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                 BUG_ON(!async_cow);
 958                 async_cow->inode = inode;
 959                 async_cow->root = root;
 960                 async_cow->locked_page = locked_page;
 961                 async_cow->start = start;
 962
 963                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 964                         cur_end = end;
 965                 else
 966                         cur_end = min(end, start + 512 * 1024 - 1);
 967
 968                 async_cow->end = cur_end;
 969                 INIT_LIST_HEAD(&async_cow->extents);
 970
 971                 async_cow->work.func = async_cow_start;
 972                 async_cow->work.ordered_func = async_cow_submit;
 973                 async_cow->work.ordered_free = async_cow_free;
 974                 async_cow->work.flags = 0;
 975
 976                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 977                         PAGE_CACHE_SHIFT;
 978                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 979
 980                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 981                                    &async_cow->work);
 982
 983                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 984                         wait_event(root->fs_info->async_submit_wait,
 985                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 986                             limit));
 987                 }
 988
 989                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 990                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 991                         wait_event(root->fs_info->async_submit_wait,
 992                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 993                            0));
 994                 }
 995
 996                 *nr_written += nr_pages;
 997                 start = cur_end + 1;
 998         }
 999         *page_started = 1;
1000         return 0;
1001 }
1002
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004                                         u64 bytenr, u64 num_bytes)
1005 {
1006         int ret;
1007         struct btrfs_ordered_sum *sums;
1008         LIST_HEAD(list);
1009
1010         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011                                        bytenr + num_bytes - 1, &list);
1012         if (ret == 0 && list_empty(&list))
1013                 return 0;
1014
1015         while (!list_empty(&list)) {
1016                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017                 list_del(&sums->list);
1018                 kfree(sums);
1019         }
1020         return 1;
1021 }
1022
1023 /*
1024  * when nowcow writeback call back.  This checks for snapshots or COW copies
1025  * of the extents that exist in the file, and COWs the file as required.
1026  *
1027  * If no cow copies or snapshots exist, we write directly to the existing
1028  * blocks on disk
1029  */
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031                                        struct page *locked_page,
1032                               u64 start, u64 end, int *page_started, int force,
1033                               unsigned long *nr_written)
1034 {
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         struct btrfs_trans_handle *trans;
1037         struct extent_buffer *leaf;
1038         struct btrfs_path *path;
1039         struct btrfs_file_extent_item *fi;
1040         struct btrfs_key found_key;
1041         u64 cow_start;
1042         u64 cur_offset;
1043         u64 extent_end;
1044         u64 extent_offset;
1045         u64 disk_bytenr;
1046         u64 num_bytes;
1047         int extent_type;
1048         int ret;
1049         int type;
1050         int nocow;
1051         int check_prev = 1;
1052         bool nolock = false;
1053
1054         path = btrfs_alloc_path();
1055         BUG_ON(!path);
1056         if (root == root->fs_info->tree_root) {
1057                 nolock = true;
1058                 trans = btrfs_join_transaction_nolock(root, 1);
1059         } else {
1060                 trans = btrfs_join_transaction(root, 1);
1061         }
1062         BUG_ON(IS_ERR(trans));
1063
1064         cow_start = (u64)-1;
1065         cur_offset = start;
1066         while (1) {
1067                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068                                                cur_offset, 0);
1069                 BUG_ON(ret < 0);
1070                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071                         leaf = path->nodes[0];
1072                         btrfs_item_key_to_cpu(leaf, &found_key,
1073                                               path->slots[0] - 1);
1074                         if (found_key.objectid == inode->i_ino &&
1075                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1076                                 path->slots[0]--;
1077                 }
1078                 check_prev = 0;
1079 next_slot:
1080                 leaf = path->nodes[0];
1081                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082                         ret = btrfs_next_leaf(root, path);
1083                         if (ret < 0)
1084                                 BUG_ON(1);
1085                         if (ret > 0)
1086                                 break;
1087                         leaf = path->nodes[0];
1088                 }
1089
1090                 nocow = 0;
1091                 disk_bytenr = 0;
1092                 num_bytes = 0;
1093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094
1095                 if (found_key.objectid > inode->i_ino ||
1096                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097                     found_key.offset > end)
1098                         break;
1099
1100                 if (found_key.offset > cur_offset) {
1101                         extent_end = found_key.offset;
1102                         extent_type = 0;
1103                         goto out_check;
1104                 }
1105
1106                 fi = btrfs_item_ptr(leaf, path->slots[0],
1107                                     struct btrfs_file_extent_item);
1108                 extent_type = btrfs_file_extent_type(leaf, fi);
1109
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_num_bytes(leaf, fi);
1116                         if (extent_end <= start) {
1117                                 path->slots[0]++;
1118                                 goto next_slot;
1119                         }
1120                         if (disk_bytenr == 0)
1121                                 goto out_check;
1122                         if (btrfs_file_extent_compression(leaf, fi) ||
1123                             btrfs_file_extent_encryption(leaf, fi) ||
1124                             btrfs_file_extent_other_encoding(leaf, fi))
1125                                 goto out_check;
1126                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127                                 goto out_check;
1128                         if (btrfs_extent_readonly(root, disk_bytenr))
1129                                 goto out_check;
1130                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131                                                   found_key.offset -
1132                                                   extent_offset, disk_bytenr))
1133                                 goto out_check;
1134                         disk_bytenr += extent_offset;
1135                         disk_bytenr += cur_offset - found_key.offset;
1136                         num_bytes = min(end + 1, extent_end) - cur_offset;
1137                         /*
1138                          * force cow if csum exists in the range.
1139                          * this ensure that csum for a given extent are
1140                          * either valid or do not exist.
1141                          */
1142                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143                                 goto out_check;
1144                         nocow = 1;
1145                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146                         extent_end = found_key.offset +
1147                                 btrfs_file_extent_inline_len(leaf, fi);
1148                         extent_end = ALIGN(extent_end, root->sectorsize);
1149                 } else {
1150                         BUG_ON(1);
1151                 }
1152 out_check:
1153                 if (extent_end <= start) {
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157                 if (!nocow) {
1158                         if (cow_start == (u64)-1)
1159                                 cow_start = cur_offset;
1160                         cur_offset = extent_end;
1161                         if (cur_offset > end)
1162                                 break;
1163                         path->slots[0]++;
1164                         goto next_slot;
1165                 }
1166
1167                 btrfs_release_path(root, path);
1168                 if (cow_start != (u64)-1) {
1169                         ret = cow_file_range(inode, locked_page, cow_start,
1170                                         found_key.offset - 1, page_started,
1171                                         nr_written, 1);
1172                         BUG_ON(ret);
1173                         cow_start = (u64)-1;
1174                 }
1175
1176                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177                         struct extent_map *em;
1178                         struct extent_map_tree *em_tree;
1179                         em_tree = &BTRFS_I(inode)->extent_tree;
1180                         em = alloc_extent_map(GFP_NOFS);
1181                         BUG_ON(!em);
1182                         em->start = cur_offset;
1183                         em->orig_start = em->start;
1184                         em->len = num_bytes;
1185                         em->block_len = num_bytes;
1186                         em->block_start = disk_bytenr;
1187                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1188                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189                         while (1) {
1190                                 write_lock(&em_tree->lock);
1191                                 ret = add_extent_mapping(em_tree, em);
1192                                 write_unlock(&em_tree->lock);
1193                                 if (ret != -EEXIST) {
1194                                         free_extent_map(em);
1195                                         break;
1196                                 }
1197                                 btrfs_drop_extent_cache(inode, em->start,
1198                                                 em->start + em->len - 1, 0);
1199                         }
1200                         type = BTRFS_ORDERED_PREALLOC;
1201                 } else {
1202                         type = BTRFS_ORDERED_NOCOW;
1203                 }
1204
1205                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206                                                num_bytes, num_bytes, type);
1207                 BUG_ON(ret);
1208
1209                 if (root->root_key.objectid ==
1210                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212                                                       num_bytes);
1213                         BUG_ON(ret);
1214                 }
1215
1216                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217                                 cur_offset, cur_offset + num_bytes - 1,
1218                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220                                 EXTENT_SET_PRIVATE2);
1221                 cur_offset = extent_end;
1222                 if (cur_offset > end)
1223                         break;
1224         }
1225         btrfs_release_path(root, path);
1226
1227         if (cur_offset <= end && cow_start == (u64)-1)
1228                 cow_start = cur_offset;
1229         if (cow_start != (u64)-1) {
1230                 ret = cow_file_range(inode, locked_page, cow_start, end,
1231                                      page_started, nr_written, 1);
1232                 BUG_ON(ret);
1233         }
1234
1235         if (nolock) {
1236                 ret = btrfs_end_transaction_nolock(trans, root);
1237                 BUG_ON(ret);
1238         } else {
1239                 ret = btrfs_end_transaction(trans, root);
1240                 BUG_ON(ret);
1241         }
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * extent_io.c call back to do delayed allocation processing
1248  */
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250                               u64 start, u64 end, int *page_started,
1251                               unsigned long *nr_written)
1252 {
1253         int ret;
1254         struct btrfs_root *root = BTRFS_I(inode)->root;
1255
1256         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258                                          page_started, 1, nr_written);
1259         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261                                          page_started, 0, nr_written);
1262         else if (!btrfs_test_opt(root, COMPRESS) &&
1263                  !(BTRFS_I(inode)->force_compress) &&
1264                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265                 ret = cow_file_range(inode, locked_page, start, end,
1266                                       page_started, nr_written, 1);
1267         else
1268                 ret = cow_file_range_async(inode, locked_page, start, end,
1269                                            page_started, nr_written);
1270         return ret;
1271 }
1272
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274                                    struct extent_state *orig, u64 split)
1275 {
1276         /* not delalloc, ignore it */
1277         if (!(orig->state & EXTENT_DELALLOC))
1278                 return 0;
1279
1280         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281         return 0;
1282 }
1283
1284 /*
1285  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286  * extents so we can keep track of new extents that are just merged onto old
1287  * extents, such as when we are doing sequential writes, so we can properly
1288  * account for the metadata space we'll need.
1289  */
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291                                    struct extent_state *new,
1292                                    struct extent_state *other)
1293 {
1294         /* not delalloc, ignore it */
1295         if (!(other->state & EXTENT_DELALLOC))
1296                 return 0;
1297
1298         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1299         return 0;
1300 }
1301
1302 /*
1303  * extent_io.c set_bit_hook, used to track delayed allocation
1304  * bytes in this file, and to maintain the list of inodes that
1305  * have pending delalloc work to be done.
1306  */
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308                               struct extent_state *state, int *bits)
1309 {
1310
1311         /*
1312          * set_bit and clear bit hooks normally require _irqsave/restore
1313          * but in this case, we are only testeing for the DELALLOC
1314          * bit, which is only set or cleared with irqs on
1315          */
1316         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317                 struct btrfs_root *root = BTRFS_I(inode)->root;
1318                 u64 len = state->end + 1 - state->start;
1319                 int do_list = (root->root_key.objectid !=
1320                                BTRFS_ROOT_TREE_OBJECTID);
1321
1322                 if (*bits & EXTENT_FIRST_DELALLOC)
1323                         *bits &= ~EXTENT_FIRST_DELALLOC;
1324                 else
1325                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326
1327                 spin_lock(&root->fs_info->delalloc_lock);
1328                 BTRFS_I(inode)->delalloc_bytes += len;
1329                 root->fs_info->delalloc_bytes += len;
1330                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332                                       &root->fs_info->delalloc_inodes);
1333                 }
1334                 spin_unlock(&root->fs_info->delalloc_lock);
1335         }
1336         return 0;
1337 }
1338
1339 /*
1340  * extent_io.c clear_bit_hook, see set_bit_hook for why
1341  */
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343                                 struct extent_state *state, int *bits)
1344 {
1345         /*
1346          * set_bit and clear bit hooks normally require _irqsave/restore
1347          * but in this case, we are only testeing for the DELALLOC
1348          * bit, which is only set or cleared with irqs on
1349          */
1350         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351                 struct btrfs_root *root = BTRFS_I(inode)->root;
1352                 u64 len = state->end + 1 - state->start;
1353                 int do_list = (root->root_key.objectid !=
1354                                BTRFS_ROOT_TREE_OBJECTID);
1355
1356                 if (*bits & EXTENT_FIRST_DELALLOC)
1357                         *bits &= ~EXTENT_FIRST_DELALLOC;
1358                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360
1361                 if (*bits & EXTENT_DO_ACCOUNTING)
1362                         btrfs_delalloc_release_metadata(inode, len);
1363
1364                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365                     && do_list)
1366                         btrfs_free_reserved_data_space(inode, len);
1367
1368                 spin_lock(&root->fs_info->delalloc_lock);
1369                 root->fs_info->delalloc_bytes -= len;
1370                 BTRFS_I(inode)->delalloc_bytes -= len;
1371
1372                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375                 }
1376                 spin_unlock(&root->fs_info->delalloc_lock);
1377         }
1378         return 0;
1379 }
1380
1381 /*
1382  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383  * we don't create bios that span stripes or chunks
1384  */
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386                          size_t size, struct bio *bio,
1387                          unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390         struct btrfs_mapping_tree *map_tree;
1391         u64 logical = (u64)bio->bi_sector << 9;
1392         u64 length = 0;
1393         u64 map_length;
1394         int ret;
1395
1396         if (bio_flags & EXTENT_BIO_COMPRESSED)
1397                 return 0;
1398
1399         length = bio->bi_size;
1400         map_tree = &root->fs_info->mapping_tree;
1401         map_length = length;
1402         ret = btrfs_map_block(map_tree, READ, logical,
1403                               &map_length, NULL, 0);
1404
1405         if (map_length < length + size)
1406                 return 1;
1407         return ret;
1408 }
1409
1410 /*
1411  * in order to insert checksums into the metadata in large chunks,
1412  * we wait until bio submission time.   All the pages in the bio are
1413  * checksummed and sums are attached onto the ordered extent record.
1414  *
1415  * At IO completion time the cums attached on the ordered extent record
1416  * are inserted into the btree
1417  */
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419                                     struct bio *bio, int mirror_num,
1420                                     unsigned long bio_flags,
1421                                     u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425
1426         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427         BUG_ON(ret);
1428         return 0;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440                           int mirror_num, unsigned long bio_flags,
1441                           u64 bio_offset)
1442 {
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 }
1446
1447 /*
1448  * extent_io.c submission hook. This does the right thing for csum calculation
1449  * on write, or reading the csums from the tree before a read
1450  */
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags,
1453                           u64 bio_offset)
1454 {
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456         int ret = 0;
1457         int skip_sum;
1458
1459         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460
1461         if (root == root->fs_info->tree_root)
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463         else
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465         BUG_ON(ret);
1466
1467         if (!(rw & REQ_WRITE)) {
1468                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469                         return btrfs_submit_compressed_read(inode, bio,
1470                                                     mirror_num, bio_flags);
1471                 } else if (!skip_sum) {
1472                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473                         if (ret)
1474                                 return ret;
1475                 }
1476                 goto mapit;
1477         } else if (!skip_sum) {
1478                 /* csum items have already been cloned */
1479                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480                         goto mapit;
1481                 /* we're doing a write, do the async checksumming */
1482                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483                                    inode, rw, bio, mirror_num,
1484                                    bio_flags, bio_offset,
1485                                    __btrfs_submit_bio_start,
1486                                    __btrfs_submit_bio_done);
1487         }
1488
1489 mapit:
1490         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1491 }
1492
1493 /*
1494  * given a list of ordered sums record them in the inode.  This happens
1495  * at IO completion time based on sums calculated at bio submission time.
1496  */
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498                              struct inode *inode, u64 file_offset,
1499                              struct list_head *list)
1500 {
1501         struct btrfs_ordered_sum *sum;
1502
1503         btrfs_set_trans_block_group(trans, inode);
1504
1505         list_for_each_entry(sum, list, list) {
1506                 btrfs_csum_file_blocks(trans,
1507                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508         }
1509         return 0;
1510 }
1511
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513                               struct extent_state **cached_state)
1514 {
1515         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516                 WARN_ON(1);
1517         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518                                    cached_state, GFP_NOFS);
1519 }
1520
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1523         struct page *page;
1524         struct btrfs_work work;
1525 };
1526
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 {
1529         struct btrfs_writepage_fixup *fixup;
1530         struct btrfs_ordered_extent *ordered;
1531         struct extent_state *cached_state = NULL;
1532         struct page *page;
1533         struct inode *inode;
1534         u64 page_start;
1535         u64 page_end;
1536
1537         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538         page = fixup->page;
1539 again:
1540         lock_page(page);
1541         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542                 ClearPageChecked(page);
1543                 goto out_page;
1544         }
1545
1546         inode = page->mapping->host;
1547         page_start = page_offset(page);
1548         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549
1550         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551                          &cached_state, GFP_NOFS);
1552
1553         /* already ordered? We're done */
1554         if (PagePrivate2(page))
1555                 goto out;
1556
1557         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558         if (ordered) {
1559                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560                                      page_end, &cached_state, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         BUG();
1567         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568         ClearPageChecked(page);
1569 out:
1570         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571                              &cached_state, GFP_NOFS);
1572 out_page:
1573         unlock_page(page);
1574         page_cache_release(page);
1575         kfree(fixup);
1576 }
1577
1578 /*
1579  * There are a few paths in the higher layers of the kernel that directly
1580  * set the page dirty bit without asking the filesystem if it is a
1581  * good idea.  This causes problems because we want to make sure COW
1582  * properly happens and the data=ordered rules are followed.
1583  *
1584  * In our case any range that doesn't have the ORDERED bit set
1585  * hasn't been properly setup for IO.  We kick off an async process
1586  * to fix it up.  The async helper will wait for ordered extents, set
1587  * the delalloc bit and make it safe to write the page.
1588  */
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         struct btrfs_writepage_fixup *fixup;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594
1595         /* this page is properly in the ordered list */
1596         if (TestClearPagePrivate2(page))
1597                 return 0;
1598
1599         if (PageChecked(page))
1600                 return -EAGAIN;
1601
1602         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603         if (!fixup)
1604                 return -EAGAIN;
1605
1606         SetPageChecked(page);
1607         page_cache_get(page);
1608         fixup->work.func = btrfs_writepage_fixup_worker;
1609         fixup->page = page;
1610         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611         return -EAGAIN;
1612 }
1613
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615                                        struct inode *inode, u64 file_pos,
1616                                        u64 disk_bytenr, u64 disk_num_bytes,
1617                                        u64 num_bytes, u64 ram_bytes,
1618                                        u8 compression, u8 encryption,
1619                                        u16 other_encoding, int extent_type)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         struct btrfs_file_extent_item *fi;
1623         struct btrfs_path *path;
1624         struct extent_buffer *leaf;
1625         struct btrfs_key ins;
1626         u64 hint;
1627         int ret;
1628
1629         path = btrfs_alloc_path();
1630         BUG_ON(!path);
1631
1632         path->leave_spinning = 1;
1633
1634         /*
1635          * we may be replacing one extent in the tree with another.
1636          * The new extent is pinned in the extent map, and we don't want
1637          * to drop it from the cache until it is completely in the btree.
1638          *
1639          * So, tell btrfs_drop_extents to leave this extent in the cache.
1640          * the caller is expected to unpin it and allow it to be merged
1641          * with the others.
1642          */
1643         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644                                  &hint, 0);
1645         BUG_ON(ret);
1646
1647         ins.objectid = inode->i_ino;
1648         ins.offset = file_pos;
1649         ins.type = BTRFS_EXTENT_DATA_KEY;
1650         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651         BUG_ON(ret);
1652         leaf = path->nodes[0];
1653         fi = btrfs_item_ptr(leaf, path->slots[0],
1654                             struct btrfs_file_extent_item);
1655         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656         btrfs_set_file_extent_type(leaf, fi, extent_type);
1657         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659         btrfs_set_file_extent_offset(leaf, fi, 0);
1660         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662         btrfs_set_file_extent_compression(leaf, fi, compression);
1663         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665
1666         btrfs_unlock_up_safe(path, 1);
1667         btrfs_set_lock_blocking(leaf);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670
1671         inode_add_bytes(inode, num_bytes);
1672
1673         ins.objectid = disk_bytenr;
1674         ins.offset = disk_num_bytes;
1675         ins.type = BTRFS_EXTENT_ITEM_KEY;
1676         ret = btrfs_alloc_reserved_file_extent(trans, root,
1677                                         root->root_key.objectid,
1678                                         inode->i_ino, file_pos, &ins);
1679         BUG_ON(ret);
1680         btrfs_free_path(path);
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * helper function for btrfs_finish_ordered_io, this
1687  * just reads in some of the csum leaves to prime them into ram
1688  * before we start the transaction.  It limits the amount of btree
1689  * reads required while inside the transaction.
1690  */
1691 /* as ordered data IO finishes, this gets called so we can finish
1692  * an ordered extent if the range of bytes in the file it covers are
1693  * fully written.
1694  */
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_trans_handle *trans = NULL;
1699         struct btrfs_ordered_extent *ordered_extent = NULL;
1700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701         struct extent_state *cached_state = NULL;
1702         int compress_type = 0;
1703         int ret;
1704         bool nolock = false;
1705
1706         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707                                              end - start + 1);
1708         if (!ret)
1709                 return 0;
1710         BUG_ON(!ordered_extent);
1711
1712         nolock = (root == root->fs_info->tree_root);
1713
1714         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715                 BUG_ON(!list_empty(&ordered_extent->list));
1716                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717                 if (!ret) {
1718                         if (nolock)
1719                                 trans = btrfs_join_transaction_nolock(root, 1);
1720                         else
1721                                 trans = btrfs_join_transaction(root, 1);
1722                         BUG_ON(IS_ERR(trans));
1723                         btrfs_set_trans_block_group(trans, inode);
1724                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725                         ret = btrfs_update_inode(trans, root, inode);
1726                         BUG_ON(ret);
1727                 }
1728                 goto out;
1729         }
1730
1731         lock_extent_bits(io_tree, ordered_extent->file_offset,
1732                          ordered_extent->file_offset + ordered_extent->len - 1,
1733                          0, &cached_state, GFP_NOFS);
1734
1735         if (nolock)
1736                 trans = btrfs_join_transaction_nolock(root, 1);
1737         else
1738                 trans = btrfs_join_transaction(root, 1);
1739         BUG_ON(IS_ERR(trans));
1740         btrfs_set_trans_block_group(trans, inode);
1741         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742
1743         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744                 compress_type = ordered_extent->compress_type;
1745         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746                 BUG_ON(compress_type);
1747                 ret = btrfs_mark_extent_written(trans, inode,
1748                                                 ordered_extent->file_offset,
1749                                                 ordered_extent->file_offset +
1750                                                 ordered_extent->len);
1751                 BUG_ON(ret);
1752         } else {
1753                 BUG_ON(root == root->fs_info->tree_root);
1754                 ret = insert_reserved_file_extent(trans, inode,
1755                                                 ordered_extent->file_offset,
1756                                                 ordered_extent->start,
1757                                                 ordered_extent->disk_len,
1758                                                 ordered_extent->len,
1759                                                 ordered_extent->len,
1760                                                 compress_type, 0, 0,
1761                                                 BTRFS_FILE_EXTENT_REG);
1762                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763                                    ordered_extent->file_offset,
1764                                    ordered_extent->len);
1765                 BUG_ON(ret);
1766         }
1767         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768                              ordered_extent->file_offset +
1769                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770
1771         add_pending_csums(trans, inode, ordered_extent->file_offset,
1772                           &ordered_extent->list);
1773
1774         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775         if (!ret) {
1776                 ret = btrfs_update_inode(trans, root, inode);
1777                 BUG_ON(ret);
1778         }
1779         ret = 0;
1780 out:
1781         if (nolock) {
1782                 if (trans)
1783                         btrfs_end_transaction_nolock(trans, root);
1784         } else {
1785                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1786                 if (trans)
1787                         btrfs_end_transaction(trans, root);
1788         }
1789
1790         /* once for us */
1791         btrfs_put_ordered_extent(ordered_extent);
1792         /* once for the tree */
1793         btrfs_put_ordered_extent(ordered_extent);
1794
1795         return 0;
1796 }
1797
1798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1799                                 struct extent_state *state, int uptodate)
1800 {
1801         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1802
1803         ClearPagePrivate2(page);
1804         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1805 }
1806
1807 /*
1808  * When IO fails, either with EIO or csum verification fails, we
1809  * try other mirrors that might have a good copy of the data.  This
1810  * io_failure_record is used to record state as we go through all the
1811  * mirrors.  If another mirror has good data, the page is set up to date
1812  * and things continue.  If a good mirror can't be found, the original
1813  * bio end_io callback is called to indicate things have failed.
1814  */
1815 struct io_failure_record {
1816         struct page *page;
1817         u64 start;
1818         u64 len;
1819         u64 logical;
1820         unsigned long bio_flags;
1821         int last_mirror;
1822 };
1823
1824 static int btrfs_io_failed_hook(struct bio *failed_bio,
1825                          struct page *page, u64 start, u64 end,
1826                          struct extent_state *state)
1827 {
1828         struct io_failure_record *failrec = NULL;
1829         u64 private;
1830         struct extent_map *em;
1831         struct inode *inode = page->mapping->host;
1832         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1833         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1834         struct bio *bio;
1835         int num_copies;
1836         int ret;
1837         int rw;
1838         u64 logical;
1839
1840         ret = get_state_private(failure_tree, start, &private);
1841         if (ret) {
1842                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1843                 if (!failrec)
1844                         return -ENOMEM;
1845                 failrec->start = start;
1846                 failrec->len = end - start + 1;
1847                 failrec->last_mirror = 0;
1848                 failrec->bio_flags = 0;
1849
1850                 read_lock(&em_tree->lock);
1851                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1852                 if (em->start > start || em->start + em->len < start) {
1853                         free_extent_map(em);
1854                         em = NULL;
1855                 }
1856                 read_unlock(&em_tree->lock);
1857
1858                 if (!em || IS_ERR(em)) {
1859                         kfree(failrec);
1860                         return -EIO;
1861                 }
1862                 logical = start - em->start;
1863                 logical = em->block_start + logical;
1864                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1865                         logical = em->block_start;
1866                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1867                         extent_set_compress_type(&failrec->bio_flags,
1868                                                  em->compress_type);
1869                 }
1870                 failrec->logical = logical;
1871                 free_extent_map(em);
1872                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1873                                 EXTENT_DIRTY, GFP_NOFS);
1874                 set_state_private(failure_tree, start,
1875                                  (u64)(unsigned long)failrec);
1876         } else {
1877                 failrec = (struct io_failure_record *)(unsigned long)private;
1878         }
1879         num_copies = btrfs_num_copies(
1880                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1881                               failrec->logical, failrec->len);
1882         failrec->last_mirror++;
1883         if (!state) {
1884                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1885                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1886                                                     failrec->start,
1887                                                     EXTENT_LOCKED);
1888                 if (state && state->start != failrec->start)
1889                         state = NULL;
1890                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1891         }
1892         if (!state || failrec->last_mirror > num_copies) {
1893                 set_state_private(failure_tree, failrec->start, 0);
1894                 clear_extent_bits(failure_tree, failrec->start,
1895                                   failrec->start + failrec->len - 1,
1896                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1897                 kfree(failrec);
1898                 return -EIO;
1899         }
1900         bio = bio_alloc(GFP_NOFS, 1);
1901         bio->bi_private = state;
1902         bio->bi_end_io = failed_bio->bi_end_io;
1903         bio->bi_sector = failrec->logical >> 9;
1904         bio->bi_bdev = failed_bio->bi_bdev;
1905         bio->bi_size = 0;
1906
1907         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1908         if (failed_bio->bi_rw & REQ_WRITE)
1909                 rw = WRITE;
1910         else
1911                 rw = READ;
1912
1913         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1914                                                       failrec->last_mirror,
1915                                                       failrec->bio_flags, 0);
1916         return ret;
1917 }
1918
1919 /*
1920  * each time an IO finishes, we do a fast check in the IO failure tree
1921  * to see if we need to process or clean up an io_failure_record
1922  */
1923 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1924 {
1925         u64 private;
1926         u64 private_failure;
1927         struct io_failure_record *failure;
1928         int ret;
1929
1930         private = 0;
1931         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1932                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1933                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1934                                         start, &private_failure);
1935                 if (ret == 0) {
1936                         failure = (struct io_failure_record *)(unsigned long)
1937                                    private_failure;
1938                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1939                                           failure->start, 0);
1940                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1941                                           failure->start,
1942                                           failure->start + failure->len - 1,
1943                                           EXTENT_DIRTY | EXTENT_LOCKED,
1944                                           GFP_NOFS);
1945                         kfree(failure);
1946                 }
1947         }
1948         return 0;
1949 }
1950
1951 /*
1952  * when reads are done, we need to check csums to verify the data is correct
1953  * if there's a match, we allow the bio to finish.  If not, we go through
1954  * the io_failure_record routines to find good copies
1955  */
1956 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1957                                struct extent_state *state)
1958 {
1959         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1960         struct inode *inode = page->mapping->host;
1961         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1962         char *kaddr;
1963         u64 private = ~(u32)0;
1964         int ret;
1965         struct btrfs_root *root = BTRFS_I(inode)->root;
1966         u32 csum = ~(u32)0;
1967
1968         if (PageChecked(page)) {
1969                 ClearPageChecked(page);
1970                 goto good;
1971         }
1972
1973         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1974                 return 0;
1975
1976         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1977             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1978                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1979                                   GFP_NOFS);
1980                 return 0;
1981         }
1982
1983         if (state && state->start == start) {
1984                 private = state->private;
1985                 ret = 0;
1986         } else {
1987                 ret = get_state_private(io_tree, start, &private);
1988         }
1989         kaddr = kmap_atomic(page, KM_USER0);
1990         if (ret)
1991                 goto zeroit;
1992
1993         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1994         btrfs_csum_final(csum, (char *)&csum);
1995         if (csum != private)
1996                 goto zeroit;
1997
1998         kunmap_atomic(kaddr, KM_USER0);
1999 good:
2000         /* if the io failure tree for this inode is non-empty,
2001          * check to see if we've recovered from a failed IO
2002          */
2003         btrfs_clean_io_failures(inode, start);
2004         return 0;
2005
2006 zeroit:
2007         if (printk_ratelimit()) {
2008                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2009                        "private %llu\n", page->mapping->host->i_ino,
2010                        (unsigned long long)start, csum,
2011                        (unsigned long long)private);
2012         }
2013         memset(kaddr + offset, 1, end - start + 1);
2014         flush_dcache_page(page);
2015         kunmap_atomic(kaddr, KM_USER0);
2016         if (private == 0)
2017                 return 0;
2018         return -EIO;
2019 }
2020
2021 struct delayed_iput {
2022         struct list_head list;
2023         struct inode *inode;
2024 };
2025
2026 void btrfs_add_delayed_iput(struct inode *inode)
2027 {
2028         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2029         struct delayed_iput *delayed;
2030
2031         if (atomic_add_unless(&inode->i_count, -1, 1))
2032                 return;
2033
2034         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2035         delayed->inode = inode;
2036
2037         spin_lock(&fs_info->delayed_iput_lock);
2038         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2039         spin_unlock(&fs_info->delayed_iput_lock);
2040 }
2041
2042 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2043 {
2044         LIST_HEAD(list);
2045         struct btrfs_fs_info *fs_info = root->fs_info;
2046         struct delayed_iput *delayed;
2047         int empty;
2048
2049         spin_lock(&fs_info->delayed_iput_lock);
2050         empty = list_empty(&fs_info->delayed_iputs);
2051         spin_unlock(&fs_info->delayed_iput_lock);
2052         if (empty)
2053                 return;
2054
2055         down_read(&root->fs_info->cleanup_work_sem);
2056         spin_lock(&fs_info->delayed_iput_lock);
2057         list_splice_init(&fs_info->delayed_iputs, &list);
2058         spin_unlock(&fs_info->delayed_iput_lock);
2059
2060         while (!list_empty(&list)) {
2061                 delayed = list_entry(list.next, struct delayed_iput, list);
2062                 list_del(&delayed->list);
2063                 iput(delayed->inode);
2064                 kfree(delayed);
2065         }
2066         up_read(&root->fs_info->cleanup_work_sem);
2067 }
2068
2069 /*
2070  * calculate extra metadata reservation when snapshotting a subvolume
2071  * contains orphan files.
2072  */
2073 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2074                                 struct btrfs_pending_snapshot *pending,
2075                                 u64 *bytes_to_reserve)
2076 {
2077         struct btrfs_root *root;
2078         struct btrfs_block_rsv *block_rsv;
2079         u64 num_bytes;
2080         int index;
2081
2082         root = pending->root;
2083         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2084                 return;
2085
2086         block_rsv = root->orphan_block_rsv;
2087
2088         /* orphan block reservation for the snapshot */
2089         num_bytes = block_rsv->size;
2090
2091         /*
2092          * after the snapshot is created, COWing tree blocks may use more
2093          * space than it frees. So we should make sure there is enough
2094          * reserved space.
2095          */
2096         index = trans->transid & 0x1;
2097         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2098                 num_bytes += block_rsv->size -
2099                              (block_rsv->reserved + block_rsv->freed[index]);
2100         }
2101
2102         *bytes_to_reserve += num_bytes;
2103 }
2104
2105 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2106                                 struct btrfs_pending_snapshot *pending)
2107 {
2108         struct btrfs_root *root = pending->root;
2109         struct btrfs_root *snap = pending->snap;
2110         struct btrfs_block_rsv *block_rsv;
2111         u64 num_bytes;
2112         int index;
2113         int ret;
2114
2115         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2116                 return;
2117
2118         /* refill source subvolume's orphan block reservation */
2119         block_rsv = root->orphan_block_rsv;
2120         index = trans->transid & 0x1;
2121         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122                 num_bytes = block_rsv->size -
2123                             (block_rsv->reserved + block_rsv->freed[index]);
2124                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2125                                               root->orphan_block_rsv,
2126                                               num_bytes);
2127                 BUG_ON(ret);
2128         }
2129
2130         /* setup orphan block reservation for the snapshot */
2131         block_rsv = btrfs_alloc_block_rsv(snap);
2132         BUG_ON(!block_rsv);
2133
2134         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2135         snap->orphan_block_rsv = block_rsv;
2136
2137         num_bytes = root->orphan_block_rsv->size;
2138         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                       block_rsv, num_bytes);
2140         BUG_ON(ret);
2141
2142 #if 0
2143         /* insert orphan item for the snapshot */
2144         WARN_ON(!root->orphan_item_inserted);
2145         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2146                                        snap->root_key.objectid);
2147         BUG_ON(ret);
2148         snap->orphan_item_inserted = 1;
2149 #endif
2150 }
2151
2152 enum btrfs_orphan_cleanup_state {
2153         ORPHAN_CLEANUP_STARTED  = 1,
2154         ORPHAN_CLEANUP_DONE     = 2,
2155 };
2156
2157 /*
2158  * This is called in transaction commmit time. If there are no orphan
2159  * files in the subvolume, it removes orphan item and frees block_rsv
2160  * structure.
2161  */
2162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2163                               struct btrfs_root *root)
2164 {
2165         int ret;
2166
2167         if (!list_empty(&root->orphan_list) ||
2168             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2169                 return;
2170
2171         if (root->orphan_item_inserted &&
2172             btrfs_root_refs(&root->root_item) > 0) {
2173                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2174                                             root->root_key.objectid);
2175                 BUG_ON(ret);
2176                 root->orphan_item_inserted = 0;
2177         }
2178
2179         if (root->orphan_block_rsv) {
2180                 WARN_ON(root->orphan_block_rsv->size > 0);
2181                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2182                 root->orphan_block_rsv = NULL;
2183         }
2184 }
2185
2186 /*
2187  * This creates an orphan entry for the given inode in case something goes
2188  * wrong in the middle of an unlink/truncate.
2189  *
2190  * NOTE: caller of this function should reserve 5 units of metadata for
2191  *       this function.
2192  */
2193 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2194 {
2195         struct btrfs_root *root = BTRFS_I(inode)->root;
2196         struct btrfs_block_rsv *block_rsv = NULL;
2197         int reserve = 0;
2198         int insert = 0;
2199         int ret;
2200
2201         if (!root->orphan_block_rsv) {
2202                 block_rsv = btrfs_alloc_block_rsv(root);
2203                 BUG_ON(!block_rsv);
2204         }
2205
2206         spin_lock(&root->orphan_lock);
2207         if (!root->orphan_block_rsv) {
2208                 root->orphan_block_rsv = block_rsv;
2209         } else if (block_rsv) {
2210                 btrfs_free_block_rsv(root, block_rsv);
2211                 block_rsv = NULL;
2212         }
2213
2214         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2215                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2216 #if 0
2217                 /*
2218                  * For proper ENOSPC handling, we should do orphan
2219                  * cleanup when mounting. But this introduces backward
2220                  * compatibility issue.
2221                  */
2222                 if (!xchg(&root->orphan_item_inserted, 1))
2223                         insert = 2;
2224                 else
2225                         insert = 1;
2226 #endif
2227                 insert = 1;
2228         }
2229
2230         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2231                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2232                 reserve = 1;
2233         }
2234         spin_unlock(&root->orphan_lock);
2235
2236         if (block_rsv)
2237                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2238
2239         /* grab metadata reservation from transaction handle */
2240         if (reserve) {
2241                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2242                 BUG_ON(ret);
2243         }
2244
2245         /* insert an orphan item to track this unlinked/truncated file */
2246         if (insert >= 1) {
2247                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2248                 BUG_ON(ret);
2249         }
2250
2251         /* insert an orphan item to track subvolume contains orphan files */
2252         if (insert >= 2) {
2253                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2254                                                root->root_key.objectid);
2255                 BUG_ON(ret);
2256         }
2257         return 0;
2258 }
2259
2260 /*
2261  * We have done the truncate/delete so we can go ahead and remove the orphan
2262  * item for this particular inode.
2263  */
2264 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2265 {
2266         struct btrfs_root *root = BTRFS_I(inode)->root;
2267         int delete_item = 0;
2268         int release_rsv = 0;
2269         int ret = 0;
2270
2271         spin_lock(&root->orphan_lock);
2272         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2273                 list_del_init(&BTRFS_I(inode)->i_orphan);
2274                 delete_item = 1;
2275         }
2276
2277         if (BTRFS_I(inode)->orphan_meta_reserved) {
2278                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2279                 release_rsv = 1;
2280         }
2281         spin_unlock(&root->orphan_lock);
2282
2283         if (trans && delete_item) {
2284                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2285                 BUG_ON(ret);
2286         }
2287
2288         if (release_rsv)
2289                 btrfs_orphan_release_metadata(inode);
2290
2291         return 0;
2292 }
2293
2294 /*
2295  * this cleans up any orphans that may be left on the list from the last use
2296  * of this root.
2297  */
2298 int btrfs_orphan_cleanup(struct btrfs_root *root)
2299 {
2300         struct btrfs_path *path;
2301         struct extent_buffer *leaf;
2302         struct btrfs_key key, found_key;
2303         struct btrfs_trans_handle *trans;
2304         struct inode *inode;
2305         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2306
2307         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2308                 return 0;
2309
2310         path = btrfs_alloc_path();
2311         if (!path) {
2312                 ret = -ENOMEM;
2313                 goto out;
2314         }
2315         path->reada = -1;
2316
2317         key.objectid = BTRFS_ORPHAN_OBJECTID;
2318         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2319         key.offset = (u64)-1;
2320
2321         while (1) {
2322                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2323                 if (ret < 0)
2324                         goto out;
2325
2326                 /*
2327                  * if ret == 0 means we found what we were searching for, which
2328                  * is weird, but possible, so only screw with path if we didn't
2329                  * find the key and see if we have stuff that matches
2330                  */
2331                 if (ret > 0) {
2332                         ret = 0;
2333                         if (path->slots[0] == 0)
2334                                 break;
2335                         path->slots[0]--;
2336                 }
2337
2338                 /* pull out the item */
2339                 leaf = path->nodes[0];
2340                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2341
2342                 /* make sure the item matches what we want */
2343                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2344                         break;
2345                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2346                         break;
2347
2348                 /* release the path since we're done with it */
2349                 btrfs_release_path(root, path);
2350
2351                 /*
2352                  * this is where we are basically btrfs_lookup, without the
2353                  * crossing root thing.  we store the inode number in the
2354                  * offset of the orphan item.
2355                  */
2356                 found_key.objectid = found_key.offset;
2357                 found_key.type = BTRFS_INODE_ITEM_KEY;
2358                 found_key.offset = 0;
2359                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2360                 if (IS_ERR(inode)) {
2361                         ret = PTR_ERR(inode);
2362                         goto out;
2363                 }
2364
2365                 /*
2366                  * add this inode to the orphan list so btrfs_orphan_del does
2367                  * the proper thing when we hit it
2368                  */
2369                 spin_lock(&root->orphan_lock);
2370                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2371                 spin_unlock(&root->orphan_lock);
2372
2373                 /*
2374                  * if this is a bad inode, means we actually succeeded in
2375                  * removing the inode, but not the orphan record, which means
2376                  * we need to manually delete the orphan since iput will just
2377                  * do a destroy_inode
2378                  */
2379                 if (is_bad_inode(inode)) {
2380                         trans = btrfs_start_transaction(root, 0);
2381                         if (IS_ERR(trans)) {
2382                                 ret = PTR_ERR(trans);
2383                                 goto out;
2384                         }
2385                         btrfs_orphan_del(trans, inode);
2386                         btrfs_end_transaction(trans, root);
2387                         iput(inode);
2388                         continue;
2389                 }
2390
2391                 /* if we have links, this was a truncate, lets do that */
2392                 if (inode->i_nlink) {
2393                         if (!S_ISREG(inode->i_mode)) {
2394                                 WARN_ON(1);
2395                                 iput(inode);
2396                                 continue;
2397                         }
2398                         nr_truncate++;
2399                         ret = btrfs_truncate(inode);
2400                 } else {
2401                         nr_unlink++;
2402                 }
2403
2404                 /* this will do delete_inode and everything for us */
2405                 iput(inode);
2406                 if (ret)
2407                         goto out;
2408         }
2409         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2410
2411         if (root->orphan_block_rsv)
2412                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2413                                         (u64)-1);
2414
2415         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2416                 trans = btrfs_join_transaction(root, 1);
2417                 if (!IS_ERR(trans))
2418                         btrfs_end_transaction(trans, root);
2419         }
2420
2421         if (nr_unlink)
2422                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2423         if (nr_truncate)
2424                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2425
2426 out:
2427         if (ret)
2428                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2429         btrfs_free_path(path);
2430         return ret;
2431 }
2432
2433 /*
2434  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2435  * don't find any xattrs, we know there can't be any acls.
2436  *
2437  * slot is the slot the inode is in, objectid is the objectid of the inode
2438  */
2439 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2440                                           int slot, u64 objectid)
2441 {
2442         u32 nritems = btrfs_header_nritems(leaf);
2443         struct btrfs_key found_key;
2444         int scanned = 0;
2445
2446         slot++;
2447         while (slot < nritems) {
2448                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2449
2450                 /* we found a different objectid, there must not be acls */
2451                 if (found_key.objectid != objectid)
2452                         return 0;
2453
2454                 /* we found an xattr, assume we've got an acl */
2455                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2456                         return 1;
2457
2458                 /*
2459                  * we found a key greater than an xattr key, there can't
2460                  * be any acls later on
2461                  */
2462                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2463                         return 0;
2464
2465                 slot++;
2466                 scanned++;
2467
2468                 /*
2469                  * it goes inode, inode backrefs, xattrs, extents,
2470                  * so if there are a ton of hard links to an inode there can
2471                  * be a lot of backrefs.  Don't waste time searching too hard,
2472                  * this is just an optimization
2473                  */
2474                 if (scanned >= 8)
2475                         break;
2476         }
2477         /* we hit the end of the leaf before we found an xattr or
2478          * something larger than an xattr.  We have to assume the inode
2479          * has acls
2480          */
2481         return 1;
2482 }
2483
2484 /*
2485  * read an inode from the btree into the in-memory inode
2486  */
2487 static void btrfs_read_locked_inode(struct inode *inode)
2488 {
2489         struct btrfs_path *path;
2490         struct extent_buffer *leaf;
2491         struct btrfs_inode_item *inode_item;
2492         struct btrfs_timespec *tspec;
2493         struct btrfs_root *root = BTRFS_I(inode)->root;
2494         struct btrfs_key location;
2495         int maybe_acls;
2496         u64 alloc_group_block;
2497         u32 rdev;
2498         int ret;
2499
2500         path = btrfs_alloc_path();
2501         BUG_ON(!path);
2502         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2503
2504         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2505         if (ret)
2506                 goto make_bad;
2507
2508         leaf = path->nodes[0];
2509         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2510                                     struct btrfs_inode_item);
2511
2512         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2513         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2514         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2515         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2516         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2517
2518         tspec = btrfs_inode_atime(inode_item);
2519         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2520         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521
2522         tspec = btrfs_inode_mtime(inode_item);
2523         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2524         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525
2526         tspec = btrfs_inode_ctime(inode_item);
2527         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2528         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2529
2530         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2531         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2532         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2533         inode->i_generation = BTRFS_I(inode)->generation;
2534         inode->i_rdev = 0;
2535         rdev = btrfs_inode_rdev(leaf, inode_item);
2536
2537         BTRFS_I(inode)->index_cnt = (u64)-1;
2538         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2539
2540         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2541
2542         /*
2543          * try to precache a NULL acl entry for files that don't have
2544          * any xattrs or acls
2545          */
2546         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547         if (!maybe_acls)
2548                 cache_no_acl(inode);
2549
2550         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2551                                                 alloc_group_block, 0);
2552         btrfs_free_path(path);
2553         inode_item = NULL;
2554
2555         switch (inode->i_mode & S_IFMT) {
2556         case S_IFREG:
2557                 inode->i_mapping->a_ops = &btrfs_aops;
2558                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2559                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2560                 inode->i_fop = &btrfs_file_operations;
2561                 inode->i_op = &btrfs_file_inode_operations;
2562                 break;
2563         case S_IFDIR:
2564                 inode->i_fop = &btrfs_dir_file_operations;
2565                 if (root == root->fs_info->tree_root)
2566                         inode->i_op = &btrfs_dir_ro_inode_operations;
2567                 else
2568                         inode->i_op = &btrfs_dir_inode_operations;
2569                 break;
2570         case S_IFLNK:
2571                 inode->i_op = &btrfs_symlink_inode_operations;
2572                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574                 break;
2575         default:
2576                 inode->i_op = &btrfs_special_inode_operations;
2577                 init_special_inode(inode, inode->i_mode, rdev);
2578                 break;
2579         }
2580
2581         btrfs_update_iflags(inode);
2582         return;
2583
2584 make_bad:
2585         btrfs_free_path(path);
2586         make_bad_inode(inode);
2587 }
2588
2589 /*
2590  * given a leaf and an inode, copy the inode fields into the leaf
2591  */
2592 static void fill_inode_item(struct btrfs_trans_handle *trans,
2593                             struct extent_buffer *leaf,
2594                             struct btrfs_inode_item *item,
2595                             struct inode *inode)
2596 {
2597         if (!leaf->map_token)
2598                 map_private_extent_buffer(leaf, (unsigned long)item,
2599                                           sizeof(struct btrfs_inode_item),
2600                                           &leaf->map_token, &leaf->kaddr,
2601                                           &leaf->map_start, &leaf->map_len,
2602                                           KM_USER1);
2603
2604         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2605         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2606         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2607         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2608         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2609
2610         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2611                                inode->i_atime.tv_sec);
2612         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2613                                 inode->i_atime.tv_nsec);
2614
2615         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2616                                inode->i_mtime.tv_sec);
2617         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2618                                 inode->i_mtime.tv_nsec);
2619
2620         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2621                                inode->i_ctime.tv_sec);
2622         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2623                                 inode->i_ctime.tv_nsec);
2624
2625         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2626         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2627         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2628         btrfs_set_inode_transid(leaf, item, trans->transid);
2629         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2630         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2631         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2632
2633         if (leaf->map_token) {
2634                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2635                 leaf->map_token = NULL;
2636         }
2637 }
2638
2639 /*
2640  * copy everything in the in-memory inode into the btree.
2641  */
2642 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2643                                 struct btrfs_root *root, struct inode *inode)
2644 {
2645         struct btrfs_inode_item *inode_item;
2646         struct btrfs_path *path;
2647         struct extent_buffer *leaf;
2648         int ret;
2649
2650         path = btrfs_alloc_path();
2651         BUG_ON(!path);
2652         path->leave_spinning = 1;
2653         ret = btrfs_lookup_inode(trans, root, path,
2654                                  &BTRFS_I(inode)->location, 1);
2655         if (ret) {
2656                 if (ret > 0)
2657                         ret = -ENOENT;
2658                 goto failed;
2659         }
2660
2661         btrfs_unlock_up_safe(path, 1);
2662         leaf = path->nodes[0];
2663         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2664                                   struct btrfs_inode_item);
2665
2666         fill_inode_item(trans, leaf, inode_item, inode);
2667         btrfs_mark_buffer_dirty(leaf);
2668         btrfs_set_inode_last_trans(trans, inode);
2669         ret = 0;
2670 failed:
2671         btrfs_free_path(path);
2672         return ret;
2673 }
2674
2675
2676 /*
2677  * unlink helper that gets used here in inode.c and in the tree logging
2678  * recovery code.  It remove a link in a directory with a given name, and
2679  * also drops the back refs in the inode to the directory
2680  */
2681 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2682                                 struct btrfs_root *root,
2683                                 struct inode *dir, struct inode *inode,
2684                                 const char *name, int name_len)
2685 {
2686         struct btrfs_path *path;
2687         int ret = 0;
2688         struct extent_buffer *leaf;
2689         struct btrfs_dir_item *di;
2690         struct btrfs_key key;
2691         u64 index;
2692
2693         path = btrfs_alloc_path();
2694         if (!path) {
2695                 ret = -ENOMEM;
2696                 goto out;
2697         }
2698
2699         path->leave_spinning = 1;
2700         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2701                                     name, name_len, -1);
2702         if (IS_ERR(di)) {
2703                 ret = PTR_ERR(di);
2704                 goto err;
2705         }
2706         if (!di) {
2707                 ret = -ENOENT;
2708                 goto err;
2709         }
2710         leaf = path->nodes[0];
2711         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2712         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2713         if (ret)
2714                 goto err;
2715         btrfs_release_path(root, path);
2716
2717         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2718                                   inode->i_ino,
2719                                   dir->i_ino, &index);
2720         if (ret) {
2721                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2722                        "inode %lu parent %lu\n", name_len, name,
2723                        inode->i_ino, dir->i_ino);
2724                 goto err;
2725         }
2726
2727         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2728                                          index, name, name_len, -1);
2729         if (IS_ERR(di)) {
2730                 ret = PTR_ERR(di);
2731                 goto err;
2732         }
2733         if (!di) {
2734                 ret = -ENOENT;
2735                 goto err;
2736         }
2737         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2738         btrfs_release_path(root, path);
2739
2740         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2741                                          inode, dir->i_ino);
2742         BUG_ON(ret != 0 && ret != -ENOENT);
2743
2744         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2745                                            dir, index);
2746         if (ret == -ENOENT)
2747                 ret = 0;
2748 err:
2749         btrfs_free_path(path);
2750         if (ret)
2751                 goto out;
2752
2753         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2754         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2755         btrfs_update_inode(trans, root, dir);
2756 out:
2757         return ret;
2758 }
2759
2760 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2761                        struct btrfs_root *root,
2762                        struct inode *dir, struct inode *inode,
2763                        const char *name, int name_len)
2764 {
2765         int ret;
2766         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2767         if (!ret) {
2768                 btrfs_drop_nlink(inode);
2769                 ret = btrfs_update_inode(trans, root, inode);
2770         }
2771         return ret;
2772 }
2773
2774
2775 /* helper to check if there is any shared block in the path */
2776 static int check_path_shared(struct btrfs_root *root,
2777                              struct btrfs_path *path)
2778 {
2779         struct extent_buffer *eb;
2780         int level;
2781         u64 refs = 1;
2782
2783         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2784                 int ret;
2785
2786                 if (!path->nodes[level])
2787                         break;
2788                 eb = path->nodes[level];
2789                 if (!btrfs_block_can_be_shared(root, eb))
2790                         continue;
2791                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2792                                                &refs, NULL);
2793                 if (refs > 1)
2794                         return 1;
2795         }
2796         return 0;
2797 }
2798
2799 /*
2800  * helper to start transaction for unlink and rmdir.
2801  *
2802  * unlink and rmdir are special in btrfs, they do not always free space.
2803  * so in enospc case, we should make sure they will free space before
2804  * allowing them to use the global metadata reservation.
2805  */
2806 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2807                                                        struct dentry *dentry)
2808 {
2809         struct btrfs_trans_handle *trans;
2810         struct btrfs_root *root = BTRFS_I(dir)->root;
2811         struct btrfs_path *path;
2812         struct btrfs_inode_ref *ref;
2813         struct btrfs_dir_item *di;
2814         struct inode *inode = dentry->d_inode;
2815         u64 index;
2816         int check_link = 1;
2817         int err = -ENOSPC;
2818         int ret;
2819
2820         trans = btrfs_start_transaction(root, 10);
2821         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2822                 return trans;
2823
2824         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2825                 return ERR_PTR(-ENOSPC);
2826
2827         /* check if there is someone else holds reference */
2828         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2829                 return ERR_PTR(-ENOSPC);
2830
2831         if (atomic_read(&inode->i_count) > 2)
2832                 return ERR_PTR(-ENOSPC);
2833
2834         if (xchg(&root->fs_info->enospc_unlink, 1))
2835                 return ERR_PTR(-ENOSPC);
2836
2837         path = btrfs_alloc_path();
2838         if (!path) {
2839                 root->fs_info->enospc_unlink = 0;
2840                 return ERR_PTR(-ENOMEM);
2841         }
2842
2843         trans = btrfs_start_transaction(root, 0);
2844         if (IS_ERR(trans)) {
2845                 btrfs_free_path(path);
2846                 root->fs_info->enospc_unlink = 0;
2847                 return trans;
2848         }
2849
2850         path->skip_locking = 1;
2851         path->search_commit_root = 1;
2852
2853         ret = btrfs_lookup_inode(trans, root, path,
2854                                 &BTRFS_I(dir)->location, 0);
2855         if (ret < 0) {
2856                 err = ret;
2857                 goto out;
2858         }
2859         if (ret == 0) {
2860                 if (check_path_shared(root, path))
2861                         goto out;
2862         } else {
2863                 check_link = 0;
2864         }
2865         btrfs_release_path(root, path);
2866
2867         ret = btrfs_lookup_inode(trans, root, path,
2868                                 &BTRFS_I(inode)->location, 0);
2869         if (ret < 0) {
2870                 err = ret;
2871                 goto out;
2872         }
2873         if (ret == 0) {
2874                 if (check_path_shared(root, path))
2875                         goto out;
2876         } else {
2877                 check_link = 0;
2878         }
2879         btrfs_release_path(root, path);
2880
2881         if (ret == 0 && S_ISREG(inode->i_mode)) {
2882                 ret = btrfs_lookup_file_extent(trans, root, path,
2883                                                inode->i_ino, (u64)-1, 0);
2884                 if (ret < 0) {
2885                         err = ret;
2886                         goto out;
2887                 }
2888                 BUG_ON(ret == 0);
2889                 if (check_path_shared(root, path))
2890                         goto out;
2891                 btrfs_release_path(root, path);
2892         }
2893
2894         if (!check_link) {
2895                 err = 0;
2896                 goto out;
2897         }
2898
2899         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2900                                 dentry->d_name.name, dentry->d_name.len, 0);
2901         if (IS_ERR(di)) {
2902                 err = PTR_ERR(di);
2903                 goto out;
2904         }
2905         if (di) {
2906                 if (check_path_shared(root, path))
2907                         goto out;
2908         } else {
2909                 err = 0;
2910                 goto out;
2911         }
2912         btrfs_release_path(root, path);
2913
2914         ref = btrfs_lookup_inode_ref(trans, root, path,
2915                                 dentry->d_name.name, dentry->d_name.len,
2916                                 inode->i_ino, dir->i_ino, 0);
2917         if (IS_ERR(ref)) {
2918                 err = PTR_ERR(ref);
2919                 goto out;
2920         }
2921         BUG_ON(!ref);
2922         if (check_path_shared(root, path))
2923                 goto out;
2924         index = btrfs_inode_ref_index(path->nodes[0], ref);
2925         btrfs_release_path(root, path);
2926
2927         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2928                                 dentry->d_name.name, dentry->d_name.len, 0);
2929         if (IS_ERR(di)) {
2930                 err = PTR_ERR(di);
2931                 goto out;
2932         }
2933         BUG_ON(ret == -ENOENT);
2934         if (check_path_shared(root, path))
2935                 goto out;
2936
2937         err = 0;
2938 out:
2939         btrfs_free_path(path);
2940         if (err) {
2941                 btrfs_end_transaction(trans, root);
2942                 root->fs_info->enospc_unlink = 0;
2943                 return ERR_PTR(err);
2944         }
2945
2946         trans->block_rsv = &root->fs_info->global_block_rsv;
2947         return trans;
2948 }
2949
2950 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2951                                struct btrfs_root *root)
2952 {
2953         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2954                 BUG_ON(!root->fs_info->enospc_unlink);
2955                 root->fs_info->enospc_unlink = 0;
2956         }
2957         btrfs_end_transaction_throttle(trans, root);
2958 }
2959
2960 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2961 {
2962         struct btrfs_root *root = BTRFS_I(dir)->root;
2963         struct btrfs_trans_handle *trans;
2964         struct inode *inode = dentry->d_inode;
2965         int ret;
2966         unsigned long nr = 0;
2967
2968         trans = __unlink_start_trans(dir, dentry);
2969         if (IS_ERR(trans))
2970                 return PTR_ERR(trans);
2971
2972         btrfs_set_trans_block_group(trans, dir);
2973
2974         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2975
2976         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2977                                  dentry->d_name.name, dentry->d_name.len);
2978         BUG_ON(ret);
2979
2980         if (inode->i_nlink == 0) {
2981                 ret = btrfs_orphan_add(trans, inode);
2982                 BUG_ON(ret);
2983         }
2984
2985         nr = trans->blocks_used;
2986         __unlink_end_trans(trans, root);
2987         btrfs_btree_balance_dirty(root, nr);
2988         return ret;
2989 }
2990
2991 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2992                         struct btrfs_root *root,
2993                         struct inode *dir, u64 objectid,
2994                         const char *name, int name_len)
2995 {
2996         struct btrfs_path *path;
2997         struct extent_buffer *leaf;
2998         struct btrfs_dir_item *di;
2999         struct btrfs_key key;
3000         u64 index;
3001         int ret;
3002
3003         path = btrfs_alloc_path();
3004         if (!path)
3005                 return -ENOMEM;
3006
3007         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3008                                    name, name_len, -1);
3009         BUG_ON(!di || IS_ERR(di));
3010
3011         leaf = path->nodes[0];
3012         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3013         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3014         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3015         BUG_ON(ret);
3016         btrfs_release_path(root, path);
3017
3018         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3019                                  objectid, root->root_key.objectid,
3020                                  dir->i_ino, &index, name, name_len);
3021         if (ret < 0) {
3022                 BUG_ON(ret != -ENOENT);
3023                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3024                                                  name, name_len);
3025                 BUG_ON(!di || IS_ERR(di));
3026
3027                 leaf = path->nodes[0];
3028                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3029                 btrfs_release_path(root, path);
3030                 index = key.offset;
3031         }
3032
3033         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3034                                          index, name, name_len, -1);
3035         BUG_ON(!di || IS_ERR(di));
3036
3037         leaf = path->nodes[0];
3038         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3039         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3040         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3041         BUG_ON(ret);
3042         btrfs_release_path(root, path);
3043
3044         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3045         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3046         ret = btrfs_update_inode(trans, root, dir);
3047         BUG_ON(ret);
3048
3049         btrfs_free_path(path);
3050         return 0;
3051 }
3052
3053 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3054 {
3055         struct inode *inode = dentry->d_inode;
3056         int err = 0;
3057         struct btrfs_root *root = BTRFS_I(dir)->root;
3058         struct btrfs_trans_handle *trans;
3059         unsigned long nr = 0;
3060
3061         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3062             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3063                 return -ENOTEMPTY;
3064
3065         trans = __unlink_start_trans(dir, dentry);
3066         if (IS_ERR(trans))
3067                 return PTR_ERR(trans);
3068
3069         btrfs_set_trans_block_group(trans, dir);
3070
3071         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3072                 err = btrfs_unlink_subvol(trans, root, dir,
3073                                           BTRFS_I(inode)->location.objectid,
3074                                           dentry->d_name.name,
3075                                           dentry->d_name.len);
3076                 goto out;
3077         }
3078
3079         err = btrfs_orphan_add(trans, inode);
3080         if (err)
3081                 goto out;
3082
3083         /* now the directory is empty */
3084         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085                                  dentry->d_name.name, dentry->d_name.len);
3086         if (!err)
3087                 btrfs_i_size_write(inode, 0);
3088 out:
3089         nr = trans->blocks_used;
3090         __unlink_end_trans(trans, root);
3091         btrfs_btree_balance_dirty(root, nr);
3092
3093         return err;
3094 }
3095
3096 #if 0
3097 /*
3098  * when truncating bytes in a file, it is possible to avoid reading
3099  * the leaves that contain only checksum items.  This can be the
3100  * majority of the IO required to delete a large file, but it must
3101  * be done carefully.
3102  *
3103  * The keys in the level just above the leaves are checked to make sure
3104  * the lowest key in a given leaf is a csum key, and starts at an offset
3105  * after the new  size.
3106  *
3107  * Then the key for the next leaf is checked to make sure it also has
3108  * a checksum item for the same file.  If it does, we know our target leaf
3109  * contains only checksum items, and it can be safely freed without reading
3110  * it.
3111  *
3112  * This is just an optimization targeted at large files.  It may do
3113  * nothing.  It will return 0 unless things went badly.
3114  */
3115 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3116                                      struct btrfs_root *root,
3117                                      struct btrfs_path *path,
3118                                      struct inode *inode, u64 new_size)
3119 {
3120         struct btrfs_key key;
3121         int ret;
3122         int nritems;
3123         struct btrfs_key found_key;
3124         struct btrfs_key other_key;
3125         struct btrfs_leaf_ref *ref;
3126         u64 leaf_gen;
3127         u64 leaf_start;
3128
3129         path->lowest_level = 1;
3130         key.objectid = inode->i_ino;
3131         key.type = BTRFS_CSUM_ITEM_KEY;
3132         key.offset = new_size;
3133 again:
3134         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3135         if (ret < 0)
3136                 goto out;
3137
3138         if (path->nodes[1] == NULL) {
3139                 ret = 0;
3140                 goto out;
3141         }
3142         ret = 0;
3143         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3144         nritems = btrfs_header_nritems(path->nodes[1]);
3145
3146         if (!nritems)
3147                 goto out;
3148
3149         if (path->slots[1] >= nritems)
3150                 goto next_node;
3151
3152         /* did we find a key greater than anything we want to delete? */
3153         if (found_key.objectid > inode->i_ino ||
3154            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3155                 goto out;
3156
3157         /* we check the next key in the node to make sure the leave contains
3158          * only checksum items.  This comparison doesn't work if our
3159          * leaf is the last one in the node
3160          */
3161         if (path->slots[1] + 1 >= nritems) {
3162 next_node:
3163                 /* search forward from the last key in the node, this
3164                  * will bring us into the next node in the tree
3165                  */
3166                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3167
3168                 /* unlikely, but we inc below, so check to be safe */
3169                 if (found_key.offset == (u64)-1)
3170                         goto out;
3171
3172                 /* search_forward needs a path with locks held, do the
3173                  * search again for the original key.  It is possible
3174                  * this will race with a balance and return a path that
3175                  * we could modify, but this drop is just an optimization
3176                  * and is allowed to miss some leaves.
3177                  */
3178                 btrfs_release_path(root, path);
3179                 found_key.offset++;
3180
3181                 /* setup a max key for search_forward */
3182                 other_key.offset = (u64)-1;
3183                 other_key.type = key.type;
3184                 other_key.objectid = key.objectid;
3185
3186                 path->keep_locks = 1;
3187                 ret = btrfs_search_forward(root, &found_key, &other_key,
3188                                            path, 0, 0);
3189                 path->keep_locks = 0;
3190                 if (ret || found_key.objectid != key.objectid ||
3191                     found_key.type != key.type) {
3192                         ret = 0;
3193                         goto out;
3194                 }
3195
3196                 key.offset = found_key.offset;
3197                 btrfs_release_path(root, path);
3198                 cond_resched();
3199                 goto again;
3200         }
3201
3202         /* we know there's one more slot after us in the tree,
3203          * read that key so we can verify it is also a checksum item
3204          */
3205         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3206
3207         if (found_key.objectid < inode->i_ino)
3208                 goto next_key;
3209
3210         if (found_key.type != key.type || found_key.offset < new_size)
3211                 goto next_key;
3212
3213         /*
3214          * if the key for the next leaf isn't a csum key from this objectid,
3215          * we can't be sure there aren't good items inside this leaf.
3216          * Bail out
3217          */
3218         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3219                 goto out;
3220
3221         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3222         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3223         /*
3224          * it is safe to delete this leaf, it contains only
3225          * csum items from this inode at an offset >= new_size
3226          */
3227         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3228         BUG_ON(ret);
3229
3230         if (root->ref_cows && leaf_gen < trans->transid) {
3231                 ref = btrfs_alloc_leaf_ref(root, 0);
3232                 if (ref) {
3233                         ref->root_gen = root->root_key.offset;
3234                         ref->bytenr = leaf_start;
3235                         ref->owner = 0;
3236                         ref->generation = leaf_gen;
3237                         ref->nritems = 0;
3238
3239                         btrfs_sort_leaf_ref(ref);
3240
3241                         ret = btrfs_add_leaf_ref(root, ref, 0);
3242                         WARN_ON(ret);
3243                         btrfs_free_leaf_ref(root, ref);
3244                 } else {
3245                         WARN_ON(1);
3246                 }
3247         }
3248 next_key:
3249         btrfs_release_path(root, path);
3250
3251         if (other_key.objectid == inode->i_ino &&
3252             other_key.type == key.type && other_key.offset > key.offset) {
3253                 key.offset = other_key.offset;
3254                 cond_resched();
3255                 goto again;
3256         }
3257         ret = 0;
3258 out:
3259         /* fixup any changes we've made to the path */
3260         path->lowest_level = 0;
3261         path->keep_locks = 0;
3262         btrfs_release_path(root, path);
3263         return ret;
3264 }
3265
3266 #endif
3267
3268 /*
3269  * this can truncate away extent items, csum items and directory items.
3270  * It starts at a high offset and removes keys until it can't find
3271  * any higher than new_size
3272  *
3273  * csum items that cross the new i_size are truncated to the new size
3274  * as well.
3275  *
3276  * min_type is the minimum key type to truncate down to.  If set to 0, this
3277  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3278  */
3279 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3280                                struct btrfs_root *root,
3281                                struct inode *inode,
3282                                u64 new_size, u32 min_type)
3283 {
3284         struct btrfs_path *path;
3285         struct extent_buffer *leaf;
3286         struct btrfs_file_extent_item *fi;
3287         struct btrfs_key key;
3288         struct btrfs_key found_key;
3289         u64 extent_start = 0;
3290         u64 extent_num_bytes = 0;
3291         u64 extent_offset = 0;
3292         u64 item_end = 0;
3293         u64 mask = root->sectorsize - 1;
3294         u32 found_type = (u8)-1;
3295         int found_extent;
3296         int del_item;
3297         int pending_del_nr = 0;
3298         int pending_del_slot = 0;
3299         int extent_type = -1;
3300         int encoding;
3301         int ret;
3302         int err = 0;
3303
3304         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3305
3306         if (root->ref_cows || root == root->fs_info->tree_root)
3307                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3308
3309         path = btrfs_alloc_path();
3310         BUG_ON(!path);
3311         path->reada = -1;
3312
3313         key.objectid = inode->i_ino;
3314         key.offset = (u64)-1;
3315         key.type = (u8)-1;
3316
3317 search_again:
3318         path->leave_spinning = 1;
3319         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3320         if (ret < 0) {
3321                 err = ret;
3322                 goto out;
3323         }
3324
3325         if (ret > 0) {
3326                 /* there are no items in the tree for us to truncate, we're
3327                  * done
3328                  */
3329                 if (path->slots[0] == 0)
3330                         goto out;
3331                 path->slots[0]--;
3332         }
3333
3334         while (1) {
3335                 fi = NULL;
3336                 leaf = path->nodes[0];
3337                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3338                 found_type = btrfs_key_type(&found_key);
3339                 encoding = 0;
3340
3341                 if (found_key.objectid != inode->i_ino)
3342                         break;
3343
3344                 if (found_type < min_type)
3345                         break;
3346
3347                 item_end = found_key.offset;
3348                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3349                         fi = btrfs_item_ptr(leaf, path->slots[0],
3350                                             struct btrfs_file_extent_item);
3351                         extent_type = btrfs_file_extent_type(leaf, fi);
3352                         encoding = btrfs_file_extent_compression(leaf, fi);
3353                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3354                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3355
3356                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3357                                 item_end +=
3358                                     btrfs_file_extent_num_bytes(leaf, fi);
3359                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3360                                 item_end += btrfs_file_extent_inline_len(leaf,
3361                                                                          fi);
3362                         }
3363                         item_end--;
3364                 }
3365                 if (found_type > min_type) {
3366                         del_item = 1;
3367                 } else {
3368                         if (item_end < new_size)
3369                                 break;
3370                         if (found_key.offset >= new_size)
3371                                 del_item = 1;
3372                         else
3373                                 del_item = 0;
3374                 }
3375                 found_extent = 0;
3376                 /* FIXME, shrink the extent if the ref count is only 1 */
3377                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3378                         goto delete;
3379
3380                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3381                         u64 num_dec;
3382                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3383                         if (!del_item && !encoding) {
3384                                 u64 orig_num_bytes =
3385                                         btrfs_file_extent_num_bytes(leaf, fi);
3386                                 extent_num_bytes = new_size -
3387                                         found_key.offset + root->sectorsize - 1;
3388                                 extent_num_bytes = extent_num_bytes &
3389                                         ~((u64)root->sectorsize - 1);
3390                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3391                                                          extent_num_bytes);
3392                                 num_dec = (orig_num_bytes -
3393                                            extent_num_bytes);
3394                                 if (root->ref_cows && extent_start != 0)
3395                                         inode_sub_bytes(inode, num_dec);
3396                                 btrfs_mark_buffer_dirty(leaf);
3397                         } else {
3398                                 extent_num_bytes =
3399                                         btrfs_file_extent_disk_num_bytes(leaf,
3400                                                                          fi);
3401                                 extent_offset = found_key.offset -
3402                                         btrfs_file_extent_offset(leaf, fi);
3403
3404                                 /* FIXME blocksize != 4096 */
3405                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3406                                 if (extent_start != 0) {
3407                                         found_extent = 1;
3408                                         if (root->ref_cows)
3409                                                 inode_sub_bytes(inode, num_dec);
3410                                 }
3411                         }
3412                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3413                         /*
3414                          * we can't truncate inline items that have had
3415                          * special encodings
3416                          */
3417                         if (!del_item &&
3418                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3419                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3420                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3421                                 u32 size = new_size - found_key.offset;
3422
3423                                 if (root->ref_cows) {
3424                                         inode_sub_bytes(inode, item_end + 1 -
3425                                                         new_size);
3426                                 }
3427                                 size =
3428                                     btrfs_file_extent_calc_inline_size(size);
3429                                 ret = btrfs_truncate_item(trans, root, path,
3430                                                           size, 1);
3431                                 BUG_ON(ret);
3432                         } else if (root->ref_cows) {
3433                                 inode_sub_bytes(inode, item_end + 1 -
3434                                                 found_key.offset);
3435                         }
3436                 }
3437 delete:
3438                 if (del_item) {
3439                         if (!pending_del_nr) {
3440                                 /* no pending yet, add ourselves */
3441                                 pending_del_slot = path->slots[0];
3442                                 pending_del_nr = 1;
3443                         } else if (pending_del_nr &&
3444                                    path->slots[0] + 1 == pending_del_slot) {
3445                                 /* hop on the pending chunk */
3446                                 pending_del_nr++;
3447                                 pending_del_slot = path->slots[0];
3448                         } else {
3449                                 BUG();
3450                         }
3451                 } else {
3452                         break;
3453                 }
3454                 if (found_extent && (root->ref_cows ||
3455                                      root == root->fs_info->tree_root)) {
3456                         btrfs_set_path_blocking(path);
3457                         ret = btrfs_free_extent(trans, root, extent_start,
3458                                                 extent_num_bytes, 0,
3459                                                 btrfs_header_owner(leaf),
3460                                                 inode->i_ino, extent_offset);
3461                         BUG_ON(ret);
3462                 }
3463
3464                 if (found_type == BTRFS_INODE_ITEM_KEY)
3465                         break;
3466
3467                 if (path->slots[0] == 0 ||
3468                     path->slots[0] != pending_del_slot) {
3469                         if (root->ref_cows) {
3470                                 err = -EAGAIN;
3471                                 goto out;
3472                         }
3473                         if (pending_del_nr) {
3474                                 ret = btrfs_del_items(trans, root, path,
3475                                                 pending_del_slot,
3476                                                 pending_del_nr);
3477                                 BUG_ON(ret);
3478                                 pending_del_nr = 0;
3479                         }
3480                         btrfs_release_path(root, path);
3481                         goto search_again;
3482                 } else {
3483                         path->slots[0]--;
3484                 }
3485         }
3486 out:
3487         if (pending_del_nr) {
3488                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3489                                       pending_del_nr);
3490                 BUG_ON(ret);
3491         }
3492         btrfs_free_path(path);
3493         return err;
3494 }
3495
3496 /*
3497  * taken from block_truncate_page, but does cow as it zeros out
3498  * any bytes left in the last page in the file.
3499  */
3500 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3501 {
3502         struct inode *inode = mapping->host;
3503         struct btrfs_root *root = BTRFS_I(inode)->root;
3504         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3505         struct btrfs_ordered_extent *ordered;
3506         struct extent_state *cached_state = NULL;
3507         char *kaddr;
3508         u32 blocksize = root->sectorsize;
3509         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3510         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3511         struct page *page;
3512         int ret = 0;
3513         u64 page_start;
3514         u64 page_end;
3515
3516         if ((offset & (blocksize - 1)) == 0)
3517                 goto out;
3518         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3519         if (ret)
3520                 goto out;
3521
3522         ret = -ENOMEM;
3523 again:
3524         page = grab_cache_page(mapping, index);
3525         if (!page) {
3526                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3527                 goto out;
3528         }
3529
3530         page_start = page_offset(page);
3531         page_end = page_start + PAGE_CACHE_SIZE - 1;
3532
3533         if (!PageUptodate(page)) {
3534                 ret = btrfs_readpage(NULL, page);
3535                 lock_page(page);
3536                 if (page->mapping != mapping) {
3537                         unlock_page(page);
3538                         page_cache_release(page);
3539                         goto again;
3540                 }
3541                 if (!PageUptodate(page)) {
3542                         ret = -EIO;
3543                         goto out_unlock;
3544                 }
3545         }
3546         wait_on_page_writeback(page);
3547
3548         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3549                          GFP_NOFS);
3550         set_page_extent_mapped(page);
3551
3552         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3553         if (ordered) {
3554                 unlock_extent_cached(io_tree, page_start, page_end,
3555                                      &cached_state, GFP_NOFS);
3556                 unlock_page(page);
3557                 page_cache_release(page);
3558                 btrfs_start_ordered_extent(inode, ordered, 1);
3559                 btrfs_put_ordered_extent(ordered);
3560                 goto again;
3561         }
3562
3563         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3564                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3565                           0, 0, &cached_state, GFP_NOFS);
3566
3567         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3568                                         &cached_state);
3569         if (ret) {
3570                 unlock_extent_cached(io_tree, page_start, page_end,
3571                                      &cached_state, GFP_NOFS);
3572                 goto out_unlock;
3573         }
3574
3575         ret = 0;
3576         if (offset != PAGE_CACHE_SIZE) {
3577                 kaddr = kmap(page);
3578                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3579                 flush_dcache_page(page);
3580                 kunmap(page);
3581         }
3582         ClearPageChecked(page);
3583         set_page_dirty(page);
3584         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3585                              GFP_NOFS);
3586
3587 out_unlock:
3588         if (ret)
3589                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3590         unlock_page(page);
3591         page_cache_release(page);
3592 out:
3593         return ret;
3594 }
3595
3596 /*
3597  * This function puts in dummy file extents for the area we're creating a hole
3598  * for.  So if we are truncating this file to a larger size we need to insert
3599  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3600  * the range between oldsize and size
3601  */
3602 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3603 {
3604         struct btrfs_trans_handle *trans;
3605         struct btrfs_root *root = BTRFS_I(inode)->root;
3606         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3607         struct extent_map *em = NULL;
3608         struct extent_state *cached_state = NULL;
3609         u64 mask = root->sectorsize - 1;
3610         u64 hole_start = (oldsize + mask) & ~mask;
3611         u64 block_end = (size + mask) & ~mask;
3612         u64 last_byte;
3613         u64 cur_offset;
3614         u64 hole_size;
3615         int err = 0;
3616
3617         if (size <= hole_start)
3618                 return 0;
3619
3620         while (1) {
3621                 struct btrfs_ordered_extent *ordered;
3622                 btrfs_wait_ordered_range(inode, hole_start,
3623                                          block_end - hole_start);
3624                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3625                                  &cached_state, GFP_NOFS);
3626                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3627                 if (!ordered)
3628                         break;
3629                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3630                                      &cached_state, GFP_NOFS);
3631                 btrfs_put_ordered_extent(ordered);
3632         }
3633
3634         cur_offset = hole_start;
3635         while (1) {
3636                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3637                                 block_end - cur_offset, 0);
3638                 BUG_ON(IS_ERR(em) || !em);
3639                 last_byte = min(extent_map_end(em), block_end);
3640                 last_byte = (last_byte + mask) & ~mask;
3641                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3642                         u64 hint_byte = 0;
3643                         hole_size = last_byte - cur_offset;
3644
3645                         trans = btrfs_start_transaction(root, 2);
3646                         if (IS_ERR(trans)) {
3647                                 err = PTR_ERR(trans);
3648                                 break;
3649                         }
3650                         btrfs_set_trans_block_group(trans, inode);
3651
3652                         err = btrfs_drop_extents(trans, inode, cur_offset,
3653                                                  cur_offset + hole_size,
3654                                                  &hint_byte, 1);
3655                         if (err)
3656                                 break;
3657
3658                         err = btrfs_insert_file_extent(trans, root,
3659                                         inode->i_ino, cur_offset, 0,
3660                                         0, hole_size, 0, hole_size,
3661                                         0, 0, 0);
3662                         if (err)
3663                                 break;
3664
3665                         btrfs_drop_extent_cache(inode, hole_start,
3666                                         last_byte - 1, 0);
3667
3668                         btrfs_end_transaction(trans, root);
3669                 }
3670                 free_extent_map(em);
3671                 em = NULL;
3672                 cur_offset = last_byte;
3673                 if (cur_offset >= block_end)
3674                         break;
3675         }
3676
3677         free_extent_map(em);
3678         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3679                              GFP_NOFS);
3680         return err;
3681 }
3682
3683 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3684 {
3685         loff_t oldsize = i_size_read(inode);
3686         int ret;
3687
3688         if (newsize == oldsize)
3689                 return 0;
3690
3691         if (newsize > oldsize) {
3692                 i_size_write(inode, newsize);
3693                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3694                 truncate_pagecache(inode, oldsize, newsize);
3695                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3696                 if (ret) {
3697                         btrfs_setsize(inode, oldsize);
3698                         return ret;
3699                 }
3700
3701                 mark_inode_dirty(inode);
3702         } else {
3703
3704                 /*
3705                  * We're truncating a file that used to have good data down to
3706                  * zero. Make sure it gets into the ordered flush list so that
3707                  * any new writes get down to disk quickly.
3708                  */
3709                 if (newsize == 0)
3710                         BTRFS_I(inode)->ordered_data_close = 1;
3711
3712                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3713                 truncate_setsize(inode, newsize);
3714                 ret = btrfs_truncate(inode);
3715         }
3716
3717         return ret;
3718 }
3719
3720 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3721 {
3722         struct inode *inode = dentry->d_inode;
3723         struct btrfs_root *root = BTRFS_I(inode)->root;
3724         int err;
3725
3726         if (btrfs_root_readonly(root))
3727                 return -EROFS;
3728
3729         err = inode_change_ok(inode, attr);
3730         if (err)
3731                 return err;
3732
3733         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3734                 err = btrfs_setsize(inode, attr->ia_size);
3735                 if (err)
3736                         return err;
3737         }
3738
3739         if (attr->ia_valid) {
3740                 setattr_copy(inode, attr);
3741                 mark_inode_dirty(inode);
3742
3743                 if (attr->ia_valid & ATTR_MODE)
3744                         err = btrfs_acl_chmod(inode);
3745         }
3746
3747         return err;
3748 }
3749
3750 void btrfs_evict_inode(struct inode *inode)
3751 {
3752         struct btrfs_trans_handle *trans;
3753         struct btrfs_root *root = BTRFS_I(inode)->root;
3754         unsigned long nr;
3755         int ret;
3756
3757         trace_btrfs_inode_evict(inode);
3758
3759         truncate_inode_pages(&inode->i_data, 0);
3760         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3761                                root == root->fs_info->tree_root))
3762                 goto no_delete;
3763
3764         if (is_bad_inode(inode)) {
3765                 btrfs_orphan_del(NULL, inode);
3766                 goto no_delete;
3767         }
3768         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3769         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3770
3771         if (root->fs_info->log_root_recovering) {
3772                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3773                 goto no_delete;
3774         }
3775
3776         if (inode->i_nlink > 0) {
3777                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3778                 goto no_delete;
3779         }
3780
3781         btrfs_i_size_write(inode, 0);
3782
3783         while (1) {
3784                 trans = btrfs_start_transaction(root, 0);
3785                 BUG_ON(IS_ERR(trans));
3786                 btrfs_set_trans_block_group(trans, inode);
3787                 trans->block_rsv = root->orphan_block_rsv;
3788
3789                 ret = btrfs_block_rsv_check(trans, root,
3790                                             root->orphan_block_rsv, 0, 5);
3791                 if (ret) {
3792                         BUG_ON(ret != -EAGAIN);
3793                         ret = btrfs_commit_transaction(trans, root);
3794                         BUG_ON(ret);
3795                         continue;
3796                 }
3797
3798                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3799                 if (ret != -EAGAIN)
3800                         break;
3801
3802                 nr = trans->blocks_used;
3803                 btrfs_end_transaction(trans, root);
3804                 trans = NULL;
3805                 btrfs_btree_balance_dirty(root, nr);
3806
3807         }
3808
3809         if (ret == 0) {
3810                 ret = btrfs_orphan_del(trans, inode);
3811                 BUG_ON(ret);
3812         }
3813
3814         nr = trans->blocks_used;
3815         btrfs_end_transaction(trans, root);
3816         btrfs_btree_balance_dirty(root, nr);
3817 no_delete:
3818         end_writeback(inode);
3819         return;
3820 }
3821
3822 /*
3823  * this returns the key found in the dir entry in the location pointer.
3824  * If no dir entries were found, location->objectid is 0.
3825  */
3826 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3827                                struct btrfs_key *location)
3828 {
3829         const char *name = dentry->d_name.name;
3830         int namelen = dentry->d_name.len;
3831         struct btrfs_dir_item *di;
3832         struct btrfs_path *path;
3833         struct btrfs_root *root = BTRFS_I(dir)->root;
3834         int ret = 0;
3835
3836         path = btrfs_alloc_path();
3837         BUG_ON(!path);
3838
3839         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3840                                     namelen, 0);
3841         if (IS_ERR(di))
3842                 ret = PTR_ERR(di);
3843
3844         if (!di || IS_ERR(di))
3845                 goto out_err;
3846
3847         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3848 out:
3849         btrfs_free_path(path);
3850         return ret;
3851 out_err:
3852         location->objectid = 0;
3853         goto out;
3854 }
3855
3856 /*
3857  * when we hit a tree root in a directory, the btrfs part of the inode
3858  * needs to be changed to reflect the root directory of the tree root.  This
3859  * is kind of like crossing a mount point.
3860  */
3861 static int fixup_tree_root_location(struct btrfs_root *root,
3862                                     struct inode *dir,
3863                                     struct dentry *dentry,
3864                                     struct btrfs_key *location,
3865                                     struct btrfs_root **sub_root)
3866 {
3867         struct btrfs_path *path;
3868         struct btrfs_root *new_root;
3869         struct btrfs_root_ref *ref;
3870         struct extent_buffer *leaf;
3871         int ret;
3872         int err = 0;
3873
3874         path = btrfs_alloc_path();
3875         if (!path) {
3876                 err = -ENOMEM;
3877                 goto out;
3878         }
3879
3880         err = -ENOENT;
3881         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3882                                   BTRFS_I(dir)->root->root_key.objectid,
3883                                   location->objectid);
3884         if (ret) {
3885                 if (ret < 0)
3886                         err = ret;
3887                 goto out;
3888         }
3889
3890         leaf = path->nodes[0];
3891         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3892         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3893             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3894                 goto out;
3895
3896         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3897                                    (unsigned long)(ref + 1),
3898                                    dentry->d_name.len);
3899         if (ret)
3900                 goto out;
3901
3902         btrfs_release_path(root->fs_info->tree_root, path);
3903
3904         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3905         if (IS_ERR(new_root)) {
3906                 err = PTR_ERR(new_root);
3907                 goto out;
3908         }
3909
3910         if (btrfs_root_refs(&new_root->root_item) == 0) {
3911                 err = -ENOENT;
3912                 goto out;
3913         }
3914
3915         *sub_root = new_root;
3916         location->objectid = btrfs_root_dirid(&new_root->root_item);
3917         location->type = BTRFS_INODE_ITEM_KEY;
3918         location->offset = 0;
3919         err = 0;
3920 out:
3921         btrfs_free_path(path);
3922         return err;
3923 }
3924
3925 static void inode_tree_add(struct inode *inode)
3926 {
3927         struct btrfs_root *root = BTRFS_I(inode)->root;
3928         struct btrfs_inode *entry;
3929         struct rb_node **p;
3930         struct rb_node *parent;
3931 again:
3932         p = &root->inode_tree.rb_node;
3933         parent = NULL;
3934
3935         if (inode_unhashed(inode))
3936                 return;
3937
3938         spin_lock(&root->inode_lock);
3939         while (*p) {
3940                 parent = *p;
3941                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3942
3943                 if (inode->i_ino < entry->vfs_inode.i_ino)
3944                         p = &parent->rb_left;
3945                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3946                         p = &parent->rb_right;
3947                 else {
3948                         WARN_ON(!(entry->vfs_inode.i_state &
3949                                   (I_WILL_FREE | I_FREEING)));
3950                         rb_erase(parent, &root->inode_tree);
3951                         RB_CLEAR_NODE(parent);
3952                         spin_unlock(&root->inode_lock);
3953                         goto again;
3954                 }
3955         }
3956         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3957         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3958         spin_unlock(&root->inode_lock);
3959 }
3960
3961 static void inode_tree_del(struct inode *inode)
3962 {
3963         struct btrfs_root *root = BTRFS_I(inode)->root;
3964         int empty = 0;
3965
3966         spin_lock(&root->inode_lock);
3967         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3968                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3969                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3970                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3971         }
3972         spin_unlock(&root->inode_lock);
3973
3974         /*
3975          * Free space cache has inodes in the tree root, but the tree root has a
3976          * root_refs of 0, so this could end up dropping the tree root as a
3977          * snapshot, so we need the extra !root->fs_info->tree_root check to
3978          * make sure we don't drop it.
3979          */
3980         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3981             root != root->fs_info->tree_root) {
3982                 synchronize_srcu(&root->fs_info->subvol_srcu);
3983                 spin_lock(&root->inode_lock);
3984                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3985                 spin_unlock(&root->inode_lock);
3986                 if (empty)
3987                         btrfs_add_dead_root(root);
3988         }
3989 }
3990
3991 int btrfs_invalidate_inodes(struct btrfs_root *root)
3992 {
3993         struct rb_node *node;
3994         struct rb_node *prev;
3995         struct btrfs_inode *entry;
3996         struct inode *inode;
3997         u64 objectid = 0;
3998
3999         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4000
4001         spin_lock(&root->inode_lock);
4002 again:
4003         node = root->inode_tree.rb_node;
4004         prev = NULL;
4005         while (node) {
4006                 prev = node;
4007                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4008
4009                 if (objectid < entry->vfs_inode.i_ino)
4010                         node = node->rb_left;
4011                 else if (objectid > entry->vfs_inode.i_ino)
4012                         node = node->rb_right;
4013                 else
4014                         break;
4015         }
4016         if (!node) {
4017                 while (prev) {
4018                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4019                         if (objectid <= entry->vfs_inode.i_ino) {
4020                                 node = prev;
4021                                 break;
4022                         }
4023                         prev = rb_next(prev);
4024                 }
4025         }
4026         while (node) {
4027                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4028                 objectid = entry->vfs_inode.i_ino + 1;
4029                 inode = igrab(&entry->vfs_inode);
4030                 if (inode) {
4031                         spin_unlock(&root->inode_lock);
4032                         if (atomic_read(&inode->i_count) > 1)
4033                                 d_prune_aliases(inode);
4034                         /*
4035                          * btrfs_drop_inode will have it removed from
4036                          * the inode cache when its usage count
4037                          * hits zero.
4038                          */
4039                         iput(inode);
4040                         cond_resched();
4041                         spin_lock(&root->inode_lock);
4042                         goto again;
4043                 }
4044
4045                 if (cond_resched_lock(&root->inode_lock))
4046                         goto again;
4047
4048                 node = rb_next(node);
4049         }
4050         spin_unlock(&root->inode_lock);
4051         return 0;
4052 }
4053
4054 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4055 {
4056         struct btrfs_iget_args *args = p;
4057         inode->i_ino = args->ino;
4058         BTRFS_I(inode)->root = args->root;
4059         btrfs_set_inode_space_info(args->root, inode);
4060         return 0;
4061 }
4062
4063 static int btrfs_find_actor(struct inode *inode, void *opaque)
4064 {
4065         struct btrfs_iget_args *args = opaque;
4066         return args->ino == inode->i_ino &&
4067                 args->root == BTRFS_I(inode)->root;
4068 }
4069
4070 static struct inode *btrfs_iget_locked(struct super_block *s,
4071                                        u64 objectid,
4072                                        struct btrfs_root *root)
4073 {
4074         struct inode *inode;
4075         struct btrfs_iget_args args;
4076         args.ino = objectid;
4077         args.root = root;
4078
4079         inode = iget5_locked(s, objectid, btrfs_find_actor,
4080                              btrfs_init_locked_inode,
4081                              (void *)&args);
4082         return inode;
4083 }
4084
4085 /* Get an inode object given its location and corresponding root.
4086  * Returns in *is_new if the inode was read from disk
4087  */
4088 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4089                          struct btrfs_root *root, int *new)
4090 {
4091         struct inode *inode;
4092
4093         inode = btrfs_iget_locked(s, location->objectid, root);
4094         if (!inode)
4095                 return ERR_PTR(-ENOMEM);
4096
4097         if (inode->i_state & I_NEW) {
4098                 BTRFS_I(inode)->root = root;
4099                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4100                 btrfs_read_locked_inode(inode);
4101                 inode_tree_add(inode);
4102                 unlock_new_inode(inode);
4103                 if (new)
4104                         *new = 1;
4105         }
4106
4107         return inode;
4108 }
4109
4110 static struct inode *new_simple_dir(struct super_block *s,
4111                                     struct btrfs_key *key,
4112                                     struct btrfs_root *root)
4113 {
4114         struct inode *inode = new_inode(s);
4115
4116         if (!inode)
4117                 return ERR_PTR(-ENOMEM);
4118
4119         BTRFS_I(inode)->root = root;
4120         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4121         BTRFS_I(inode)->dummy_inode = 1;
4122
4123         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4124         inode->i_op = &simple_dir_inode_operations;
4125         inode->i_fop = &simple_dir_operations;
4126         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4127         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4128
4129         return inode;
4130 }
4131
4132 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4133 {
4134         struct inode *inode;
4135         struct btrfs_root *root = BTRFS_I(dir)->root;
4136         struct btrfs_root *sub_root = root;
4137         struct btrfs_key location;
4138         int index;
4139         int ret;
4140
4141         if (dentry->d_name.len > BTRFS_NAME_LEN)
4142                 return ERR_PTR(-ENAMETOOLONG);
4143
4144         ret = btrfs_inode_by_name(dir, dentry, &location);
4145
4146         if (ret < 0)
4147                 return ERR_PTR(ret);
4148
4149         if (location.objectid == 0)
4150                 return NULL;
4151
4152         if (location.type == BTRFS_INODE_ITEM_KEY) {
4153                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4154                 return inode;
4155         }
4156
4157         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4158
4159         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4160         ret = fixup_tree_root_location(root, dir, dentry,
4161                                        &location, &sub_root);
4162         if (ret < 0) {
4163                 if (ret != -ENOENT)
4164                         inode = ERR_PTR(ret);
4165                 else
4166                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4167         } else {
4168                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4169         }
4170         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4171
4172         if (!IS_ERR(inode) && root != sub_root) {
4173                 down_read(&root->fs_info->cleanup_work_sem);
4174                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4175                         ret = btrfs_orphan_cleanup(sub_root);
4176                 up_read(&root->fs_info->cleanup_work_sem);
4177                 if (ret)
4178                         inode = ERR_PTR(ret);
4179         }
4180
4181         return inode;
4182 }
4183
4184 static int btrfs_dentry_delete(const struct dentry *dentry)
4185 {
4186         struct btrfs_root *root;
4187
4188         if (!dentry->d_inode && !IS_ROOT(dentry))
4189                 dentry = dentry->d_parent;
4190
4191         if (dentry->d_inode) {
4192                 root = BTRFS_I(dentry->d_inode)->root;
4193                 if (btrfs_root_refs(&root->root_item) == 0)
4194                         return 1;
4195         }
4196         return 0;
4197 }
4198
4199 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4200                                    struct nameidata *nd)
4201 {
4202         struct inode *inode;
4203
4204         inode = btrfs_lookup_dentry(dir, dentry);
4205         if (IS_ERR(inode))
4206                 return ERR_CAST(inode);
4207
4208         return d_splice_alias(inode, dentry);
4209 }
4210
4211 static unsigned char btrfs_filetype_table[] = {
4212         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4213 };
4214
4215 static int btrfs_real_readdir(struct file *filp, void *dirent,
4216                               filldir_t filldir)
4217 {
4218         struct inode *inode = filp->f_dentry->d_inode;
4219         struct btrfs_root *root = BTRFS_I(inode)->root;
4220         struct btrfs_item *item;
4221         struct btrfs_dir_item *di;
4222         struct btrfs_key key;
4223         struct btrfs_key found_key;
4224         struct btrfs_path *path;
4225         int ret;
4226         struct extent_buffer *leaf;
4227         int slot;
4228         unsigned char d_type;
4229         int over = 0;
4230         u32 di_cur;
4231         u32 di_total;
4232         u32 di_len;
4233         int key_type = BTRFS_DIR_INDEX_KEY;
4234         char tmp_name[32];
4235         char *name_ptr;
4236         int name_len;
4237
4238         /* FIXME, use a real flag for deciding about the key type */
4239         if (root->fs_info->tree_root == root)
4240                 key_type = BTRFS_DIR_ITEM_KEY;
4241
4242         /* special case for "." */
4243         if (filp->f_pos == 0) {
4244                 over = filldir(dirent, ".", 1,
4245                                1, inode->i_ino,
4246                                DT_DIR);
4247                 if (over)
4248                         return 0;
4249                 filp->f_pos = 1;
4250         }
4251         /* special case for .., just use the back ref */
4252         if (filp->f_pos == 1) {
4253                 u64 pino = parent_ino(filp->f_path.dentry);
4254                 over = filldir(dirent, "..", 2,
4255                                2, pino, DT_DIR);
4256                 if (over)
4257                         return 0;
4258                 filp->f_pos = 2;
4259         }
4260         path = btrfs_alloc_path();
4261         path->reada = 2;
4262
4263         btrfs_set_key_type(&key, key_type);
4264         key.offset = filp->f_pos;
4265         key.objectid = inode->i_ino;
4266
4267         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4268         if (ret < 0)
4269                 goto err;
4270
4271         while (1) {
4272                 leaf = path->nodes[0];
4273                 slot = path->slots[0];
4274                 if (slot >= btrfs_header_nritems(leaf)) {
4275                         ret = btrfs_next_leaf(root, path);
4276                         if (ret < 0)
4277                                 goto err;
4278                         else if (ret > 0)
4279                                 break;
4280                         continue;
4281                 }
4282
4283                 item = btrfs_item_nr(leaf, slot);
4284                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4285
4286                 if (found_key.objectid != key.objectid)
4287                         break;
4288                 if (btrfs_key_type(&found_key) != key_type)
4289                         break;
4290                 if (found_key.offset < filp->f_pos)
4291                         goto next;
4292
4293                 filp->f_pos = found_key.offset;
4294
4295                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4296                 di_cur = 0;
4297                 di_total = btrfs_item_size(leaf, item);
4298
4299                 while (di_cur < di_total) {
4300                         struct btrfs_key location;
4301
4302                         if (verify_dir_item(root, leaf, di))
4303                                 break;
4304
4305                         name_len = btrfs_dir_name_len(leaf, di);
4306                         if (name_len <= sizeof(tmp_name)) {
4307                                 name_ptr = tmp_name;
4308                         } else {
4309                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4310                                 if (!name_ptr) {
4311                                         ret = -ENOMEM;
4312                                         goto err;
4313                                 }
4314                         }
4315                         read_extent_buffer(leaf, name_ptr,
4316                                            (unsigned long)(di + 1), name_len);
4317
4318                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4319                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4320
4321                         /* is this a reference to our own snapshot? If so
4322                          * skip it
4323                          */
4324                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4325                             location.objectid == root->root_key.objectid) {
4326                                 over = 0;
4327                                 goto skip;
4328                         }
4329                         over = filldir(dirent, name_ptr, name_len,
4330                                        found_key.offset, location.objectid,
4331                                        d_type);
4332
4333 skip:
4334                         if (name_ptr != tmp_name)
4335                                 kfree(name_ptr);
4336
4337                         if (over)
4338                                 goto nopos;
4339                         di_len = btrfs_dir_name_len(leaf, di) +
4340                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4341                         di_cur += di_len;
4342                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4343                 }
4344 next:
4345                 path->slots[0]++;
4346         }
4347
4348         /* Reached end of directory/root. Bump pos past the last item. */
4349         if (key_type == BTRFS_DIR_INDEX_KEY)
4350                 /*
4351                  * 32-bit glibc will use getdents64, but then strtol -
4352                  * so the last number we can serve is this.
4353                  */
4354                 filp->f_pos = 0x7fffffff;
4355         else
4356                 filp->f_pos++;
4357 nopos:
4358         ret = 0;
4359 err:
4360         btrfs_free_path(path);
4361         return ret;
4362 }
4363
4364 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4365 {
4366         struct btrfs_root *root = BTRFS_I(inode)->root;
4367         struct btrfs_trans_handle *trans;
4368         int ret = 0;
4369         bool nolock = false;
4370
4371         if (BTRFS_I(inode)->dummy_inode)
4372                 return 0;
4373
4374         smp_mb();
4375         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4376
4377         if (wbc->sync_mode == WB_SYNC_ALL) {
4378                 if (nolock)
4379                         trans = btrfs_join_transaction_nolock(root, 1);
4380                 else
4381                         trans = btrfs_join_transaction(root, 1);
4382                 if (IS_ERR(trans))
4383                         return PTR_ERR(trans);
4384                 btrfs_set_trans_block_group(trans, inode);
4385                 if (nolock)
4386                         ret = btrfs_end_transaction_nolock(trans, root);
4387                 else
4388                         ret = btrfs_commit_transaction(trans, root);
4389         }
4390         return ret;
4391 }
4392
4393 /*
4394  * This is somewhat expensive, updating the tree every time the
4395  * inode changes.  But, it is most likely to find the inode in cache.
4396  * FIXME, needs more benchmarking...there are no reasons other than performance
4397  * to keep or drop this code.
4398  */
4399 void btrfs_dirty_inode(struct inode *inode)
4400 {
4401         struct btrfs_root *root = BTRFS_I(inode)->root;
4402         struct btrfs_trans_handle *trans;
4403         int ret;
4404
4405         if (BTRFS_I(inode)->dummy_inode)
4406                 return;
4407
4408         trans = btrfs_join_transaction(root, 1);
4409         BUG_ON(IS_ERR(trans));
4410         btrfs_set_trans_block_group(trans, inode);
4411
4412         ret = btrfs_update_inode(trans, root, inode);
4413         if (ret && ret == -ENOSPC) {
4414                 /* whoops, lets try again with the full transaction */
4415                 btrfs_end_transaction(trans, root);
4416                 trans = btrfs_start_transaction(root, 1);
4417                 if (IS_ERR(trans)) {
4418                         if (printk_ratelimit()) {
4419                                 printk(KERN_ERR "btrfs: fail to "
4420                                        "dirty  inode %lu error %ld\n",
4421                                        inode->i_ino, PTR_ERR(trans));
4422                         }
4423                         return;
4424                 }
4425                 btrfs_set_trans_block_group(trans, inode);
4426
4427                 ret = btrfs_update_inode(trans, root, inode);
4428                 if (ret) {
4429                         if (printk_ratelimit()) {
4430                                 printk(KERN_ERR "btrfs: fail to "
4431                                        "dirty  inode %lu error %d\n",
4432                                        inode->i_ino, ret);
4433                         }
4434                 }
4435         }
4436         btrfs_end_transaction(trans, root);
4437 }
4438
4439 /*
4440  * find the highest existing sequence number in a directory
4441  * and then set the in-memory index_cnt variable to reflect
4442  * free sequence numbers
4443  */
4444 static int btrfs_set_inode_index_count(struct inode *inode)
4445 {
4446         struct btrfs_root *root = BTRFS_I(inode)->root;
4447         struct btrfs_key key, found_key;
4448         struct btrfs_path *path;
4449         struct extent_buffer *leaf;
4450         int ret;
4451
4452         key.objectid = inode->i_ino;
4453         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4454         key.offset = (u64)-1;
4455
4456         path = btrfs_alloc_path();
4457         if (!path)
4458                 return -ENOMEM;
4459
4460         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4461         if (ret < 0)
4462                 goto out;
4463         /* FIXME: we should be able to handle this */
4464         if (ret == 0)
4465                 goto out;
4466         ret = 0;
4467
4468         /*
4469          * MAGIC NUMBER EXPLANATION:
4470          * since we search a directory based on f_pos we have to start at 2
4471          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4472          * else has to start at 2
4473          */
4474         if (path->slots[0] == 0) {
4475                 BTRFS_I(inode)->index_cnt = 2;
4476                 goto out;
4477         }
4478
4479         path->slots[0]--;
4480
4481         leaf = path->nodes[0];
4482         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4483
4484         if (found_key.objectid != inode->i_ino ||
4485             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4486                 BTRFS_I(inode)->index_cnt = 2;
4487                 goto out;
4488         }
4489
4490         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4491 out:
4492         btrfs_free_path(path);
4493         return ret;
4494 }
4495
4496 /*
4497  * helper to find a free sequence number in a given directory.  This current
4498  * code is very simple, later versions will do smarter things in the btree
4499  */
4500 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4501 {
4502         int ret = 0;
4503
4504         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4505                 ret = btrfs_set_inode_index_count(dir);
4506                 if (ret)
4507                         return ret;
4508         }
4509
4510         *index = BTRFS_I(dir)->index_cnt;
4511         BTRFS_I(dir)->index_cnt++;
4512
4513         return ret;
4514 }
4515
4516 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4517                                      struct btrfs_root *root,
4518                                      struct inode *dir,
4519                                      const char *name, int name_len,
4520                                      u64 ref_objectid, u64 objectid,
4521                                      u64 alloc_hint, int mode, u64 *index)
4522 {
4523         struct inode *inode;
4524         struct btrfs_inode_item *inode_item;
4525         struct btrfs_key *location;
4526         struct btrfs_path *path;
4527         struct btrfs_inode_ref *ref;
4528         struct btrfs_key key[2];
4529         u32 sizes[2];
4530         unsigned long ptr;
4531         int ret;
4532         int owner;
4533
4534         path = btrfs_alloc_path();
4535         BUG_ON(!path);
4536
4537         inode = new_inode(root->fs_info->sb);
4538         if (!inode) {
4539                 btrfs_free_path(path);
4540                 return ERR_PTR(-ENOMEM);
4541         }
4542
4543         if (dir) {
4544                 trace_btrfs_inode_request(dir);
4545
4546                 ret = btrfs_set_inode_index(dir, index);
4547                 if (ret) {
4548                         btrfs_free_path(path);
4549                         iput(inode);
4550                         return ERR_PTR(ret);
4551                 }
4552         }
4553         /*
4554          * index_cnt is ignored for everything but a dir,
4555          * btrfs_get_inode_index_count has an explanation for the magic
4556          * number
4557          */
4558         BTRFS_I(inode)->index_cnt = 2;
4559         BTRFS_I(inode)->root = root;
4560         BTRFS_I(inode)->generation = trans->transid;
4561         inode->i_generation = BTRFS_I(inode)->generation;
4562         btrfs_set_inode_space_info(root, inode);
4563
4564         if (mode & S_IFDIR)
4565                 owner = 0;
4566         else
4567                 owner = 1;
4568         BTRFS_I(inode)->block_group =
4569                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4570
4571         key[0].objectid = objectid;
4572         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4573         key[0].offset = 0;
4574
4575         key[1].objectid = objectid;
4576         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4577         key[1].offset = ref_objectid;
4578
4579         sizes[0] = sizeof(struct btrfs_inode_item);
4580         sizes[1] = name_len + sizeof(*ref);
4581
4582         path->leave_spinning = 1;
4583         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4584         if (ret != 0)
4585                 goto fail;
4586
4587         inode_init_owner(inode, dir, mode);
4588         inode->i_ino = objectid;
4589         inode_set_bytes(inode, 0);
4590         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4591         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4592                                   struct btrfs_inode_item);
4593         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4594
4595         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4596                              struct btrfs_inode_ref);
4597         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4598         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4599         ptr = (unsigned long)(ref + 1);
4600         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4601
4602         btrfs_mark_buffer_dirty(path->nodes[0]);
4603         btrfs_free_path(path);
4604
4605         location = &BTRFS_I(inode)->location;
4606         location->objectid = objectid;
4607         location->offset = 0;
4608         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4609
4610         btrfs_inherit_iflags(inode, dir);
4611
4612         if ((mode & S_IFREG)) {
4613                 if (btrfs_test_opt(root, NODATASUM))
4614                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4615                 if (btrfs_test_opt(root, NODATACOW) ||
4616                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4617                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4618         }
4619
4620         insert_inode_hash(inode);
4621         inode_tree_add(inode);
4622
4623         trace_btrfs_inode_new(inode);
4624
4625         return inode;
4626 fail:
4627         if (dir)
4628                 BTRFS_I(dir)->index_cnt--;
4629         btrfs_free_path(path);
4630         iput(inode);
4631         return ERR_PTR(ret);
4632 }
4633
4634 static inline u8 btrfs_inode_type(struct inode *inode)
4635 {
4636         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4637 }
4638
4639 /*
4640  * utility function to add 'inode' into 'parent_inode' with
4641  * a give name and a given sequence number.
4642  * if 'add_backref' is true, also insert a backref from the
4643  * inode to the parent directory.
4644  */
4645 int btrfs_add_link(struct btrfs_trans_handle *trans,
4646                    struct inode *parent_inode, struct inode *inode,
4647                    const char *name, int name_len, int add_backref, u64 index)
4648 {
4649         int ret = 0;
4650         struct btrfs_key key;
4651         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4652
4653         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4654                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4655         } else {
4656                 key.objectid = inode->i_ino;
4657                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4658                 key.offset = 0;
4659         }
4660
4661         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4662                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4663                                          key.objectid, root->root_key.objectid,
4664                                          parent_inode->i_ino,
4665                                          index, name, name_len);
4666         } else if (add_backref) {
4667                 ret = btrfs_insert_inode_ref(trans, root,
4668                                              name, name_len, inode->i_ino,
4669                                              parent_inode->i_ino, index);
4670         }
4671
4672         if (ret == 0) {
4673                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4674                                             parent_inode->i_ino, &key,
4675                                             btrfs_inode_type(inode), index);
4676                 BUG_ON(ret);
4677
4678                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4679                                    name_len * 2);
4680                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4681                 ret = btrfs_update_inode(trans, root, parent_inode);
4682         }
4683         return ret;
4684 }
4685
4686 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4687                             struct inode *dir, struct dentry *dentry,
4688                             struct inode *inode, int backref, u64 index)
4689 {
4690         int err = btrfs_add_link(trans, dir, inode,
4691                                  dentry->d_name.name, dentry->d_name.len,
4692                                  backref, index);
4693         if (!err) {
4694                 d_instantiate(dentry, inode);
4695                 return 0;
4696         }
4697         if (err > 0)
4698                 err = -EEXIST;
4699         return err;
4700 }
4701
4702 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4703                         int mode, dev_t rdev)
4704 {
4705         struct btrfs_trans_handle *trans;
4706         struct btrfs_root *root = BTRFS_I(dir)->root;
4707         struct inode *inode = NULL;
4708         int err;
4709         int drop_inode = 0;
4710         u64 objectid;
4711         unsigned long nr = 0;
4712         u64 index = 0;
4713
4714         if (!new_valid_dev(rdev))
4715                 return -EINVAL;
4716
4717         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4718         if (err)
4719                 return err;
4720
4721         /*
4722          * 2 for inode item and ref
4723          * 2 for dir items
4724          * 1 for xattr if selinux is on
4725          */
4726         trans = btrfs_start_transaction(root, 5);
4727         if (IS_ERR(trans))
4728                 return PTR_ERR(trans);
4729
4730         btrfs_set_trans_block_group(trans, dir);
4731
4732         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4733                                 dentry->d_name.len, dir->i_ino, objectid,
4734                                 BTRFS_I(dir)->block_group, mode, &index);
4735         if (IS_ERR(inode)) {
4736                 err = PTR_ERR(inode);
4737                 goto out_unlock;
4738         }
4739
4740         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4741         if (err) {
4742                 drop_inode = 1;
4743                 goto out_unlock;
4744         }
4745
4746         btrfs_set_trans_block_group(trans, inode);
4747         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4748         if (err)
4749                 drop_inode = 1;
4750         else {
4751                 inode->i_op = &btrfs_special_inode_operations;
4752                 init_special_inode(inode, inode->i_mode, rdev);
4753                 btrfs_update_inode(trans, root, inode);
4754         }
4755         btrfs_update_inode_block_group(trans, inode);
4756         btrfs_update_inode_block_group(trans, dir);
4757 out_unlock:
4758         nr = trans->blocks_used;
4759         btrfs_end_transaction_throttle(trans, root);
4760         btrfs_btree_balance_dirty(root, nr);
4761         if (drop_inode) {
4762                 inode_dec_link_count(inode);
4763                 iput(inode);
4764         }
4765         return err;
4766 }
4767
4768 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4769                         int mode, struct nameidata *nd)
4770 {
4771         struct btrfs_trans_handle *trans;
4772         struct btrfs_root *root = BTRFS_I(dir)->root;
4773         struct inode *inode = NULL;
4774         int drop_inode = 0;
4775         int err;
4776         unsigned long nr = 0;
4777         u64 objectid;
4778         u64 index = 0;
4779
4780         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4781         if (err)
4782                 return err;
4783         /*
4784          * 2 for inode item and ref
4785          * 2 for dir items
4786          * 1 for xattr if selinux is on
4787          */
4788         trans = btrfs_start_transaction(root, 5);
4789         if (IS_ERR(trans))
4790                 return PTR_ERR(trans);
4791
4792         btrfs_set_trans_block_group(trans, dir);
4793
4794         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4795                                 dentry->d_name.len, dir->i_ino, objectid,
4796                                 BTRFS_I(dir)->block_group, mode, &index);
4797         if (IS_ERR(inode)) {
4798                 err = PTR_ERR(inode);
4799                 goto out_unlock;
4800         }
4801
4802         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4803         if (err) {
4804                 drop_inode = 1;
4805                 goto out_unlock;
4806         }
4807
4808         btrfs_set_trans_block_group(trans, inode);
4809         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4810         if (err)
4811                 drop_inode = 1;
4812         else {
4813                 inode->i_mapping->a_ops = &btrfs_aops;
4814                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4815                 inode->i_fop = &btrfs_file_operations;
4816                 inode->i_op = &btrfs_file_inode_operations;
4817                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4818         }
4819         btrfs_update_inode_block_group(trans, inode);
4820         btrfs_update_inode_block_group(trans, dir);
4821 out_unlock:
4822         nr = trans->blocks_used;
4823         btrfs_end_transaction_throttle(trans, root);
4824         if (drop_inode) {
4825                 inode_dec_link_count(inode);
4826                 iput(inode);
4827         }
4828         btrfs_btree_balance_dirty(root, nr);
4829         return err;
4830 }
4831
4832 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4833                       struct dentry *dentry)
4834 {
4835         struct btrfs_trans_handle *trans;
4836         struct btrfs_root *root = BTRFS_I(dir)->root;
4837         struct inode *inode = old_dentry->d_inode;
4838         u64 index;
4839         unsigned long nr = 0;
4840         int err;
4841         int drop_inode = 0;
4842
4843         /* do not allow sys_link's with other subvols of the same device */
4844         if (root->objectid != BTRFS_I(inode)->root->objectid)
4845                 return -EXDEV;
4846
4847         if (inode->i_nlink == ~0U)
4848                 return -EMLINK;
4849
4850         err = btrfs_set_inode_index(dir, &index);
4851         if (err)
4852                 goto fail;
4853
4854         /*
4855          * 2 items for inode and inode ref
4856          * 2 items for dir items
4857          * 1 item for parent inode
4858          */
4859         trans = btrfs_start_transaction(root, 5);
4860         if (IS_ERR(trans)) {
4861                 err = PTR_ERR(trans);
4862                 goto fail;
4863         }
4864
4865         btrfs_inc_nlink(inode);
4866         inode->i_ctime = CURRENT_TIME;
4867
4868         btrfs_set_trans_block_group(trans, dir);
4869         ihold(inode);
4870
4871         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4872
4873         if (err) {
4874                 drop_inode = 1;
4875         } else {
4876                 struct dentry *parent = dget_parent(dentry);
4877                 btrfs_update_inode_block_group(trans, dir);
4878                 err = btrfs_update_inode(trans, root, inode);
4879                 BUG_ON(err);
4880                 btrfs_log_new_name(trans, inode, NULL, parent);
4881                 dput(parent);
4882         }
4883
4884         nr = trans->blocks_used;
4885         btrfs_end_transaction_throttle(trans, root);
4886 fail:
4887         if (drop_inode) {
4888                 inode_dec_link_count(inode);
4889                 iput(inode);
4890         }
4891         btrfs_btree_balance_dirty(root, nr);
4892         return err;
4893 }
4894
4895 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4896 {
4897         struct inode *inode = NULL;
4898         struct btrfs_trans_handle *trans;
4899         struct btrfs_root *root = BTRFS_I(dir)->root;
4900         int err = 0;
4901         int drop_on_err = 0;
4902         u64 objectid = 0;
4903         u64 index = 0;
4904         unsigned long nr = 1;
4905
4906         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4907         if (err)
4908                 return err;
4909
4910         /*
4911          * 2 items for inode and ref
4912          * 2 items for dir items
4913          * 1 for xattr if selinux is on
4914          */
4915         trans = btrfs_start_transaction(root, 5);
4916         if (IS_ERR(trans))
4917                 return PTR_ERR(trans);
4918         btrfs_set_trans_block_group(trans, dir);
4919
4920         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4921                                 dentry->d_name.len, dir->i_ino, objectid,
4922                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4923                                 &index);
4924         if (IS_ERR(inode)) {
4925                 err = PTR_ERR(inode);
4926                 goto out_fail;
4927         }
4928
4929         drop_on_err = 1;
4930
4931         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4932         if (err)
4933                 goto out_fail;
4934
4935         inode->i_op = &btrfs_dir_inode_operations;
4936         inode->i_fop = &btrfs_dir_file_operations;
4937         btrfs_set_trans_block_group(trans, inode);
4938
4939         btrfs_i_size_write(inode, 0);
4940         err = btrfs_update_inode(trans, root, inode);
4941         if (err)
4942                 goto out_fail;
4943
4944         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4945                              dentry->d_name.len, 0, index);
4946         if (err)
4947                 goto out_fail;
4948
4949         d_instantiate(dentry, inode);
4950         drop_on_err = 0;
4951         btrfs_update_inode_block_group(trans, inode);
4952         btrfs_update_inode_block_group(trans, dir);
4953
4954 out_fail:
4955         nr = trans->blocks_used;
4956         btrfs_end_transaction_throttle(trans, root);
4957         if (drop_on_err)
4958                 iput(inode);
4959         btrfs_btree_balance_dirty(root, nr);
4960         return err;
4961 }
4962
4963 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4964  * and an extent that you want to insert, deal with overlap and insert
4965  * the new extent into the tree.
4966  */
4967 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4968                                 struct extent_map *existing,
4969                                 struct extent_map *em,
4970                                 u64 map_start, u64 map_len)
4971 {
4972         u64 start_diff;
4973
4974         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4975         start_diff = map_start - em->start;
4976         em->start = map_start;
4977         em->len = map_len;
4978         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4979             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4980                 em->block_start += start_diff;
4981                 em->block_len -= start_diff;
4982         }
4983         return add_extent_mapping(em_tree, em);
4984 }
4985
4986 static noinline int uncompress_inline(struct btrfs_path *path,
4987                                       struct inode *inode, struct page *page,
4988                                       size_t pg_offset, u64 extent_offset,
4989                                       struct btrfs_file_extent_item *item)
4990 {
4991         int ret;
4992         struct extent_buffer *leaf = path->nodes[0];
4993         char *tmp;
4994         size_t max_size;
4995         unsigned long inline_size;
4996         unsigned long ptr;
4997         int compress_type;
4998
4999         WARN_ON(pg_offset != 0);
5000         compress_type = btrfs_file_extent_compression(leaf, item);
5001         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5002         inline_size = btrfs_file_extent_inline_item_len(leaf,
5003                                         btrfs_item_nr(leaf, path->slots[0]));
5004         tmp = kmalloc(inline_size, GFP_NOFS);
5005         if (!tmp)
5006                 return -ENOMEM;
5007         ptr = btrfs_file_extent_inline_start(item);
5008
5009         read_extent_buffer(leaf, tmp, ptr, inline_size);
5010
5011         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5012         ret = btrfs_decompress(compress_type, tmp, page,
5013                                extent_offset, inline_size, max_size);
5014         if (ret) {
5015                 char *kaddr = kmap_atomic(page, KM_USER0);
5016                 unsigned long copy_size = min_t(u64,
5017                                   PAGE_CACHE_SIZE - pg_offset,
5018                                   max_size - extent_offset);
5019                 memset(kaddr + pg_offset, 0, copy_size);
5020                 kunmap_atomic(kaddr, KM_USER0);
5021         }
5022         kfree(tmp);
5023         return 0;
5024 }
5025
5026 /*
5027  * a bit scary, this does extent mapping from logical file offset to the disk.
5028  * the ugly parts come from merging extents from the disk with the in-ram
5029  * representation.  This gets more complex because of the data=ordered code,
5030  * where the in-ram extents might be locked pending data=ordered completion.
5031  *
5032  * This also copies inline extents directly into the page.
5033  */
5034
5035 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5036                                     size_t pg_offset, u64 start, u64 len,
5037                                     int create)
5038 {
5039         int ret;
5040         int err = 0;
5041         u64 bytenr;
5042         u64 extent_start = 0;
5043         u64 extent_end = 0;
5044         u64 objectid = inode->i_ino;
5045         u32 found_type;
5046         struct btrfs_path *path = NULL;
5047         struct btrfs_root *root = BTRFS_I(inode)->root;
5048         struct btrfs_file_extent_item *item;
5049         struct extent_buffer *leaf;
5050         struct btrfs_key found_key;
5051         struct extent_map *em = NULL;
5052         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5053         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5054         struct btrfs_trans_handle *trans = NULL;
5055         int compress_type;
5056
5057 again:
5058         read_lock(&em_tree->lock);
5059         em = lookup_extent_mapping(em_tree, start, len);
5060         if (em)
5061                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5062         read_unlock(&em_tree->lock);
5063
5064         if (em) {
5065                 if (em->start > start || em->start + em->len <= start)
5066                         free_extent_map(em);
5067                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5068                         free_extent_map(em);
5069                 else
5070                         goto out;
5071         }
5072         em = alloc_extent_map(GFP_NOFS);
5073         if (!em) {
5074                 err = -ENOMEM;
5075                 goto out;
5076         }
5077         em->bdev = root->fs_info->fs_devices->latest_bdev;
5078         em->start = EXTENT_MAP_HOLE;
5079         em->orig_start = EXTENT_MAP_HOLE;
5080         em->len = (u64)-1;
5081         em->block_len = (u64)-1;
5082
5083         if (!path) {
5084                 path = btrfs_alloc_path();
5085                 BUG_ON(!path);
5086         }
5087
5088         ret = btrfs_lookup_file_extent(trans, root, path,
5089                                        objectid, start, trans != NULL);
5090         if (ret < 0) {
5091                 err = ret;
5092                 goto out;
5093         }
5094
5095         if (ret != 0) {
5096                 if (path->slots[0] == 0)
5097                         goto not_found;
5098                 path->slots[0]--;
5099         }
5100
5101         leaf = path->nodes[0];
5102         item = btrfs_item_ptr(leaf, path->slots[0],
5103                               struct btrfs_file_extent_item);
5104         /* are we inside the extent that was found? */
5105         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5106         found_type = btrfs_key_type(&found_key);
5107         if (found_key.objectid != objectid ||
5108             found_type != BTRFS_EXTENT_DATA_KEY) {
5109                 goto not_found;
5110         }
5111
5112         found_type = btrfs_file_extent_type(leaf, item);
5113         extent_start = found_key.offset;
5114         compress_type = btrfs_file_extent_compression(leaf, item);
5115         if (found_type == BTRFS_FILE_EXTENT_REG ||
5116             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5117                 extent_end = extent_start +
5118                        btrfs_file_extent_num_bytes(leaf, item);
5119         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5120                 size_t size;
5121                 size = btrfs_file_extent_inline_len(leaf, item);
5122                 extent_end = (extent_start + size + root->sectorsize - 1) &
5123                         ~((u64)root->sectorsize - 1);
5124         }
5125
5126         if (start >= extent_end) {
5127                 path->slots[0]++;
5128                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5129                         ret = btrfs_next_leaf(root, path);
5130                         if (ret < 0) {
5131                                 err = ret;
5132                                 goto out;
5133                         }
5134                         if (ret > 0)
5135                                 goto not_found;
5136                         leaf = path->nodes[0];
5137                 }
5138                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5139                 if (found_key.objectid != objectid ||
5140                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5141                         goto not_found;
5142                 if (start + len <= found_key.offset)
5143                         goto not_found;
5144                 em->start = start;
5145                 em->len = found_key.offset - start;
5146                 goto not_found_em;
5147         }
5148
5149         if (found_type == BTRFS_FILE_EXTENT_REG ||
5150             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5151                 em->start = extent_start;
5152                 em->len = extent_end - extent_start;
5153                 em->orig_start = extent_start -
5154                                  btrfs_file_extent_offset(leaf, item);
5155                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5156                 if (bytenr == 0) {
5157                         em->block_start = EXTENT_MAP_HOLE;
5158                         goto insert;
5159                 }
5160                 if (compress_type != BTRFS_COMPRESS_NONE) {
5161                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5162                         em->compress_type = compress_type;
5163                         em->block_start = bytenr;
5164                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5165                                                                          item);
5166                 } else {
5167                         bytenr += btrfs_file_extent_offset(leaf, item);
5168                         em->block_start = bytenr;
5169                         em->block_len = em->len;
5170                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5171                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5172                 }
5173                 goto insert;
5174         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5175                 unsigned long ptr;
5176                 char *map;
5177                 size_t size;
5178                 size_t extent_offset;
5179                 size_t copy_size;
5180
5181                 em->block_start = EXTENT_MAP_INLINE;
5182                 if (!page || create) {
5183                         em->start = extent_start;
5184                         em->len = extent_end - extent_start;
5185                         goto out;
5186                 }
5187
5188                 size = btrfs_file_extent_inline_len(leaf, item);
5189                 extent_offset = page_offset(page) + pg_offset - extent_start;
5190                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5191                                 size - extent_offset);
5192                 em->start = extent_start + extent_offset;
5193                 em->len = (copy_size + root->sectorsize - 1) &
5194                         ~((u64)root->sectorsize - 1);
5195                 em->orig_start = EXTENT_MAP_INLINE;
5196                 if (compress_type) {
5197                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5198                         em->compress_type = compress_type;
5199                 }
5200                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5201                 if (create == 0 && !PageUptodate(page)) {
5202                         if (btrfs_file_extent_compression(leaf, item) !=
5203                             BTRFS_COMPRESS_NONE) {
5204                                 ret = uncompress_inline(path, inode, page,
5205                                                         pg_offset,
5206                                                         extent_offset, item);
5207                                 BUG_ON(ret);
5208                         } else {
5209                                 map = kmap(page);
5210                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5211                                                    copy_size);
5212                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5213                                         memset(map + pg_offset + copy_size, 0,
5214                                                PAGE_CACHE_SIZE - pg_offset -
5215                                                copy_size);
5216                                 }
5217                                 kunmap(page);
5218                         }
5219                         flush_dcache_page(page);
5220                 } else if (create && PageUptodate(page)) {
5221                         WARN_ON(1);
5222                         if (!trans) {
5223                                 kunmap(page);
5224                                 free_extent_map(em);
5225                                 em = NULL;
5226                                 btrfs_release_path(root, path);
5227                                 trans = btrfs_join_transaction(root, 1);
5228                                 if (IS_ERR(trans))
5229                                         return ERR_CAST(trans);
5230                                 goto again;
5231                         }
5232                         map = kmap(page);
5233                         write_extent_buffer(leaf, map + pg_offset, ptr,
5234                                             copy_size);
5235                         kunmap(page);
5236                         btrfs_mark_buffer_dirty(leaf);
5237                 }
5238                 set_extent_uptodate(io_tree, em->start,
5239                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5240                 goto insert;
5241         } else {
5242                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5243                 WARN_ON(1);
5244         }
5245 not_found:
5246         em->start = start;
5247         em->len = len;
5248 not_found_em:
5249         em->block_start = EXTENT_MAP_HOLE;
5250         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5251 insert:
5252         btrfs_release_path(root, path);
5253         if (em->start > start || extent_map_end(em) <= start) {
5254                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5255                        "[%llu %llu]\n", (unsigned long long)em->start,
5256                        (unsigned long long)em->len,
5257                        (unsigned long long)start,
5258                        (unsigned long long)len);
5259                 err = -EIO;
5260                 goto out;
5261         }
5262
5263         err = 0;
5264         write_lock(&em_tree->lock);
5265         ret = add_extent_mapping(em_tree, em);
5266         /* it is possible that someone inserted the extent into the tree
5267          * while we had the lock dropped.  It is also possible that
5268          * an overlapping map exists in the tree
5269          */
5270         if (ret == -EEXIST) {
5271                 struct extent_map *existing;
5272
5273                 ret = 0;
5274
5275                 existing = lookup_extent_mapping(em_tree, start, len);
5276                 if (existing && (existing->start > start ||
5277                     existing->start + existing->len <= start)) {
5278                         free_extent_map(existing);
5279                         existing = NULL;
5280                 }
5281                 if (!existing) {
5282                         existing = lookup_extent_mapping(em_tree, em->start,
5283                                                          em->len);
5284                         if (existing) {
5285                                 err = merge_extent_mapping(em_tree, existing,
5286                                                            em, start,
5287                                                            root->sectorsize);
5288                                 free_extent_map(existing);
5289                                 if (err) {
5290                                         free_extent_map(em);
5291                                         em = NULL;
5292                                 }
5293                         } else {
5294                                 err = -EIO;
5295                                 free_extent_map(em);
5296                                 em = NULL;
5297                         }
5298                 } else {
5299                         free_extent_map(em);
5300                         em = existing;
5301                         err = 0;
5302                 }
5303         }
5304         write_unlock(&em_tree->lock);
5305 out:
5306
5307         trace_btrfs_get_extent(root, em);
5308
5309         if (path)
5310                 btrfs_free_path(path);
5311         if (trans) {
5312                 ret = btrfs_end_transaction(trans, root);
5313                 if (!err)
5314                         err = ret;
5315         }
5316         if (err) {
5317                 free_extent_map(em);
5318                 return ERR_PTR(err);
5319         }
5320         return em;
5321 }
5322
5323 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5324                                            size_t pg_offset, u64 start, u64 len,
5325                                            int create)
5326 {
5327         struct extent_map *em;
5328         struct extent_map *hole_em = NULL;
5329         u64 range_start = start;
5330         u64 end;
5331         u64 found;
5332         u64 found_end;
5333         int err = 0;
5334
5335         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5336         if (IS_ERR(em))
5337                 return em;
5338         if (em) {
5339                 /*
5340                  * if our em maps to a hole, there might
5341                  * actually be delalloc bytes behind it
5342                  */
5343                 if (em->block_start != EXTENT_MAP_HOLE)
5344                         return em;
5345                 else
5346                         hole_em = em;
5347         }
5348
5349         /* check to see if we've wrapped (len == -1 or similar) */
5350         end = start + len;
5351         if (end < start)
5352                 end = (u64)-1;
5353         else
5354                 end -= 1;
5355
5356         em = NULL;
5357
5358         /* ok, we didn't find anything, lets look for delalloc */
5359         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5360                                  end, len, EXTENT_DELALLOC, 1);
5361         found_end = range_start + found;
5362         if (found_end < range_start)
5363                 found_end = (u64)-1;
5364
5365         /*
5366          * we didn't find anything useful, return
5367          * the original results from get_extent()
5368          */
5369         if (range_start > end || found_end <= start) {
5370                 em = hole_em;
5371                 hole_em = NULL;
5372                 goto out;
5373         }
5374
5375         /* adjust the range_start to make sure it doesn't
5376          * go backwards from the start they passed in
5377          */
5378         range_start = max(start,range_start);
5379         found = found_end - range_start;
5380
5381         if (found > 0) {
5382                 u64 hole_start = start;
5383                 u64 hole_len = len;
5384
5385                 em = alloc_extent_map(GFP_NOFS);
5386                 if (!em) {
5387                         err = -ENOMEM;
5388                         goto out;
5389                 }
5390                 /*
5391                  * when btrfs_get_extent can't find anything it
5392                  * returns one huge hole
5393                  *
5394                  * make sure what it found really fits our range, and
5395                  * adjust to make sure it is based on the start from
5396                  * the caller
5397                  */
5398                 if (hole_em) {
5399                         u64 calc_end = extent_map_end(hole_em);
5400
5401                         if (calc_end <= start || (hole_em->start > end)) {
5402                                 free_extent_map(hole_em);
5403                                 hole_em = NULL;
5404                         } else {
5405                                 hole_start = max(hole_em->start, start);
5406                                 hole_len = calc_end - hole_start;
5407                         }
5408                 }
5409                 em->bdev = NULL;
5410                 if (hole_em && range_start > hole_start) {
5411                         /* our hole starts before our delalloc, so we
5412                          * have to return just the parts of the hole
5413                          * that go until  the delalloc starts
5414                          */
5415                         em->len = min(hole_len,
5416                                       range_start - hole_start);
5417                         em->start = hole_start;
5418                         em->orig_start = hole_start;
5419                         /*
5420                          * don't adjust block start at all,
5421                          * it is fixed at EXTENT_MAP_HOLE
5422                          */
5423                         em->block_start = hole_em->block_start;
5424                         em->block_len = hole_len;
5425                 } else {
5426                         em->start = range_start;
5427                         em->len = found;
5428                         em->orig_start = range_start;
5429                         em->block_start = EXTENT_MAP_DELALLOC;
5430                         em->block_len = found;
5431                 }
5432         } else if (hole_em) {
5433                 return hole_em;
5434         }
5435 out:
5436
5437         free_extent_map(hole_em);
5438         if (err) {
5439                 free_extent_map(em);
5440                 return ERR_PTR(err);
5441         }
5442         return em;
5443 }
5444
5445 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5446                                                   struct extent_map *em,
5447                                                   u64 start, u64 len)
5448 {
5449         struct btrfs_root *root = BTRFS_I(inode)->root;
5450         struct btrfs_trans_handle *trans;
5451         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5452         struct btrfs_key ins;
5453         u64 alloc_hint;
5454         int ret;
5455         bool insert = false;
5456
5457         /*
5458          * Ok if the extent map we looked up is a hole and is for the exact
5459          * range we want, there is no reason to allocate a new one, however if
5460          * it is not right then we need to free this one and drop the cache for
5461          * our range.
5462          */
5463         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5464             em->len != len) {
5465                 free_extent_map(em);
5466                 em = NULL;
5467                 insert = true;
5468                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5469         }
5470
5471         trans = btrfs_join_transaction(root, 0);
5472         if (IS_ERR(trans))
5473                 return ERR_CAST(trans);
5474
5475         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5476
5477         alloc_hint = get_extent_allocation_hint(inode, start, len);
5478         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5479                                    alloc_hint, (u64)-1, &ins, 1);
5480         if (ret) {
5481                 em = ERR_PTR(ret);
5482                 goto out;
5483         }
5484
5485         if (!em) {
5486                 em = alloc_extent_map(GFP_NOFS);
5487                 if (!em) {
5488                         em = ERR_PTR(-ENOMEM);
5489                         goto out;
5490                 }
5491         }
5492
5493         em->start = start;
5494         em->orig_start = em->start;
5495         em->len = ins.offset;
5496
5497         em->block_start = ins.objectid;
5498         em->block_len = ins.offset;
5499         em->bdev = root->fs_info->fs_devices->latest_bdev;
5500
5501         /*
5502          * We need to do this because if we're using the original em we searched
5503          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5504          */
5505         em->flags = 0;
5506         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5507
5508         while (insert) {
5509                 write_lock(&em_tree->lock);
5510                 ret = add_extent_mapping(em_tree, em);
5511                 write_unlock(&em_tree->lock);
5512                 if (ret != -EEXIST)
5513                         break;
5514                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5515         }
5516
5517         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5518                                            ins.offset, ins.offset, 0);
5519         if (ret) {
5520                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5521                 em = ERR_PTR(ret);
5522         }
5523 out:
5524         btrfs_end_transaction(trans, root);
5525         return em;
5526 }
5527
5528 /*
5529  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5530  * block must be cow'd
5531  */
5532 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5533                                       struct inode *inode, u64 offset, u64 len)
5534 {
5535         struct btrfs_path *path;
5536         int ret;
5537         struct extent_buffer *leaf;
5538         struct btrfs_root *root = BTRFS_I(inode)->root;
5539         struct btrfs_file_extent_item *fi;
5540         struct btrfs_key key;
5541         u64 disk_bytenr;
5542         u64 backref_offset;
5543         u64 extent_end;
5544         u64 num_bytes;
5545         int slot;
5546         int found_type;
5547
5548         path = btrfs_alloc_path();
5549         if (!path)
5550                 return -ENOMEM;
5551
5552         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5553                                        offset, 0);
5554         if (ret < 0)
5555                 goto out;
5556
5557         slot = path->slots[0];
5558         if (ret == 1) {
5559                 if (slot == 0) {
5560                         /* can't find the item, must cow */
5561                         ret = 0;
5562                         goto out;
5563                 }
5564                 slot--;
5565         }
5566         ret = 0;
5567         leaf = path->nodes[0];
5568         btrfs_item_key_to_cpu(leaf, &key, slot);
5569         if (key.objectid != inode->i_ino ||
5570             key.type != BTRFS_EXTENT_DATA_KEY) {
5571                 /* not our file or wrong item type, must cow */
5572                 goto out;
5573         }
5574
5575         if (key.offset > offset) {
5576                 /* Wrong offset, must cow */
5577                 goto out;
5578         }
5579
5580         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5581         found_type = btrfs_file_extent_type(leaf, fi);
5582         if (found_type != BTRFS_FILE_EXTENT_REG &&
5583             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5584                 /* not a regular extent, must cow */
5585                 goto out;
5586         }
5587         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5588         backref_offset = btrfs_file_extent_offset(leaf, fi);
5589
5590         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5591         if (extent_end < offset + len) {
5592                 /* extent doesn't include our full range, must cow */
5593                 goto out;
5594         }
5595
5596         if (btrfs_extent_readonly(root, disk_bytenr))
5597                 goto out;
5598
5599         /*
5600          * look for other files referencing this extent, if we
5601          * find any we must cow
5602          */
5603         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5604                                   key.offset - backref_offset, disk_bytenr))
5605                 goto out;
5606
5607         /*
5608          * adjust disk_bytenr and num_bytes to cover just the bytes
5609          * in this extent we are about to write.  If there
5610          * are any csums in that range we have to cow in order
5611          * to keep the csums correct
5612          */
5613         disk_bytenr += backref_offset;
5614         disk_bytenr += offset - key.offset;
5615         num_bytes = min(offset + len, extent_end) - offset;
5616         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5617                                 goto out;
5618         /*
5619          * all of the above have passed, it is safe to overwrite this extent
5620          * without cow
5621          */
5622         ret = 1;
5623 out:
5624         btrfs_free_path(path);
5625         return ret;
5626 }
5627
5628 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5629                                    struct buffer_head *bh_result, int create)
5630 {
5631         struct extent_map *em;
5632         struct btrfs_root *root = BTRFS_I(inode)->root;
5633         u64 start = iblock << inode->i_blkbits;
5634         u64 len = bh_result->b_size;
5635         struct btrfs_trans_handle *trans;
5636
5637         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5638         if (IS_ERR(em))
5639                 return PTR_ERR(em);
5640
5641         /*
5642          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5643          * io.  INLINE is special, and we could probably kludge it in here, but
5644          * it's still buffered so for safety lets just fall back to the generic
5645          * buffered path.
5646          *
5647          * For COMPRESSED we _have_ to read the entire extent in so we can
5648          * decompress it, so there will be buffering required no matter what we
5649          * do, so go ahead and fallback to buffered.
5650          *
5651          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5652          * to buffered IO.  Don't blame me, this is the price we pay for using