out;
3140         }
3141         ret = 0;
3142         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3143         nritems = btrfs_header_nritems(path->nodes[1]);
3144
3145         if (!nritems)
3146                 goto out;
3147
3148         if (path->slots[1] >= nritems)
3149                 goto next_node;
3150
3151         /* did we find a key greater than anything we want to delete? */
3152         if (found_key.objectid > inode->i_ino ||
3153            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3154                 goto out;
3155
3156         /* we check the next key in the node to make sure the leave contains
3157          * only checksum items.  This comparison doesn't work if our
3158          * leaf is the last one in the node
3159          */
3160         if (path->slots[1] + 1 >= nritems) {
3161 next_node:
3162                 /* search forward from the last key in the node, this
3163                  * will bring us into the next node in the tree
3164                  */
3165                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3166
3167                 /* unlikely, but we inc below, so check to be safe */
3168                 if (found_key.offset == (u64)-1)
3169                         goto out;
3170
3171                 /* search_forward needs a path with locks held, do the
3172                  * search again for the original key.  It is possible
3173                  * this will race with a balance and return a path that
3174                  * we could modify, but this drop is just an optimization
3175                  * and is allowed to miss some leaves.
3176                  */
3177                 btrfs_release_path(root, path);
3178                 found_key.offset++;
3179
3180                 /* setup a max key for search_forward */
3181                 other_key.offset = (u64)-1;
3182                 other_key.type = key.type;
3183                 other_key.objectid = key.objectid;
3184
3185                 path->keep_locks = 1;
3186                 ret = btrfs_search_forward(root, &found_key, &other_key,
3187                                            path, 0, 0);
3188                 path->keep_locks = 0;
3189                 if (ret || found_key.objectid != key.objectid ||
3190                     found_key.type != key.type) {
3191                         ret = 0;
3192                         goto out;
3193                 }
3194
3195                 key.offset = found_key.offset;
3196                 btrfs_release_path(root, path);
3197                 cond_resched();
3198                 goto again;
3199         }
3200
3201         /* we know there's one more slot after us in the tree,
3202          * read that key so we can verify it is also a checksum item
3203          */
3204         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3205
3206         if (found_key.objectid < inode->i_ino)
3207                 goto next_key;
3208
3209         if (found_key.type != key.type || found_key.offset < new_size)
3210                 goto next_key;
3211
3212         /*
3213          * if the key for the next leaf isn't a csum key from this objectid,
3214          * we can't be sure there aren't good items inside this leaf.
3215          * Bail out
3216          */
3217         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3218                 goto out;
3219
3220         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3221         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3222         /*
3223          * it is safe to delete this leaf, it contains only
3224          * csum items from this inode at an offset >= new_size
3225          */
3226         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3227         BUG_ON(ret);
3228
3229         if (root->ref_cows && leaf_gen < trans->transid) {
3230                 ref = btrfs_alloc_leaf_ref(root, 0);
3231                 if (ref) {
3232                         ref->root_gen = root->root_key.offset;
3233                         ref->bytenr = leaf_start;
3234                         ref->owner = 0;
3235                         ref->generation = leaf_gen;
3236                         ref->nritems = 0;
3237
3238                         btrfs_sort_leaf_ref(ref);
3239
3240                         ret = btrfs_add_leaf_ref(root, ref, 0);
3241                         WARN_ON(ret);
3242                         btrfs_free_leaf_ref(root, ref);
3243                 } else {
3244                         WARN_ON(1);
3245                 }
3246         }
3247 next_key:
3248         btrfs_release_path(root, path);
3249
3250         if (other_key.objectid == inode->i_ino &&
3251             other_key.type == key.type && other_key.offset > key.offset) {
3252                 key.offset = other_key.offset;
3253                 cond_resched();
3254                 goto again;
3255         }
3256         ret = 0;
3257 out:
3258         /* fixup any changes we've made to the path */
3259         path->lowest_level = 0;
3260         path->keep_locks = 0;
3261         btrfs_release_path(root, path);
3262         return ret;
3263 }
3264
3265 #endif
3266
3267 /*
3268  * this can truncate away extent items, csum items and directory items.
3269  * It starts at a high offset and removes keys until it can't find
3270  * any higher than new_size
3271  *
3272  * csum items that cross the new i_size are truncated to the new size
3273  * as well.
3274  *
3275  * min_type is the minimum key type to truncate down to.  If set to 0, this
3276  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3277  */
3278 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3279                                struct btrfs_root *root,
3280                                struct inode *inode,
3281                                u64 new_size, u32 min_type)
3282 {
3283         struct btrfs_path *path;
3284         struct extent_buffer *leaf;
3285         struct btrfs_file_extent_item *fi;
3286         struct btrfs_key key;
3287         struct btrfs_key found_key;
3288         u64 extent_start = 0;
3289         u64 extent_num_bytes = 0;
3290         u64 extent_offset = 0;
3291         u64 item_end = 0;
3292         u64 mask = root->sectorsize - 1;
3293         u32 found_type = (u8)-1;
3294         int found_extent;
3295         int del_item;
3296         int pending_del_nr = 0;
3297         int pending_del_slot = 0;
3298         int extent_type = -1;
3299         int encoding;
3300         int ret;
3301         int err = 0;
3302
3303         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3304
3305         if (root->ref_cows || root == root->fs_info->tree_root)
3306                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3307
3308         path = btrfs_alloc_path();
3309         BUG_ON(!path);
3310         path->reada = -1;
3311
3312         key.objectid = inode->i_ino;
3313         key.offset = (u64)-1;
3314         key.type = (u8)-1;
3315
3316 search_again:
3317         path->leave_spinning = 1;
3318         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3319         if (ret < 0) {
3320                 err = ret;
3321                 goto out;
3322         }
3323
3324         if (ret > 0) {
3325                 /* there are no items in the tree for us to truncate, we're
3326                  * done
3327                  */
3328                 if (path->slots[0] == 0)
3329                         goto out;
3330                 path->slots[0]--;
3331         }
3332
3333         while (1) {
3334                 fi = NULL;
3335                 leaf = path->nodes[0];
3336                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3337                 found_type = btrfs_key_type(&found_key);
3338                 encoding = 0;
3339
3340                 if (found_key.objectid != inode->i_ino)
3341                         break;
3342
3343                 if (found_type < min_type)
3344                         break;
3345
3346                 item_end = found_key.offset;
3347                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3348                         fi = btrfs_item_ptr(leaf, path->slots[0],
3349                                             struct btrfs_file_extent_item);
3350                         extent_type = btrfs_file_extent_type(leaf, fi);
3351                         encoding = btrfs_file_extent_compression(leaf, fi);
3352                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3353                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3354
3355                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3356                                 item_end +=
3357                                     btrfs_file_extent_num_bytes(leaf, fi);
3358                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3359                                 item_end += btrfs_file_extent_inline_len(leaf,
3360                                                                          fi);
3361                         }
3362                         item_end--;
3363                 }
3364                 if (found_type > min_type) {
3365                         del_item = 1;
3366                 } else {
3367                         if (item_end < new_size)
3368                                 break;
3369                         if (found_key.offset >= new_size)
3370                                 del_item = 1;
3371                         else
3372                                 del_item = 0;
3373                 }
3374                 found_extent = 0;
3375                 /* FIXME, shrink the extent if the ref count is only 1 */
3376                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3377                         goto delete;
3378
3379                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3380                         u64 num_dec;
3381                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3382                         if (!del_item && !encoding) {
3383                                 u64 orig_num_bytes =
3384                                         btrfs_file_extent_num_bytes(leaf, fi);
3385                                 extent_num_bytes = new_size -
3386                                         found_key.offset + root->sectorsize - 1;
3387                                 extent_num_bytes = extent_num_bytes &
3388                                         ~((u64)root->sectorsize - 1);
3389                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3390                                                          extent_num_bytes);
3391                                 num_dec = (orig_num_bytes -
3392                                            extent_num_bytes);
3393                                 if (root->ref_cows && extent_start != 0)
3394                                         inode_sub_bytes(inode, num_dec);
3395                                 btrfs_mark_buffer_dirty(leaf);
3396                         } else {
3397                                 extent_num_bytes =
3398                                         btrfs_file_extent_disk_num_bytes(leaf,
3399                                                                          fi);
3400                                 extent_offset = found_key.offset -
3401                                         btrfs_file_extent_offset(leaf, fi);
3402
3403                                 /* FIXME blocksize != 4096 */
3404                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3405                                 if (extent_start != 0) {
3406                                         found_extent = 1;
3407                                         if (root->ref_cows)
3408                                                 inode_sub_bytes(inode, num_dec);
3409                                 }
3410                         }
3411                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3412                         /*
3413                          * we can't truncate inline items that have had
3414                          * special encodings
3415                          */
3416                         if (!del_item &&
3417                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3418                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3419                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3420                                 u32 size = new_size - found_key.offset;
3421
3422                                 if (root->ref_cows) {
3423                                         inode_sub_bytes(inode, item_end + 1 -
3424                                                         new_size);
3425                                 }
3426                                 size =
3427                                     btrfs_file_extent_calc_inline_size(size);
3428                                 ret = btrfs_truncate_item(trans, root, path,
3429                                                           size, 1);
3430                                 BUG_ON(ret);
3431                         } else if (root->ref_cows) {
3432                                 inode_sub_bytes(inode, item_end + 1 -
3433                                                 found_key.offset);
3434                         }
3435                 }
3436 delete:
3437                 if (del_item) {
3438                         if (!pending_del_nr) {
3439                                 /* no pending yet, add ourselves */
3440                                 pending_del_slot = path->slots[0];
3441                                 pending_del_nr = 1;
3442                         } else if (pending_del_nr &&
3443                                    path->slots[0] + 1 == pending_del_slot) {
3444                                 /* hop on the pending chunk */
3445                                 pending_del_nr++;
3446                                 pending_del_slot = path->slots[0];
3447                         } else {
3448                                 BUG();
3449                         }
3450                 } else {
3451                         break;
3452                 }
3453                 if (found_extent && (root->ref_cows ||
3454                                      root == root->fs_info->tree_root)) {
3455                         btrfs_set_path_blocking(path);
3456                         ret = btrfs_free_extent(trans, root, extent_start,
3457                                                 extent_num_bytes, 0,
3458                                                 btrfs_header_owner(leaf),
3459                                                 inode->i_ino, extent_offset);
3460                         BUG_ON(ret);
3461                 }
3462
3463                 if (found_type == BTRFS_INODE_ITEM_KEY)
3464                         break;
3465
3466                 if (path->slots[0] == 0 ||
3467                     path->slots[0] != pending_del_slot) {
3468                         if (root->ref_cows) {
3469                                 err = -EAGAIN;
3470                                 goto out;
3471                         }
3472                         if (pending_del_nr) {
3473                                 ret = btrfs_del_items(trans, root, path,
3474                                                 pending_del_slot,
3475                                                 pending_del_nr);
3476                                 BUG_ON(ret);
3477                                 pending_del_nr = 0;
3478                         }
3479                         btrfs_release_path(root, path);
3480                         goto search_again;
3481                 } else {
3482                         path->slots[0]--;
3483                 }
3484         }
3485 out:
3486         if (pending_del_nr) {
3487                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3488                                       pending_del_nr);
3489                 BUG_ON(ret);
3490         }
3491         btrfs_free_path(path);
3492         return err;
3493 }
3494
3495 /*
3496  * taken from block_truncate_page, but does cow as it zeros out
3497  * any bytes left in the last page in the file.
3498  */
3499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3500 {
3501         struct inode *inode = mapping->host;
3502         struct btrfs_root *root = BTRFS_I(inode)->root;
3503         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3504         struct btrfs_ordered_extent *ordered;
3505         struct extent_state *cached_state = NULL;
3506         char *kaddr;
3507         u32 blocksize = root->sectorsize;
3508         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3509         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3510         struct page *page;
3511         int ret = 0;
3512         u64 page_start;
3513         u64 page_end;
3514
3515         if ((offset & (blocksize - 1)) == 0)
3516                 goto out;
3517         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3518         if (ret)
3519                 goto out;
3520
3521         ret = -ENOMEM;
3522 again:
3523         page = grab_cache_page(mapping, index);
3524         if (!page) {
3525                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3526                 goto out;
3527         }
3528
3529         page_start = page_offset(page);
3530         page_end = page_start + PAGE_CACHE_SIZE - 1;
3531
3532         if (!PageUptodate(page)) {
3533                 ret = btrfs_readpage(NULL, page);
3534                 lock_page(page);
3535                 if (page->mapping != mapping) {
3536                         unlock_page(page);
3537                         page_cache_release(page);
3538                         goto again;
3539                 }
3540                 if (!PageUptodate(page)) {
3541                         ret = -EIO;
3542                         goto out_unlock;
3543                 }
3544         }
3545         wait_on_page_writeback(page);
3546
3547         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3548                          GFP_NOFS);
3549         set_page_extent_mapped(page);
3550
3551         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3552         if (ordered) {
3553                 unlock_extent_cached(io_tree, page_start, page_end,
3554                                      &cached_state, GFP_NOFS);
3555                 unlock_page(page);
3556                 page_cache_release(page);
3557                 btrfs_start_ordered_extent(inode, ordered, 1);
3558                 btrfs_put_ordered_extent(ordered);
3559                 goto again;
3560         }
3561
3562         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3563                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3564                           0, 0, &cached_state, GFP_NOFS);
3565
3566         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3567                                         &cached_state);
3568         if (ret) {
3569                 unlock_extent_cached(io_tree, page_start, page_end,
3570                                      &cached_state, GFP_NOFS);
3571                 goto out_unlock;
3572         }
3573
3574         ret = 0;
3575         if (offset != PAGE_CACHE_SIZE) {
3576                 kaddr = kmap(page);
3577                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3578                 flush_dcache_page(page);
3579                 kunmap(page);
3580         }
3581         ClearPageChecked(page);
3582         set_page_dirty(page);
3583         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3584                              GFP_NOFS);
3585
3586 out_unlock:
3587         if (ret)
3588                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3589         unlock_page(page);
3590         page_cache_release(page);
3591 out:
3592         return ret;
3593 }
3594
3595 /*
3596  * This function puts in dummy file extents for the area we're creating a hole
3597  * for.  So if we are truncating this file to a larger size we need to insert
3598  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3599  * the range between oldsize and size
3600  */
3601 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3602 {
3603         struct btrfs_trans_handle *trans;
3604         struct btrfs_root *root = BTRFS_I(inode)->root;
3605         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3606         struct extent_map *em = NULL;
3607         struct extent_state *cached_state = NULL;
3608         u64 mask = root->sectorsize - 1;
3609         u64 hole_start = (oldsize + mask) & ~mask;
3610         u64 block_end = (size + mask) & ~mask;
3611         u64 last_byte;
3612         u64 cur_offset;
3613         u64 hole_size;
3614         int err = 0;
3615
3616         if (size <= hole_start)
3617                 return 0;
3618
3619         while (1) {
3620                 struct btrfs_ordered_extent *ordered;
3621                 btrfs_wait_ordered_range(inode, hole_start,
3622                                          block_end - hole_start);
3623                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3624                                  &cached_state, GFP_NOFS);
3625                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3626                 if (!ordered)
3627                         break;
3628                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3629                                      &cached_state, GFP_NOFS);
3630                 btrfs_put_ordered_extent(ordered);
3631         }
3632
3633         cur_offset = hole_start;
3634         while (1) {
3635                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3636                                 block_end - cur_offset, 0);
3637                 BUG_ON(IS_ERR(em) || !em);
3638                 last_byte = min(extent_map_end(em), block_end);
3639                 last_byte = (last_byte + mask) & ~mask;
3640                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3641                         u64 hint_byte = 0;
3642                         hole_size = last_byte - cur_offset;
3643
3644                         trans = btrfs_start_transaction(root, 2);
3645                         if (IS_ERR(trans)) {
3646                                 err = PTR_ERR(trans);
3647                                 break;
3648                         }
3649                         btrfs_set_trans_block_group(trans, inode);
3650
3651                         err = btrfs_drop_extents(trans, inode, cur_offset,
3652                                                  cur_offset + hole_size,
3653                                                  &hint_byte, 1);
3654                         if (err)
3655                                 break;
3656
3657                         err = btrfs_insert_file_extent(trans, root,
3658                                         inode->i_ino, cur_offset, 0,
3659                                         0, hole_size, 0, hole_size,
3660                                         0, 0, 0);
3661                         if (err)
3662                                 break;
3663
3664                         btrfs_drop_extent_cache(inode, hole_start,
3665                                         last_byte - 1, 0);
3666
3667                         btrfs_end_transaction(trans, root);
3668                 }
3669                 free_extent_map(em);
3670                 em = NULL;
3671                 cur_offset = last_byte;
3672                 if (cur_offset >= block_end)
3673                         break;
3674         }
3675
3676         free_extent_map(em);
3677         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3678                              GFP_NOFS);
3679         return err;
3680 }
3681
3682 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3683 {
3684         loff_t oldsize = i_size_read(inode);
3685         int ret;
3686
3687         if (newsize == oldsize)
3688                 return 0;
3689
3690         if (newsize > oldsize) {
3691                 i_size_write(inode, newsize);
3692                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3693                 truncate_pagecache(inode, oldsize, newsize);
3694                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3695                 if (ret) {
3696                         btrfs_setsize(inode, oldsize);
3697                         return ret;
3698                 }
3699
3700                 mark_inode_dirty(inode);
3701         } else {
3702
3703                 /*
3704                  * We're truncating a file that used to have good data down to
3705                  * zero. Make sure it gets into the ordered flush list so that
3706                  * any new writes get down to disk quickly.
3707                  */
3708                 if (newsize == 0)
3709                         BTRFS_I(inode)->ordered_data_close = 1;
3710
3711                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3712                 truncate_setsize(inode, newsize);
3713                 ret = btrfs_truncate(inode);
3714         }
3715
3716         return ret;
3717 }
3718
3719 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3720 {
3721         struct inode *inode = dentry->d_inode;
3722         struct btrfs_root *root = BTRFS_I(inode)->root;
3723         int err;
3724
3725         if (btrfs_root_readonly(root))
3726                 return -EROFS;
3727
3728         err = inode_change_ok(inode, attr);
3729         if (err)
3730                 return err;
3731
3732         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3733                 err = btrfs_setsize(inode, attr->ia_size);
3734                 if (err)
3735                         return err;
3736         }
3737
3738         if (attr->ia_valid) {
3739                 setattr_copy(inode, attr);
3740                 mark_inode_dirty(inode);
3741
3742                 if (attr->ia_valid & ATTR_MODE)
3743                         err = btrfs_acl_chmod(inode);
3744         }
3745
3746         return err;
3747 }
3748
3749 void btrfs_evict_inode(struct inode *inode)
3750 {
3751         struct btrfs_trans_handle *trans;
3752         struct btrfs_root *root = BTRFS_I(inode)->root;
3753         unsigned long nr;
3754         int ret;
3755
3756         trace_btrfs_inode_evict(inode);
3757
3758         truncate_inode_pages(&inode->i_data, 0);
3759         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3760                                root == root->fs_info->tree_root))
3761                 goto no_delete;
3762
3763         if (is_bad_inode(inode)) {
3764                 btrfs_orphan_del(NULL, inode);
3765                 goto no_delete;
3766         }
3767         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3768         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3769
3770         if (root->fs_info->log_root_recovering) {
3771                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3772                 goto no_delete;
3773         }
3774
3775         if (inode->i_nlink > 0) {
3776                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3777                 goto no_delete;
3778         }
3779
3780         btrfs_i_size_write(inode, 0);
3781
3782         while (1) {
3783                 trans = btrfs_start_transaction(root, 0);
3784                 BUG_ON(IS_ERR(trans));
3785                 btrfs_set_trans_block_group(trans, inode);
3786                 trans->block_rsv = root->orphan_block_rsv;
3787
3788                 ret = btrfs_block_rsv_check(trans, root,
3789                                             root->orphan_block_rsv, 0, 5);
3790                 if (ret) {
3791                         BUG_ON(ret != -EAGAIN);
3792                         ret = btrfs_commit_transaction(trans, root);
3793                         BUG_ON(ret);
3794                         continue;
3795                 }
3796
3797                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3798                 if (ret != -EAGAIN)
3799                         break;
3800
3801                 nr = trans->blocks_used;
3802                 btrfs_end_transaction(trans, root);
3803                 trans = NULL;
3804                 btrfs_btree_balance_dirty(root, nr);
3805
3806         }
3807
3808         if (ret == 0) {
3809                 ret = btrfs_orphan_del(trans, inode);
3810                 BUG_ON(ret);
3811         }
3812
3813         nr = trans->blocks_used;
3814         btrfs_end_transaction(trans, root);
3815         btrfs_btree_balance_dirty(root, nr);
3816 no_delete:
3817         end_writeback(inode);
3818         return;
3819 }
3820
3821 /*
3822  * this returns the key found in the dir entry in the location pointer.
3823  * If no dir entries were found, location->objectid is 0.
3824  */
3825 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3826                                struct btrfs_key *location)
3827 {
3828         const char *name = dentry->d_name.name;
3829         int namelen = dentry->d_name.len;
3830         struct btrfs_dir_item *di;
3831         struct btrfs_path *path;
3832         struct btrfs_root *root = BTRFS_I(dir)->root;
3833         int ret = 0;
3834
3835         path = btrfs_alloc_path();
3836         BUG_ON(!path);
3837
3838         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3839                                     namelen, 0);
3840         if (IS_ERR(di))
3841                 ret = PTR_ERR(di);
3842
3843         if (!di || IS_ERR(di))
3844                 goto out_err;
3845
3846         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3847 out:
3848         btrfs_free_path(path);
3849         return ret;
3850 out_err:
3851         location->objectid = 0;
3852         goto out;
3853 }
3854
3855 /*
3856  * when we hit a tree root in a directory, the btrfs part of the inode
3857  * needs to be changed to reflect the root directory of the tree root.  This
3858  * is kind of like crossing a mount point.
3859  */
3860 static int fixup_tree_root_location(struct btrfs_root *root,
3861                                     struct inode *dir,
3862                                     struct dentry *dentry,
3863                                     struct btrfs_key *location,
3864                                     struct btrfs_root **sub_root)
3865 {
3866         struct btrfs_path *path;
3867         struct btrfs_root *new_root;
3868         struct btrfs_root_ref *ref;
3869         struct extent_buffer *leaf;
3870         int ret;
3871         int err = 0;
3872
3873         path = btrfs_alloc_path();
3874         if (!path) {
3875                 err = -ENOMEM;
3876                 goto out;
3877         }
3878
3879         err = -ENOENT;
3880         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3881                                   BTRFS_I(dir)->root->root_key.objectid,
3882                                   location->objectid);
3883         if (ret) {
3884                 if (ret < 0)
3885                         err = ret;
3886                 goto out;
3887         }
3888
3889         leaf = path->nodes[0];
3890         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3891         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3892             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3893                 goto out;
3894
3895         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3896                                    (unsigned long)(ref + 1),
3897                                    dentry->d_name.len);
3898         if (ret)
3899                 goto out;
3900
3901         btrfs_release_path(root->fs_info->tree_root, path);
3902
3903         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3904         if (IS_ERR(new_root)) {
3905                 err = PTR_ERR(new_root);
3906                 goto out;
3907         }
3908
3909         if (btrfs_root_refs(&new_root->root_item) == 0) {
3910                 err = -ENOENT;
3911                 goto out;
3912         }
3913
3914         *sub_root = new_root;
3915         location->objectid = btrfs_root_dirid(&new_root->root_item);
3916         location->type = BTRFS_INODE_ITEM_KEY;
3917         location->offset = 0;
3918         err = 0;
3919 out:
3920         btrfs_free_path(path);
3921         return err;
3922 }
3923
3924 static void inode_tree_add(struct inode *inode)
3925 {
3926         struct btrfs_root *root = BTRFS_I(inode)->root;
3927         struct btrfs_inode *entry;
3928         struct rb_node **p;
3929         struct rb_node *parent;
3930 again:
3931         p = &root->inode_tree.rb_node;
3932         parent = NULL;
3933
3934         if (inode_unhashed(inode))
3935                 return;
3936
3937         spin_lock(&root->inode_lock);
3938         while (*p) {
3939                 parent = *p;
3940                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3941
3942                 if (inode->i_ino < entry->vfs_inode.i_ino)
3943                         p = &parent->rb_left;
3944                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3945                         p = &parent->rb_right;
3946                 else {
3947                         WARN_ON(!(entry->vfs_inode.i_state &
3948                                   (I_WILL_FREE | I_FREEING)));
3949                         rb_erase(parent, &root->inode_tree);
3950                         RB_CLEAR_NODE(parent);
3951                         spin_unlock(&root->inode_lock);
3952                         goto again;
3953                 }
3954         }
3955         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3956         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3957         spin_unlock(&root->inode_lock);
3958 }
3959
3960 static void inode_tree_del(struct inode *inode)
3961 {
3962         struct btrfs_root *root = BTRFS_I(inode)->root;
3963         int empty = 0;
3964
3965         spin_lock(&root->inode_lock);
3966         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3967                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3968                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3969                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3970         }
3971         spin_unlock(&root->inode_lock);
3972
3973         /*
3974          * Free space cache has inodes in the tree root, but the tree root has a
3975          * root_refs of 0, so this could end up dropping the tree root as a
3976          * snapshot, so we need the extra !root->fs_info->tree_root check to
3977          * make sure we don't drop it.
3978          */
3979         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3980             root != root->fs_info->tree_root) {
3981                 synchronize_srcu(&root->fs_info->subvol_srcu);
3982                 spin_lock(&root->inode_lock);
3983                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3984                 spin_unlock(&root->inode_lock);
3985                 if (empty)
3986                         btrfs_add_dead_root(root);
3987         }
3988 }
3989
3990 int btrfs_invalidate_inodes(struct btrfs_root *root)
3991 {
3992         struct rb_node *node;
3993         struct rb_node *prev;
3994         struct btrfs_inode *entry;
3995         struct inode *inode;
3996         u64 objectid = 0;
3997
3998         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3999
4000         spin_lock(&root->inode_lock);
4001 again:
4002         node = root->inode_tree.rb_node;
4003         prev = NULL;
4004         while (node) {
4005                 prev = node;
4006                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4007
4008                 if (objectid < entry->vfs_inode.i_ino)
4009                         node = node->rb_left;
4010                 else if (objectid > entry->vfs_inode.i_ino)
4011                         node = node->rb_right;
4012                 else
4013                         break;
4014         }
4015         if (!node) {
4016                 while (prev) {
4017                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4018                         if (objectid <= entry->vfs_inode.i_ino) {
4019                                 node = prev;
4020                                 break;
4021                         }
4022                         prev = rb_next(prev);
4023                 }
4024         }
4025         while (node) {
4026                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4027                 objectid = entry->vfs_inode.i_ino + 1;
4028                 inode = igrab(&entry->vfs_inode);
4029                 if (inode) {
4030                         spin_unlock(&root->inode_lock);
4031                         if (atomic_read(&inode->i_count) > 1)
4032                                 d_prune_aliases(inode);
4033                         /*
4034                          * btrfs_drop_inode will have it removed from
4035                          * the inode cache when its usage count
4036                          * hits zero.
4037                          */
4038                         iput(inode);
4039                         cond_resched();
4040                         spin_lock(&root->inode_lock);
4041                         goto again;
4042                 }
4043
4044                 if (cond_resched_lock(&root->inode_lock))
4045                         goto again;
4046
4047                 node = rb_next(node);
4048         }
4049         spin_unlock(&root->inode_lock);
4050         return 0;
4051 }
4052
4053 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4054 {
4055         struct btrfs_iget_args *args = p;
4056         inode->i_ino = args->ino;
4057         BTRFS_I(inode)->root = args->root;
4058         btrfs_set_inode_space_info(args->root, inode);
4059         return 0;
4060 }
4061
4062 static int btrfs_find_actor(struct inode *inode, void *opaque)
4063 {
4064         struct btrfs_iget_args *args = opaque;
4065         return args->ino == inode->i_ino &&
4066                 args->root == BTRFS_I(inode)->root;
4067 }
4068
4069 static struct inode *btrfs_iget_locked(struct super_block *s,
4070                                        u64 objectid,
4071                                        struct btrfs_root *root)
4072 {
4073         struct inode *inode;
4074         struct btrfs_iget_args args;
4075         args.ino = objectid;
4076         args.root = root;
4077
4078         inode = iget5_locked(s, objectid, btrfs_find_actor,
4079                              btrfs_init_locked_inode,
4080                              (void *)&args);
4081         return inode;
4082 }
4083
4084 /* Get an inode object given its location and corresponding root.
4085  * Returns in *is_new if the inode was read from disk
4086  */
4087 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4088                          struct btrfs_root *root, int *new)
4089 {
4090         struct inode *inode;
4091
4092         inode = btrfs_iget_locked(s, location->objectid, root);
4093         if (!inode)
4094                 return ERR_PTR(-ENOMEM);
4095
4096         if (inode->i_state & I_NEW) {
4097                 BTRFS_I(inode)->root = root;
4098                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4099                 btrfs_read_locked_inode(inode);
4100                 inode_tree_add(inode);
4101                 unlock_new_inode(inode);
4102                 if (new)
4103                         *new = 1;
4104         }
4105
4106         return inode;
4107 }
4108
4109 static struct inode *new_simple_dir(struct super_block *s,
4110                                     struct btrfs_key *key,
4111                                     struct btrfs_root *root)
4112 {
4113         struct inode *inode = new_inode(s);
4114
4115         if (!inode)
4116                 return ERR_PTR(-ENOMEM);
4117
4118         BTRFS_I(inode)->root = root;
4119         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4120         BTRFS_I(inode)->dummy_inode = 1;
4121
4122         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4123         inode->i_op = &simple_dir_inode_operations;
4124         inode->i_fop = &simple_dir_operations;
4125         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4126         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4127
4128         return inode;
4129 }
4130
4131 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4132 {
4133         struct inode *inode;
4134         struct btrfs_root *root = BTRFS_I(dir)->root;
4135         struct btrfs_root *sub_root = root;
4136         struct btrfs_key location;
4137         int index;
4138         int ret;
4139
4140         if (dentry->d_name.len > BTRFS_NAME_LEN)
4141                 return ERR_PTR(-ENAMETOOLONG);
4142
4143         ret = btrfs_inode_by_name(dir, dentry, &location);
4144
4145         if (ret < 0)
4146                 return ERR_PTR(ret);
4147
4148         if (location.objectid == 0)
4149                 return NULL;
4150
4151         if (location.type == BTRFS_INODE_ITEM_KEY) {
4152                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4153                 return inode;
4154         }
4155
4156         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4157
4158         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4159         ret = fixup_tree_root_location(root, dir, dentry,
4160                                        &location, &sub_root);
4161         if (ret < 0) {
4162                 if (ret != -ENOENT)
4163                         inode = ERR_PTR(ret);
4164                 else
4165                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4166         } else {
4167                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4168         }
4169         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4170
4171         if (!IS_ERR(inode) && root != sub_root) {
4172                 down_read(&root->fs_info->cleanup_work_sem);
4173                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4174                         ret = btrfs_orphan_cleanup(sub_root);
4175                 up_read(&root->fs_info->cleanup_work_sem);
4176                 if (ret)
4177                         inode = ERR_PTR(ret);
4178         }
4179
4180         return inode;
4181 }
4182
4183 static int btrfs_dentry_delete(const struct dentry *dentry)
4184 {
4185         struct btrfs_root *root;
4186
4187         if (!dentry->d_inode && !IS_ROOT(dentry))
4188                 dentry = dentry->d_parent;
4189
4190         if (dentry->d_inode) {
4191                 root = BTRFS_I(dentry->d_inode)->root;
4192                 if (btrfs_root_refs(&root->root_item) == 0)
4193                         return 1;
4194         }
4195         return 0;
4196 }
4197
4198 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4199                                    struct nameidata *nd)
4200 {
4201         struct inode *inode;
4202
4203         inode = btrfs_lookup_dentry(dir, dentry);
4204         if (IS_ERR(inode))
4205                 return ERR_CAST(inode);
4206
4207         return d_splice_alias(inode, dentry);
4208 }
4209
4210 static unsigned char btrfs_filetype_table[] = {
4211         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4212 };
4213
4214 static int btrfs_real_readdir(struct file *filp, void *dirent,
4215                               filldir_t filldir)
4216 {
4217         struct inode *inode = filp->f_dentry->d_inode;
4218         struct btrfs_root *root = BTRFS_I(inode)->root;
4219         struct btrfs_item *item;
4220         struct btrfs_dir_item *di;
4221         struct btrfs_key key;
4222         struct btrfs_key found_key;
4223         struct btrfs_path *path;
4224         int ret;
4225         struct extent_buffer *leaf;
4226         int slot;
4227         unsigned char d_type;
4228         int over = 0;
4229         u32 di_cur;
4230         u32 di_total;
4231         u32 di_len;
4232         int key_type = BTRFS_DIR_INDEX_KEY;
4233         char tmp_name[32];
4234         char *name_ptr;
4235         int name_len;
4236
4237         /* FIXME, use a real flag for deciding about the key type */
4238         if (root->fs_info->tree_root == root)
4239                 key_type = BTRFS_DIR_ITEM_KEY;
4240
4241         /* special case for "." */
4242         if (filp->f_pos == 0) {
4243                 over = filldir(dirent, ".", 1,
4244                                1, inode->i_ino,
4245                                DT_DIR);
4246                 if (over)
4247                         return 0;
4248                 filp->f_pos = 1;
4249         }
4250         /* special case for .., just use the back ref */
4251         if (filp->f_pos == 1) {
4252                 u64 pino = parent_ino(filp->f_path.dentry);
4253                 over = filldir(dirent, "..", 2,
4254                                2, pino, DT_DIR);
4255                 if (over)
4256                         return 0;
4257                 filp->f_pos = 2;
4258         }
4259         path = btrfs_alloc_path();
4260         path->reada = 2;
4261
4262         btrfs_set_key_type(&key, key_type);
4263         key.offset = filp->f_pos;
4264         key.objectid = inode->i_ino;
4265
4266         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4267         if (ret < 0)
4268                 goto err;
4269
4270         while (1) {
4271                 leaf = path->nodes[0];
4272                 slot = path->slots[0];
4273                 if (slot >= btrfs_header_nritems(leaf)) {
4274                         ret = btrfs_next_leaf(root, path);
4275                         if (ret < 0)
4276                                 goto err;
4277                         else if (ret > 0)
4278                                 break;
4279                         continue;
4280                 }
4281
4282                 item = btrfs_item_nr(leaf, slot);
4283                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4284
4285                 if (found_key.objectid != key.objectid)
4286                         break;
4287                 if (btrfs_key_type(&found_key) != key_type)
4288                         break;
4289                 if (found_key.offset < filp->f_pos)
4290                         goto next;
4291
4292                 filp->f_pos = found_key.offset;
4293
4294                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4295                 di_cur = 0;
4296                 di_total = btrfs_item_size(leaf, item);
4297
4298                 while (di_cur < di_total) {
4299                         struct btrfs_key location;
4300
4301                         if (verify_dir_item(root, leaf, di))
4302                                 break;
4303
4304                         name_len = btrfs_dir_name_len(leaf, di);
4305                         if (name_len <= sizeof(tmp_name)) {
4306                                 name_ptr = tmp_name;
4307                         } else {
4308                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4309                                 if (!name_ptr) {
4310                                         ret = -ENOMEM;
4311                                         goto err;
4312                                 }
4313                         }
4314                         read_extent_buffer(leaf, name_ptr,
4315                                            (unsigned long)(di + 1), name_len);
4316
4317                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4318                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4319
4320                         /* is this a reference to our own snapshot? If so
4321                          * skip it
4322                          */
4323                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4324                             location.objectid == root->root_key.objectid) {
4325                                 over = 0;
4326                                 goto skip;
4327                         }
4328                         over = filldir(dirent, name_ptr, name_len,
4329                                        found_key.offset, location.objectid,
4330                                        d_type);
4331
4332 skip:
4333                         if (name_ptr != tmp_name)
4334                                 kfree(name_ptr);
4335
4336                         if (over)
4337                                 goto nopos;
4338                         di_len = btrfs_dir_name_len(leaf, di) +
4339                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4340                         di_cur += di_len;
4341                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4342                 }
4343 next:
4344                 path->slots[0]++;
4345         }
4346
4347         /* Reached end of directory/root. Bump pos past the last item. */
4348         if (key_type == BTRFS_DIR_INDEX_KEY)
4349                 /*
4350                  * 32-bit glibc will use getdents64, but then strtol -
4351                  * so the last number we can serve is this.
4352                  */
4353                 filp->f_pos = 0x7fffffff;
4354         else
4355                 filp->f_pos++;
4356 nopos:
4357         ret = 0;
4358 err:
4359         btrfs_free_path(path);
4360         return ret;
4361 }
4362
4363 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4364 {
4365         struct btrfs_root *root = BTRFS_I(inode)->root;
4366         struct btrfs_trans_handle *trans;
4367         int ret = 0;
4368         bool nolock = false;
4369
4370         if (BTRFS_I(inode)->dummy_inode)
4371                 return 0;
4372
4373         smp_mb();
4374         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4375
4376         if (wbc->sync_mode == WB_SYNC_ALL) {
4377                 if (nolock)
4378                         trans = btrfs_join_transaction_nolock(root, 1);
4379                 else
4380                         trans = btrfs_join_transaction(root, 1);
4381                 if (IS_ERR(trans))
4382                         return PTR_ERR(trans);
4383                 btrfs_set_trans_block_group(trans, inode);
4384                 if (nolock)
4385                         ret = btrfs_end_transaction_nolock(trans, root);
4386                 else
4387                         ret = btrfs_commit_transaction(trans, root);
4388         }
4389         return ret;
4390 }
4391
4392 /*
4393  * This is somewhat expensive, updating the tree every time the
4394  * inode changes.  But, it is most likely to find the inode in cache.
4395  * FIXME, needs more benchmarking...there are no reasons other than performance
4396  * to keep or drop this code.
4397  */
4398 void btrfs_dirty_inode(struct inode *inode)
4399 {
4400         struct btrfs_root *root = BTRFS_I(inode)->root;
4401         struct btrfs_trans_handle *trans;
4402         int ret;
4403
4404         if (BTRFS_I(inode)->dummy_inode)
4405                 return;
4406
4407         trans = btrfs_join_transaction(root, 1);
4408         BUG_ON(IS_ERR(trans));
4409         btrfs_set_trans_block_group(trans, inode);
4410
4411         ret = btrfs_update_inode(trans, root, inode);
4412         if (ret && ret == -ENOSPC) {
4413                 /* whoops, lets try again with the full transaction */
4414                 btrfs_end_transaction(trans, root);
4415                 trans = btrfs_start_transaction(root, 1);
4416                 if (IS_ERR(trans)) {
4417                         if (printk_ratelimit()) {
4418                                 printk(KERN_ERR "btrfs: fail to "
4419                                        "dirty  inode %lu error %ld\n",
4420                                        inode->i_ino, PTR_ERR(trans));
4421                         }
4422                         return;
4423                 }
4424                 btrfs_set_trans_block_group(trans, inode);
4425
4426                 ret = btrfs_update_inode(trans, root, inode);
4427                 if (ret) {
4428                         if (printk_ratelimit()) {
4429                                 printk(KERN_ERR "btrfs: fail to "
4430                                        "dirty  inode %lu error %d\n",
4431                                        inode->i_ino, ret);
4432                         }
4433                 }
4434         }
4435         btrfs_end_transaction(trans, root);
4436 }
4437
4438 /*
4439  * find the highest existing sequence number in a directory
4440  * and then set the in-memory index_cnt variable to reflect
4441  * free sequence numbers
4442  */
4443 static int btrfs_set_inode_index_count(struct inode *inode)
4444 {
4445         struct btrfs_root *root = BTRFS_I(inode)->root;
4446         struct btrfs_key key, found_key;
4447         struct btrfs_path *path;
4448         struct extent_buffer *leaf;
4449         int ret;
4450
4451         key.objectid = inode->i_ino;
4452         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4453         key.offset = (u64)-1;
4454
4455         path = btrfs_alloc_path();
4456         if (!path)
4457                 return -ENOMEM;
4458
4459         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4460         if (ret < 0)
4461                 goto out;
4462         /* FIXME: we should be able to handle this */
4463         if (ret == 0)
4464                 goto out;
4465         ret = 0;
4466
4467         /*
4468          * MAGIC NUMBER EXPLANATION:
4469          * since we search a directory based on f_pos we have to start at 2
4470          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4471          * else has to start at 2
4472          */
4473         if (path->slots[0] == 0) {
4474                 BTRFS_I(inode)->index_cnt = 2;
4475                 goto out;
4476         }
4477
4478         path->slots[0]--;
4479
4480         leaf = path->nodes[0];
4481         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4482
4483         if (found_key.objectid != inode->i_ino ||
4484             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4485                 BTRFS_I(inode)->index_cnt = 2;
4486                 goto out;
4487         }
4488
4489         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4490 out:
4491         btrfs_free_path(path);
4492         return ret;
4493 }
4494
4495 /*
4496  * helper to find a free sequence number in a given directory.  This current
4497  * code is very simple, later versions will do smarter things in the btree
4498  */
4499 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4500 {
4501         int ret = 0;
4502
4503         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4504                 ret = btrfs_set_inode_index_count(dir);
4505                 if (ret)
4506                         return ret;
4507         }
4508
4509         *index = BTRFS_I(dir)->index_cnt;
4510         BTRFS_I(dir)->index_cnt++;
4511
4512         return ret;
4513 }
4514
4515 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4516                                      struct btrfs_root *root,
4517                                      struct inode *dir,
4518                                      const char *name, int name_len,
4519                                      u64 ref_objectid, u64 objectid,
4520                                      u64 alloc_hint, int mode, u64 *index)
4521 {
4522         struct inode *inode;
4523         struct btrfs_inode_item *inode_item;
4524         struct btrfs_key *location;
4525         struct btrfs_path *path;
4526         struct btrfs_inode_ref *ref;
4527         struct btrfs_key key[2];
4528         u32 sizes[2];
4529         unsigned long ptr;
4530         int ret;
4531         int owner;
4532
4533         path = btrfs_alloc_path();
4534         BUG_ON(!path);
4535
4536         inode = new_inode(root->fs_info->sb);
4537         if (!inode) {
4538                 btrfs_free_path(path);
4539                 return ERR_PTR(-ENOMEM);
4540         }
4541
4542         if (dir) {
4543                 trace_btrfs_inode_request(dir);
4544
4545                 ret = btrfs_set_inode_index(dir, index);
4546                 if (ret) {
4547                         btrfs_free_path(path);
4548                         iput(inode);
4549                         return ERR_PTR(ret);
4550                 }
4551         }
4552         /*
4553          * index_cnt is ignored for everything but a dir,
4554          * btrfs_get_inode_index_count has an explanation for the magic
4555          * number
4556          */
4557         BTRFS_I(inode)->index_cnt = 2;
4558         BTRFS_I(inode)->root = root;
4559         BTRFS_I(inode)->generation = trans->transid;
4560         inode->i_generation = BTRFS_I(inode)->generation;
4561         btrfs_set_inode_space_info(root, inode);
4562
4563         if (mode & S_IFDIR)
4564                 owner = 0;
4565         else
4566                 owner = 1;
4567         BTRFS_I(inode)->block_group =
4568                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4569
4570         key[0].objectid = objectid;
4571         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4572         key[0].offset = 0;
4573
4574         key[1].objectid = objectid;
4575         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4576         key[1].offset = ref_objectid;
4577
4578         sizes[0] = sizeof(struct btrfs_inode_item);
4579         sizes[1] = name_len + sizeof(*ref);
4580
4581         path->leave_spinning = 1;
4582         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4583         if (ret != 0)
4584                 goto fail;
4585
4586         inode_init_owner(inode, dir, mode);
4587         inode->i_ino = objectid;
4588         inode_set_bytes(inode, 0);
4589         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4590         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4591                                   struct btrfs_inode_item);
4592         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4593
4594         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4595                              struct btrfs_inode_ref);
4596         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4597         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4598         ptr = (unsigned long)(ref + 1);
4599         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4600
4601         btrfs_mark_buffer_dirty(path->nodes[0]);
4602         btrfs_free_path(path);
4603
4604         location = &BTRFS_I(inode)->location;
4605         location->objectid = objectid;
4606         location->offset = 0;
4607         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4608
4609         btrfs_inherit_iflags(inode, dir);
4610
4611         if ((mode & S_IFREG)) {
4612                 if (btrfs_test_opt(root, NODATASUM))
4613                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4614                 if (btrfs_test_opt(root, NODATACOW) ||
4615                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4616                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4617         }
4618
4619         insert_inode_hash(inode);
4620         inode_tree_add(inode);
4621
4622         trace_btrfs_inode_new(inode);
4623
4624         return inode;
4625 fail:
4626         if (dir)
4627                 BTRFS_I(dir)->index_cnt--;
4628         btrfs_free_path(path);
4629         iput(inode);
4630         return ERR_PTR(ret);
4631 }
4632
4633 static inline u8 btrfs_inode_type(struct inode *inode)
4634 {
4635         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4636 }
4637
4638 /*
4639  * utility function to add 'inode' into 'parent_inode' with
4640  * a give name and a given sequence number.
4641  * if 'add_backref' is true, also insert a backref from the
4642  * inode to the parent directory.
4643  */
4644 int btrfs_add_link(struct btrfs_trans_handle *trans,
4645                    struct inode *parent_inode, struct inode *inode,
4646                    const char *name, int name_len, int add_backref, u64 index)
4647 {
4648         int ret = 0;
4649         struct btrfs_key key;
4650         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4651
4652         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4653                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4654         } else {
4655                 key.objectid = inode->i_ino;
4656                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4657                 key.offset = 0;
4658         }
4659
4660         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4661                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4662                                          key.objectid, root->root_key.objectid,
4663                                          parent_inode->i_ino,
4664                                          index, name, name_len);
4665         } else if (add_backref) {
4666                 ret = btrfs_insert_inode_ref(trans, root,
4667                                              name, name_len, inode->i_ino,
4668                                              parent_inode->i_ino, index);
4669         }
4670
4671         if (ret == 0) {
4672                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4673                                             parent_inode->i_ino, &key,
4674                                             btrfs_inode_type(inode), index);
4675                 BUG_ON(ret);
4676
4677                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4678                                    name_len * 2);
4679                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4680                 ret = btrfs_update_inode(trans, root, parent_inode);
4681         }
4682         return ret;
4683 }
4684
4685 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4686                             struct inode *dir, struct dentry *dentry,
4687                             struct inode *inode, int backref, u64 index)
4688 {
4689         int err = btrfs_add_link(trans, dir, inode,
4690                                  dentry->d_name.name, dentry->d_name.len,
4691                                  backref, index);
4692         if (!err) {
4693                 d_instantiate(dentry, inode);
4694                 return 0;
4695         }
4696         if (err > 0)
4697                 err = -EEXIST;
4698         return err;
4699 }
4700
4701 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4702                         int mode, dev_t rdev)
4703 {
4704         struct btrfs_trans_handle *trans;
4705         struct btrfs_root *root = BTRFS_I(dir)->root;
4706         struct inode *inode = NULL;
4707         int err;
4708         int drop_inode = 0;
4709         u64 objectid;
4710         unsigned long nr = 0;
4711         u64 index = 0;
4712
4713         if (!new_valid_dev(rdev))
4714                 return -EINVAL;
4715
4716         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4717         if (err)
4718                 return err;
4719
4720         /*
4721          * 2 for inode item and ref
4722          * 2 for dir items
4723          * 1 for xattr if selinux is on
4724          */
4725         trans = btrfs_start_transaction(root, 5);
4726         if (IS_ERR(trans))
4727                 return PTR_ERR(trans);
4728
4729         btrfs_set_trans_block_group(trans, dir);
4730
4731         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4732                                 dentry->d_name.len, dir->i_ino, objectid,
4733                                 BTRFS_I(dir)->block_group, mode, &index);
4734         err = PTR_ERR(inode);
4735         if (IS_ERR(inode))
4736                 goto out_unlock;
4737
4738         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4739         if (err) {
4740                 drop_inode = 1;
4741                 goto out_unlock;
4742         }
4743
4744         btrfs_set_trans_block_group(trans, inode);
4745         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4746         if (err)
4747                 drop_inode = 1;
4748         else {
4749                 inode->i_op = &btrfs_special_inode_operations;
4750                 init_special_inode(inode, inode->i_mode, rdev);
4751                 btrfs_update_inode(trans, root, inode);
4752         }
4753         btrfs_update_inode_block_group(trans, inode);
4754         btrfs_update_inode_block_group(trans, dir);
4755 out_unlock:
4756         nr = trans->blocks_used;
4757         btrfs_end_transaction_throttle(trans, root);
4758         btrfs_btree_balance_dirty(root, nr);
4759         if (drop_inode) {
4760                 inode_dec_link_count(inode);
4761                 iput(inode);
4762         }
4763         return err;
4764 }
4765
4766 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4767                         int mode, struct nameidata *nd)
4768 {
4769         struct btrfs_trans_handle *trans;
4770         struct btrfs_root *root = BTRFS_I(dir)->root;
4771         struct inode *inode = NULL;
4772         int drop_inode = 0;
4773         int err;
4774         unsigned long nr = 0;
4775         u64 objectid;
4776         u64 index = 0;
4777
4778         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4779         if (err)
4780                 return err;
4781         /*
4782          * 2 for inode item and ref
4783          * 2 for dir items
4784          * 1 for xattr if selinux is on
4785          */
4786         trans = btrfs_start_transaction(root, 5);
4787         if (IS_ERR(trans))
4788                 return PTR_ERR(trans);
4789
4790         btrfs_set_trans_block_group(trans, dir);
4791
4792         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4793                                 dentry->d_name.len, dir->i_ino, objectid,
4794                                 BTRFS_I(dir)->block_group, mode, &index);
4795         err = PTR_ERR(inode);
4796         if (IS_ERR(inode))
4797                 goto out_unlock;
4798
4799         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4800         if (err) {
4801                 drop_inode = 1;
4802                 goto out_unlock;
4803         }
4804
4805         btrfs_set_trans_block_group(trans, inode);
4806         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4807         if (err)
4808                 drop_inode = 1;
4809         else {
4810                 inode->i_mapping->a_ops = &btrfs_aops;
4811                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4812                 inode->i_fop = &btrfs_file_operations;
4813                 inode->i_op = &btrfs_file_inode_operations;
4814                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4815         }
4816         btrfs_update_inode_block_group(trans, inode);
4817         btrfs_update_inode_block_group(trans, dir);
4818 out_unlock:
4819         nr = trans->blocks_used;
4820         btrfs_end_transaction_throttle(trans, root);
4821         if (drop_inode) {
4822                 inode_dec_link_count(inode);
4823                 iput(inode);
4824         }
4825         btrfs_btree_balance_dirty(root, nr);
4826         return err;
4827 }
4828
4829 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4830                       struct dentry *dentry)
4831 {
4832         struct btrfs_trans_handle *trans;
4833         struct btrfs_root *root = BTRFS_I(dir)->root;
4834         struct inode *inode = old_dentry->d_inode;
4835         u64 index;
4836         unsigned long nr = 0;
4837         int err;
4838         int drop_inode = 0;
4839
4840         /* do not allow sys_link's with other subvols of the same device */
4841         if (root->objectid != BTRFS_I(inode)->root->objectid)
4842                 return -EXDEV;
4843
4844         if (inode->i_nlink == ~0U)
4845                 return -EMLINK;
4846
4847         err = btrfs_set_inode_index(dir, &index);
4848         if (err)
4849                 goto fail;
4850
4851         /*
4852          * 2 items for inode and inode ref
4853          * 2 items for dir items
4854          * 1 item for parent inode
4855          */
4856         trans = btrfs_start_transaction(root, 5);
4857         if (IS_ERR(trans)) {
4858                 err = PTR_ERR(trans);
4859                 goto fail;
4860         }
4861
4862         btrfs_inc_nlink(inode);
4863         inode->i_ctime = CURRENT_TIME;
4864
4865         btrfs_set_trans_block_group(trans, dir);
4866         ihold(inode);
4867
4868         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4869
4870         if (err) {
4871                 drop_inode = 1;
4872         } else {
4873                 struct dentry *parent = dget_parent(dentry);
4874                 btrfs_update_inode_block_group(trans, dir);
4875                 err = btrfs_update_inode(trans, root, inode);
4876                 BUG_ON(err);
4877                 btrfs_log_new_name(trans, inode, NULL, parent);
4878                 dput(parent);
4879         }
4880
4881         nr = trans->blocks_used;
4882         btrfs_end_transaction_throttle(trans, root);
4883 fail:
4884         if (drop_inode) {
4885                 inode_dec_link_count(inode);
4886                 iput(inode);
4887         }
4888         btrfs_btree_balance_dirty(root, nr);
4889         return err;
4890 }
4891
4892 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4893 {
4894         struct inode *inode = NULL;
4895         struct btrfs_trans_handle *trans;
4896         struct btrfs_root *root = BTRFS_I(dir)->root;
4897         int err = 0;
4898         int drop_on_err = 0;
4899         u64 objectid = 0;
4900         u64 index = 0;
4901         unsigned long nr = 1;
4902
4903         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4904         if (err)
4905                 return err;
4906
4907         /*
4908          * 2 items for inode and ref
4909          * 2 items for dir items
4910          * 1 for xattr if selinux is on
4911          */
4912         trans = btrfs_start_transaction(root, 5);
4913         if (IS_ERR(trans))
4914                 return PTR_ERR(trans);
4915         btrfs_set_trans_block_group(trans, dir);
4916
4917         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4918                                 dentry->d_name.len, dir->i_ino, objectid,
4919                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4920                                 &index);
4921         if (IS_ERR(inode)) {
4922                 err = PTR_ERR(inode);
4923                 goto out_fail;
4924         }
4925
4926         drop_on_err = 1;
4927
4928         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4929         if (err)
4930                 goto out_fail;
4931
4932         inode->i_op = &btrfs_dir_inode_operations;
4933         inode->i_fop = &btrfs_dir_file_operations;
4934         btrfs_set_trans_block_group(trans, inode);
4935
4936         btrfs_i_size_write(inode, 0);
4937         err = btrfs_update_inode(trans, root, inode);
4938         if (err)
4939                 goto out_fail;
4940
4941         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4942                              dentry->d_name.len, 0, index);
4943         if (err)
4944                 goto out_fail;
4945
4946         d_instantiate(dentry, inode);
4947         drop_on_err = 0;
4948         btrfs_update_inode_block_group(trans, inode);
4949         btrfs_update_inode_block_group(trans, dir);
4950
4951 out_fail:
4952         nr = trans->blocks_used;
4953         btrfs_end_transaction_throttle(trans, root);
4954         if (drop_on_err)
4955                 iput(inode);
4956         btrfs_btree_balance_dirty(root, nr);
4957         return err;
4958 }
4959
4960 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4961  * and an extent that you want to insert, deal with overlap and insert
4962  * the new extent into the tree.
4963  */
4964 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4965                                 struct extent_map *existing,
4966                                 struct extent_map *em,
4967                                 u64 map_start, u64 map_len)
4968 {
4969         u64 start_diff;
4970
4971         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4972         start_diff = map_start - em->start;
4973         em->start = map_start;
4974         em->len = map_len;
4975         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4976             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4977                 em->block_start += start_diff;
4978                 em->block_len -= start_diff;
4979         }
4980         return add_extent_mapping(em_tree, em);
4981 }
4982
4983 static noinline int uncompress_inline(struct btrfs_path *path,
4984                                       struct inode *inode, struct page *page,
4985                                       size_t pg_offset, u64 extent_offset,
4986                                       struct btrfs_file_extent_item *item)
4987 {
4988         int ret;
4989         struct extent_buffer *leaf = path->nodes[0];
4990         char *tmp;
4991         size_t max_size;
4992         unsigned long inline_size;
4993         unsigned long ptr;
4994         int compress_type;
4995
4996         WARN_ON(pg_offset != 0);
4997         compress_type = btrfs_file_extent_compression(leaf, item);
4998         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4999         inline_size = btrfs_file_extent_inline_item_len(leaf,
5000                                         btrfs_item_nr(leaf, path->slots[0]));
5001         tmp = kmalloc(inline_size, GFP_NOFS);
5002         ptr = btrfs_file_extent_inline_start(item);
5003
5004         read_extent_buffer(leaf, tmp, ptr, inline_size);
5005
5006         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5007         ret = btrfs_decompress(compress_type, tmp, page,
5008                                extent_offset, inline_size, max_size);
5009         if (ret) {
5010                 char *kaddr = kmap_atomic(page, KM_USER0);
5011                 unsigned long copy_size = min_t(u64,
5012                                   PAGE_CACHE_SIZE - pg_offset,
5013                                   max_size - extent_offset);
5014                 memset(kaddr + pg_offset, 0, copy_size);
5015                 kunmap_atomic(kaddr, KM_USER0);
5016         }
5017         kfree(tmp);
5018         return 0;
5019 }
5020
5021 /*
5022  * a bit scary, this does extent mapping from logical file offset to the disk.
5023  * the ugly parts come from merging extents from the disk with the in-ram
5024  * representation.  This gets more complex because of the data=ordered code,
5025  * where the in-ram extents might be locked pending data=ordered completion.
5026  *
5027  * This also copies inline extents directly into the page.
5028  */
5029
5030 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5031                                     size_t pg_offset, u64 start, u64 len,
5032                                     int create)
5033 {
5034         int ret;
5035         int err = 0;
5036         u64 bytenr;
5037         u64 extent_start = 0;
5038         u64 extent_end = 0;
5039         u64 objectid = inode->i_ino;
5040         u32 found_type;
5041         struct btrfs_path *path = NULL;
5042         struct btrfs_root *root = BTRFS_I(inode)->root;
5043         struct btrfs_file_extent_item *item;
5044         struct extent_buffer *leaf;
5045         struct btrfs_key found_key;
5046         struct extent_map *em = NULL;
5047         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5048         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5049         struct btrfs_trans_handle *trans = NULL;
5050         int compress_type;
5051
5052 again:
5053         read_lock(&em_tree->lock);
5054         em = lookup_extent_mapping(em_tree, start, len);
5055         if (em)
5056                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5057         read_unlock(&em_tree->lock);
5058
5059         if (em) {
5060                 if (em->start > start || em->start + em->len <= start)
5061                         free_extent_map(em);
5062                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5063                         free_extent_map(em);
5064                 else
5065                         goto out;
5066         }
5067         em = alloc_extent_map(GFP_NOFS);
5068         if (!em) {
5069                 err = -ENOMEM;
5070                 goto out;
5071         }
5072         em->bdev = root->fs_info->fs_devices->latest_bdev;
5073         em->start = EXTENT_MAP_HOLE;
5074         em->orig_start = EXTENT_MAP_HOLE;
5075         em->len = (u64)-1;
5076         em->block_len = (u64)-1;
5077
5078         if (!path) {
5079                 path = btrfs_alloc_path();
5080                 BUG_ON(!path);
5081         }
5082
5083         ret = btrfs_lookup_file_extent(trans, root, path,
5084                                        objectid, start, trans != NULL);
5085         if (ret < 0) {
5086                 err = ret;
5087                 goto out;
5088         }
5089
5090         if (ret != 0) {
5091                 if (path->slots[0] == 0)
5092                         goto not_found;
5093                 path->slots[0]--;
5094         }
5095
5096         leaf = path->nodes[0];
5097         item = btrfs_item_ptr(leaf, path->slots[0],
5098                               struct btrfs_file_extent_item);
5099         /* are we inside the extent that was found? */
5100         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5101         found_type = btrfs_key_type(&found_key);
5102         if (found_key.objectid != objectid ||
5103             found_type != BTRFS_EXTENT_DATA_KEY) {
5104                 goto not_found;
5105         }
5106
5107         found_type = btrfs_file_extent_type(leaf, item);
5108         extent_start = found_key.offset;
5109         compress_type = btrfs_file_extent_compression(leaf, item);
5110         if (found_type == BTRFS_FILE_EXTENT_REG ||
5111             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5112                 extent_end = extent_start +
5113                        btrfs_file_extent_num_bytes(leaf, item);
5114         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5115                 size_t size;
5116                 size = btrfs_file_extent_inline_len(leaf, item);
5117                 extent_end = (extent_start + size + root->sectorsize - 1) &
5118                         ~((u64)root->sectorsize - 1);
5119         }
5120
5121         if (start >= extent_end) {
5122                 path->slots[0]++;
5123                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5124                         ret = btrfs_next_leaf(root, path);
5125                         if (ret < 0) {
5126                                 err = ret;
5127                                 goto out;
5128                         }
5129                         if (ret > 0)
5130                                 goto not_found;
5131                         leaf = path->nodes[0];
5132                 }
5133                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5134                 if (found_key.objectid != objectid ||
5135                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5136                         goto not_found;
5137                 if (start + len <= found_key.offset)
5138                         goto not_found;
5139                 em->start = start;
5140                 em->len = found_key.offset - start;
5141                 goto not_found_em;
5142         }
5143
5144         if (found_type == BTRFS_FILE_EXTENT_REG ||
5145             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5146                 em->start = extent_start;
5147                 em->len = extent_end - extent_start;
5148                 em->orig_start = extent_start -
5149                                  btrfs_file_extent_offset(leaf, item);
5150                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5151                 if (bytenr == 0) {
5152                         em->block_start = EXTENT_MAP_HOLE;
5153                         goto insert;
5154                 }
5155                 if (compress_type != BTRFS_COMPRESS_NONE) {
5156                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5157                         em->compress_type = compress_type;
5158                         em->block_start = bytenr;
5159                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5160                                                                          item);
5161                 } else {
5162                         bytenr += btrfs_file_extent_offset(leaf, item);
5163                         em->block_start = bytenr;
5164                         em->block_len = em->len;
5165                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5166                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5167                 }
5168                 goto insert;
5169         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5170                 unsigned long ptr;
5171                 char *map;
5172                 size_t size;
5173                 size_t extent_offset;
5174                 size_t copy_size;
5175
5176                 em->block_start = EXTENT_MAP_INLINE;
5177                 if (!page || create) {
5178                         em->start = extent_start;
5179                         em->len = extent_end - extent_start;
5180                         goto out;
5181                 }
5182
5183                 size = btrfs_file_extent_inline_len(leaf, item);
5184                 extent_offset = page_offset(page) + pg_offset - extent_start;
5185                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5186                                 size - extent_offset);
5187                 em->start = extent_start + extent_offset;
5188                 em->len = (copy_size + root->sectorsize - 1) &
5189                         ~((u64)root->sectorsize - 1);
5190                 em->orig_start = EXTENT_MAP_INLINE;
5191                 if (compress_type) {
5192                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5193                         em->compress_type = compress_type;
5194                 }
5195                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5196                 if (create == 0 && !PageUptodate(page)) {
5197                         if (btrfs_file_extent_compression(leaf, item) !=
5198                             BTRFS_COMPRESS_NONE) {
5199                                 ret = uncompress_inline(path, inode, page,
5200                                                         pg_offset,
5201                                                         extent_offset, item);
5202                                 BUG_ON(ret);
5203                         } else {
5204                                 map = kmap(page);
5205                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5206                                                    copy_size);
5207                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5208                                         memset(map + pg_offset + copy_size, 0,
5209                                                PAGE_CACHE_SIZE - pg_offset -
5210                                                copy_size);
5211                                 }
5212                                 kunmap(page);
5213                         }
5214                         flush_dcache_page(page);
5215                 } else if (create && PageUptodate(page)) {
5216                         WARN_ON(1);
5217                         if (!trans) {
5218                                 kunmap(page);
5219                                 free_extent_map(em);
5220                                 em = NULL;
5221                                 btrfs_release_path(root, path);
5222                                 trans = btrfs_join_transaction(root, 1);
5223                                 if (IS_ERR(trans))
5224                                         return ERR_CAST(trans);
5225                                 goto again;
5226                         }
5227                         map = kmap(page);
5228                         write_extent_buffer(leaf, map + pg_offset, ptr,
5229                                             copy_size);
5230                         kunmap(page);
5231                         btrfs_mark_buffer_dirty(leaf);
5232                 }
5233                 set_extent_uptodate(io_tree, em->start,
5234                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5235                 goto insert;
5236         } else {
5237                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5238                 WARN_ON(1);
5239         }
5240 not_found:
5241         em->start = start;
5242         em->len = len;
5243 not_found_em:
5244         em->block_start = EXTENT_MAP_HOLE;
5245         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5246 insert:
5247         btrfs_release_path(root, path);
5248         if (em->start > start || extent_map_end(em) <= start) {
5249                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5250                        "[%llu %llu]\n", (unsigned long long)em->start,
5251                        (unsigned long long)em->len,
5252                        (unsigned long long)start,
5253                        (unsigned long long)len);
5254                 err = -EIO;
5255                 goto out;
5256         }
5257
5258         err = 0;
5259         write_lock(&em_tree->lock);
5260         ret = add_extent_mapping(em_tree, em);
5261         /* it is possible that someone inserted the extent into the tree
5262          * while we had the lock dropped.  It is also possible that
5263          * an overlapping map exists in the tree
5264          */
5265         if (ret == -EEXIST) {
5266                 struct extent_map *existing;
5267
5268                 ret = 0;
5269
5270                 existing = lookup_extent_mapping(em_tree, start, len);
5271                 if (existing && (existing->start > start ||
5272                     existing->start + existing->len <= start)) {
5273                         free_extent_map(existing);
5274                         existing = NULL;
5275                 }
5276                 if (!existing) {
5277                         existing = lookup_extent_mapping(em_tree, em->start,
5278                                                          em->len);
5279                         if (existing) {
5280                                 err = merge_extent_mapping(em_tree, existing,
5281                                                            em, start,
5282                                                            root->sectorsize);
5283                                 free_extent_map(existing);
5284                                 if (err) {
5285                                         free_extent_map(em);
5286                                         em = NULL;
5287                                 }
5288                         } else {
5289                                 err = -EIO;
5290                                 free_extent_map(em);
5291                                 em = NULL;
5292                         }
5293                 } else {
5294                         free_extent_map(em);
5295                         em = existing;
5296                         err = 0;
5297                 }
5298         }
5299         write_unlock(&em_tree->lock);
5300 out:
5301
5302         trace_btrfs_get_extent(root, em);
5303
5304         if (path)
5305                 btrfs_free_path(path);
5306         if (trans) {
5307                 ret = btrfs_end_transaction(trans, root);
5308                 if (!err)
5309                         err = ret;
5310         }
5311         if (err) {
5312                 free_extent_map(em);
5313                 return ERR_PTR(err);
5314         }
5315         return em;
5316 }
5317
5318 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5319                                            size_t pg_offset, u64 start, u64 len,
5320                                            int create)
5321 {
5322         struct extent_map *em;
5323         struct extent_map *hole_em = NULL;
5324         u64 range_start = start;
5325         u64 end;
5326         u64 found;
5327         u64 found_end;
5328         int err = 0;
5329
5330         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5331         if (IS_ERR(em))
5332                 return em;
5333         if (em) {
5334                 /*
5335                  * if our em maps to a hole, there might
5336                  * actually be delalloc bytes behind it
5337                  */
5338                 if (em->block_start != EXTENT_MAP_HOLE)
5339                         return em;
5340                 else
5341                         hole_em = em;
5342         }
5343
5344         /* check to see if we've wrapped (len == -1 or similar) */
5345         end = start + len;
5346         if (end < start)
5347                 end = (u64)-1;
5348         else
5349                 end -= 1;
5350
5351         em = NULL;
5352
5353         /* ok, we didn't find anything, lets look for delalloc */
5354         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5355                                  end, len, EXTENT_DELALLOC, 1);
5356         found_end = range_start + found;
5357         if (found_end < range_start)
5358                 found_end = (u64)-1;
5359
5360         /*
5361          * we didn't find anything useful, return
5362          * the original results from get_extent()
5363          */
5364         if (range_start > end || found_end <= start) {
5365                 em = hole_em;
5366                 hole_em = NULL;
5367                 goto out;
5368         }
5369
5370         /* adjust the range_start to make sure it doesn't
5371          * go backwards from the start they passed in
5372          */
5373         range_start = max(start,range_start);
5374         found = found_end - range_start;
5375
5376         if (found > 0) {
5377                 u64 hole_start = start;
5378                 u64 hole_len = len;
5379
5380                 em = alloc_extent_map(GFP_NOFS);
5381                 if (!em) {
5382                         err = -ENOMEM;
5383                         goto out;
5384                 }
5385                 /*
5386                  * when btrfs_get_extent can't find anything it
5387                  * returns one huge hole
5388                  *
5389                  * make sure what it found really fits our range, and
5390                  * adjust to make sure it is based on the start from
5391                  * the caller
5392                  */
5393                 if (hole_em) {
5394                         u64 calc_end = extent_map_end(hole_em);
5395
5396                         if (calc_end <= start || (hole_em->start > end)) {
5397                                 free_extent_map(hole_em);
5398                                 hole_em = NULL;
5399                         } else {
5400                                 hole_start = max(hole_em->start, start);
5401                                 hole_len = calc_end - hole_start;
5402                         }
5403                 }
5404                 em->bdev = NULL;
5405                 if (hole_em && range_start > hole_start) {
5406                         /* our hole starts before our delalloc, so we
5407                          * have to return just the parts of the hole
5408                          * that go until  the delalloc starts
5409                          */
5410                         em->len = min(hole_len,
5411                                       range_start - hole_start);
5412                         em->start = hole_start;
5413                         em->orig_start = hole_start;
5414                         /*
5415                          * don't adjust block start at all,
5416                          * it is fixed at EXTENT_MAP_HOLE
5417                          */
5418                         em->block_start = hole_em->block_start;
5419                         em->block_len = hole_len;
5420                 } else {
5421                         em->start = range_start;
5422                         em->len = found;
5423                         em->orig_start = range_start;
5424                         em->block_start = EXTENT_MAP_DELALLOC;
5425                         em->block_len = found;
5426                 }
5427         } else if (hole_em) {
5428                 return hole_em;
5429         }
5430 out:
5431
5432         free_extent_map(hole_em);
5433         if (err) {
5434                 free_extent_map(em);
5435                 return ERR_PTR(err);
5436         }
5437         return em;
5438 }
5439
5440 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5441                                                   struct extent_map *em,
5442                                                   u64 start, u64 len)
5443 {
5444         struct btrfs_root *root = BTRFS_I(inode)->root;
5445         struct btrfs_trans_handle *trans;
5446         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5447         struct btrfs_key ins;
5448         u64 alloc_hint;
5449         int ret;
5450         bool insert = false;
5451
5452         /*
5453          * Ok if the extent map we looked up is a hole and is for the exact
5454          * range we want, there is no reason to allocate a new one, however if
5455          * it is not right then we need to free this one and drop the cache for
5456          * our range.
5457          */
5458         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5459             em->len != len) {
5460                 free_extent_map(em);
5461                 em = NULL;
5462                 insert = true;
5463                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5464         }
5465
5466         trans = btrfs_join_transaction(root, 0);
5467         if (IS_ERR(trans))
5468                 return ERR_CAST(trans);
5469
5470         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5471
5472         alloc_hint = get_extent_allocation_hint(inode, start, len);
5473         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5474                                    alloc_hint, (u64)-1, &ins, 1);
5475         if (ret) {
5476                 em = ERR_PTR(ret);
5477                 goto out;
5478         }
5479
5480         if (!em) {
5481                 em = alloc_extent_map(GFP_NOFS);
5482                 if (!em) {
5483                         em = ERR_PTR(-ENOMEM);
5484                         goto out;
5485                 }
5486         }
5487
5488         em->start = start;
5489         em->orig_start = em->start;
5490         em->len = ins.offset;
5491
5492         em->block_start = ins.objectid;
5493         em->block_len = ins.offset;
5494         em->bdev = root->fs_info->fs_devices->latest_bdev;
5495
5496         /*
5497          * We need to do this because if we're using the original em we searched
5498          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5499          */
5500         em->flags = 0;
5501         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5502
5503         while (insert) {
5504                 write_lock(&em_tree->lock);
5505                 ret = add_extent_mapping(em_tree, em);
5506                 write_unlock(&em_tree->lock);
5507                 if (ret != -EEXIST)
5508                         break;
5509                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5510         }
5511
5512         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5513                                            ins.offset, ins.offset, 0);
5514         if (ret) {
5515                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5516                 em = ERR_PTR(ret);
5517         }
5518 out:
5519         btrfs_end_transaction(trans, root);
5520         return em;
5521 }
5522
5523 /*
5524  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5525  * block must be cow'd
5526  */
5527 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5528                                       struct inode *inode, u64 offset, u64 len)
5529 {
5530         struct btrfs_path *path;
5531         int ret;
5532         struct extent_buffer *leaf;
5533         struct btrfs_root *root = BTRFS_I(inode)->root;
5534         struct btrfs_file_extent_item *fi;
5535         struct btrfs_key key;
5536         u64 disk_bytenr;
5537         u64 backref_offset;
5538         u64 extent_end;
5539         u64 num_bytes;
5540         int slot;
5541         int found_type;
5542
5543         path = btrfs_alloc_path();
5544         if (!path)
5545                 return -ENOMEM;
5546
5547         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5548                                        offset, 0);
5549         if (ret < 0)
5550                 goto out;
5551
5552         slot = path->slots[0];
5553         if (ret == 1) {
5554                 if (slot == 0) {
5555                         /* can't find the item, must cow */
5556                         ret = 0;
5557                         goto out;
5558                 }
5559                 slot--;
5560         }
5561         ret = 0;
5562         leaf = path->nodes[0];
5563         btrfs_item_key_to_cpu(leaf, &key, slot);
5564         if (key.objectid != inode->i_ino ||
5565             key.type != BTRFS_EXTENT_DATA_KEY) {
5566                 /* not our file or wrong item type, must cow */
5567                 goto out;
5568         }
5569
5570         if (key.offset > offset) {
5571                 /* Wrong offset, must cow */
5572                 goto out;
5573         }
5574
5575         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5576         found_type = btrfs_file_extent_type(leaf, fi);
5577         if (found_type != BTRFS_FILE_EXTENT_REG &&
5578             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5579                 /* not a regular extent, must cow */
5580                 goto out;
5581         }
5582         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5583         backref_offset = btrfs_file_extent_offset(leaf, fi);
5584
5585         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5586         if (extent_end < offset + len) {
5587                 /* extent doesn't include our full range, must cow */
5588                 goto out;
5589         }
5590
5591         if (btrfs_extent_readonly(root, disk_bytenr))
5592                 goto out;
5593
5594         /*
5595          * look for other files referencing this extent, if we
5596          * find any we must cow
5597          */
5598         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5599                                   key.offset - backref_offset, disk_bytenr))
5600                 goto out;
5601
5602         /*
5603          * adjust disk_bytenr and num_bytes to cover just the bytes
5604          * in this extent we are about to write.  If there
5605          * are any csums in that range we have to cow in order
5606          * to keep the csums correct
5607          */
5608         disk_bytenr += backref_offset;
5609         disk_bytenr += offset - key.offset;
5610         num_bytes = min(offset + len, extent_end) - offset;
5611         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5612                                 goto out;
5613         /*
5614          * all of the above have passed, it is safe to overwrite this extent
5615          * without cow
5616          */
5617         ret = 1;
5618 out:
5619         btrfs_free_path(path);
5620         return ret;
5621 }
5622
5623 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5624                                    struct buffer_head *bh_result, int create)
5625 {
5626         struct extent_map *em;
5627         struct btrfs_root *root = BTRFS_I(inode)->root;
5628         u64 start = iblock << inode->i_blkbits;
5629         u64 len = bh_result->b_size;
5630         struct btrfs_trans_handle *trans;
5631
5632         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5633         if (IS_ERR(em))
5634                 return PTR_ERR(em);
5635
5636         /*
5637          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5638          * io.  INLINE is special, and we could probably kludge it in here, but
5639          * it's still buffered so for safety lets just fall back to the generic
5640          * buffered path.
5641          *
5642          * For COMPRESSED we _have_ to read the entire extent in so we can
5643          * decompress it, so there will be buffering required no matter what we
5644          * do, so go ahead and fallback to buffered.
5645          *
5646          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5647          * to buffered IO.  Don't blame me, this is the price we pay for using
5648          * the generic code.
5649          */
5650         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5651             em->block_start == EXTENT_MAP_INLINE) {
5652                 free_extent_map(em);
5653                 return -ENOTBLK;
5654         }
5655
5656         /* Just a good old fashioned hole, return */
5657         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5658                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5659                 free_extent_map(em);
5660                 /* DIO will do one hole at a time, so just unlock a sector */
5661                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5662                               start + root->sectorsize - 1, GFP_NOFS);
5663                 return 0;
5664         }
5665
5666         /*
5667          * We don't allocate a new extent in the following cases
5668          *
5669          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5670          * existing extent.
5671          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5672          * just use the extent.
5673          *
5674          */
5675         if (!create) {
5676                 len = em->len - (start - em->start);
5677                 goto map;
5678         }
5679
5680         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5681             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5682              em->block_start != EXTENT_MAP_HOLE)) {
5683                 int type;
5684                 int ret;
5685                 u64 block_start;
5686
5687                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5688                         type = BTRFS_ORDERED_PREALLOC;
5689                 else
5690                         type = BTRFS_ORDERED_NOCOW;
5691                 len = min(len, em->len - (start - em->start));
5692                 block_start = em->block_start + (start - em->start);
5693
5694                 /*
5695                  * we're not going to log anything, but we do need
5696                  * to make sure the current transaction stays open
5697                  * while we look for nocow cross refs
5698                  */
5699                 trans = btrfs_join_transaction(root, 0);
5700                 if (IS_ERR(trans))
5701                         goto must_cow;
5702
5703                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5704                         ret = btrfs_add_ordered_extent_dio(inode, start,
5705                                            block_start, len, len, type);
5706                         btrfs_end_transaction(trans, root);
5707                         if (ret) {
5708                                 free_extent_map(em);
5709                                 return ret;
5710                         }
5711                         goto unlock;
5712                 }
5713                 btrfs_end_transaction(trans, root);
5714         }
5715 must_cow:
5716         /*
5717          * this will cow the extent, reset the len in case we changed
5718          * it above
5719          */
5720         len = bh_result->b_size;
5721         em = btrfs_new_extent_direct(inode, em, start, len);
5722         if (IS_ERR(em))
5723                 return PTR_ERR(em);
5724         len = min(len, em->len - (start - em->start));
5725 unlock:
5726         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5727                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5728                           0, NULL, GFP_NOFS);
5729 map:
5730         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5731                 inode->i_blkbits;
5732         bh_result->b_size = len;
5733         bh_result->b_bdev = em->bdev;
5734         set_buffer_mapped(bh_result);
5735         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5736                 set_buffer_new(bh_result);
5737
5738         free_extent_map(em);
5739
5740         return 0;
5741 }
5742
5743 struct btrfs_dio_private {
5744         struct inode *inode;
5745         u64 logical_offset;
5746         u64 disk_bytenr;
5747         u64 bytes;
5748         u32 *csums;
5749         void *private;
5750
5751         /* number of bios pending for this dio */
5752         atomic_t pending_bios;
5753
5754         /* IO errors */
5755         int errors;
5756
5757         struct bio *orig_bio;
5758 };
5759
5760 static void btrfs_endio_direct_read(struct bio *bio, int err)
5761 {
5762         struct btrfs_dio_private *dip = bio->bi_private;
5763         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5764         struct bio_vec *bvec = bio->bi_io_vec;
5765         struct inode *inode = dip->inode;
5766         struct btrfs_root *root = BTRFS_I(inode)->root;
5767         u64 start;
5768         u32 *private = dip->csums;
5769
5770         start = dip->logical_offset;
5771         do {
5772                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5773                         struct page *page = bvec->bv_page;
5774                         char *kaddr;
5775                         u32 csum = ~(u32)0;
5776                         unsigned long flags;
5777
5778                         local_irq_save(flags);
5779                         kaddr = kmap_atomic(page, KM_IRQ0);
5780                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5781                                                csum, bvec->bv_len);
5782                         btrfs_csum_final(csum, (char *)&csum);
5783                         kunmap_atomic(kaddr, KM_IRQ0);
5784                         local_irq_restore(flags);
5785
5786                         flush_dcache_page(bvec->bv_page);
5787                         if (csum != *private) {
5788                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5789                                       " %llu csum %u private %u\n",
5790                                       inode->i_ino, (unsigned long long)start,
5791                                       csum, *private);
5792                                 err = -EIO;
5793                         }
5794                 }
5795
5796                 start += bvec->bv_len;
5797                 private++;
5798                 bvec++;
5799         } while (bvec <= bvec_end);
5800
5801         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5802                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5803         bio->bi_private = dip->private;
5804
5805         kfree(dip->csums);
5806         kfree(dip);
5807
5808         /* If we had a csum failure make sure to clear the uptodate flag */
5809         if (err)
5810                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5811         dio_end_io(bio, err);
5812 }
5813
5814 static void btrfs_endio_direct_write(struct bio *bio, int err)
5815 {
5816         struct btrfs_dio_private *dip = bio->bi_private;
5817         struct inode *inode = dip->inode;
5818         struct btrfs_root *root = BTRFS_I(inode)->root;
5819         struct btrfs_trans_handle *trans;
5820         struct btrfs_ordered_extent *ordered = NULL;
5821         struct extent_state *cached_state = NULL;
5822         u64 ordered_offset = dip->logical_offset;
5823         u64 ordered_bytes = dip->bytes;
5824         int ret;
5825
5826         if (err)
5827                 goto out_done;
5828 again:
5829         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5830                                                    &ordered_offset,
5831                                                    ordered_bytes);
5832         if (!ret)
5833                 goto out_test;
5834
5835         BUG_ON(!ordered);
5836
5837         trans = btrfs_join_transaction(root, 1);
5838         if (IS_ERR(trans)) {
5839                 err = -ENOMEM;
5840                 goto out;
5841         }
5842         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5843
5844         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5845                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5846                 if (!ret)
5847                         ret = btrfs_update_inode(trans, root, inode);
5848                 err = ret;
5849                 goto out;
5850         }
5851
5852         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5853                          ordered->file_offset + ordered->len - 1, 0,
5854                          &cached_state, GFP_NOFS);
5855
5856         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5857                 ret = btrfs_mark_extent_written(trans, inode,
5858                                                 ordered->file_offset,
5859                                                 ordered->file_offset +
5860                                                 ordered->len);
5861                 if (ret) {
5862                         err = ret;
5863                         goto out_unlock;
5864                 }
5865         } else {
5866                 ret = insert_reserved_file_extent(trans, inode,
5867                                                   ordered->file_offset,
5868                                                   ordered->start,
5869                                                   ordered->disk_len,
5870                                                   ordered->len,
5871                                                   ordered->len,
5872                                                   0, 0, 0,
5873                                                   BTRFS_FILE_EXTENT_REG);
5874                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5875                                    ordered->file_offset, ordered->len);
5876                 if (ret) {
5877                         err = ret;
5878                         WARN_ON(1);
5879                         goto out_unlock;
5880                 }
5881         }
5882
5883         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5884         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5885         if (!ret)
5886                 btrfs_update_inode(trans, root, inode);
5887         ret = 0;
5888 out_unlock:
5889         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5890                              ordered->file_offset + ordered->len - 1,
5891                              &cached_state, GFP_NOFS);
5892 out:
5893         btrfs_delalloc_release_metadata(inode, ordered->len);
5894         btrfs_end_transaction(trans, root);
5895         ordered_offset = ordered->file_offset + ordered->len;
5896         btrfs_put_ordered_extent(ordered);
5897         btrfs_put_ordered_extent(ordered);
5898
5899 out_test:
5900         /*
5901          * our bio might span multiple ordered extents.  If we haven't
5902          * completed the accounting for the whole dio, go back and try again
5903          */
5904         if (ordered_offset < dip->logical_offset + dip->bytes) {
5905                 ordered_bytes = dip->logical_offset + dip->bytes -
5906                         ordered_offset;
5907                 goto again;
5908         }
5909 out_done:
5910         bio->bi_private = dip->private;
5911
5912         kfree(dip->csums);
5913         kfree(dip);
5914
5915         /* If we had an error make sure to clear the uptodate flag */
5916         if (err)
5917                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5918         dio_end_io(bio, err);
5919 }
5920
5921 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5922                                     struct bio *bio, int mirror_num,
5923                                     unsigned long bio_flags, u64 offset)
5924 {
5925         int ret;
5926         struct btrfs_root *root = BTRFS_I(inode)->root;
5927         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5928         BUG_ON(ret);
5929         return 0;
5930 }
5931
5932 static void btrfs_end_dio_bio(struct bio *bio, int err)
5933 {
5934         struct btrfs_dio_private *dip = bio->bi_private;
5935
5936         if (err) {
5937                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5938                       "sector %#Lx len %u err no %d\n",
5939                       dip->inode->i_ino, bio->bi_rw,
5940                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5941                 dip->errors = 1;
5942
5943                 /*
5944                  * before atomic variable goto zero, we must make sure
5945                  * dip->errors is perceived to be set.
5946                  */
5947                 smp_mb__before_atomic_dec();
5948         }
5949
5950         /* if there are more bios still pending for this dio, just exit */
5951         if (!atomic_dec_and_test(&dip->pending_bios))
5952                 goto out;
5953
5954         if (dip->errors)
5955                 bio_io_error(dip->orig_bio);
5956         else {
5957                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5958                 bio_endio(dip->orig_bio, 0);
5959         }
5960 out:
5961         bio_put(bio);
5962 }
5963
5964 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5965                                        u64 first_sector, gfp_t gfp_flags)
5966 {
5967         int nr_vecs = bio_get_nr_vecs(bdev);
5968         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5969 }
5970
5971 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5972                                          int rw, u64 file_offset, int skip_sum,
5973                                          u32 *csums, int async_submit)
5974 {
5975         int write = rw & REQ_WRITE;
5976         struct btrfs_root *root = BTRFS_I(inode)->root;
5977         int ret;
5978
5979         bio_get(bio);
5980         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5981         if (ret)
5982                 goto err;
5983
5984         if (skip_sum)
5985                 goto map;
5986
5987         if (write && async_submit) {
5988                 ret = btrfs_wq_submit_bio(root->fs_info,
5989                                    inode, rw, bio, 0, 0,
5990                                    file_offset,
5991                                    __btrfs_submit_bio_start_direct_io,
5992                                    __btrfs_submit_bio_done);
5993                 goto err;
5994         } else if (write) {
5995                 /*
5996                  * If we aren't doing async submit, calculate the csum of the
5997                  * bio now.
5998                  */
5999                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6000                 if (ret)
6001                         goto err;
6002         } else if (!skip_sum) {
6003                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6004                                           file_offset, csums);
6005                 if (ret)
6006                         goto err;
6007         }
6008
6009 map:
6010         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6011 err:
6012         bio_put(bio);
6013         return ret;
6014 }
6015
6016 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6017                                     int skip_sum)
6018 {
6019         struct inode *inode = dip->inode;
6020         struct btrfs_root *root = BTRFS_I(inode)->root;
6021         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6022         struct bio *bio;
6023         struct bio *orig_bio = dip->orig_bio;
6024         struct bio_vec *bvec = orig_bio->bi_io_vec;
6025         u64 start_sector = orig_bio->bi_sector;
6026         u64 file_offset = dip->logical_offset;
6027         u64 submit_len = 0;
6028         u64 map_length;
6029         int nr_pages = 0;
6030         u32 *csums = dip->csums;
6031         int ret = 0;
6032         int async_submit = 0;
6033         int write = rw & REQ_WRITE;
6034
6035         map_length = orig_bio->bi_size;
6036         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6037                               &map_length, NULL, 0);
6038         if (ret) {
6039                 bio_put(bio);
6040                 return -EIO;
6041         }
6042
6043         if (map_length >= orig_bio->bi_size) {
6044                 bio = orig_bio;
6045                 goto submit;
6046         }
6047
6048         async_submit = 1;
6049         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6050         if (!bio)
6051                 return -ENOMEM;
6052         bio->bi_private = dip;
6053         bio->bi_end_io = btrfs_end_dio_bio;
6054         atomic_inc(&dip->pending_bios);
6055
6056         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6057                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6058                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6059                                  bvec->bv_offset) < bvec->bv_len)) {
6060                         /*
6061                          * inc the count before we submit the bio so
6062                          * we know the end IO handler won't happen before
6063                          * we inc the count. Otherwise, the dip might get freed
6064                          * before we're done setting it up
6065                          */
6066                         atomic_inc(&dip->pending_bios);
6067                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6068                                                      file_offset, skip_sum,
6069                                                      csums, async_submit);
6070                         if (ret) {
6071                                 bio_put(bio);
6072                                 atomic_dec(&dip->pending_bios);
6073                                 goto out_err;
6074                         }
6075
6076                         /* Write's use the ordered csums */
6077                         if (!write && !skip_sum)
6078                                 csums = csums + nr_pages;
6079                         start_sector += submit_len >> 9;
6080                         file_offset += submit_len;
6081
6082                         submit_len = 0;
6083                         nr_pages = 0;
6084
6085                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6086                                                   start_sector, GFP_NOFS);
6087                         if (!bio)
6088                                 goto out_err;
6089                         bio->bi_private = dip;
6090                         bio->bi_end_io = btrfs_end_dio_bio;
6091
6092                         map_length = orig_bio->bi_size;
6093                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6094                                               &map_length, NULL, 0);
6095                         if (ret) {
6096                                 bio_put(bio);
6097                                 goto out_err;
6098                         }
6099                 } else {
6100                         submit_len += bvec->bv_len;
6101                         nr_pages ++;
6102                         bvec++;
6103                 }
6104         }
6105
6106 submit:
6107         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6108                                      csums, async_submit);
6109         if (!ret)
6110                 return 0;
6111
6112         bio_put(bio);
6113 out_err:
6114         dip->errors = 1;
6115         /*
6116          * before atomic variable goto zero, we must
6117          * make sure dip->errors is perceived to be set.
6118          */
6119         smp_mb__before_atomic_dec();
6120         if (atomic_dec_and_test(&dip->pending_bios))
6121                 bio_io_error(dip->orig_bio);
6122
6123         /* bio_end_io() will handle error, so we needn't return it */
6124         return 0;
6125 }
6126
6127 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6128                                 loff_t file_offset)
6129 {
6130         struct btrfs_root *root = BTRFS_I(inode)->root;
6131         struct btrfs_dio_private *dip;
6132         struct bio_vec *bvec = bio->bi_io_vec;
6133         int skip_sum;
6134         int write = rw & REQ_WRITE;
6135         int ret = 0;
6136
6137         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6138
6139         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6140         if (!dip) {
6141                 ret = -ENOMEM;
6142                 goto free_ordered;
6143         }
6144         dip->csums = NULL;
6145
6146         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6147         if (!write && !skip_sum) {
6148                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6149                 if (!dip->csums) {
6150                         kfree(dip);
6151                         ret = -ENOMEM;
6152                         goto free_ordered;
6153                 }
6154         }
6155
6156         dip->private = bio->bi_private;
6157         dip->inode = inode;
6158         dip->logical_offset = file_offset;
6159
6160         dip->bytes = 0;
6161         do {
6162                 dip->bytes += bvec->bv_len;
6163                 bvec++;
6164         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6165
6166         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6167         bio->bi_private = dip;
6168         dip->errors = 0;
6169         dip->orig_bio = bio;
6170         atomic_set(&dip->pending_bios, 0);
6171
6172         if (write)
6173                 bio->bi_end_io = btrfs_endio_direct_write;
6174         else
6175                 bio->bi_end_io = btrfs_endio_direct_read;
6176
6177         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6178         if (!ret)
6179                 return;
6180 free_ordered:
6181         /*
6182          * If this is a write, we need to clean up the reserved space and kill
6183          * the ordered extent.
6184          */
6185         if (write) {
6186                 struct btrfs_ordered_extent *ordered;
6187                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6188                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6189                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6190                         btrfs_free_reserved_extent(root, ordered->start,
6191                                                    ordered->disk_len);
6192                 btrfs_put_ordered_extent(ordered);
6193                 btrfs_put_ordered_extent(ordered);
6194         }
6195         bio_endio(bio, ret);
6196 }
6197
6198 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6199                         const struct iovec *iov, loff_t offset,
6200                         unsigned long nr_segs)
6201 {
6202         int seg;
6203         int i;
6204         size_t size;
6205         unsigned long addr;
6206         unsigned blocksize_mask = root->sectorsize - 1;
6207         ssize_t retval = -EINVAL;
6208         loff_t end = offset;
6209
6210         if (offset & blocksize_mask)
6211                 goto out;
6212
6213         /* Check the memory alignment.  Blocks cannot straddle pages */
6214         for (seg = 0; seg < nr_segs; seg++) {
6215                 addr = (unsigned long)iov[seg].iov_base;
6216                 size = iov[seg].iov_len;
6217                 end += size;
6218                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6219                         goto out;
6220
6221                 /* If this is a write we don't need to check anymore */
6222                 if (rw & WRITE)
6223                         continue;
6224
6225                 /*
6226                  * Check to make sure we don't have duplicate iov_base's in this
6227                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6228                  * when reading back.
6229                  */
6230                 for (i = seg + 1; i < nr_segs; i++) {
6231                         if (iov[seg].iov_base == iov[i].iov_base)
6232                                 goto out;
6233                 }
6234         }
6235         retval = 0;
6236 out:
6237         return retval;
6238 }
6239 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6240                         const struct iovec *iov, loff_t offset,
6241                         unsigned long nr_segs)
6242 {
6243         struct file *file = iocb->ki_filp;
6244         struct inode *inode = file->f_mapping->host;
6245         struct btrfs_ordered_extent *ordered;
6246         struct extent_state *cached_state = NULL;
6247         u64 lockstart, lockend;
6248         ssize_t ret;
6249         int writing = rw & WRITE;
6250         int write_bits = 0;
6251         size_t count = iov_length(iov, nr_segs);
6252
6253         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6254                             offset, nr_segs)) {
6255                 return 0;
6256         }
6257
6258         lockstart = offset;
6259         lockend = offset + count - 1;
6260
6261         if (writing) {
6262                 ret = btrfs_delalloc_reserve_space(inode, count);
6263                 if (ret)
6264                         goto out;
6265         }
6266
6267         while (1) {
6268                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6269                                  0, &cached_state, GFP_NOFS);
6270                 /*
6271                  * We're concerned with the entire range that we're going to be
6272                  * doing DIO to, so we need to make sure theres no ordered
6273                  * extents in this range.
6274                  */
6275                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6276                                                      lockend - lockstart + 1);
6277                 if (!ordered)
6278                         break;
6279                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6280                                      &cached_state, GFP_NOFS);
6281                 btrfs_start_ordered_extent(inode, ordered, 1);
6282                 btrfs_put_ordered_extent(ordered);
6283                 cond_resched();
6284         }
6285
6286         /*
6287          * we don't use btrfs_set_extent_delalloc because we don't want
6288          * the dirty or uptodate bits
6289          */
6290         if (writing) {
6291                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6292                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6293                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6294                                      GFP_NOFS);
6295                 if (ret) {
6296                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6297                                          lockend, EXTENT_LOCKED | write_bits,
6298                                          1, 0, &cached_state, GFP_NOFS);
6299                         goto out;
6300                 }
6301         }
6302
6303         free_extent_state(cached_state);
6304         cached_state = NULL;
6305
6306         ret = __blockdev_direct_IO(rw, iocb, inode,
6307                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6308                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6309                    btrfs_submit_direct, 0);
6310
6311         if (ret < 0 && ret != -EIOCBQUEUED) {
6312                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6313                               offset + iov_length(iov, nr_segs) - 1,
6314                               EXTENT_LOCKED | write_bits, 1, 0,
6315                               &cached_state, GFP_NOFS);
6316         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6317                 /*
6318                  * We're falling back to buffered, unlock the section we didn't
6319                  * do IO on.
6320                  */
6321                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6322                               offset + iov_length(iov, nr_segs) - 1,
6323                               EXTENT_LOCKED | write_bits, 1, 0,
6324                               &cached_state, GFP_NOFS);
6325         }
6326 out:
6327         free_extent_state(cached_state);
6328         return ret;
6329 }
6330
6331 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6332                 __u64 start, __u64 len)
6333 {
6334         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6335 }
6336
6337 int btrfs_readpage(struct file *file, struct page *page)
6338 {
6339         struct extent_io_tree *tree;
6340         tree = &BTRFS_I(page->mapping->host)->io_tree;
6341         return extent_read_full_page(tree, page, btrfs_get_extent);
6342 }
6343
6344 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6345 {
6346         struct extent_io_tree *tree;
6347
6348
6349         if (current->flags & PF_MEMALLOC) {
6350                 redirty_page_for_writepage(wbc, page);
6351                 unlock_page(page);
6352                 return 0;
6353         }
6354         tree = &BTRFS_I(page->mapping->host)->io_tree;
6355         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6356 }
6357
6358 int btrfs_writepages(struct address_space *mapping,
6359                      struct writeback_control *wbc)
6360 {
6361         struct extent_io_tree *tree;
6362
6363         tree = &BTRFS_I(mapping->host)->io_tree;
6364         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6365 }
6366
6367 static int
6368 btrfs_readpages(struct file *file, struct address_space *mapping,
6369                 struct list_head *pages, unsigned nr_pages)
6370 {
6371         struct extent_io_tree *tree;
6372         tree = &BTRFS_I(mapping->host)->io_tree;
6373         return extent_readpages(tree, mapping, pages, nr_pages,
6374                                 btrfs_get_extent);
6375 }
6376 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6377 {
6378         struct extent_io_tree *tree;
6379         struct extent_map_tree *map;
6380         int ret;
6381
6382         tree = &BTRFS_I(page->mapping->host)->io_tree;
6383         map = &BTRFS_I(page->mapping->host)->extent_tree;
6384         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6385         if (ret == 1) {
6386                 ClearPagePrivate(page);
6387                 set_page_private(page, 0);
6388                 page_cache_release(page);
6389         }
6390         return ret;
6391 }
6392
6393 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6394 {
6395         if (PageWriteback(page) || PageDirty(page))
6396                 return 0;
6397         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6398 }
6399
6400 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6401 {
6402         struct extent_io_tree *tree;
6403         struct btrfs_ordered_extent *ordered;
6404         struct extent_state *cached_state = NULL;
6405         u64 page_start = page_offset(page);
6406         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6407
6408
6409         /*
6410          * we have the page locked, so new writeback can't start,
6411          * and the dirty bit won't be cleared while we are here.
6412          *
6413          * Wait for IO on this page so that we can safely clear
6414          * the PagePrivate2 bit and do ordered accounting
6415          */
6416         wait_on_page_writeback(page);
6417
6418         tree = &BTRFS_I(page->mapping->host)->io_tree;
6419         if (offset) {
6420                 btrfs_releasepage(page, GFP_NOFS);
6421                 return;
6422         }
6423         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6424                          GFP_NOFS);
6425         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6426                                            page_offset(page));
6427         if (ordered) {
6428                 /*
6429                  * IO on this page will never be started, so we need
6430                  * to account for any ordered extents now
6431                  */
6432                 clear_extent_bit(tree, page_start, page_end,
6433                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6434                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6435                                  &cached_state, GFP_NOFS);
6436                 /*
6437                  * whoever cleared the private bit is responsible
6438                  * for the finish_ordered_io
6439                  */
6440                 if (TestClearPagePrivate2(page)) {
6441                         btrfs_finish_ordered_io(page->mapping->host,
6442                                                 page_start, page_end);
6443                 }
6444                 btrfs_put_ordered_extent(ordered);
6445                 cached_state = NULL;
6446                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6447                                  GFP_NOFS);
6448         }
6449         clear_extent_bit(tree, page_start, page_end,
6450                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6451                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6452         __btrfs_releasepage(page, GFP_NOFS);
6453
6454         ClearPageChecked(page);
6455         if (PagePrivate(page)) {
6456                 ClearPagePrivate(page);
6457                 set_page_private(page, 0);
6458                 page_cache_release(page);
6459         }
6460 }
6461
6462 /*
6463  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6464  * called from a page fault handler when a page is first dirtied. Hence we must
6465  * be careful to check for EOF conditions here. We set the page up correctly
6466  * for a written page which means we get ENOSPC checking when writing into
6467  * holes and correct delalloc and unwritten extent mapping on filesystems that
6468  * support these features.
6469  *
6470  * We are not allowed to take the i_mutex here so we have to play games to
6471  * protect against truncate races as the page could now be beyond EOF.  Because
6472  * vmtruncate() writes the inode size before removing pages, once we have the
6473  * page lock we can determine safely if the page is beyond EOF. If it is not
6474  * beyond EOF, then the page is guaranteed safe against truncation until we
6475  * unlock the page.
6476  */
6477 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6478 {
6479         struct page *page = vmf->page;
6480         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6481         struct btrfs_root *root = BTRFS_I(inode)->root;
6482         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6483         struct btrfs_ordered_extent *ordered;
6484         struct extent_state *cached_state = NULL;
6485         char *kaddr;
6486         unsigned long zero_start;
6487         loff_t size;
6488         int ret;
6489         u64 page_start;
6490         u64 page_end;
6491
6492         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6493         if (ret) {
6494                 if (ret == -ENOMEM)
6495                         ret = VM_FAULT_OOM;
6496                 else /* -ENOSPC, -EIO, etc */
6497                         ret = VM_FAULT_SIGBUS;
6498                 goto out;
6499         }
6500
6501         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6502 again:
6503         lock_page(page);
6504         size = i_size_read(inode);
6505         page_start = page_offset(page);
6506         page_end = page_start + PAGE_CACHE_SIZE - 1;
6507
6508         if ((page->mapping != inode->i_mapping) ||
6509             (page_start >= size)) {
6510                 /* page got truncated out from underneath us */
6511                 goto out_unlock;
6512         }
6513         wait_on_page_writeback(page);
6514
6515         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6516                          GFP_NOFS);
6517         set_page_extent_mapped(page);
6518
6519         /*
6520          * we can't set the delalloc bits if there are pending ordered
6521          * extents.  Drop our locks and wait for them to finish
6522          */
6523         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6524         if (ordered) {
6525                 unlock_extent_cached(io_tree, page_start, page_end,
6526                                      &cached_state, GFP_NOFS);
6527                 unlock_page(page);
6528                 btrfs_start_ordered_extent(inode, ordered, 1);
6529                 btrfs_put_ordered_extent(ordered);
6530                 goto again;
6531         }
6532
6533         /*
6534          * XXX - page_mkwrite gets called every time the page is dirtied, even
6535          * if it was already dirty, so for space accounting reasons we need to
6536          * clear any delalloc bits for the range we are fixing to save.  There
6537          * is probably a better way to do this, but for now keep consistent with
6538          * prepare_pages in the normal write path.
6539          */
6540         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6541                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6542                           0, 0, &cached_state, GFP_NOFS);
6543
6544         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6545                                         &cached_state);
6546         if (ret) {
6547                 unlock_extent_cached(io_tree, page_start, page_end,
6548                                      &cached_state, GFP_NOFS);
6549                 ret = VM_FAULT_SIGBUS;
6550                 goto out_unlock;
6551         }
6552         ret = 0;
6553
6554         /* page is wholly or partially inside EOF */
6555         if (page_start + PAGE_CACHE_SIZE > size)
6556                 zero_start = size & ~PAGE_CACHE_MASK;
6557         else
6558                 zero_start = PAGE_CACHE_SIZE;
6559
6560         if (zero_start != PAGE_CACHE_SIZE) {
6561                 kaddr = kmap(page);
6562                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6563                 flush_dcache_page(page);
6564                 kunmap(page);
6565         }
6566         ClearPageChecked(page);
6567         set_page_dirty(page);
6568         SetPageUptodate(page);
6569
6570         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6571         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6572
6573         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6574
6575 out_unlock:
6576         if (!ret)
6577                 return VM_FAULT_LOCKED;
6578         unlock_page(page);
6579         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6580 out:
6581         return ret;
6582 }
6583
6584 static int btrfs_truncate(struct inode *inode)
6585 {
6586         struct btrfs_root *root = BTRFS_I(inode)->root;
6587         int ret;
6588         int err = 0;
6589         struct btrfs_trans_handle *trans;
6590         unsigned long nr;
6591         u64 mask = root->sectorsize - 1;
6592
6593         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6594         if (ret)
6595                 return ret;
6596
6597         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6598         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6599
6600         trans = btrfs_start_transaction(root, 5);
6601         if (IS_ERR(trans))
6602                 return PTR_ERR(trans);
6603
6604         btrfs_set_trans_block_group(trans, inode);
6605
6606         ret = btrfs_orphan_add(trans, inode);
6607         if (ret) {
6608                 btrfs_end_transaction(trans, root);
6609                 return ret;
6610         }
6611
6612         nr = trans->blocks_used;
6613         btrfs_end_transaction(trans, root);
6614         btrfs_btree_balance_dirty(root, nr);
6615
6616         /* Now start a transaction for the truncate */
6617         trans = btrfs_start_transaction(root, 0);
6618         if (IS_ERR(trans))
6619                 return PTR_ERR(trans);
6620         btrfs_set_trans_block_group(trans, inode);
6621         trans->block_rsv = root->orphan_block_rsv;
6622
6623         /*
6624          * setattr is responsible for setting the ordered_data_close flag,
6625          * but that is only tested during the last file release.  That
6626          * could happen well after the next commit, leaving a great big
6627          * window where new writes may get lost if someone chooses to write
6628          * to this file after truncating to zero
6629          *
6630          * The inode doesn't have any dirty data here, and so if we commit
6631          * this is a noop.  If someone immediately starts writing to the inode
6632          * it is very likely we'll catch some of their writes in this
6633          * transaction, and the commit will find this file on the ordered
6634          * data list with good things to send down.
6635          *
6636          * This is a best effort solution, there is still a window where
6637          * using truncate to replace the contents of the file will
6638          * end up with a zero length file after a crash.
6639          */
6640         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6641                 btrfs_add_ordered_operation(trans, root, inode);
6642
6643         while (1) {
6644                 if (!trans) {
6645                         trans = btrfs_start_transaction(root, 0);
6646                         if (IS_ERR(trans))
6647                                 return PTR_ERR(trans);
6648                         btrfs_set_trans_block_group(trans, inode);
6649                         trans->block_rsv = root->orphan_block_rsv;
6650                 }
6651
6652                 ret = btrfs_block_rsv_check(trans, root,
6653                                             root->orphan_block_rsv, 0, 5);
6654                 if (ret == -EAGAIN) {
6655                         ret = btrfs_commit_transaction(trans, root);
6656                         if (ret)
6657                                 return ret;
6658                         trans = NULL;
6659                         continue;
6660                 } else if (ret) {
6661                         err = ret;
6662                         break;
6663                 }
6664
6665                 ret = btrfs_truncate_inode_items(trans, root, inode,
6666                                                  inode->i_size,
6667                                                  BTRFS_EXTENT_DATA_KEY);
6668                 if (ret != -EAGAIN) {
6669                         err = ret;
6670                         break;
6671                 }
6672
6673                 ret = btrfs_update_inode(trans, root, inode);
6674                 if (ret) {
6675                         err = ret;
6676                         break;
6677                 }
6678
6679                 nr = trans->blocks_used;
6680                 btrfs_end_transaction(trans, root);
6681                 trans = NULL;
6682                 btrfs_btree_balance_dirty(root, nr);
6683         }
6684
6685         if (ret == 0 && inode->i_nlink > 0) {
6686                 ret = btrfs_orphan_del(trans, inode);
6687                 if (ret)
6688                         err = ret;
6689         } else if (ret && inode->i_nlink > 0) {
6690                 /*
6691                  * Failed to do the truncate, remove us from the in memory
6692                  * orphan list.
6693                  */
6694                 ret = btrfs_orphan_del(NULL, inode);
6695         }
6696
6697         ret = btrfs_update_inode(trans, root, inode);
6698         if (ret && !err)
6699                 err = ret;
6700
6701         nr = trans->blocks_used;
6702         ret = btrfs_end_transaction_throttle(trans, root);
6703         if (ret && !err)
6704                 err = ret;
6705         btrfs_btree_balance_dirty(root, nr);
6706
6707         return err;
6708 }
6709
6710 /*
6711  * create a new subvolume directory/inode (helper for the ioctl).
6712  */
6713 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6714                              struct btrfs_root *new_root,
6715                              u64 new_dirid, u64 alloc_hint)
6716 {
6717         struct inode *inode;
6718         int err;
6719         u64 index = 0;
6720
6721         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6722                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6723         if (IS_ERR(inode))
6724                 return PTR_ERR(inode);
6725         inode->i_op = &btrfs_dir_inode_operations;
6726         inode->i_fop = &btrfs_dir_file_operations;
6727
6728         inode->i_nlink = 1;
6729         btrfs_i_size_write(inode, 0);
6730
6731         err = btrfs_update_inode(trans, new_root, inode);
6732         BUG_ON(err);
6733
6734         iput(inode);
6735         return 0;
6736 }
6737
6738 /* helper function for file defrag and space balancing.  This
6739  * forces readahead on a given range of bytes in an inode
6740  */
6741 unsigned long btrfs_force_ra(struct address_space *mapping,
6742                               struct file_ra_state *ra, struct file *file,
6743                               pgoff_t offset, pgoff_t last_index)
6744 {
6745         pgoff_t req_size = last_index - offset + 1;
6746
6747         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6748         return offset + req_size;
6749 }
6750
6751 struct inode *btrfs_alloc_inode(struct super_block *sb)
6752 {
6753         struct btrfs_inode *ei;
6754         struct inode *inode;
6755
6756         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6757         if (!ei)
6758                 return NULL;
6759
6760         ei->root = NULL;
6761         ei->space_info = NULL;
6762         ei->generation = 0;
6763         ei->sequence = 0;
6764         ei->last_trans = 0;
6765         ei->last_sub_trans = 0;
6766         ei->logged_trans = 0;
6767         ei->delalloc_bytes = 0;
6768         ei->reserved_bytes = 0;
6769         ei->disk_i_size = 0;
6770         ei->flags = 0;
6771         ei->index_cnt = (u64)-1;
6772         ei->last_unlink_trans = 0;
6773
6774         atomic_set(&ei->outstanding_extents, 0);
6775         atomic_set(&ei->reserved_extents, 0);
6776
6777         ei->ordered_data_close = 0;
6778         ei->orphan_meta_reserved = 0;
6779         ei->dummy_inode = 0;
6780         ei->force_compress = BTRFS_COMPRESS_NONE;
6781
6782         inode = &ei->vfs_inode;
6783         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6784         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6785         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6786         mutex_init(&ei->log_mutex);
6787         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6788         INIT_LIST_HEAD(&ei->i_orphan);
6789         INIT_LIST_HEAD(&ei->delalloc_inodes);
6790         INIT_LIST_HEAD(&ei->ordered_operations);
6791         RB_CLEAR_NODE(&ei->rb_node);
6792
6793         return inode;
6794 }
6795
6796 static void btrfs_i_callback(struct rcu_head *head)
6797 {
6798         struct inode *inode = container_of(head, struct inode, i_rcu);
6799         INIT_LIST_HEAD(&inode->i_dentry);
6800         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6801 }
6802
6803 void btrfs_destroy_inode(struct inode *inode)
6804 {
6805         struct btrfs_ordered_extent *ordered;
6806         struct btrfs_root *root = BTRFS_I(inode)->root;
6807
6808         WARN_ON(!list_empty(&inode->i_dentry));
6809         WARN_ON(inode->i_data.nrpages);
6810         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6811         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6812
6813         /*
6814          * This can happen where we create an inode, but somebody else also
6815          * created the same inode and we need to destroy the one we already
6816          * created.
6817          */
6818         if (!root)
6819                 goto free;
6820
6821         /*
6822          * Make sure we're properly removed from the ordered operation
6823          * lists.
6824          */
6825         smp_mb();
6826         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6827                 spin_lock(&root->fs_info->ordered_extent_lock);
6828                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6829                 spin_unlock(&root->fs_info->ordered_extent_lock);
6830         }
6831
6832         if (root == root->fs_info->tree_root) {
6833                 struct btrfs_block_group_cache *block_group;
6834
6835                 block_group = btrfs_lookup_block_group(root->fs_info,
6836                                                 BTRFS_I(inode)->block_group);
6837                 if (block_group && block_group->inode == inode) {
6838                         spin_lock(&block_group->lock);
6839                         block_group->inode = NULL;
6840                         spin_unlock(&block_group->lock);
6841                         btrfs_put_block_group(block_group);
6842                 } else if (block_group) {
6843                         btrfs_put_block_group(block_group);
6844                 }
6845         }
6846
6847         spin_lock(&root->orphan_lock);
6848         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6849                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6850                        inode->i_ino);
6851                 list_del_init(&BTRFS_I(inode)->i_orphan);
6852         }
6853         spin_unlock(&root->orphan_lock);
6854
6855         while (1) {
6856                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6857                 if (!ordered)
6858                         break;
6859                 else {
6860                         printk(KERN_ERR "btrfs found ordered "
6861                                "extent %llu %llu on inode cleanup\n",
6862                                (unsigned long long)ordered->file_offset,
6863                                (unsigned long long)ordered->len);
6864                         btrfs_remove_ordered_extent(inode, ordered);
6865                         btrfs_put_ordered_extent(ordered);
6866                         btrfs_put_ordered_extent(ordered);
6867                 }
6868         }
6869         inode_tree_del(inode);
6870         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6871 free:
6872         call_rcu(&inode->i_rcu, btrfs_i_callback);
6873 }
6874
6875 int btrfs_drop_inode(struct inode *inode)
6876 {
6877         struct btrfs_root *root = BTRFS_I(inode)->root;
6878
6879         if (btrfs_root_refs(&root->root_item) == 0 &&
6880             root != root->fs_info->tree_root)
6881                 return 1;
6882         else
6883                 return generic_drop_inode(inode);
6884 }
6885
6886 static void init_once(void *foo)
6887 {
6888         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6889
6890         inode_init_once(&ei->vfs_inode);
6891 }
6892
6893 void btrfs_destroy_cachep(void)
6894 {
6895         if (btrfs_inode_cachep)
6896                 kmem_cache_destroy(btrfs_inode_cachep);
6897         if (btrfs_trans_handle_cachep)
6898                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6899         if (btrfs_transaction_cachep)
6900                 kmem_cache_destroy(btrfs_transaction_cachep);
6901         if (btrfs_path_cachep)
6902                 kmem_cache_destroy(btrfs_path_cachep);
6903         if (btrfs_free_space_cachep)
6904                 kmem_cache_destroy(btrfs_free_space_cachep);
6905 }
6906
6907 int btrfs_init_cachep(void)
6908 {
6909         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6910                         sizeof(struct btrfs_inode), 0,
6911                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6912         if (!btrfs_inode_cachep)
6913                 goto fail;
6914
6915         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6916                         sizeof(struct btrfs_trans_handle), 0,
6917                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6918         if (!btrfs_trans_handle_cachep)
6919                 goto fail;
6920
6921         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6922                         sizeof(struct btrfs_transaction), 0,
6923                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6924         if (!btrfs_transaction_cachep)
6925                 goto fail;
6926
6927         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6928                         sizeof(struct btrfs_path), 0,
6929                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6930         if (!btrfs_path_cachep)
6931                 goto fail;
6932
6933         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6934                         sizeof(struct btrfs_free_space), 0,
6935                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6936         if (!btrfs_free_space_cachep)
6937                 goto fail;
6938
6939         return 0;
6940 fail:
6941         btrfs_destroy_cachep();
6942         return -ENOMEM;
6943 }
6944
6945 static int btrfs_getattr(struct vfsmount *mnt,
6946                          struct dentry *dentry, struct kstat *stat)
6947 {
6948         struct inode *inode = dentry->d_inode;
6949         generic_fillattr(inode, stat);
6950         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6951         stat->blksize = PAGE_CACHE_SIZE;
6952         stat->blocks = (inode_get_bytes(inode) +
6953                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6954         return 0;
6955 }
6956
6957 /*
6958  * If a file is moved, it will inherit the cow and compression flags of the new
6959  * directory.
6960  */
6961 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6962 {
6963         struct btrfs_inode *b_dir = BTRFS_I(dir);
6964         struct btrfs_inode *b_inode = BTRFS_I(inode);
6965
6966         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6967                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6968         else
6969                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6970
6971         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6972                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6973         else
6974                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6975 }
6976
6977 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6978                            struct inode *new_dir, struct dentry *new_dentry)
6979 {
6980         struct btrfs_trans_handle *trans;
6981         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6982         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6983         struct inode *new_inode = new_dentry->d_inode;
6984         struct inode *old_inode = old_dentry->d_inode;
6985         struct timespec ctime = CURRENT_TIME;
6986         u64 index = 0;
6987         u64 root_objectid;
6988         int ret;
6989
6990         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6991                 return -EPERM;
6992
6993         /* we only allow rename subvolume link between subvolumes */
6994         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6995                 return -EXDEV;
6996
6997         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6998             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6999                 return -ENOTEMPTY;
7000
7001         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7002             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7003                 return -ENOTEMPTY;
7004         /*
7005          * we're using rename to replace one file with another.
7006          * and the replacement file is large.  Start IO on it now so
7007          * we don't add too much work to the end of the transaction
7008          */
7009         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7010             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7011                 filemap_flush(old_inode->i_mapping);
7012
7013         /* close the racy window with snapshot create/destroy ioctl */
7014         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7015                 down_read(&root->fs_info->subvol_sem);
7016         /*
7017          * We want to reserve the absolute worst case amount of items.  So if
7018          * both inodes are subvols and we need to unlink them then that would
7019          * require 4 item modifications, but if they are both normal inodes it
7020          * would require 5 item modifications, so we'll assume their normal
7021          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7022          * should cover the worst case number of items we'll modify.
7023          */
7024         trans = btrfs_start_transaction(root, 20);
7025         if (IS_ERR(trans)) {
7026                 ret = PTR_ERR(trans);
7027                 goto out_notrans;
7028         }
7029
7030         btrfs_set_trans_block_group(trans, new_dir);
7031
7032         if (dest != root)
7033                 btrfs_record_root_in_trans(trans, dest);
7034
7035         ret = btrfs_set_inode_index(new_dir, &index);
7036         if (ret)
7037                 goto out_fail;
7038
7039         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7040                 /* force full log commit if subvolume involved. */
7041                 root->fs_info->last_trans_log_full_commit = trans->transid;
7042         } else {
7043                 ret = btrfs_insert_inode_ref(trans, dest,
7044                                              new_dentry->d_name.name,
7045                                              new_dentry->d_name.len,
7046                                              old_inode->i_ino,
7047                                              new_dir->i_ino, index);
7048                 if (ret)
7049                         goto out_fail;
7050                 /*
7051                  * this is an ugly little race, but the rename is required
7052                  * to make sure that if we crash, the inode is either at the
7053                  * old name or the new one.  pinning the log transaction lets
7054                  * us make sure we don't allow a log commit to come in after
7055                  * we unlink the name but before we add the new name back in.
7056                  */
7057                 btrfs_pin_log_trans(root);
7058         }
7059         /*
7060          * make sure the inode gets flushed if it is replacing
7061          * something.
7062          */
7063         if (new_inode && new_inode->i_size &&
7064             old_inode && S_ISREG(old_inode->i_mode)) {
7065                 btrfs_add_ordered_operation(trans, root, old_inode);
7066         }
7067
7068         old_dir->i_ctime = old_dir->i_mtime = ctime;
7069         new_dir->i_ctime = new_dir->i_mtime = ctime;
7070         old_inode->i_ctime = ctime;
7071
7072         if (old_dentry->d_parent != new_dentry->d_parent)
7073                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7074
7075         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7076                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7077                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7078                                         old_dentry->d_name.name,
7079                                         old_dentry->d_name.len);
7080         } else {
7081                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7082                                         old_dentry->d_inode,
7083                                         old_dentry->d_name.name,
7084                                         old_dentry->d_name.len);
7085                 if (!ret)
7086                         ret = btrfs_update_inode(trans, root, old_inode);
7087         }
7088         BUG_ON(ret);
7089
7090         if (new_inode) {
7091                 new_inode->i_ctime = CURRENT_TIME;
7092                 if (unlikely(new_inode->i_ino ==
7093                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7094                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7095                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7096                                                 root_objectid,
7097                                                 new_dentry->d_name.name,
7098                                                 new_dentry->d_name.len);
7099                         BUG_ON(new_inode->i_nlink == 0);
7100                 } else {
7101                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7102                                                  new_dentry->d_inode,
7103                                                  new_dentry->d_name.name,
7104                                                  new_dentry->d_name.len);
7105                 }
7106                 BUG_ON(ret);
7107                 if (new_inode->i_nlink == 0) {
7108                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7109                         BUG_ON(ret);
7110                 }
7111         }
7112
7113         fixup_inode_flags(new_dir, old_inode);
7114
7115         ret = btrfs_add_link(trans, new_dir, old_inode,
7116                              new_dentry->d_name.name,
7117                              new_dentry->d_name.len, 0, index);
7118         BUG_ON(ret);
7119
7120         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7121                 struct dentry *parent = dget_parent(new_dentry);
7122                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7123                 dput(parent);
7124                 btrfs_end_log_trans(root);
7125         }
7126 out_fail:
7127         btrfs_end_transaction_throttle(trans, root);
7128 out_notrans:
7129         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7130                 up_read(&root->fs_info->subvol_sem);
7131
7132         return ret;
7133 }
7134
7135 /*
7136  * some fairly slow code that needs optimization. This walks the list
7137  * of all the inodes with pending delalloc and forces them to disk.
7138  */
7139 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7140 {
7141         struct list_head *head = &root->fs_info->delalloc_inodes;
7142         struct btrfs_inode *binode;
7143         struct inode *inode;
7144
7145         if (root->fs_info->sb->s_flags & MS_RDONLY)
7146                 return -EROFS;
7147
7148         spin_lock(&root->fs_info->delalloc_lock);
7149         while (!list_empty(head)) {
7150                 binode = list_entry(head->next, struct btrfs_inode,
7151                                     delalloc_inodes);
7152                 inode = igrab(&binode->vfs_inode);
7153                 if (!inode)
7154                         list_del_init(&binode->delalloc_inodes);
7155                 spin_unlock(&root->fs_info->delalloc_lock);
7156                 if (inode) {
7157                         filemap_flush(inode->i_mapping);
7158                         if (delay_iput)
7159                                 btrfs_add_delayed_iput(inode);
7160                         else
7161                                 iput(inode);
7162                 }
7163                 cond_resched();
7164                 spin_lock(&root->fs_info->delalloc_lock);
7165         }
7166         spin_unlock(&root->fs_info->delalloc_lock);
7167
7168         /* the filemap_flush will queue IO into the worker threads, but
7169          * we have to make sure the IO is actually started and that
7170          * ordered extents get created before we return
7171          */
7172         atomic_inc(&root->fs_info->async_submit_draining);
7173         while (atomic_read(&root->fs_info->nr_async_submits) ||
7174               atomic_read(&root->fs_info->async_delalloc_pages)) {
7175                 wait_event(root->fs_info->async_submit_wait,
7176                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7177                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7178         }
7179         atomic_dec(&root->fs_info->async_submit_draining);
7180         return 0;
7181 }
7182
7183 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7184                                    int sync)
7185 {
7186         struct btrfs_inode *binode;
7187         struct inode *inode = NULL;
7188
7189         spin_lock(&root->fs_info->delalloc_lock);
7190         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7191                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7192                                     struct btrfs_inode, delalloc_inodes);
7193                 inode = igrab(&binode->vfs_inode);
7194                 if (inode) {
7195                         list_move_tail(&binode->delalloc_inodes,
7196                                        &root->fs_info->delalloc_inodes);
7197                         break;
7198                 }
7199
7200                 list_del_init(&binode->delalloc_inodes);
7201                 cond_resched_lock(&root->fs_info->delalloc_lock);
7202         }
7203         spin_unlock(&root->fs_info->delalloc_lock);
7204
7205         if (inode) {
7206                 if (sync) {
7207                         filemap_write_and_wait(inode->i_mapping);
7208                         /*
7209                          * We have to do this because compression doesn't
7210                          * actually set PG_writeback until it submits the pages
7211                          * for IO, which happens in an async thread, so we could
7212                          * race and not actually wait for any writeback pages
7213                          * because they've not been submitted yet.  Technically
7214                          * this could still be the case for the ordered stuff
7215                          * since the async thread may not have started to do its
7216                          * work yet.  If this becomes the case then we need to
7217                          * figure out a way to make sure that in writepage we
7218                          * wait for any async pages to be submitted before
7219                          * returning so that fdatawait does what its supposed to
7220                          * do.
7221                          */
7222                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7223                 } else {
7224                         filemap_flush(inode->i_mapping);
7225                 }
7226                 if (delay_iput)
7227                         btrfs_add_delayed_iput(inode);
7228                 else
7229                         iput(inode);
7230                 return 1;
7231         }
7232         return 0;
7233 }
7234
7235 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7236                          const char *symname)
7237 {
7238         struct btrfs_trans_handle *trans;
7239         struct btrfs_root *root = BTRFS_I(dir)->root;
7240         struct btrfs_path *path;
7241         struct btrfs_key key;
7242         struct inode *inode = NULL;
7243         int err;
7244         int drop_inode = 0;
7245         u64 objectid;
7246         u64 index = 0 ;
7247         int name_len;
7248         int datasize;
7249         unsigned long ptr;
7250         struct btrfs_file_extent_item *ei;
7251         struct extent_buffer *leaf;
7252         unsigned long nr = 0;
7253
7254         name_len = strlen(symname) + 1;
7255         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7256                 return -ENAMETOOLONG;
7257
7258         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7259         if (err)
7260                 return err;
7261         /*
7262          * 2 items for inode item and ref
7263          * 2 items for dir items
7264          * 1 item for xattr if selinux is on
7265          */
7266         trans = btrfs_start_transaction(root, 5);
7267         if (IS_ERR(trans))
7268                 return PTR_ERR(trans);
7269
7270         btrfs_set_trans_block_group(trans, dir);
7271
7272         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7273                                 dentry->d_name.len, dir->i_ino, objectid,
7274                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7275                                 &index);
7276         err = PTR_ERR(inode);
7277         if (IS_ERR(inode))
7278                 goto out_unlock;
7279
7280         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7281         if (err) {
7282                 drop_inode = 1;
7283                 goto out_unlock;
7284         }
7285
7286         btrfs_set_trans_block_group(trans, inode);
7287         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7288         if (err)
7289                 drop_inode = 1;
7290         else {
7291                 inode->i_mapping->a_ops = &btrfs_aops;
7292                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7293                 inode->i_fop = &btrfs_file_operations;
7294                 inode->i_op = &btrfs_file_inode_operations;
7295                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7296         }
7297         btrfs_update_inode_block_group(trans, inode);
7298         btrfs_update_inode_block_group(trans, dir);
7299         if (drop_inode)
7300                 goto out_unlock;
7301
7302         path = btrfs_alloc_path();
7303         BUG_ON(!path);
7304         key.objectid = inode->i_ino;
7305         key.offset = 0;
7306         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7307         datasize = btrfs_file_extent_calc_inline_size(name_len);
7308         err = btrfs_insert_empty_item(trans, root, path, &key,
7309                                       datasize);
7310         if (err) {
7311                 drop_inode = 1;
7312                 goto out_unlock;
7313         }
7314         leaf = path->nodes[0];
7315         ei = btrfs_item_ptr(leaf, path->slots[0],
7316                             struct btrfs_file_extent_item);
7317         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7318         btrfs_set_file_extent_type(leaf, ei,
7319                                    BTRFS_FILE_EXTENT_INLINE);
7320         btrfs_set_file_extent_encryption(leaf, ei, 0);
7321         btrfs_set_file_extent_compression(leaf, ei, 0);
7322         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7323         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7324
7325         ptr = btrfs_file_extent_inline_start(ei);
7326         write_extent_buffer(leaf, symname, ptr, name_len);
7327         btrfs_mark_buffer_dirty(leaf);
7328         btrfs_free_path(path);
7329
7330         inode->i_op = &btrfs_symlink_inode_operations;
7331         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7332         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7333         inode_set_bytes(inode, name_len);
7334         btrfs_i_size_write(inode, name_len - 1);
7335         err = btrfs_update_inode(trans, root, inode);
7336         if (err)
7337                 drop_inode = 1;
7338
7339 out_unlock:
7340         nr = trans->blocks_used;
7341         btrfs_end_transaction_throttle(trans, root);
7342         if (drop_inode) {
7343                 inode_dec_link_count(inode);
7344                 iput(inode);
7345         }
7346         btrfs_btree_balance_dirty(root, nr);
7347         return err;
7348 }
7349
7350 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7351                                        u64 start, u64 num_bytes, u64 min_size,
7352                                        loff_t actual_len, u64 *alloc_hint,
7353                                        struct btrfs_trans_handle *trans)
7354 {
7355         struct btrfs_root *root = BTRFS_I(inode)->root;
7356         struct btrfs_key ins;
7357         u64 cur_offset = start;
7358         u64 i_size;
7359         int ret = 0;
7360         bool own_trans = true;
7361
7362         if (trans)
7363                 own_trans = false;
7364         while (num_bytes > 0) {
7365                 if (own_trans) {
7366                         trans = btrfs_start_transaction(root, 3);
7367                         if (IS_ERR(trans)) {
7368                                 ret = PTR_ERR(trans);
7369                                 break;
7370                         }
7371                 }
7372
7373                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7374                                            0, *alloc_hint, (u64)-1, &ins, 1);
7375                 if (ret) {
7376                         if (own_trans)
7377                                 btrfs_end_transaction(trans, root);
7378                         break;
7379                 }
7380
7381                 ret = insert_reserved_file_extent(trans, inode,
7382                                                   cur_offset, ins.objectid,
7383                                                   ins.offset, ins.offset,
7384                                                   ins.offset, 0, 0, 0,
7385                                                   BTRFS_FILE_EXTENT_PREALLOC);
7386                 BUG_ON(ret);
7387                 btrfs_drop_extent_cache(inode, cur_offset,
7388                                         cur_offset + ins.offset -1, 0);
7389
7390                 num_bytes -= ins.offset;
7391                 cur_offset += ins.offset;
7392                 *alloc_hint = ins.objectid + ins.offset;
7393
7394                 inode->i_ctime = CURRENT_TIME;
7395                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7396                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7397                     (actual_len > inode->i_size) &&
7398                     (cur_offset > inode->i_size)) {
7399                         if (cur_offset > actual_len)
7400                                 i_size = actual_len;
7401                         else
7402                                 i_size = cur_offset;
7403                         i_size_write(inode, i_size);
7404                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7405                 }
7406
7407                 ret = btrfs_update_inode(trans, root, inode);
7408                 BUG_ON(ret);
7409
7410                 if (own_trans)
7411                         btrfs_end_transaction(trans, root);
7412         }
7413         return ret;
7414 }
7415
7416 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7417                               u64 start, u64 num_bytes, u64 min_size,
7418                               loff_t actual_len, u64 *alloc_hint)
7419 {
7420         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7421                                            min_size, actual_len, alloc_hint,
7422                                            NULL);
7423 }
7424
7425 int btrfs_prealloc_file_range_trans(struct inode *inode,
7426                                     struct btrfs_trans_handle *trans, int mode,
7427                                     u64 start, u64 num_bytes, u64 min_size,
7428                                     loff_t actual_len, u64 *alloc_hint)
7429 {
7430         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7431                                            min_size, actual_len, alloc_hint, trans);
7432 }
7433
7434 static int btrfs_set_page_dirty(struct page *page)
7435 {
7436         return __set_page_dirty_nobuffers(page);
7437 }
7438
7439 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7440 {
7441         struct btrfs_root *root = BTRFS_I(inode)->root;
7442
7443         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7444                 return -EROFS;
7445         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7446                 return -EACCES;
7447         return generic_permission(inode, mask, flags, btrfs_check_acl);
7448 }
7449
7450 static const struct inode_operations btrfs_dir_inode_operations = {
7451         .getattr        = btrfs_getattr,
7452         .lookup         = btrfs_lookup,
7453         .create         = btrfs_create,
7454         .unlink         = btrfs_unlink,
7455         .link           = btrfs_link,
7456         .mkdir          = btrfs_mkdir,
7457         .rmdir          = btrfs_rmdir,
7458         .rename         = btrfs_rename,
7459         .symlink        = btrfs_symlink,
7460         .setattr        = btrfs_setattr,
7461         .mknod          = btrfs_mknod,
7462         .setxattr       = btrfs_setxattr,
7463         .getxattr       = btrfs_getxattr,
7464         .listxattr      = btrfs_listxattr,
7465         .removexattr    = btrfs_removexattr,
7466         .permission     = btrfs_permission,
7467 };
7468 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7469         .lookup         = btrfs_lookup,
7470         .permission     = btrfs_permission,
7471 };
7472
7473 static const struct file_operations btrfs_dir_file_operations = {
7474         .llseek         = generic_file_llseek,
7475         .read           = generic_read_dir,
7476         .readdir        = btrfs_real_readdir,
7477         .unlocked_ioctl = btrfs_ioctl,
7478 #ifdef CONFIG_COMPAT
7479         .compat_ioctl   = btrfs_ioctl,
7480 #endif
7481         .release        = btrfs_release_file,
7482         .fsync          = btrfs_sync_file,
7483 };
7484
7485 static struct extent_io_ops btrfs_extent_io_ops = {
7486         .fill_delalloc = run_delalloc_range,
7487         .submit_bio_hook = btrfs_submit_bio_hook,
7488         .merge_bio_hook = btrfs_merge_bio_hook,
7489         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7490         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7491         .writepage_start_hook = btrfs_writepage_start_hook,
7492         .readpage_io_failed_hook = btrfs_io_failed_hook,
7493         .set_bit_hook = btrfs_set_bit_hook,
7494         .clear_bit_hook = btrfs_clear_bit_hook,
7495         .merge_extent_hook = btrfs_merge_extent_hook,
7496         .split_extent_hook = btrfs_split_extent_hook,
7497 };
7498
7499 /*
7500  * btrfs doesn't support the bmap operation because swapfiles
7501  * use bmap to make a mapping of extents in the file.  They assume
7502  * these extents won't change over the life of the file and they
7503  * use the bmap result to do IO directly to the drive.
7504  *
7505  * the btrfs bmap call would return logical addresses that aren't
7506  * suitable for IO and they also will change frequently as COW
7507  * operations happen.  So, swapfile + btrfs == corruption.
7508  *
7509  * For now we're avoiding this by dropping bmap.
7510  */
7511 static const struct address_space_operations btrfs_aops = {
7512         .readpage       = btrfs_readpage,
7513         .writepage      = btrfs_writepage,
7514         .writepages     = btrfs_writepages,
7515         .readpages      = btrfs_readpages,
7516         .direct_IO      = btrfs_direct_IO,
7517         .invalidatepage = btrfs_invalidatepage,
7518         .releasepage    = btrfs_releasepage,
7519         .set_page_dirty = btrfs_set_page_dirty,
7520         .error_remove_page = generic_error_remove_page,
7521 };
7522
7523 static const struct address_space_operations btrfs_symlink_aops = {
7524         .readpage       = btrfs_readpage,
7525         .writepage      = btrfs_writepage,
7526         .invalidatepage = btrfs_invalidatepage,
7527         .releasepage    = btrfs_releasepage,
7528 };
7529
7530 static const struct inode_operations btrfs_file_inode_operations = {
7531         .getattr        = btrfs_getattr,
7532         .setattr        = btrfs_setattr,
7533         .setxattr       = btrfs_setxattr,
7534         .getxattr       = btrfs_getxattr,
7535         .listxattr      = btrfs_listxattr,
7536         .removexattr    = btrfs_removexattr,
7537         .permission     = btrfs_permission,
7538         .fiemap         = btrfs_fiemap,
7539 };
7540 static const struct inode_operations btrfs_special_inode_operations = {
7541         .getattr        = btrfs_getattr,
7542         .setattr        = btrfs_setattr,
7543         .permission     = btrfs_permission,
7544         .setxattr       = btrfs_setxattr,
7545         .getxattr       = btrfs_getxattr,
7546         .listxattr      = btrfs_listxattr,
7547         .removexattr    = btrfs_removexattr,
7548 };
7549 static const struct inode_operations btrfs_symlink_inode_operations = {
7550         .readlink       = generic_readlink,
7551         .follow_link    = page_follow_link_light,
7552         .put_link       = page_put_link,
7553         .getattr        = btrfs_getattr,
7554         .permission     = btrfs_permission,
7555         .setxattr       = btrfs_setxattr,
7556         .getxattr       = btrfs_getxattr,
7557         .listxattr      = btrfs_listxattr,
7558         .removexattr    = btrfs_removexattr,
7559 };
7560
7561 const struct dentry_operations btrfs_dentry_operations = {
7562         .d_delete       = btrfs_dentry_delete,
7563 };