projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: reset intwrite on transaction abort
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 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
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 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>
39279cc3
CM		 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>
a27bb332 35#include <linux/aio.h>
9ebefb18 36#include <linux/bit_spinlock.h>
5103e947 37#include <linux/xattr.h>
33268eaf 38#include <linux/posix_acl.h>
d899e052 39#include <linux/falloc.h>
5a0e3ad6 40#include <linux/slab.h>
7a36ddec 41#include <linux/ratelimit.h>
22c44fe6 42#include <linux/mount.h>
55e301fd 43#include <linux/btrfs.h>
53b381b3 44#include <linux/blkdev.h>
f23b5a59 45#include <linux/posix_acl_xattr.h>
4b4e25f2 46#include "compat.h"
39279cc3
CM		 47#include "ctree.h"
48#include "disk-io.h"
49#include "transaction.h"
50#include "btrfs_inode.h"
39279cc3 51#include "print-tree.h"
e6dcd2dc 52#include "ordered-data.h"
95819c05 53#include "xattr.h"
e02119d5 54#include "tree-log.h"
4a54c8c1 55#include "volumes.h"
c8b97818 56#include "compression.h"
b4ce94de 57#include "locking.h"
dc89e982 58#include "free-space-cache.h"
581bb050 59#include "inode-map.h"
38c227d8 60#include "backref.h"
f23b5a59 61#include "hash.h"
39279cc3
CM		 62
63struct btrfs_iget_args {
64 u64 ino;
65 struct btrfs_root *root;
66};
67
6e1d5dcc
AD		 68static const struct inode_operations btrfs_dir_inode_operations;
69static const struct inode_operations btrfs_symlink_inode_operations;
70static const struct inode_operations btrfs_dir_ro_inode_operations;
71static const struct inode_operations btrfs_special_inode_operations;
72static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 73static const struct address_space_operations btrfs_aops;
74static const struct address_space_operations btrfs_symlink_aops;
828c0950 75static const struct file_operations btrfs_dir_file_operations;
d1310b2e 76static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 77
78static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 79static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 80struct kmem_cache *btrfs_trans_handle_cachep;
81struct kmem_cache *btrfs_transaction_cachep;
39279cc3 82struct kmem_cache *btrfs_path_cachep;
dc89e982 83struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 84
85#define S_SHIFT 12
86static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
94};
95
3972f260 96static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 97static int btrfs_truncate(struct inode *inode);
5fd02043 98static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 99static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
70c8a91c
JB		 103static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
cc95bef6
JB		 106 u64 orig_block_len, u64 ram_bytes,
107 int type);
7b128766 108
48a3b636 109static int btrfs_dirty_inode(struct inode *inode);
7b128766 110
f34f57a3 111static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
0279b4cd
JO		 114{
115 int err;
116
f34f57a3 117 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 118 if (!err)
2a7dba39 119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 120 return err;
121}
122
c8b97818
CM		 123/*
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
d397712b 128static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
fe3f566c 131 int compress_type,
c8b97818
CM		 132 struct page **compressed_pages)
133{
134 struct btrfs_key key;
135 struct btrfs_path *path;
136 struct extent_buffer *leaf;
137 struct page *page = NULL;
138 char *kaddr;
139 unsigned long ptr;
140 struct btrfs_file_extent_item *ei;
141 int err = 0;
142 int ret;
143 size_t cur_size = size;
144 size_t datasize;
145 unsigned long offset;
c8b97818 146
fe3f566c 147 if (compressed_size && compressed_pages)
c8b97818 148 cur_size = compressed_size;
c8b97818 149
d397712b
CM		 150 path = btrfs_alloc_path();
151 if (!path)
c8b97818
CM		 152 return -ENOMEM;
153
b9473439 154 path->leave_spinning = 1;
c8b97818 155
33345d01 156 key.objectid = btrfs_ino(inode);
c8b97818
CM		 157 key.offset = start;
158 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 159 datasize = btrfs_file_extent_calc_inline_size(cur_size);
160
161 inode_add_bytes(inode, size);
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
163 datasize);
c8b97818
CM		 164 if (ret) {
165 err = ret;
c8b97818
CM		 166 goto fail;
167 }
168 leaf = path->nodes[0];
169 ei = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_file_extent_item);
171 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
172 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
173 btrfs_set_file_extent_encryption(leaf, ei, 0);
174 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
175 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
176 ptr = btrfs_file_extent_inline_start(ei);
177
261507a0 178 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 179 struct page *cpage;
180 int i = 0;
d397712b 181 while (compressed_size > 0) {
c8b97818 182 cpage = compressed_pages[i];
5b050f04 183 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 184 PAGE_CACHE_SIZE);
185
7ac687d9 186 kaddr = kmap_atomic(cpage);
c8b97818 187 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 188 kunmap_atomic(kaddr);
c8b97818
CM		 189
190 i++;
191 ptr += cur_size;
192 compressed_size -= cur_size;
193 }
194 btrfs_set_file_extent_compression(leaf, ei,
261507a0 195 compress_type);
c8b97818
CM		 196 } else {
197 page = find_get_page(inode->i_mapping,
198 start >> PAGE_CACHE_SHIFT);
199 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 200 kaddr = kmap_atomic(page);
c8b97818
CM		 201 offset = start & (PAGE_CACHE_SIZE - 1);
202 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 203 kunmap_atomic(kaddr);
c8b97818
CM		 204 page_cache_release(page);
205 }
206 btrfs_mark_buffer_dirty(leaf);
207 btrfs_free_path(path);
208
c2167754
YZ		 209 /*
210 * we're an inline extent, so nobody can
211 * extend the file past i_size without locking
212 * a page we already have locked.
213 *
214 * We must do any isize and inode updates
215 * before we unlock the pages. Otherwise we
216 * could end up racing with unlink.
217 */
c8b97818 218 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 219 ret = btrfs_update_inode(trans, root, inode);
c2167754 220
79787eaa 221 return ret;
c8b97818
CM		 222fail:
223 btrfs_free_path(path);
224 return err;
225}
226
227
228/*
229 * conditionally insert an inline extent into the file. This
230 * does the checks required to make sure the data is small enough
231 * to fit as an inline extent.
232 */
00361589
JB		 233static noinline int cow_file_range_inline(struct btrfs_root *root,
234 struct inode *inode, u64 start,
235 u64 end, size_t compressed_size,
236 int compress_type,
237 struct page **compressed_pages)
c8b97818 238{
00361589 239 struct btrfs_trans_handle *trans;
c8b97818
CM		 240 u64 isize = i_size_read(inode);
241 u64 actual_end = min(end + 1, isize);
242 u64 inline_len = actual_end - start;
fda2832f 243 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 244 u64 data_len = inline_len;
245 int ret;
246
247 if (compressed_size)
248 data_len = compressed_size;
249
250 if (start > 0 ||
70b99e69 251 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 252 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
253 (!compressed_size &&
254 (actual_end & (root->sectorsize - 1)) == 0) ||
255 end + 1 < isize ||
256 data_len > root->fs_info->max_inline) {
257 return 1;
258 }
259
00361589
JB		 260 trans = btrfs_join_transaction(root);
261 if (IS_ERR(trans))
262 return PTR_ERR(trans);
263 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
264
2671485d 265 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
00361589
JB		 266 if (ret) {
267 btrfs_abort_transaction(trans, root, ret);
268 goto out;
269 }
c8b97818
CM		 270
271 if (isize > actual_end)
272 inline_len = min_t(u64, isize, actual_end);
273 ret = insert_inline_extent(trans, root, inode, start,
274 inline_len, compressed_size,
fe3f566c 275 compress_type, compressed_pages);
2adcac1a 276 if (ret && ret != -ENOSPC) {
79787eaa 277 btrfs_abort_transaction(trans, root, ret);
00361589 278 goto out;
2adcac1a 279 } else if (ret == -ENOSPC) {
00361589
JB		 280 ret = 1;
281 goto out;
79787eaa 282 }
2adcac1a 283
bdc20e67 284 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 285 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 286 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589
JB		 287out:
288 btrfs_end_transaction(trans, root);
289 return ret;
c8b97818
CM		 290}
291
771ed689
CM		 292struct async_extent {
293 u64 start;
294 u64 ram_size;
295 u64 compressed_size;
296 struct page **pages;
297 unsigned long nr_pages;
261507a0 298 int compress_type;
771ed689
CM		 299 struct list_head list;
300};
301
302struct async_cow {
303 struct inode *inode;
304 struct btrfs_root *root;
305 struct page *locked_page;
306 u64 start;
307 u64 end;
308 struct list_head extents;
309 struct btrfs_work work;
310};
311
312static noinline int add_async_extent(struct async_cow *cow,
313 u64 start, u64 ram_size,
314 u64 compressed_size,
315 struct page **pages,
261507a0
LZ		 316 unsigned long nr_pages,
317 int compress_type)
771ed689
CM		 318{
319 struct async_extent *async_extent;
320
321 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 322 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 323 async_extent->start = start;
324 async_extent->ram_size = ram_size;
325 async_extent->compressed_size = compressed_size;
326 async_extent->pages = pages;
327 async_extent->nr_pages = nr_pages;
261507a0 328 async_extent->compress_type = compress_type;
771ed689
CM		 329 list_add_tail(&async_extent->list, &cow->extents);
330 return 0;
331}
332
d352ac68 333/*
771ed689
CM		 334 * we create compressed extents in two phases. The first
335 * phase compresses a range of pages that have already been
336 * locked (both pages and state bits are locked).
c8b97818 337 *
771ed689
CM		 338 * This is done inside an ordered work queue, and the compression
339 * is spread across many cpus. The actual IO submission is step
340 * two, and the ordered work queue takes care of making sure that
341 * happens in the same order things were put onto the queue by
342 * writepages and friends.
c8b97818 343 *
771ed689
CM		 344 * If this code finds it can't get good compression, it puts an
345 * entry onto the work queue to write the uncompressed bytes. This
346 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 347 * are written in the same order that the flusher thread sent them
348 * down.
d352ac68 349 */
771ed689
CM		 350static noinline int compress_file_range(struct inode *inode,
351 struct page *locked_page,
352 u64 start, u64 end,
353 struct async_cow *async_cow,
354 int *num_added)
b888db2b
CM		 355{
356 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 357 u64 num_bytes;
db94535d 358 u64 blocksize = root->sectorsize;
c8b97818 359 u64 actual_end;
42dc7bab 360 u64 isize = i_size_read(inode);
e6dcd2dc 361 int ret = 0;
c8b97818
CM		 362 struct page **pages = NULL;
363 unsigned long nr_pages;
364 unsigned long nr_pages_ret = 0;
365 unsigned long total_compressed = 0;
366 unsigned long total_in = 0;
367 unsigned long max_compressed = 128 * 1024;
771ed689 368 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 369 int i;
370 int will_compress;
261507a0 371 int compress_type = root->fs_info->compress_type;
4adaa611 372 int redirty = 0;
b888db2b 373
4cb13e5d
LB		 374 /* if this is a small write inside eof, kick off a defrag */
375 if ((end - start + 1) < 16 * 1024 &&
376 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 377 btrfs_add_inode_defrag(NULL, inode);
378
42dc7bab 379 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 380again:
381 will_compress = 0;
382 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
383 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 384
f03d9301
CM		 385 /*
386 * we don't want to send crud past the end of i_size through
387 * compression, that's just a waste of CPU time. So, if the
388 * end of the file is before the start of our current
389 * requested range of bytes, we bail out to the uncompressed
390 * cleanup code that can deal with all of this.
391 *
392 * It isn't really the fastest way to fix things, but this is a
393 * very uncommon corner.
394 */
395 if (actual_end <= start)
396 goto cleanup_and_bail_uncompressed;
397
c8b97818
CM		 398 total_compressed = actual_end - start;
399
400 /* we want to make sure that amount of ram required to uncompress
401 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 402 * of a compressed extent to 128k. This is a crucial number
403 * because it also controls how easily we can spread reads across
404 * cpus for decompression.
405 *
406 * We also want to make sure the amount of IO required to do
407 * a random read is reasonably small, so we limit the size of
408 * a compressed extent to 128k.
c8b97818
CM		 409 */
410 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 411 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 412 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 413 total_in = 0;
414 ret = 0;
db94535d 415
771ed689
CM		 416 /*
417 * we do compression for mount -o compress and when the
418 * inode has not been flagged as nocompress. This flag can
419 * change at any time if we discover bad compression ratios.
c8b97818 420 */
6cbff00f 421 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 422 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 423 (BTRFS_I(inode)->force_compress) ||
424 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 425 WARN_ON(pages);
cfbc246e 426 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 427 if (!pages) {
428 /* just bail out to the uncompressed code */
429 goto cont;
430 }
c8b97818 431
261507a0
LZ		 432 if (BTRFS_I(inode)->force_compress)
433 compress_type = BTRFS_I(inode)->force_compress;
434
4adaa611
CM		 435 /*
436 * we need to call clear_page_dirty_for_io on each
437 * page in the range. Otherwise applications with the file
438 * mmap'd can wander in and change the page contents while
439 * we are compressing them.
440 *
441 * If the compression fails for any reason, we set the pages
442 * dirty again later on.
443 */
444 extent_range_clear_dirty_for_io(inode, start, end);
445 redirty = 1;
261507a0
LZ		 446 ret = btrfs_compress_pages(compress_type,
447 inode->i_mapping, start,
448 total_compressed, pages,
449 nr_pages, &nr_pages_ret,
450 &total_in,
451 &total_compressed,
452 max_compressed);
c8b97818
CM		 453
454 if (!ret) {
455 unsigned long offset = total_compressed &
456 (PAGE_CACHE_SIZE - 1);
457 struct page *page = pages[nr_pages_ret - 1];
458 char *kaddr;
459
460 /* zero the tail end of the last page, we might be
461 * sending it down to disk
462 */
463 if (offset) {
7ac687d9 464 kaddr = kmap_atomic(page);
c8b97818
CM		 465 memset(kaddr + offset, 0,
466 PAGE_CACHE_SIZE - offset);
7ac687d9 467 kunmap_atomic(kaddr);
c8b97818
CM		 468 }
469 will_compress = 1;
470 }
471 }
560f7d75 472cont:
c8b97818
CM		 473 if (start == 0) {
474 /* lets try to make an inline extent */
771ed689 475 if (ret || total_in < (actual_end - start)) {
c8b97818 476 /* we didn't compress the entire range, try
771ed689 477 * to make an uncompressed inline extent.
c8b97818 478 */
00361589
JB		 479 ret = cow_file_range_inline(root, inode, start, end,
480 0, 0, NULL);
c8b97818 481 } else {
771ed689 482 /* try making a compressed inline extent */
00361589 483 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 484 total_compressed,
485 compress_type, pages);
c8b97818 486 }
79787eaa 487 if (ret <= 0) {
151a41bc
JB		 488 unsigned long clear_flags = EXTENT_DELALLOC |
489 EXTENT_DEFRAG;
490 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
491
771ed689 492 /*
79787eaa
JM		 493 * inline extent creation worked or returned error,
494 * we don't need to create any more async work items.
495 * Unlock and free up our temp pages.
771ed689 496 */
c2790a2e 497 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 498 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 499 PAGE_CLEAR_DIRTY |
500 PAGE_SET_WRITEBACK |
501 PAGE_END_WRITEBACK);
c8b97818
CM		 502 goto free_pages_out;
503 }
504 }
505
506 if (will_compress) {
507 /*
508 * we aren't doing an inline extent round the compressed size
509 * up to a block size boundary so the allocator does sane
510 * things
511 */
fda2832f 512 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 513
514 /*
515 * one last check to make sure the compression is really a
516 * win, compare the page count read with the blocks on disk
517 */
fda2832f 518 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 519 if (total_compressed >= total_in) {
520 will_compress = 0;
521 } else {
c8b97818
CM		 522 num_bytes = total_in;
523 }
524 }
525 if (!will_compress && pages) {
526 /*
527 * the compression code ran but failed to make things smaller,
528 * free any pages it allocated and our page pointer array
529 */
530 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 531 WARN_ON(pages[i]->mapping);
c8b97818
CM		 532 page_cache_release(pages[i]);
533 }
534 kfree(pages);
535 pages = NULL;
536 total_compressed = 0;
537 nr_pages_ret = 0;
538
539 /* flag the file so we don't compress in the future */
1e701a32
CM		 540 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
541 !(BTRFS_I(inode)->force_compress)) {
a555f810 542 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 543 }
c8b97818 544 }
771ed689
CM		 545 if (will_compress) {
546 *num_added += 1;
c8b97818 547
771ed689
CM		 548 /* the async work queues will take care of doing actual
549 * allocation on disk for these compressed pages,
550 * and will submit them to the elevator.
551 */
552 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 553 total_compressed, pages, nr_pages_ret,
554 compress_type);
179e29e4 555
24ae6365 556 if (start + num_bytes < end) {
771ed689
CM		 557 start += num_bytes;
558 pages = NULL;
559 cond_resched();
560 goto again;
561 }
562 } else {
f03d9301 563cleanup_and_bail_uncompressed:
771ed689
CM		 564 /*
565 * No compression, but we still need to write the pages in
566 * the file we've been given so far. redirty the locked
567 * page if it corresponds to our extent and set things up
568 * for the async work queue to run cow_file_range to do
569 * the normal delalloc dance
570 */
571 if (page_offset(locked_page) >= start &&
572 page_offset(locked_page) <= end) {
573 __set_page_dirty_nobuffers(locked_page);
574 /* unlocked later on in the async handlers */
575 }
4adaa611
CM		 576 if (redirty)
577 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 578 add_async_extent(async_cow, start, end - start + 1,
579 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 580 *num_added += 1;
581 }
3b951516 582
771ed689 583out:
79787eaa 584 return ret;
771ed689
CM		 585
586free_pages_out:
587 for (i = 0; i < nr_pages_ret; i++) {
588 WARN_ON(pages[i]->mapping);
589 page_cache_release(pages[i]);
590 }
d397712b 591 kfree(pages);
771ed689
CM		 592
593 goto out;
594}
595
596/*
597 * phase two of compressed writeback. This is the ordered portion
598 * of the code, which only gets called in the order the work was
599 * queued. We walk all the async extents created by compress_file_range
600 * and send them down to the disk.
601 */
602static noinline int submit_compressed_extents(struct inode *inode,
603 struct async_cow *async_cow)
604{
605 struct async_extent *async_extent;
606 u64 alloc_hint = 0;
771ed689
CM		 607 struct btrfs_key ins;
608 struct extent_map *em;
609 struct btrfs_root *root = BTRFS_I(inode)->root;
610 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
611 struct extent_io_tree *io_tree;
f5a84ee3 612 int ret = 0;
771ed689
CM		 613
614 if (list_empty(&async_cow->extents))
615 return 0;
616
3e04e7f1 617again:
d397712b 618 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 619 async_extent = list_entry(async_cow->extents.next,
620 struct async_extent, list);
621 list_del(&async_extent->list);
c8b97818 622
771ed689
CM		 623 io_tree = &BTRFS_I(inode)->io_tree;
624
f5a84ee3 625retry:
771ed689
CM		 626 /* did the compression code fall back to uncompressed IO? */
627 if (!async_extent->pages) {
628 int page_started = 0;
629 unsigned long nr_written = 0;
630
631 lock_extent(io_tree, async_extent->start,
2ac55d41 632 async_extent->start +
d0082371 633 async_extent->ram_size - 1);
771ed689
CM		 634
635 /* allocate blocks */
f5a84ee3
JB		 636 ret = cow_file_range(inode, async_cow->locked_page,
637 async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1,
640 &page_started, &nr_written, 0);
771ed689 641
79787eaa
JM		 642 /* JDM XXX */
643
771ed689
CM		 644 /*
645 * if page_started, cow_file_range inserted an
646 * inline extent and took care of all the unlocking
647 * and IO for us. Otherwise, we need to submit
648 * all those pages down to the drive.
649 */
f5a84ee3 650 if (!page_started && !ret)
771ed689
CM		 651 extent_write_locked_range(io_tree,
652 inode, async_extent->start,
d397712b 653 async_extent->start +
771ed689
CM		 654 async_extent->ram_size - 1,
655 btrfs_get_extent,
656 WB_SYNC_ALL);
3e04e7f1
JB		 657 else if (ret)
658 unlock_page(async_cow->locked_page);
771ed689
CM		 659 kfree(async_extent);
660 cond_resched();
661 continue;
662 }
663
664 lock_extent(io_tree, async_extent->start,
d0082371 665 async_extent->start + async_extent->ram_size - 1);
771ed689 666
00361589 667 ret = btrfs_reserve_extent(root,
771ed689
CM		 668 async_extent->compressed_size,
669 async_extent->compressed_size,
81c9ad23 670 0, alloc_hint, &ins, 1);
f5a84ee3
JB		 671 if (ret) {
672 int i;
3e04e7f1 673
f5a84ee3
JB		 674 for (i = 0; i < async_extent->nr_pages; i++) {
675 WARN_ON(async_extent->pages[i]->mapping);
676 page_cache_release(async_extent->pages[i]);
677 }
678 kfree(async_extent->pages);
679 async_extent->nr_pages = 0;
680 async_extent->pages = NULL;
3e04e7f1 681
fdf8e2ea
JB		 682 if (ret == -ENOSPC) {
683 unlock_extent(io_tree, async_extent->start,
684 async_extent->start +
685 async_extent->ram_size - 1);
79787eaa 686 goto retry;
fdf8e2ea 687 }
3e04e7f1 688 goto out_free;
f5a84ee3
JB		 689 }
690
c2167754
YZ		 691 /*
692 * here we're doing allocation and writeback of the
693 * compressed pages
694 */
695 btrfs_drop_extent_cache(inode, async_extent->start,
696 async_extent->start +
697 async_extent->ram_size - 1, 0);
698
172ddd60 699 em = alloc_extent_map();
b9aa55be
LB		 700 if (!em) {
701 ret = -ENOMEM;
3e04e7f1 702 goto out_free_reserve;
b9aa55be 703 }
771ed689
CM		 704 em->start = async_extent->start;
705 em->len = async_extent->ram_size;
445a6944 706 em->orig_start = em->start;
2ab28f32
JB		 707 em->mod_start = em->start;
708 em->mod_len = em->len;
c8b97818 709
771ed689
CM		 710 em->block_start = ins.objectid;
711 em->block_len = ins.offset;
b4939680 712 em->orig_block_len = ins.offset;
cc95bef6 713 em->ram_bytes = async_extent->ram_size;
771ed689 714 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 715 em->compress_type = async_extent->compress_type;
771ed689
CM		 716 set_bit(EXTENT_FLAG_PINNED, &em->flags);
717 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 718 em->generation = -1;
771ed689 719
d397712b 720 while (1) {
890871be 721 write_lock(&em_tree->lock);
09a2a8f9 722 ret = add_extent_mapping(em_tree, em, 1);
890871be 723 write_unlock(&em_tree->lock);
771ed689
CM		 724 if (ret != -EEXIST) {
725 free_extent_map(em);
726 break;
727 }
728 btrfs_drop_extent_cache(inode, async_extent->start,
729 async_extent->start +
730 async_extent->ram_size - 1, 0);
731 }
732
3e04e7f1
JB		 733 if (ret)
734 goto out_free_reserve;
735
261507a0
LZ		 736 ret = btrfs_add_ordered_extent_compress(inode,
737 async_extent->start,
738 ins.objectid,
739 async_extent->ram_size,
740 ins.offset,
741 BTRFS_ORDERED_COMPRESSED,
742 async_extent->compress_type);
3e04e7f1
JB		 743 if (ret)
744 goto out_free_reserve;
771ed689 745
771ed689
CM		 746 /*
747 * clear dirty, set writeback and unlock the pages.
748 */
c2790a2e 749 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 750 async_extent->start +
751 async_extent->ram_size - 1,
151a41bc
JB		 752 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
753 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 754 PAGE_SET_WRITEBACK);
771ed689 755 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 756 async_extent->start,
757 async_extent->ram_size,
758 ins.objectid,
759 ins.offset, async_extent->pages,
760 async_extent->nr_pages);
771ed689
CM		 761 alloc_hint = ins.objectid + ins.offset;
762 kfree(async_extent);
3e04e7f1
JB		 763 if (ret)
764 goto out;
771ed689
CM		 765 cond_resched();
766 }
79787eaa
JM		 767 ret = 0;
768out:
769 return ret;
3e04e7f1
JB		 770out_free_reserve:
771 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 772out_free:
c2790a2e 773 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 774 async_extent->start +
775 async_extent->ram_size - 1,
c2790a2e 776 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 777 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
778 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
779 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 780 kfree(async_extent);
3e04e7f1 781 goto again;
771ed689
CM		 782}
783
4b46fce2
JB		 784static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
785 u64 num_bytes)
786{
787 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
788 struct extent_map *em;
789 u64 alloc_hint = 0;
790
791 read_lock(&em_tree->lock);
792 em = search_extent_mapping(em_tree, start, num_bytes);
793 if (em) {
794 /*
795 * if block start isn't an actual block number then find the
796 * first block in this inode and use that as a hint. If that
797 * block is also bogus then just don't worry about it.
798 */
799 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
800 free_extent_map(em);
801 em = search_extent_mapping(em_tree, 0, 0);
802 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
803 alloc_hint = em->block_start;
804 if (em)
805 free_extent_map(em);
806 } else {
807 alloc_hint = em->block_start;
808 free_extent_map(em);
809 }
810 }
811 read_unlock(&em_tree->lock);
812
813 return alloc_hint;
814}
815
771ed689
CM		 816/*
817 * when extent_io.c finds a delayed allocation range in the file,
818 * the call backs end up in this code. The basic idea is to
819 * allocate extents on disk for the range, and create ordered data structs
820 * in ram to track those extents.
821 *
822 * locked_page is the page that writepage had locked already. We use
823 * it to make sure we don't do extra locks or unlocks.
824 *
825 * *page_started is set to one if we unlock locked_page and do everything
826 * required to start IO on it. It may be clean and already done with
827 * IO when we return.
828 */
00361589
JB		 829static noinline int cow_file_range(struct inode *inode,
830 struct page *locked_page,
831 u64 start, u64 end, int *page_started,
832 unsigned long *nr_written,
833 int unlock)
771ed689 834{
00361589 835 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 836 u64 alloc_hint = 0;
837 u64 num_bytes;
838 unsigned long ram_size;
839 u64 disk_num_bytes;
840 u64 cur_alloc_size;
841 u64 blocksize = root->sectorsize;
771ed689
CM		 842 struct btrfs_key ins;
843 struct extent_map *em;
844 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
845 int ret = 0;
846
83eea1f1 847 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 848
fda2832f 849 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 850 num_bytes = max(blocksize, num_bytes);
851 disk_num_bytes = num_bytes;
771ed689 852
4cb5300b 853 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 854 if (num_bytes < 64 * 1024 &&
855 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 856 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 857
771ed689
CM		 858 if (start == 0) {
859 /* lets try to make an inline extent */
00361589
JB		 860 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
861 NULL);
771ed689 862 if (ret == 0) {
c2790a2e
JB		 863 extent_clear_unlock_delalloc(inode, start, end, NULL,
864 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 865 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 866 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
867 PAGE_END_WRITEBACK);
c2167754 868
771ed689
CM		 869 *nr_written = *nr_written +
870 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
871 *page_started = 1;
771ed689 872 goto out;
79787eaa 873 } else if (ret < 0) {
79787eaa 874 goto out_unlock;
771ed689
CM		 875 }
876 }
877
878 BUG_ON(disk_num_bytes >
6c41761f 879 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 880
4b46fce2 881 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 882 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
883
d397712b 884 while (disk_num_bytes > 0) {
a791e35e
CM		 885 unsigned long op;
886
287a0ab9 887 cur_alloc_size = disk_num_bytes;
00361589 888 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 889 root->sectorsize, 0, alloc_hint,
81c9ad23 890 &ins, 1);
00361589 891 if (ret < 0)
79787eaa 892 goto out_unlock;
d397712b 893
172ddd60 894 em = alloc_extent_map();
b9aa55be
LB		 895 if (!em) {
896 ret = -ENOMEM;
ace68bac 897 goto out_reserve;
b9aa55be 898 }
e6dcd2dc 899 em->start = start;
445a6944 900 em->orig_start = em->start;
771ed689
CM		 901 ram_size = ins.offset;
902 em->len = ins.offset;
2ab28f32
JB		 903 em->mod_start = em->start;
904 em->mod_len = em->len;
c8b97818 905
e6dcd2dc 906 em->block_start = ins.objectid;
c8b97818 907 em->block_len = ins.offset;
b4939680 908 em->orig_block_len = ins.offset;
cc95bef6 909 em->ram_bytes = ram_size;
e6dcd2dc 910 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 911 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 912 em->generation = -1;
c8b97818 913
d397712b 914 while (1) {
890871be 915 write_lock(&em_tree->lock);
09a2a8f9 916 ret = add_extent_mapping(em_tree, em, 1);
890871be 917 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 918 if (ret != -EEXIST) {
919 free_extent_map(em);
920 break;
921 }
922 btrfs_drop_extent_cache(inode, start,
c8b97818 923 start + ram_size - 1, 0);
e6dcd2dc 924 }
ace68bac
LB		 925 if (ret)
926 goto out_reserve;
e6dcd2dc 927
98d20f67 928 cur_alloc_size = ins.offset;
e6dcd2dc 929 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 930 ram_size, cur_alloc_size, 0);
ace68bac
LB		 931 if (ret)
932 goto out_reserve;
c8b97818 933
17d217fe
YZ		 934 if (root->root_key.objectid ==
935 BTRFS_DATA_RELOC_TREE_OBJECTID) {
936 ret = btrfs_reloc_clone_csums(inode, start,
937 cur_alloc_size);
00361589 938 if (ret)
ace68bac 939 goto out_reserve;
17d217fe
YZ		 940 }
941
d397712b 942 if (disk_num_bytes < cur_alloc_size)
3b951516 943 break;
d397712b 944
c8b97818
CM		 945 /* we're not doing compressed IO, don't unlock the first
946 * page (which the caller expects to stay locked), don't
947 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 948 *
949 * Do set the Private2 bit so we know this page was properly
950 * setup for writepage
c8b97818 951 */
c2790a2e
JB		 952 op = unlock ? PAGE_UNLOCK : 0;
953 op |= PAGE_SET_PRIVATE2;
a791e35e 954
c2790a2e
JB		 955 extent_clear_unlock_delalloc(inode, start,
956 start + ram_size - 1, locked_page,
957 EXTENT_LOCKED | EXTENT_DELALLOC,
958 op);
c8b97818 959 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 960 num_bytes -= cur_alloc_size;
961 alloc_hint = ins.objectid + ins.offset;
962 start += cur_alloc_size;
b888db2b 963 }
79787eaa 964out:
be20aa9d 965 return ret;
b7d5b0a8 966
ace68bac
LB		 967out_reserve:
968 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 969out_unlock:
c2790a2e 970 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 971 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
972 EXTENT_DELALLOC | EXTENT_DEFRAG,
973 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
974 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 975 goto out;
771ed689 976}
c8b97818 977
771ed689
CM		 978/*
979 * work queue call back to started compression on a file and pages
980 */
981static noinline void async_cow_start(struct btrfs_work *work)
982{
983 struct async_cow *async_cow;
984 int num_added = 0;
985 async_cow = container_of(work, struct async_cow, work);
986
987 compress_file_range(async_cow->inode, async_cow->locked_page,
988 async_cow->start, async_cow->end, async_cow,
989 &num_added);
8180ef88 990 if (num_added == 0) {
cb77fcd8 991 btrfs_add_delayed_iput(async_cow->inode);
771ed689 992 async_cow->inode = NULL;
8180ef88 993 }
771ed689
CM		 994}
995
996/*
997 * work queue call back to submit previously compressed pages
998 */
999static noinline void async_cow_submit(struct btrfs_work *work)
1000{
1001 struct async_cow *async_cow;
1002 struct btrfs_root *root;
1003 unsigned long nr_pages;
1004
1005 async_cow = container_of(work, struct async_cow, work);
1006
1007 root = async_cow->root;
1008 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1009 PAGE_CACHE_SHIFT;
1010
66657b31 1011 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1012 5 * 1024 * 1024 &&
771ed689
CM		 1013 waitqueue_active(&root->fs_info->async_submit_wait))
1014 wake_up(&root->fs_info->async_submit_wait);
1015
d397712b 1016 if (async_cow->inode)
771ed689 1017 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1018}
c8b97818 1019
771ed689
CM		 1020static noinline void async_cow_free(struct btrfs_work *work)
1021{
1022 struct async_cow *async_cow;
1023 async_cow = container_of(work, struct async_cow, work);
8180ef88 1024 if (async_cow->inode)
cb77fcd8 1025 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1026 kfree(async_cow);
1027}
1028
1029static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1030 u64 start, u64 end, int *page_started,
1031 unsigned long *nr_written)
1032{
1033 struct async_cow *async_cow;
1034 struct btrfs_root *root = BTRFS_I(inode)->root;
1035 unsigned long nr_pages;
1036 u64 cur_end;
287082b0 1037 int limit = 10 * 1024 * 1024;
771ed689 1038
a3429ab7
CM		 1039 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1040 1, 0, NULL, GFP_NOFS);
d397712b 1041 while (start < end) {
771ed689 1042 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1043 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1044 async_cow->inode = igrab(inode);
771ed689
CM		 1045 async_cow->root = root;
1046 async_cow->locked_page = locked_page;
1047 async_cow->start = start;
1048
6cbff00f 1049 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1050 cur_end = end;
1051 else
1052 cur_end = min(end, start + 512 * 1024 - 1);
1053
1054 async_cow->end = cur_end;
1055 INIT_LIST_HEAD(&async_cow->extents);
1056
1057 async_cow->work.func = async_cow_start;
1058 async_cow->work.ordered_func = async_cow_submit;
1059 async_cow->work.ordered_free = async_cow_free;
1060 async_cow->work.flags = 0;
1061
771ed689
CM		 1062 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1063 PAGE_CACHE_SHIFT;
1064 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1065
1066 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1067 &async_cow->work);
1068
1069 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1070 wait_event(root->fs_info->async_submit_wait,
1071 (atomic_read(&root->fs_info->async_delalloc_pages) <
1072 limit));
1073 }
1074
d397712b 1075 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1076 atomic_read(&root->fs_info->async_delalloc_pages)) {
1077 wait_event(root->fs_info->async_submit_wait,
1078 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1079 0));
1080 }
1081
1082 *nr_written += nr_pages;
1083 start = cur_end + 1;
1084 }
1085 *page_started = 1;
1086 return 0;
be20aa9d
CM		 1087}
1088
d397712b 1089static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1090 u64 bytenr, u64 num_bytes)
1091{
1092 int ret;
1093 struct btrfs_ordered_sum *sums;
1094 LIST_HEAD(list);
1095
07d400a6 1096 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1097 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1098 if (ret == 0 && list_empty(&list))
1099 return 0;
1100
1101 while (!list_empty(&list)) {
1102 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1103 list_del(&sums->list);
1104 kfree(sums);
1105 }
1106 return 1;
1107}
1108
d352ac68
CM		 1109/*
1110 * when nowcow writeback call back. This checks for snapshots or COW copies
1111 * of the extents that exist in the file, and COWs the file as required.
1112 *
1113 * If no cow copies or snapshots exist, we write directly to the existing
1114 * blocks on disk
1115 */
7f366cfe
CM		 1116static noinline int run_delalloc_nocow(struct inode *inode,
1117 struct page *locked_page,
771ed689
CM		 1118 u64 start, u64 end, int *page_started, int force,
1119 unsigned long *nr_written)
be20aa9d 1120{
be20aa9d 1121 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1122 struct btrfs_trans_handle *trans;
be20aa9d 1123 struct extent_buffer *leaf;
be20aa9d 1124 struct btrfs_path *path;
80ff3856 1125 struct btrfs_file_extent_item *fi;
be20aa9d 1126 struct btrfs_key found_key;
80ff3856
YZ		 1127 u64 cow_start;
1128 u64 cur_offset;
1129 u64 extent_end;
5d4f98a2 1130 u64 extent_offset;
80ff3856
YZ		 1131 u64 disk_bytenr;
1132 u64 num_bytes;
b4939680 1133 u64 disk_num_bytes;
cc95bef6 1134 u64 ram_bytes;
80ff3856 1135 int extent_type;
79787eaa 1136 int ret, err;
d899e052 1137 int type;
80ff3856
YZ		 1138 int nocow;
1139 int check_prev = 1;
82d5902d 1140 bool nolock;
33345d01 1141 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1142
1143 path = btrfs_alloc_path();
17ca04af 1144 if (!path) {
c2790a2e
JB		 1145 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1146 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1147 EXTENT_DO_ACCOUNTING |
1148 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1149 PAGE_CLEAR_DIRTY |
1150 PAGE_SET_WRITEBACK |
1151 PAGE_END_WRITEBACK);
d8926bb3 1152 return -ENOMEM;
17ca04af 1153 }
82d5902d 1154
83eea1f1 1155 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1156
1157 if (nolock)
7a7eaa40 1158 trans = btrfs_join_transaction_nolock(root);
82d5902d 1159 else
7a7eaa40 1160 trans = btrfs_join_transaction(root);
ff5714cc 1161
79787eaa 1162 if (IS_ERR(trans)) {
c2790a2e
JB		 1163 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1164 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1165 EXTENT_DO_ACCOUNTING |
1166 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1167 PAGE_CLEAR_DIRTY |
1168 PAGE_SET_WRITEBACK |
1169 PAGE_END_WRITEBACK);
79787eaa
JM		 1170 btrfs_free_path(path);
1171 return PTR_ERR(trans);
1172 }
1173
74b21075 1174 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1175
80ff3856
YZ		 1176 cow_start = (u64)-1;
1177 cur_offset = start;
1178 while (1) {
33345d01 1179 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1180 cur_offset, 0);
79787eaa
JM		 1181 if (ret < 0) {
1182 btrfs_abort_transaction(trans, root, ret);
1183 goto error;
1184 }
80ff3856
YZ		 1185 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key,
1188 path->slots[0] - 1);
33345d01 1189 if (found_key.objectid == ino &&
80ff3856
YZ		 1190 found_key.type == BTRFS_EXTENT_DATA_KEY)
1191 path->slots[0]--;
1192 }
1193 check_prev = 0;
1194next_slot:
1195 leaf = path->nodes[0];
1196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1198 if (ret < 0) {
1199 btrfs_abort_transaction(trans, root, ret);
1200 goto error;
1201 }
80ff3856
YZ		 1202 if (ret > 0)
1203 break;
1204 leaf = path->nodes[0];
1205 }
be20aa9d 1206
80ff3856
YZ		 1207 nocow = 0;
1208 disk_bytenr = 0;
17d217fe 1209 num_bytes = 0;
80ff3856
YZ		 1210 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1211
33345d01 1212 if (found_key.objectid > ino ||
80ff3856
YZ		 1213 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1214 found_key.offset > end)
1215 break;
1216
1217 if (found_key.offset > cur_offset) {
1218 extent_end = found_key.offset;
e9061e21 1219 extent_type = 0;
80ff3856
YZ		 1220 goto out_check;
1221 }
1222
1223 fi = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_file_extent_item);
1225 extent_type = btrfs_file_extent_type(leaf, fi);
1226
cc95bef6 1227 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1228 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1230 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1231 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1232 extent_end = found_key.offset +
1233 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1234 disk_num_bytes =
1235 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1236 if (extent_end <= start) {
1237 path->slots[0]++;
1238 goto next_slot;
1239 }
17d217fe
YZ		 1240 if (disk_bytenr == 0)
1241 goto out_check;
80ff3856
YZ		 1242 if (btrfs_file_extent_compression(leaf, fi) ||
1243 btrfs_file_extent_encryption(leaf, fi) ||
1244 btrfs_file_extent_other_encoding(leaf, fi))
1245 goto out_check;
d899e052
YZ		 1246 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247 goto out_check;
d2fb3437 1248 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1249 goto out_check;
33345d01 1250 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1251 found_key.offset -
1252 extent_offset, disk_bytenr))
17d217fe 1253 goto out_check;
5d4f98a2 1254 disk_bytenr += extent_offset;
17d217fe
YZ		 1255 disk_bytenr += cur_offset - found_key.offset;
1256 num_bytes = min(end + 1, extent_end) - cur_offset;
1257 /*
1258 * force cow if csum exists in the range.
1259 * this ensure that csum for a given extent are
1260 * either valid or do not exist.
1261 */
1262 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263 goto out_check;
80ff3856
YZ		 1264 nocow = 1;
1265 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266 extent_end = found_key.offset +
1267 btrfs_file_extent_inline_len(leaf, fi);
1268 extent_end = ALIGN(extent_end, root->sectorsize);
1269 } else {
1270 BUG_ON(1);
1271 }
1272out_check:
1273 if (extent_end <= start) {
1274 path->slots[0]++;
1275 goto next_slot;
1276 }
1277 if (!nocow) {
1278 if (cow_start == (u64)-1)
1279 cow_start = cur_offset;
1280 cur_offset = extent_end;
1281 if (cur_offset > end)
1282 break;
1283 path->slots[0]++;
1284 goto next_slot;
7ea394f1
YZ		 1285 }
1286
b3b4aa74 1287 btrfs_release_path(path);
80ff3856 1288 if (cow_start != (u64)-1) {
00361589
JB		 1289 ret = cow_file_range(inode, locked_page,
1290 cow_start, found_key.offset - 1,
1291 page_started, nr_written, 1);
79787eaa
JM		 1292 if (ret) {
1293 btrfs_abort_transaction(trans, root, ret);
1294 goto error;
1295 }
80ff3856 1296 cow_start = (u64)-1;
7ea394f1 1297 }
80ff3856 1298
d899e052
YZ		 1299 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300 struct extent_map *em;
1301 struct extent_map_tree *em_tree;
1302 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1303 em = alloc_extent_map();
79787eaa 1304 BUG_ON(!em); /* -ENOMEM */
d899e052 1305 em->start = cur_offset;
70c8a91c 1306 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1307 em->len = num_bytes;
1308 em->block_len = num_bytes;
1309 em->block_start = disk_bytenr;
b4939680 1310 em->orig_block_len = disk_num_bytes;
cc95bef6 1311 em->ram_bytes = ram_bytes;
d899e052 1312 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1313 em->mod_start = em->start;
1314 em->mod_len = em->len;
d899e052 1315 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1316 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1317 em->generation = -1;
d899e052 1318 while (1) {
890871be 1319 write_lock(&em_tree->lock);
09a2a8f9 1320 ret = add_extent_mapping(em_tree, em, 1);
890871be 1321 write_unlock(&em_tree->lock);
d899e052
YZ		 1322 if (ret != -EEXIST) {
1323 free_extent_map(em);
1324 break;
1325 }
1326 btrfs_drop_extent_cache(inode, em->start,
1327 em->start + em->len - 1, 0);
1328 }
1329 type = BTRFS_ORDERED_PREALLOC;
1330 } else {
1331 type = BTRFS_ORDERED_NOCOW;
1332 }
80ff3856
YZ		 1333
1334 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1335 num_bytes, num_bytes, type);
79787eaa 1336 BUG_ON(ret); /* -ENOMEM */
771ed689 1337
efa56464
YZ		 1338 if (root->root_key.objectid ==
1339 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1340 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1341 num_bytes);
79787eaa
JM		 1342 if (ret) {
1343 btrfs_abort_transaction(trans, root, ret);
1344 goto error;
1345 }
efa56464
YZ		 1346 }
1347
c2790a2e
JB		 1348 extent_clear_unlock_delalloc(inode, cur_offset,
1349 cur_offset + num_bytes - 1,
1350 locked_page, EXTENT_LOCKED |
1351 EXTENT_DELALLOC, PAGE_UNLOCK |
1352 PAGE_SET_PRIVATE2);
80ff3856
YZ		 1353 cur_offset = extent_end;
1354 if (cur_offset > end)
1355 break;
be20aa9d 1356 }
b3b4aa74 1357 btrfs_release_path(path);
80ff3856 1358
17ca04af 1359 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1360 cow_start = cur_offset;
17ca04af
JB		 1361 cur_offset = end;
1362 }
1363
80ff3856 1364 if (cow_start != (u64)-1) {
00361589
JB		 1365 ret = cow_file_range(inode, locked_page, cow_start, end,
1366 page_started, nr_written, 1);
79787eaa
JM		 1367 if (ret) {
1368 btrfs_abort_transaction(trans, root, ret);
1369 goto error;
1370 }
80ff3856
YZ		 1371 }
1372
79787eaa 1373error:
a698d075 1374 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1375 if (!ret)
1376 ret = err;
1377
17ca04af 1378 if (ret && cur_offset < end)
c2790a2e
JB		 1379 extent_clear_unlock_delalloc(inode, cur_offset, end,
1380 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1381 EXTENT_DELALLOC | EXTENT_DEFRAG |
1382 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1383 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1384 PAGE_SET_WRITEBACK |
1385 PAGE_END_WRITEBACK);
7ea394f1 1386 btrfs_free_path(path);
79787eaa 1387 return ret;
be20aa9d
CM		 1388}
1389
d352ac68
CM		 1390/*
1391 * extent_io.c call back to do delayed allocation processing
1392 */
c8b97818 1393static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1394 u64 start, u64 end, int *page_started,
1395 unsigned long *nr_written)
be20aa9d 1396{
be20aa9d 1397 int ret;
7f366cfe 1398 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1399
7ddf5a42 1400 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1401 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1402 page_started, 1, nr_written);
7ddf5a42 1403 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1404 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1405 page_started, 0, nr_written);
7ddf5a42
JB		 1406 } else if (!btrfs_test_opt(root, COMPRESS) &&
1407 !(BTRFS_I(inode)->force_compress) &&
1408 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1409 ret = cow_file_range(inode, locked_page, start, end,
1410 page_started, nr_written, 1);
7ddf5a42
JB		 1411 } else {
1412 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1413 &BTRFS_I(inode)->runtime_flags);
771ed689 1414 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1415 page_started, nr_written);
7ddf5a42 1416 }
b888db2b
CM		 1417 return ret;
1418}
1419
1bf85046
JM		 1420static void btrfs_split_extent_hook(struct inode *inode,
1421 struct extent_state *orig, u64 split)
9ed74f2d 1422{
0ca1f7ce 1423 /* not delalloc, ignore it */
9ed74f2d 1424 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1425 return;
9ed74f2d 1426
9e0baf60
JB		 1427 spin_lock(&BTRFS_I(inode)->lock);
1428 BTRFS_I(inode)->outstanding_extents++;
1429 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1430}
1431
1432/*
1433 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1434 * extents so we can keep track of new extents that are just merged onto old
1435 * extents, such as when we are doing sequential writes, so we can properly
1436 * account for the metadata space we'll need.
1437 */
1bf85046
JM		 1438static void btrfs_merge_extent_hook(struct inode *inode,
1439 struct extent_state *new,
1440 struct extent_state *other)
9ed74f2d 1441{
9ed74f2d
JB		 1442 /* not delalloc, ignore it */
1443 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1444 return;
9ed74f2d 1445
9e0baf60
JB		 1446 spin_lock(&BTRFS_I(inode)->lock);
1447 BTRFS_I(inode)->outstanding_extents--;
1448 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1449}
1450
eb73c1b7
MX		 1451static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1452 struct inode *inode)
1453{
1454 spin_lock(&root->delalloc_lock);
1455 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1456 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1457 &root->delalloc_inodes);
1458 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1459 &BTRFS_I(inode)->runtime_flags);
1460 root->nr_delalloc_inodes++;
1461 if (root->nr_delalloc_inodes == 1) {
1462 spin_lock(&root->fs_info->delalloc_root_lock);
1463 BUG_ON(!list_empty(&root->delalloc_root));
1464 list_add_tail(&root->delalloc_root,
1465 &root->fs_info->delalloc_roots);
1466 spin_unlock(&root->fs_info->delalloc_root_lock);
1467 }
1468 }
1469 spin_unlock(&root->delalloc_lock);
1470}
1471
1472static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1473 struct inode *inode)
1474{
1475 spin_lock(&root->delalloc_lock);
1476 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1477 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1478 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1479 &BTRFS_I(inode)->runtime_flags);
1480 root->nr_delalloc_inodes--;
1481 if (!root->nr_delalloc_inodes) {
1482 spin_lock(&root->fs_info->delalloc_root_lock);
1483 BUG_ON(list_empty(&root->delalloc_root));
1484 list_del_init(&root->delalloc_root);
1485 spin_unlock(&root->fs_info->delalloc_root_lock);
1486 }
1487 }
1488 spin_unlock(&root->delalloc_lock);
1489}
1490
d352ac68
CM		 1491/*
1492 * extent_io.c set_bit_hook, used to track delayed allocation
1493 * bytes in this file, and to maintain the list of inodes that
1494 * have pending delalloc work to be done.
1495 */
1bf85046 1496static void btrfs_set_bit_hook(struct inode *inode,
41074888 1497 struct extent_state *state, unsigned long *bits)
291d673e 1498{
9ed74f2d 1499
75eff68e
CM		 1500 /*
1501 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1502 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1503 * bit, which is only set or cleared with irqs on
1504 */
0ca1f7ce 1505 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1506 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1507 u64 len = state->end + 1 - state->start;
83eea1f1 1508 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1509
9e0baf60 1510 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1511 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1512 } else {
1513 spin_lock(&BTRFS_I(inode)->lock);
1514 BTRFS_I(inode)->outstanding_extents++;
1515 spin_unlock(&BTRFS_I(inode)->lock);
1516 }
287a0ab9 1517
963d678b
MX		 1518 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1519 root->fs_info->delalloc_batch);
df0af1a5 1520 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1521 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5 1522 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1523 &BTRFS_I(inode)->runtime_flags))
1524 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1525 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1526 }
291d673e
CM		 1527}
1528
d352ac68
CM		 1529/*
1530 * extent_io.c clear_bit_hook, see set_bit_hook for why
1531 */
1bf85046 1532static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1533 struct extent_state *state,
1534 unsigned long *bits)
291d673e 1535{
75eff68e
CM		 1536 /*
1537 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1538 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1539 * bit, which is only set or cleared with irqs on
1540 */
0ca1f7ce 1541 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1542 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1543 u64 len = state->end + 1 - state->start;
83eea1f1 1544 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1545
9e0baf60 1546 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1547 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1548 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1549 spin_lock(&BTRFS_I(inode)->lock);
1550 BTRFS_I(inode)->outstanding_extents--;
1551 spin_unlock(&BTRFS_I(inode)->lock);
1552 }
0ca1f7ce
YZ		 1553
1554 if (*bits & EXTENT_DO_ACCOUNTING)
1555 btrfs_delalloc_release_metadata(inode, len);
1556
0cb59c99 1557 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1558 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1559 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1560
963d678b
MX		 1561 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1562 root->fs_info->delalloc_batch);
df0af1a5 1563 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1564 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1565 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1566 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1567 &BTRFS_I(inode)->runtime_flags))
1568 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1569 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1570 }
291d673e
CM		 1571}
1572
d352ac68
CM		 1573/*
1574 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1575 * we don't create bios that span stripes or chunks
1576 */
64a16701 1577int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1578 size_t size, struct bio *bio,
1579 unsigned long bio_flags)
239b14b3
CM		 1580{
1581 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1582 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1583 u64 length = 0;
1584 u64 map_length;
239b14b3
CM		 1585 int ret;
1586
771ed689
CM		 1587 if (bio_flags & EXTENT_BIO_COMPRESSED)
1588 return 0;
1589
f2d8d74d 1590 length = bio->bi_size;
239b14b3 1591 map_length = length;
64a16701 1592 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1593 &map_length, NULL, 0);
3ec706c8 1594 /* Will always return 0 with map_multi == NULL */
3444a972 1595 BUG_ON(ret < 0);
d397712b 1596 if (map_length < length + size)
239b14b3 1597 return 1;
3444a972 1598 return 0;
239b14b3
CM		 1599}
1600
d352ac68
CM		 1601/*
1602 * in order to insert checksums into the metadata in large chunks,
1603 * we wait until bio submission time. All the pages in the bio are
1604 * checksummed and sums are attached onto the ordered extent record.
1605 *
1606 * At IO completion time the cums attached on the ordered extent record
1607 * are inserted into the btree
1608 */
d397712b
CM		 1609static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1610 struct bio *bio, int mirror_num,
eaf25d93
CM		 1611 unsigned long bio_flags,
1612 u64 bio_offset)
065631f6 1613{
065631f6 1614 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1615 int ret = 0;
e015640f 1616
d20f7043 1617 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1618 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1619 return 0;
1620}
e015640f 1621
4a69a410
CM		 1622/*
1623 * in order to insert checksums into the metadata in large chunks,
1624 * we wait until bio submission time. All the pages in the bio are
1625 * checksummed and sums are attached onto the ordered extent record.
1626 *
1627 * At IO completion time the cums attached on the ordered extent record
1628 * are inserted into the btree
1629 */
b2950863 1630static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1631 int mirror_num, unsigned long bio_flags,
1632 u64 bio_offset)
4a69a410
CM		 1633{
1634 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1635 int ret;
1636
1637 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1638 if (ret)
1639 bio_endio(bio, ret);
1640 return ret;
44b8bd7e
CM		 1641}
1642
d352ac68 1643/*
cad321ad
CM		 1644 * extent_io.c submission hook. This does the right thing for csum calculation
1645 * on write, or reading the csums from the tree before a read
d352ac68 1646 */
b2950863 1647static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1648 int mirror_num, unsigned long bio_flags,
1649 u64 bio_offset)
44b8bd7e
CM		 1650{
1651 struct btrfs_root *root = BTRFS_I(inode)->root;
1652 int ret = 0;
19b9bdb0 1653 int skip_sum;
0417341e 1654 int metadata = 0;
b812ce28 1655 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1656
6cbff00f 1657 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1658
83eea1f1 1659 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1660 metadata = 2;
1661
7b6d91da 1662 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1663 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1664 if (ret)
61891923 1665 goto out;
5fd02043 1666
d20f7043 1667 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1668 ret = btrfs_submit_compressed_read(inode, bio,
1669 mirror_num,
1670 bio_flags);
1671 goto out;
c2db1073
TI		 1672 } else if (!skip_sum) {
1673 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1674 if (ret)
61891923 1675 goto out;
c2db1073 1676 }
4d1b5fb4 1677 goto mapit;
b812ce28 1678 } else if (async && !skip_sum) {
17d217fe
YZ		 1679 /* csum items have already been cloned */
1680 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1681 goto mapit;
19b9bdb0 1682 /* we're doing a write, do the async checksumming */
61891923 1683 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1684 inode, rw, bio, mirror_num,
eaf25d93
CM		 1685 bio_flags, bio_offset,
1686 __btrfs_submit_bio_start,
4a69a410 1687 __btrfs_submit_bio_done);
61891923 1688 goto out;
b812ce28
JB		 1689 } else if (!skip_sum) {
1690 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1691 if (ret)
1692 goto out;
19b9bdb0
CM		 1693 }
1694
0b86a832 1695mapit:
61891923
SB		 1696 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1697
1698out:
1699 if (ret < 0)
1700 bio_endio(bio, ret);
1701 return ret;
065631f6 1702}
6885f308 1703
d352ac68
CM		 1704/*
1705 * given a list of ordered sums record them in the inode. This happens
1706 * at IO completion time based on sums calculated at bio submission time.
1707 */
ba1da2f4 1708static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1709 struct inode *inode, u64 file_offset,
1710 struct list_head *list)
1711{
e6dcd2dc
CM		 1712 struct btrfs_ordered_sum *sum;
1713
c6e30871 1714 list_for_each_entry(sum, list, list) {
39847c4d 1715 trans->adding_csums = 1;
d20f7043
CM		 1716 btrfs_csum_file_blocks(trans,
1717 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1718 trans->adding_csums = 0;
e6dcd2dc
CM		 1719 }
1720 return 0;
1721}
1722
2ac55d41
JB		 1723int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1724 struct extent_state **cached_state)
ea8c2819 1725{
6c1500f2 1726 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1727 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1728 cached_state, GFP_NOFS);
ea8c2819
CM		 1729}
1730
d352ac68 1731/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1732struct btrfs_writepage_fixup {
1733 struct page *page;
1734 struct btrfs_work work;
1735};
1736
b2950863 1737static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1738{
1739 struct btrfs_writepage_fixup *fixup;
1740 struct btrfs_ordered_extent *ordered;
2ac55d41 1741 struct extent_state *cached_state = NULL;
247e743c
CM		 1742 struct page *page;
1743 struct inode *inode;
1744 u64 page_start;
1745 u64 page_end;
87826df0 1746 int ret;
247e743c
CM		 1747
1748 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1749 page = fixup->page;
4a096752 1750again:
247e743c
CM		 1751 lock_page(page);
1752 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1753 ClearPageChecked(page);
1754 goto out_page;
1755 }
1756
1757 inode = page->mapping->host;
1758 page_start = page_offset(page);
1759 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1760
2ac55d41 1761 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1762 &cached_state);
4a096752
CM		 1763
1764 /* already ordered? We're done */
8b62b72b 1765 if (PagePrivate2(page))
247e743c 1766 goto out;
4a096752
CM		 1767
1768 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1769 if (ordered) {
2ac55d41
JB		 1770 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1771 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1772 unlock_page(page);
1773 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1774 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1775 goto again;
1776 }
247e743c 1777
87826df0
JM		 1778 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1779 if (ret) {
1780 mapping_set_error(page->mapping, ret);
1781 end_extent_writepage(page, ret, page_start, page_end);
1782 ClearPageChecked(page);
1783 goto out;
1784 }
1785
2ac55d41 1786 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1787 ClearPageChecked(page);
87826df0 1788 set_page_dirty(page);
247e743c 1789out:
2ac55d41
JB		 1790 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1791 &cached_state, GFP_NOFS);
247e743c
CM		 1792out_page:
1793 unlock_page(page);
1794 page_cache_release(page);
b897abec 1795 kfree(fixup);
247e743c
CM		 1796}
1797
1798/*
1799 * There are a few paths in the higher layers of the kernel that directly
1800 * set the page dirty bit without asking the filesystem if it is a
1801 * good idea. This causes problems because we want to make sure COW
1802 * properly happens and the data=ordered rules are followed.
1803 *
c8b97818 1804 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1805 * hasn't been properly setup for IO. We kick off an async process
1806 * to fix it up. The async helper will wait for ordered extents, set
1807 * the delalloc bit and make it safe to write the page.
1808 */
b2950863 1809static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1810{
1811 struct inode *inode = page->mapping->host;
1812 struct btrfs_writepage_fixup *fixup;
1813 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1814
8b62b72b
CM		 1815 /* this page is properly in the ordered list */
1816 if (TestClearPagePrivate2(page))
247e743c
CM		 1817 return 0;
1818
1819 if (PageChecked(page))
1820 return -EAGAIN;
1821
1822 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1823 if (!fixup)
1824 return -EAGAIN;
f421950f 1825
247e743c
CM		 1826 SetPageChecked(page);
1827 page_cache_get(page);
1828 fixup->work.func = btrfs_writepage_fixup_worker;
1829 fixup->page = page;
1830 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1831 return -EBUSY;
247e743c
CM		 1832}
1833
d899e052
YZ		 1834static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1835 struct inode *inode, u64 file_pos,
1836 u64 disk_bytenr, u64 disk_num_bytes,
1837 u64 num_bytes, u64 ram_bytes,
1838 u8 compression, u8 encryption,
1839 u16 other_encoding, int extent_type)
1840{
1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 struct btrfs_file_extent_item *fi;
1843 struct btrfs_path *path;
1844 struct extent_buffer *leaf;
1845 struct btrfs_key ins;
d899e052
YZ		 1846 int ret;
1847
1848 path = btrfs_alloc_path();
d8926bb3
MF		 1849 if (!path)
1850 return -ENOMEM;
d899e052 1851
b9473439 1852 path->leave_spinning = 1;
a1ed835e
CM		 1853
1854 /*
1855 * we may be replacing one extent in the tree with another.
1856 * The new extent is pinned in the extent map, and we don't want
1857 * to drop it from the cache until it is completely in the btree.
1858 *
1859 * So, tell btrfs_drop_extents to leave this extent in the cache.
1860 * the caller is expected to unpin it and allow it to be merged
1861 * with the others.
1862 */
5dc562c5 1863 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1864 file_pos + num_bytes, 0);
79787eaa
JM		 1865 if (ret)
1866 goto out;
d899e052 1867
33345d01 1868 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1869 ins.offset = file_pos;
1870 ins.type = BTRFS_EXTENT_DATA_KEY;
1871 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1872 if (ret)
1873 goto out;
d899e052
YZ		 1874 leaf = path->nodes[0];
1875 fi = btrfs_item_ptr(leaf, path->slots[0],
1876 struct btrfs_file_extent_item);
1877 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1878 btrfs_set_file_extent_type(leaf, fi, extent_type);
1879 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1880 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1881 btrfs_set_file_extent_offset(leaf, fi, 0);
1882 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1883 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1884 btrfs_set_file_extent_compression(leaf, fi, compression);
1885 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1886 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1887
d899e052 1888 btrfs_mark_buffer_dirty(leaf);
ce195332 1889 btrfs_release_path(path);
d899e052
YZ		 1890
1891 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1892
1893 ins.objectid = disk_bytenr;
1894 ins.offset = disk_num_bytes;
1895 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1896 ret = btrfs_alloc_reserved_file_extent(trans, root,
1897 root->root_key.objectid,
33345d01 1898 btrfs_ino(inode), file_pos, &ins);
79787eaa 1899out:
d899e052 1900 btrfs_free_path(path);
b9473439 1901
79787eaa 1902 return ret;
d899e052
YZ		 1903}
1904
38c227d8
LB		 1905/* snapshot-aware defrag */
1906struct sa_defrag_extent_backref {
1907 struct rb_node node;
1908 struct old_sa_defrag_extent *old;
1909 u64 root_id;
1910 u64 inum;
1911 u64 file_pos;
1912 u64 extent_offset;
1913 u64 num_bytes;
1914 u64 generation;
1915};
1916
1917struct old_sa_defrag_extent {
1918 struct list_head list;
1919 struct new_sa_defrag_extent *new;
1920
1921 u64 extent_offset;
1922 u64 bytenr;
1923 u64 offset;
1924 u64 len;
1925 int count;
1926};
1927
1928struct new_sa_defrag_extent {
1929 struct rb_root root;
1930 struct list_head head;
1931 struct btrfs_path *path;
1932 struct inode *inode;
1933 u64 file_pos;
1934 u64 len;
1935 u64 bytenr;
1936 u64 disk_len;
1937 u8 compress_type;
1938};
1939
1940static int backref_comp(struct sa_defrag_extent_backref *b1,
1941 struct sa_defrag_extent_backref *b2)
1942{
1943 if (b1->root_id < b2->root_id)
1944 return -1;
1945 else if (b1->root_id > b2->root_id)
1946 return 1;
1947
1948 if (b1->inum < b2->inum)
1949 return -1;
1950 else if (b1->inum > b2->inum)
1951 return 1;
1952
1953 if (b1->file_pos < b2->file_pos)
1954 return -1;
1955 else if (b1->file_pos > b2->file_pos)
1956 return 1;
1957
1958 /*
1959 * [------------------------------] ===> (a range of space)
1960 * |<--->| |<---->| =============> (fs/file tree A)
1961 * |<---------------------------->| ===> (fs/file tree B)
1962 *
1963 * A range of space can refer to two file extents in one tree while
1964 * refer to only one file extent in another tree.
1965 *
1966 * So we may process a disk offset more than one time(two extents in A)
1967 * and locate at the same extent(one extent in B), then insert two same
1968 * backrefs(both refer to the extent in B).
1969 */
1970 return 0;
1971}
1972
1973static void backref_insert(struct rb_root *root,
1974 struct sa_defrag_extent_backref *backref)
1975{
1976 struct rb_node **p = &root->rb_node;
1977 struct rb_node *parent = NULL;
1978 struct sa_defrag_extent_backref *entry;
1979 int ret;
1980
1981 while (*p) {
1982 parent = *p;
1983 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1984
1985 ret = backref_comp(backref, entry);
1986 if (ret < 0)
1987 p = &(*p)->rb_left;
1988 else
1989 p = &(*p)->rb_right;
1990 }
1991
1992 rb_link_node(&backref->node, parent, p);
1993 rb_insert_color(&backref->node, root);
1994}
1995
1996/*
1997 * Note the backref might has changed, and in this case we just return 0.
1998 */
1999static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2000 void *ctx)
2001{
2002 struct btrfs_file_extent_item *extent;
2003 struct btrfs_fs_info *fs_info;
2004 struct old_sa_defrag_extent *old = ctx;
2005 struct new_sa_defrag_extent *new = old->new;
2006 struct btrfs_path *path = new->path;
2007 struct btrfs_key key;
2008 struct btrfs_root *root;
2009 struct sa_defrag_extent_backref *backref;
2010 struct extent_buffer *leaf;
2011 struct inode *inode = new->inode;
2012 int slot;
2013 int ret;
2014 u64 extent_offset;
2015 u64 num_bytes;
2016
2017 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2018 inum == btrfs_ino(inode))
2019 return 0;
2020
2021 key.objectid = root_id;
2022 key.type = BTRFS_ROOT_ITEM_KEY;
2023 key.offset = (u64)-1;
2024
2025 fs_info = BTRFS_I(inode)->root->fs_info;
2026 root = btrfs_read_fs_root_no_name(fs_info, &key);
2027 if (IS_ERR(root)) {
2028 if (PTR_ERR(root) == -ENOENT)
2029 return 0;
2030 WARN_ON(1);
2031 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2032 inum, offset, root_id);
2033 return PTR_ERR(root);
2034 }
2035
2036 key.objectid = inum;
2037 key.type = BTRFS_EXTENT_DATA_KEY;
2038 if (offset > (u64)-1 << 32)
2039 key.offset = 0;
2040 else
2041 key.offset = offset;
2042
2043 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2044 if (ret < 0) {
2045 WARN_ON(1);
2046 return ret;
2047 }
50f1319c 2048 ret = 0;
38c227d8
LB		 2049
2050 while (1) {
2051 cond_resched();
2052
2053 leaf = path->nodes[0];
2054 slot = path->slots[0];
2055
2056 if (slot >= btrfs_header_nritems(leaf)) {
2057 ret = btrfs_next_leaf(root, path);
2058 if (ret < 0) {
2059 goto out;
2060 } else if (ret > 0) {
2061 ret = 0;
2062 goto out;
2063 }
2064 continue;
2065 }
2066
2067 path->slots[0]++;
2068
2069 btrfs_item_key_to_cpu(leaf, &key, slot);
2070
2071 if (key.objectid > inum)
2072 goto out;
2073
2074 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2075 continue;
2076
2077 extent = btrfs_item_ptr(leaf, slot,
2078 struct btrfs_file_extent_item);
2079
2080 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2081 continue;
2082
e68afa49
LB		 2083 /*
2084 * 'offset' refers to the exact key.offset,
2085 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2086 * (key.offset - extent_offset).
2087 */
2088 if (key.offset != offset)
38c227d8
LB		 2089 continue;
2090
e68afa49 2091 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2092 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2093
38c227d8
LB		 2094 if (extent_offset >= old->extent_offset + old->offset +
2095 old->len || extent_offset + num_bytes <=
2096 old->extent_offset + old->offset)
2097 continue;
38c227d8
LB		 2098 break;
2099 }
2100
2101 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2102 if (!backref) {
2103 ret = -ENOENT;
2104 goto out;
2105 }
2106
2107 backref->root_id = root_id;
2108 backref->inum = inum;
e68afa49 2109 backref->file_pos = offset;
38c227d8
LB		 2110 backref->num_bytes = num_bytes;
2111 backref->extent_offset = extent_offset;
2112 backref->generation = btrfs_file_extent_generation(leaf, extent);
2113 backref->old = old;
2114 backref_insert(&new->root, backref);
2115 old->count++;
2116out:
2117 btrfs_release_path(path);
2118 WARN_ON(ret);
2119 return ret;
2120}
2121
2122static noinline bool record_extent_backrefs(struct btrfs_path *path,
2123 struct new_sa_defrag_extent *new)
2124{
2125 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2126 struct old_sa_defrag_extent *old, *tmp;
2127 int ret;
2128
2129 new->path = path;
2130
2131 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2132 ret = iterate_inodes_from_logical(old->bytenr +
2133 old->extent_offset, fs_info,
38c227d8
LB		 2134 path, record_one_backref,
2135 old);
2136 BUG_ON(ret < 0 && ret != -ENOENT);
2137
2138 /* no backref to be processed for this extent */
2139 if (!old->count) {
2140 list_del(&old->list);
2141 kfree(old);
2142 }
2143 }
2144
2145 if (list_empty(&new->head))
2146 return false;
2147
2148 return true;
2149}
2150
2151static int relink_is_mergable(struct extent_buffer *leaf,
2152 struct btrfs_file_extent_item *fi,
116e0024 2153 struct new_sa_defrag_extent *new)
38c227d8 2154{
116e0024 2155 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2156 return 0;
2157
2158 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2159 return 0;
2160
116e0024
LB		 2161 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2162 return 0;
2163
2164 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2165 btrfs_file_extent_other_encoding(leaf, fi))
2166 return 0;
2167
2168 return 1;
2169}
2170
2171/*
2172 * Note the backref might has changed, and in this case we just return 0.
2173 */
2174static noinline int relink_extent_backref(struct btrfs_path *path,
2175 struct sa_defrag_extent_backref *prev,
2176 struct sa_defrag_extent_backref *backref)
2177{
2178 struct btrfs_file_extent_item *extent;
2179 struct btrfs_file_extent_item *item;
2180 struct btrfs_ordered_extent *ordered;
2181 struct btrfs_trans_handle *trans;
2182 struct btrfs_fs_info *fs_info;
2183 struct btrfs_root *root;
2184 struct btrfs_key key;
2185 struct extent_buffer *leaf;
2186 struct old_sa_defrag_extent *old = backref->old;
2187 struct new_sa_defrag_extent *new = old->new;
2188 struct inode *src_inode = new->inode;
2189 struct inode *inode;
2190 struct extent_state *cached = NULL;
2191 int ret = 0;
2192 u64 start;
2193 u64 len;
2194 u64 lock_start;
2195 u64 lock_end;
2196 bool merge = false;
2197 int index;
2198
2199 if (prev && prev->root_id == backref->root_id &&
2200 prev->inum == backref->inum &&
2201 prev->file_pos + prev->num_bytes == backref->file_pos)
2202 merge = true;
2203
2204 /* step 1: get root */
2205 key.objectid = backref->root_id;
2206 key.type = BTRFS_ROOT_ITEM_KEY;
2207 key.offset = (u64)-1;
2208
2209 fs_info = BTRFS_I(src_inode)->root->fs_info;
2210 index = srcu_read_lock(&fs_info->subvol_srcu);
2211
2212 root = btrfs_read_fs_root_no_name(fs_info, &key);
2213 if (IS_ERR(root)) {
2214 srcu_read_unlock(&fs_info->subvol_srcu, index);
2215 if (PTR_ERR(root) == -ENOENT)
2216 return 0;
2217 return PTR_ERR(root);
2218 }
38c227d8
LB		 2219
2220 /* step 2: get inode */
2221 key.objectid = backref->inum;
2222 key.type = BTRFS_INODE_ITEM_KEY;
2223 key.offset = 0;
2224
2225 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2226 if (IS_ERR(inode)) {
2227 srcu_read_unlock(&fs_info->subvol_srcu, index);
2228 return 0;
2229 }
2230
2231 srcu_read_unlock(&fs_info->subvol_srcu, index);
2232
2233 /* step 3: relink backref */
2234 lock_start = backref->file_pos;
2235 lock_end = backref->file_pos + backref->num_bytes - 1;
2236 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2237 0, &cached);
2238
2239 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2240 if (ordered) {
2241 btrfs_put_ordered_extent(ordered);
2242 goto out_unlock;
2243 }
2244
2245 trans = btrfs_join_transaction(root);
2246 if (IS_ERR(trans)) {
2247 ret = PTR_ERR(trans);
2248 goto out_unlock;
2249 }
2250
2251 key.objectid = backref->inum;
2252 key.type = BTRFS_EXTENT_DATA_KEY;
2253 key.offset = backref->file_pos;
2254
2255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2256 if (ret < 0) {
2257 goto out_free_path;
2258 } else if (ret > 0) {
2259 ret = 0;
2260 goto out_free_path;
2261 }
2262
2263 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2264 struct btrfs_file_extent_item);
2265
2266 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2267 backref->generation)
2268 goto out_free_path;
2269
2270 btrfs_release_path(path);
2271
2272 start = backref->file_pos;
2273 if (backref->extent_offset < old->extent_offset + old->offset)
2274 start += old->extent_offset + old->offset -
2275 backref->extent_offset;
2276
2277 len = min(backref->extent_offset + backref->num_bytes,
2278 old->extent_offset + old->offset + old->len);
2279 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2280
2281 ret = btrfs_drop_extents(trans, root, inode, start,
2282 start + len, 1);
2283 if (ret)
2284 goto out_free_path;
2285again:
2286 key.objectid = btrfs_ino(inode);
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2288 key.offset = start;
2289
a09a0a70 2290 path->leave_spinning = 1;
38c227d8
LB		 2291 if (merge) {
2292 struct btrfs_file_extent_item *fi;
2293 u64 extent_len;
2294 struct btrfs_key found_key;
2295
2296 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2297 if (ret < 0)
2298 goto out_free_path;
2299
2300 path->slots[0]--;
2301 leaf = path->nodes[0];
2302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2303
2304 fi = btrfs_item_ptr(leaf, path->slots[0],
2305 struct btrfs_file_extent_item);
2306 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2307
116e0024
LB		 2308 if (extent_len + found_key.offset == start &&
2309 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2310 btrfs_set_file_extent_num_bytes(leaf, fi,
2311 extent_len + len);
2312 btrfs_mark_buffer_dirty(leaf);
2313 inode_add_bytes(inode, len);
2314
2315 ret = 1;
2316 goto out_free_path;
2317 } else {
2318 merge = false;
2319 btrfs_release_path(path);
2320 goto again;
2321 }
2322 }
2323
2324 ret = btrfs_insert_empty_item(trans, root, path, &key,
2325 sizeof(*extent));
2326 if (ret) {
2327 btrfs_abort_transaction(trans, root, ret);
2328 goto out_free_path;
2329 }
2330
2331 leaf = path->nodes[0];
2332 item = btrfs_item_ptr(leaf, path->slots[0],
2333 struct btrfs_file_extent_item);
2334 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2335 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2336 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2337 btrfs_set_file_extent_num_bytes(leaf, item, len);
2338 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2339 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2340 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2341 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2342 btrfs_set_file_extent_encryption(leaf, item, 0);
2343 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2344
2345 btrfs_mark_buffer_dirty(leaf);
2346 inode_add_bytes(inode, len);
a09a0a70 2347 btrfs_release_path(path);
38c227d8
LB		 2348
2349 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2350 new->disk_len, 0,
2351 backref->root_id, backref->inum,
2352 new->file_pos, 0); /* start - extent_offset */
2353 if (ret) {
2354 btrfs_abort_transaction(trans, root, ret);
2355 goto out_free_path;
2356 }
2357
2358 ret = 1;
2359out_free_path:
2360 btrfs_release_path(path);
a09a0a70 2361 path->leave_spinning = 0;
38c227d8
LB		 2362 btrfs_end_transaction(trans, root);
2363out_unlock:
2364 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2365 &cached, GFP_NOFS);
2366 iput(inode);
2367 return ret;
2368}
2369
2370static void relink_file_extents(struct new_sa_defrag_extent *new)
2371{
2372 struct btrfs_path *path;
2373 struct old_sa_defrag_extent *old, *tmp;
2374 struct sa_defrag_extent_backref *backref;
2375 struct sa_defrag_extent_backref *prev = NULL;
2376 struct inode *inode;
2377 struct btrfs_root *root;
2378 struct rb_node *node;
2379 int ret;
2380
2381 inode = new->inode;
2382 root = BTRFS_I(inode)->root;
2383
2384 path = btrfs_alloc_path();
2385 if (!path)
2386 return;
2387
2388 if (!record_extent_backrefs(path, new)) {
2389 btrfs_free_path(path);
2390 goto out;
2391 }
2392 btrfs_release_path(path);
2393
2394 while (1) {
2395 node = rb_first(&new->root);
2396 if (!node)
2397 break;
2398 rb_erase(node, &new->root);
2399
2400 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2401
2402 ret = relink_extent_backref(path, prev, backref);
2403 WARN_ON(ret < 0);
2404
2405 kfree(prev);
2406
2407 if (ret == 1)
2408 prev = backref;
2409 else
2410 prev = NULL;
2411 cond_resched();
2412 }
2413 kfree(prev);
2414
2415 btrfs_free_path(path);
2416
2417 list_for_each_entry_safe(old, tmp, &new->head, list) {
2418 list_del(&old->list);
2419 kfree(old);
2420 }
2421out:
2422 atomic_dec(&root->fs_info->defrag_running);
2423 wake_up(&root->fs_info->transaction_wait);
2424
2425 kfree(new);
2426}
2427
2428static struct new_sa_defrag_extent *
2429record_old_file_extents(struct inode *inode,
2430 struct btrfs_ordered_extent *ordered)
2431{
2432 struct btrfs_root *root = BTRFS_I(inode)->root;
2433 struct btrfs_path *path;
2434 struct btrfs_key key;
2435 struct old_sa_defrag_extent *old, *tmp;
2436 struct new_sa_defrag_extent *new;
2437 int ret;
2438
2439 new = kmalloc(sizeof(*new), GFP_NOFS);
2440 if (!new)
2441 return NULL;
2442
2443 new->inode = inode;
2444 new->file_pos = ordered->file_offset;
2445 new->len = ordered->len;
2446 new->bytenr = ordered->start;
2447 new->disk_len = ordered->disk_len;
2448 new->compress_type = ordered->compress_type;
2449 new->root = RB_ROOT;
2450 INIT_LIST_HEAD(&new->head);
2451
2452 path = btrfs_alloc_path();
2453 if (!path)
2454 goto out_kfree;
2455
2456 key.objectid = btrfs_ino(inode);
2457 key.type = BTRFS_EXTENT_DATA_KEY;
2458 key.offset = new->file_pos;
2459
2460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2461 if (ret < 0)
2462 goto out_free_path;
2463 if (ret > 0 && path->slots[0] > 0)
2464 path->slots[0]--;
2465
2466 /* find out all the old extents for the file range */
2467 while (1) {
2468 struct btrfs_file_extent_item *extent;
2469 struct extent_buffer *l;
2470 int slot;
2471 u64 num_bytes;
2472 u64 offset;
2473 u64 end;
2474 u64 disk_bytenr;
2475 u64 extent_offset;
2476
2477 l = path->nodes[0];
2478 slot = path->slots[0];
2479
2480 if (slot >= btrfs_header_nritems(l)) {
2481 ret = btrfs_next_leaf(root, path);
2482 if (ret < 0)
2483 goto out_free_list;
2484 else if (ret > 0)
2485 break;
2486 continue;
2487 }
2488
2489 btrfs_item_key_to_cpu(l, &key, slot);
2490
2491 if (key.objectid != btrfs_ino(inode))
2492 break;
2493 if (key.type != BTRFS_EXTENT_DATA_KEY)
2494 break;
2495 if (key.offset >= new->file_pos + new->len)
2496 break;
2497
2498 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2499
2500 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2501 if (key.offset + num_bytes < new->file_pos)
2502 goto next;
2503
2504 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2505 if (!disk_bytenr)
2506 goto next;
2507
2508 extent_offset = btrfs_file_extent_offset(l, extent);
2509
2510 old = kmalloc(sizeof(*old), GFP_NOFS);
2511 if (!old)
2512 goto out_free_list;
2513
2514 offset = max(new->file_pos, key.offset);
2515 end = min(new->file_pos + new->len, key.offset + num_bytes);
2516
2517 old->bytenr = disk_bytenr;
2518 old->extent_offset = extent_offset;
2519 old->offset = offset - key.offset;
2520 old->len = end - offset;
2521 old->new = new;
2522 old->count = 0;
2523 list_add_tail(&old->list, &new->head);
2524next:
2525 path->slots[0]++;
2526 cond_resched();
2527 }
2528
2529 btrfs_free_path(path);
2530 atomic_inc(&root->fs_info->defrag_running);
2531
2532 return new;
2533
2534out_free_list:
2535 list_for_each_entry_safe(old, tmp, &new->head, list) {
2536 list_del(&old->list);
2537 kfree(old);
2538 }
2539out_free_path:
2540 btrfs_free_path(path);
2541out_kfree:
2542 kfree(new);
2543 return NULL;
2544}
2545
5d13a98f
CM		 2546/*
2547 * helper function for btrfs_finish_ordered_io, this
2548 * just reads in some of the csum leaves to prime them into ram
2549 * before we start the transaction. It limits the amount of btree
2550 * reads required while inside the transaction.
2551 */
d352ac68
CM		 2552/* as ordered data IO finishes, this gets called so we can finish
2553 * an ordered extent if the range of bytes in the file it covers are
2554 * fully written.
2555 */
5fd02043 2556static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2557{
5fd02043 2558 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2559 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2560 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2561 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2562 struct extent_state *cached_state = NULL;
38c227d8 2563 struct new_sa_defrag_extent *new = NULL;
261507a0 2564 int compress_type = 0;
77cef2ec
JB		 2565 int ret = 0;
2566 u64 logical_len = ordered_extent->len;
82d5902d 2567 bool nolock;
77cef2ec 2568 bool truncated = false;
e6dcd2dc 2569
83eea1f1 2570 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2571
5fd02043
JB		 2572 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2573 ret = -EIO;
2574 goto out;
2575 }
2576
77cef2ec
JB		 2577 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2578 truncated = true;
2579 logical_len = ordered_extent->truncated_len;
2580 /* Truncated the entire extent, don't bother adding */
2581 if (!logical_len)
2582 goto out;
2583 }
2584
c2167754 2585 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2586 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2587 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2588 if (nolock)
2589 trans = btrfs_join_transaction_nolock(root);
2590 else
2591 trans = btrfs_join_transaction(root);
2592 if (IS_ERR(trans)) {
2593 ret = PTR_ERR(trans);
2594 trans = NULL;
2595 goto out;
c2167754 2596 }
6c760c07
JB		 2597 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2598 ret = btrfs_update_inode_fallback(trans, root, inode);
2599 if (ret) /* -ENOMEM or corruption */
2600 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2601 goto out;
2602 }
e6dcd2dc 2603
2ac55d41
JB		 2604 lock_extent_bits(io_tree, ordered_extent->file_offset,
2605 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2606 0, &cached_state);
e6dcd2dc 2607
38c227d8
LB		 2608 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2609 ordered_extent->file_offset + ordered_extent->len - 1,
2610 EXTENT_DEFRAG, 1, cached_state);
2611 if (ret) {
2612 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2613 if (last_snapshot >= BTRFS_I(inode)->generation)
2614 /* the inode is shared */
2615 new = record_old_file_extents(inode, ordered_extent);
2616
2617 clear_extent_bit(io_tree, ordered_extent->file_offset,
2618 ordered_extent->file_offset + ordered_extent->len - 1,
2619 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2620 }
2621
0cb59c99 2622 if (nolock)
7a7eaa40 2623 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2624 else
7a7eaa40 2625 trans = btrfs_join_transaction(root);
79787eaa
JM		 2626 if (IS_ERR(trans)) {
2627 ret = PTR_ERR(trans);
2628 trans = NULL;
2629 goto out_unlock;
2630 }
0ca1f7ce 2631 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2632
c8b97818 2633 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2634 compress_type = ordered_extent->compress_type;
d899e052 2635 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2636 BUG_ON(compress_type);
920bbbfb 2637 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2638 ordered_extent->file_offset,
2639 ordered_extent->file_offset +
77cef2ec 2640 logical_len);
d899e052 2641 } else {
0af3d00b 2642 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2643 ret = insert_reserved_file_extent(trans, inode,
2644 ordered_extent->file_offset,
2645 ordered_extent->start,
2646 ordered_extent->disk_len,
77cef2ec 2647 logical_len, logical_len,
261507a0 2648 compress_type, 0, 0,
d899e052 2649 BTRFS_FILE_EXTENT_REG);
d899e052 2650 }
5dc562c5
JB		 2651 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2652 ordered_extent->file_offset, ordered_extent->len,
2653 trans->transid);
79787eaa
JM		 2654 if (ret < 0) {
2655 btrfs_abort_transaction(trans, root, ret);
5fd02043 2656 goto out_unlock;
79787eaa 2657 }
2ac55d41 2658
e6dcd2dc
CM		 2659 add_pending_csums(trans, inode, ordered_extent->file_offset,
2660 &ordered_extent->list);
2661
6c760c07
JB		 2662 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2663 ret = btrfs_update_inode_fallback(trans, root, inode);
2664 if (ret) { /* -ENOMEM or corruption */
2665 btrfs_abort_transaction(trans, root, ret);
2666 goto out_unlock;
1ef30be1
JB		 2667 }
2668 ret = 0;
5fd02043
JB		 2669out_unlock:
2670 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2671 ordered_extent->file_offset +
2672 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2673out:
5b0e95bf 2674 if (root != root->fs_info->tree_root)
0cb59c99 2675 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2676 if (trans)
2677 btrfs_end_transaction(trans, root);
0cb59c99 2678
77cef2ec
JB		 2679 if (ret || truncated) {
2680 u64 start, end;
2681
2682 if (truncated)
2683 start = ordered_extent->file_offset + logical_len;
2684 else
2685 start = ordered_extent->file_offset;
2686 end = ordered_extent->file_offset + ordered_extent->len - 1;
2687 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2688
2689 /* Drop the cache for the part of the extent we didn't write. */
2690 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2691
0bec9ef5
JB		 2692 /*
2693 * If the ordered extent had an IOERR or something else went
2694 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2695 * back to the allocator. We only free the extent in the
2696 * truncated case if we didn't write out the extent at all.
0bec9ef5 2697 */
77cef2ec
JB		 2698 if ((ret || !logical_len) &&
2699 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2700 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2701 btrfs_free_reserved_extent(root, ordered_extent->start,
2702 ordered_extent->disk_len);
2703 }
2704
2705
5fd02043 2706 /*
8bad3c02
LB		 2707 * This needs to be done to make sure anybody waiting knows we are done
2708 * updating everything for this ordered extent.
5fd02043
JB		 2709 */
2710 btrfs_remove_ordered_extent(inode, ordered_extent);
2711
38c227d8
LB		 2712 /* for snapshot-aware defrag */
2713 if (new)
2714 relink_file_extents(new);
2715
e6dcd2dc
CM		 2716 /* once for us */
2717 btrfs_put_ordered_extent(ordered_extent);
2718 /* once for the tree */
2719 btrfs_put_ordered_extent(ordered_extent);
2720
5fd02043
JB		 2721 return ret;
2722}
2723
2724static void finish_ordered_fn(struct btrfs_work *work)
2725{
2726 struct btrfs_ordered_extent *ordered_extent;
2727 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2728 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2729}
2730
b2950863 2731static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2732 struct extent_state *state, int uptodate)
2733{
5fd02043
JB		 2734 struct inode *inode = page->mapping->host;
2735 struct btrfs_root *root = BTRFS_I(inode)->root;
2736 struct btrfs_ordered_extent *ordered_extent = NULL;
2737 struct btrfs_workers *workers;
2738
1abe9b8a 2739 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2740
8b62b72b 2741 ClearPagePrivate2(page);
5fd02043
JB		 2742 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2743 end - start + 1, uptodate))
2744 return 0;
2745
2746 ordered_extent->work.func = finish_ordered_fn;
2747 ordered_extent->work.flags = 0;
2748
83eea1f1 2749 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2750 workers = &root->fs_info->endio_freespace_worker;
2751 else
2752 workers = &root->fs_info->endio_write_workers;
2753 btrfs_queue_worker(workers, &ordered_extent->work);
2754
2755 return 0;
211f90e6
CM		 2756}
2757
d352ac68
CM		 2758/*
2759 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2760 * if there's a match, we allow the bio to finish. If not, the code in
2761 * extent_io.c will try to find good copies for us.
d352ac68 2762 */
facc8a22
MX		 2763static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2764 u64 phy_offset, struct page *page,
2765 u64 start, u64 end, int mirror)
07157aac 2766{
4eee4fa4 2767 size_t offset = start - page_offset(page);
07157aac 2768 struct inode *inode = page->mapping->host;
d1310b2e 2769 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2770 char *kaddr;
ff79f819 2771 struct btrfs_root *root = BTRFS_I(inode)->root;
facc8a22 2772 u32 csum_expected;
ff79f819 2773 u32 csum = ~(u32)0;
c2cf52eb
SK		 2774 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2775 DEFAULT_RATELIMIT_BURST);
d1310b2e 2776
d20f7043
CM		 2777 if (PageChecked(page)) {
2778 ClearPageChecked(page);
2779 goto good;
2780 }
6cbff00f
CH		 2781
2782 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2783 goto good;
17d217fe
YZ		 2784
2785 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2786 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2787 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2788 GFP_NOFS);
b6cda9bc 2789 return 0;
17d217fe 2790 }
d20f7043 2791
facc8a22
MX		 2792 phy_offset >>= inode->i_sb->s_blocksize_bits;
2793 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
d397712b 2794
facc8a22 2795 kaddr = kmap_atomic(page);
b0496686 2796 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2797 btrfs_csum_final(csum, (char *)&csum);
facc8a22 2798 if (csum != csum_expected)
07157aac 2799 goto zeroit;
d397712b 2800
7ac687d9 2801 kunmap_atomic(kaddr);
d20f7043 2802good:
07157aac
CM		 2803 return 0;
2804
2805zeroit:
c2cf52eb 2806 if (__ratelimit(&_rs))
facc8a22 2807 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c 2808 btrfs_ino(page->mapping->host), start, csum, csum_expected);
db94535d
CM		 2809 memset(kaddr + offset, 1, end - start + 1);
2810 flush_dcache_page(page);
7ac687d9 2811 kunmap_atomic(kaddr);
facc8a22 2812 if (csum_expected == 0)
3b951516 2813 return 0;
7e38326f 2814 return -EIO;
07157aac 2815}
b888db2b 2816
24bbcf04
YZ		 2817struct delayed_iput {
2818 struct list_head list;
2819 struct inode *inode;
2820};
2821
79787eaa
JM		 2822/* JDM: If this is fs-wide, why can't we add a pointer to
2823 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2824void btrfs_add_delayed_iput(struct inode *inode)
2825{
2826 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2827 struct delayed_iput *delayed;
2828
2829 if (atomic_add_unless(&inode->i_count, -1, 1))
2830 return;
2831
2832 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2833 delayed->inode = inode;
2834
2835 spin_lock(&fs_info->delayed_iput_lock);
2836 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2837 spin_unlock(&fs_info->delayed_iput_lock);
2838}
2839
2840void btrfs_run_delayed_iputs(struct btrfs_root *root)
2841{
2842 LIST_HEAD(list);
2843 struct btrfs_fs_info *fs_info = root->fs_info;
2844 struct delayed_iput *delayed;
2845 int empty;
2846
2847 spin_lock(&fs_info->delayed_iput_lock);
2848 empty = list_empty(&fs_info->delayed_iputs);
2849 spin_unlock(&fs_info->delayed_iput_lock);
2850 if (empty)
2851 return;
2852
24bbcf04
YZ		 2853 spin_lock(&fs_info->delayed_iput_lock);
2854 list_splice_init(&fs_info->delayed_iputs, &list);
2855 spin_unlock(&fs_info->delayed_iput_lock);
2856
2857 while (!list_empty(&list)) {
2858 delayed = list_entry(list.next, struct delayed_iput, list);
2859 list_del(&delayed->list);
2860 iput(delayed->inode);
2861 kfree(delayed);
2862 }
24bbcf04
YZ		 2863}
2864
d68fc57b 2865/*
42b2aa86 2866 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2867 * files in the subvolume, it removes orphan item and frees block_rsv
2868 * structure.
2869 */
2870void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root)
2872{
90290e19 2873 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2874 int ret;
2875
8a35d95f 2876 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2877 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2878 return;
2879
90290e19 2880 spin_lock(&root->orphan_lock);
8a35d95f 2881 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2882 spin_unlock(&root->orphan_lock);
2883 return;
2884 }
2885
2886 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2887 spin_unlock(&root->orphan_lock);
2888 return;
2889 }
2890
2891 block_rsv = root->orphan_block_rsv;
2892 root->orphan_block_rsv = NULL;
2893 spin_unlock(&root->orphan_lock);
2894
d68fc57b
YZ		 2895 if (root->orphan_item_inserted &&
2896 btrfs_root_refs(&root->root_item) > 0) {
2897 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2898 root->root_key.objectid);
4ef31a45
JB		 2899 if (ret)
2900 btrfs_abort_transaction(trans, root, ret);
2901 else
2902 root->orphan_item_inserted = 0;
d68fc57b
YZ		 2903 }
2904
90290e19
JB		 2905 if (block_rsv) {
2906 WARN_ON(block_rsv->size > 0);
2907 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2908 }
2909}
2910
7b128766
JB		 2911/*
2912 * This creates an orphan entry for the given inode in case something goes
2913 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2914 *
2915 * NOTE: caller of this function should reserve 5 units of metadata for
2916 * this function.
7b128766
JB		 2917 */
2918int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2919{
2920 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2921 struct btrfs_block_rsv *block_rsv = NULL;
2922 int reserve = 0;
2923 int insert = 0;
2924 int ret;
7b128766 2925
d68fc57b 2926 if (!root->orphan_block_rsv) {
66d8f3dd 2927 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2928 if (!block_rsv)
2929 return -ENOMEM;
d68fc57b 2930 }
7b128766 2931
d68fc57b
YZ		 2932 spin_lock(&root->orphan_lock);
2933 if (!root->orphan_block_rsv) {
2934 root->orphan_block_rsv = block_rsv;
2935 } else if (block_rsv) {
2936 btrfs_free_block_rsv(root, block_rsv);
2937 block_rsv = NULL;
7b128766 2938 }
7b128766 2939
8a35d95f
JB		 2940 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2941 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 2942#if 0
2943 /*
2944 * For proper ENOSPC handling, we should do orphan
2945 * cleanup when mounting. But this introduces backward
2946 * compatibility issue.
2947 */
2948 if (!xchg(&root->orphan_item_inserted, 1))
2949 insert = 2;
2950 else
2951 insert = 1;
2952#endif
2953 insert = 1;
321f0e70 2954 atomic_inc(&root->orphan_inodes);
7b128766
JB		 2955 }
2956
72ac3c0d
JB		 2957 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2958 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2959 reserve = 1;
d68fc57b 2960 spin_unlock(&root->orphan_lock);
7b128766 2961
d68fc57b
YZ		 2962 /* grab metadata reservation from transaction handle */
2963 if (reserve) {
2964 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 2965 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 2966 }
7b128766 2967
d68fc57b
YZ		 2968 /* insert an orphan item to track this unlinked/truncated file */
2969 if (insert >= 1) {
33345d01 2970 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 2971 if (ret) {
703c88e0 2972 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 2973 if (reserve) {
2974 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2975 &BTRFS_I(inode)->runtime_flags);
2976 btrfs_orphan_release_metadata(inode);
2977 }
2978 if (ret != -EEXIST) {
e8e7cff6
JB		 2979 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 2981 btrfs_abort_transaction(trans, root, ret);
2982 return ret;
2983 }
79787eaa
JM		 2984 }
2985 ret = 0;
d68fc57b
YZ		 2986 }
2987
2988 /* insert an orphan item to track subvolume contains orphan files */
2989 if (insert >= 2) {
2990 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2991 root->root_key.objectid);
79787eaa
JM		 2992 if (ret && ret != -EEXIST) {
2993 btrfs_abort_transaction(trans, root, ret);
2994 return ret;
2995 }
d68fc57b
YZ		 2996 }
2997 return 0;
7b128766
JB		 2998}
2999
3000/*
3001 * We have done the truncate/delete so we can go ahead and remove the orphan
3002 * item for this particular inode.
3003 */
48a3b636
ES		 3004static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3005 struct inode *inode)
7b128766
JB		 3006{
3007 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3008 int delete_item = 0;
3009 int release_rsv = 0;
7b128766
JB		 3010 int ret = 0;
3011
d68fc57b 3012 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3013 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3014 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3015 delete_item = 1;
7b128766 3016
72ac3c0d
JB		 3017 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3018 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3019 release_rsv = 1;
d68fc57b 3020 spin_unlock(&root->orphan_lock);
7b128766 3021
703c88e0 3022 if (delete_item) {
8a35d95f 3023 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3024 if (trans)
3025 ret = btrfs_del_orphan_item(trans, root,
3026 btrfs_ino(inode));
8a35d95f 3027 }
7b128766 3028
703c88e0
FDBM		 3029 if (release_rsv)
3030 btrfs_orphan_release_metadata(inode);
3031
4ef31a45 3032 return ret;
7b128766
JB		 3033}
3034
3035/*
3036 * this cleans up any orphans that may be left on the list from the last use
3037 * of this root.
3038 */
66b4ffd1 3039int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3040{
3041 struct btrfs_path *path;
3042 struct extent_buffer *leaf;
7b128766
JB		 3043 struct btrfs_key key, found_key;
3044 struct btrfs_trans_handle *trans;
3045 struct inode *inode;
8f6d7f4f 3046 u64 last_objectid = 0;
7b128766
JB		 3047 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3048
d68fc57b 3049 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3050 return 0;
c71bf099
YZ		 3051
3052 path = btrfs_alloc_path();
66b4ffd1
JB		 3053 if (!path) {
3054 ret = -ENOMEM;
3055 goto out;
3056 }
7b128766
JB		 3057 path->reada = -1;
3058
3059 key.objectid = BTRFS_ORPHAN_OBJECTID;
3060 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3061 key.offset = (u64)-1;
3062
7b128766
JB		 3063 while (1) {
3064 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3065 if (ret < 0)
3066 goto out;
7b128766
JB		 3067
3068 /*
3069 * if ret == 0 means we found what we were searching for, which
25985edc 3070 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3071 * find the key and see if we have stuff that matches
3072 */
3073 if (ret > 0) {
66b4ffd1 3074 ret = 0;
7b128766
JB		 3075 if (path->slots[0] == 0)
3076 break;
3077 path->slots[0]--;
3078 }
3079
3080 /* pull out the item */
3081 leaf = path->nodes[0];
7b128766
JB		 3082 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3083
3084 /* make sure the item matches what we want */
3085 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3086 break;
3087 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3088 break;
3089
3090 /* release the path since we're done with it */
b3b4aa74 3091 btrfs_release_path(path);
7b128766
JB		 3092
3093 /*
3094 * this is where we are basically btrfs_lookup, without the
3095 * crossing root thing. we store the inode number in the
3096 * offset of the orphan item.
3097 */
8f6d7f4f
JB		 3098
3099 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3100 btrfs_err(root->fs_info,
3101 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3102 ret = -EINVAL;
3103 goto out;
3104 }
3105
3106 last_objectid = found_key.offset;
3107
5d4f98a2
YZ		 3108 found_key.objectid = found_key.offset;
3109 found_key.type = BTRFS_INODE_ITEM_KEY;
3110 found_key.offset = 0;
73f73415 3111 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3112 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3113 if (ret && ret != -ESTALE)
66b4ffd1 3114 goto out;
7b128766 3115
f8e9e0b0
AJ		 3116 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3117 struct btrfs_root *dead_root;
3118 struct btrfs_fs_info *fs_info = root->fs_info;
3119 int is_dead_root = 0;
3120
3121 /*
3122 * this is an orphan in the tree root. Currently these
3123 * could come from 2 sources:
3124 * a) a snapshot deletion in progress
3125 * b) a free space cache inode
3126 * We need to distinguish those two, as the snapshot
3127 * orphan must not get deleted.
3128 * find_dead_roots already ran before us, so if this
3129 * is a snapshot deletion, we should find the root
3130 * in the dead_roots list
3131 */
3132 spin_lock(&fs_info->trans_lock);
3133 list_for_each_entry(dead_root, &fs_info->dead_roots,
3134 root_list) {
3135 if (dead_root->root_key.objectid ==
3136 found_key.objectid) {
3137 is_dead_root = 1;
3138 break;
3139 }
3140 }
3141 spin_unlock(&fs_info->trans_lock);
3142 if (is_dead_root) {
3143 /* prevent this orphan from being found again */
3144 key.offset = found_key.objectid - 1;
3145 continue;
3146 }
3147 }
7b128766 3148 /*
a8c9e576
JB		 3149 * Inode is already gone but the orphan item is still there,
3150 * kill the orphan item.
7b128766 3151 */
a8c9e576
JB		 3152 if (ret == -ESTALE) {
3153 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3154 if (IS_ERR(trans)) {
3155 ret = PTR_ERR(trans);
3156 goto out;
3157 }
c2cf52eb
SK		 3158 btrfs_debug(root->fs_info, "auto deleting %Lu",
3159 found_key.objectid);
a8c9e576
JB		 3160 ret = btrfs_del_orphan_item(trans, root,
3161 found_key.objectid);
5b21f2ed 3162 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3163 if (ret)
3164 goto out;
7b128766
JB		 3165 continue;
3166 }
3167
a8c9e576
JB		 3168 /*
3169 * add this inode to the orphan list so btrfs_orphan_del does
3170 * the proper thing when we hit it
3171 */
8a35d95f
JB		 3172 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3173 &BTRFS_I(inode)->runtime_flags);
925396ec 3174 atomic_inc(&root->orphan_inodes);
a8c9e576 3175
7b128766
JB		 3176 /* if we have links, this was a truncate, lets do that */
3177 if (inode->i_nlink) {
a41ad394
JB		 3178 if (!S_ISREG(inode->i_mode)) {
3179 WARN_ON(1);
3180 iput(inode);
3181 continue;
3182 }
7b128766 3183 nr_truncate++;
f3fe820c
JB		 3184
3185 /* 1 for the orphan item deletion. */
3186 trans = btrfs_start_transaction(root, 1);
3187 if (IS_ERR(trans)) {
c69b26b0 3188 iput(inode);
f3fe820c
JB		 3189 ret = PTR_ERR(trans);
3190 goto out;
3191 }
3192 ret = btrfs_orphan_add(trans, inode);
3193 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3194 if (ret) {
3195 iput(inode);
f3fe820c 3196 goto out;
c69b26b0 3197 }
f3fe820c 3198
66b4ffd1 3199 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3200 if (ret)
3201 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3202 } else {
3203 nr_unlink++;
3204 }
3205
3206 /* this will do delete_inode and everything for us */
3207 iput(inode);
66b4ffd1
JB		 3208 if (ret)
3209 goto out;
7b128766 3210 }
3254c876
MX		 3211 /* release the path since we're done with it */
3212 btrfs_release_path(path);
3213
d68fc57b
YZ		 3214 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3215
3216 if (root->orphan_block_rsv)
3217 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3218 (u64)-1);
3219
3220 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3221 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3222 if (!IS_ERR(trans))
3223 btrfs_end_transaction(trans, root);
d68fc57b 3224 }
7b128766
JB		 3225
3226 if (nr_unlink)
4884b476 3227 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3228 if (nr_truncate)
4884b476 3229 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3230
3231out:
3232 if (ret)
c2cf52eb
SK		 3233 btrfs_crit(root->fs_info,
3234 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3235 btrfs_free_path(path);
3236 return ret;
7b128766
JB		 3237}
3238
46a53cca
CM		 3239/*
3240 * very simple check to peek ahead in the leaf looking for xattrs. If we
3241 * don't find any xattrs, we know there can't be any acls.
3242 *
3243 * slot is the slot the inode is in, objectid is the objectid of the inode
3244 */
3245static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3246 int slot, u64 objectid)
3247{
3248 u32 nritems = btrfs_header_nritems(leaf);
3249 struct btrfs_key found_key;
f23b5a59
JB		 3250 static u64 xattr_access = 0;
3251 static u64 xattr_default = 0;
46a53cca
CM		 3252 int scanned = 0;
3253
f23b5a59
JB		 3254 if (!xattr_access) {
3255 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3256 strlen(POSIX_ACL_XATTR_ACCESS));
3257 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3258 strlen(POSIX_ACL_XATTR_DEFAULT));
3259 }
3260
46a53cca
CM		 3261 slot++;
3262 while (slot < nritems) {
3263 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3264
3265 /* we found a different objectid, there must not be acls */
3266 if (found_key.objectid != objectid)
3267 return 0;
3268
3269 /* we found an xattr, assume we've got an acl */
f23b5a59
JB		 3270 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3271 if (found_key.offset == xattr_access ||
3272 found_key.offset == xattr_default)
3273 return 1;
3274 }
46a53cca
CM		 3275
3276 /*
3277 * we found a key greater than an xattr key, there can't
3278 * be any acls later on
3279 */
3280 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3281 return 0;
3282
3283 slot++;
3284 scanned++;
3285
3286 /*
3287 * it goes inode, inode backrefs, xattrs, extents,
3288 * so if there are a ton of hard links to an inode there can
3289 * be a lot of backrefs. Don't waste time searching too hard,
3290 * this is just an optimization
3291 */
3292 if (scanned >= 8)
3293 break;
3294 }
3295 /* we hit the end of the leaf before we found an xattr or
3296 * something larger than an xattr. We have to assume the inode
3297 * has acls
3298 */
3299 return 1;
3300}
3301
d352ac68
CM		 3302/*
3303 * read an inode from the btree into the in-memory inode
3304 */
5d4f98a2 3305static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3306{
3307 struct btrfs_path *path;
5f39d397 3308 struct extent_buffer *leaf;
39279cc3 3309 struct btrfs_inode_item *inode_item;
0b86a832 3310 struct btrfs_timespec *tspec;
39279cc3
CM		 3311 struct btrfs_root *root = BTRFS_I(inode)->root;
3312 struct btrfs_key location;
46a53cca 3313 int maybe_acls;
618e21d5 3314 u32 rdev;
39279cc3 3315 int ret;
2f7e33d4
MX		 3316 bool filled = false;
3317
3318 ret = btrfs_fill_inode(inode, &rdev);
3319 if (!ret)
3320 filled = true;
39279cc3
CM		 3321
3322 path = btrfs_alloc_path();
1748f843
MF		 3323 if (!path)
3324 goto make_bad;
3325
d90c7321 3326 path->leave_spinning = 1;
39279cc3 3327 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3328
39279cc3 3329 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3330 if (ret)
39279cc3 3331 goto make_bad;
39279cc3 3332
5f39d397 3333 leaf = path->nodes[0];
2f7e33d4
MX		 3334
3335 if (filled)
3336 goto cache_acl;
3337
5f39d397
CM		 3338 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3339 struct btrfs_inode_item);
5f39d397 3340 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3341 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3342 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3343 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3344 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3345
3346 tspec = btrfs_inode_atime(inode_item);
3347 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3348 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3349
3350 tspec = btrfs_inode_mtime(inode_item);
3351 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3352 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3353
3354 tspec = btrfs_inode_ctime(inode_item);
3355 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3356 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3357
a76a3cd4 3358 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3359 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3360 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3361
3362 /*
3363 * If we were modified in the current generation and evicted from memory
3364 * and then re-read we need to do a full sync since we don't have any
3365 * idea about which extents were modified before we were evicted from
3366 * cache.
3367 */
3368 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3369 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3370 &BTRFS_I(inode)->runtime_flags);
3371
0c4d2d95 3372 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3373 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3374 inode->i_rdev = 0;
5f39d397
CM		 3375 rdev = btrfs_inode_rdev(leaf, inode_item);
3376
aec7477b 3377 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3378 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3379cache_acl:
46a53cca
CM		 3380 /*
3381 * try to precache a NULL acl entry for files that don't have
3382 * any xattrs or acls
3383 */
33345d01
LZ		 3384 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3385 btrfs_ino(inode));
72c04902
AV		 3386 if (!maybe_acls)
3387 cache_no_acl(inode);
46a53cca 3388
39279cc3 3389 btrfs_free_path(path);
39279cc3 3390
39279cc3 3391 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3392 case S_IFREG:
3393 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3394 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3395 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3396 inode->i_fop = &btrfs_file_operations;
3397 inode->i_op = &btrfs_file_inode_operations;
3398 break;
3399 case S_IFDIR:
3400 inode->i_fop = &btrfs_dir_file_operations;
3401 if (root == root->fs_info->tree_root)
3402 inode->i_op = &btrfs_dir_ro_inode_operations;
3403 else
3404 inode->i_op = &btrfs_dir_inode_operations;
3405 break;
3406 case S_IFLNK:
3407 inode->i_op = &btrfs_symlink_inode_operations;
3408 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3409 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3410 break;
618e21d5 3411 default:
0279b4cd 3412 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3413 init_special_inode(inode, inode->i_mode, rdev);
3414 break;
39279cc3 3415 }
6cbff00f
CH		 3416
3417 btrfs_update_iflags(inode);
39279cc3
CM		 3418 return;
3419
3420make_bad:
39279cc3 3421 btrfs_free_path(path);
39279cc3
CM		 3422 make_bad_inode(inode);
3423}
3424
d352ac68
CM		 3425/*
3426 * given a leaf and an inode, copy the inode fields into the leaf
3427 */
e02119d5
CM		 3428static void fill_inode_item(struct btrfs_trans_handle *trans,
3429 struct extent_buffer *leaf,
5f39d397 3430 struct btrfs_inode_item *item,
39279cc3
CM		 3431 struct inode *inode)
3432{
51fab693
LB		 3433 struct btrfs_map_token token;
3434
3435 btrfs_init_map_token(&token);
5f39d397 3436
51fab693
LB		 3437 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3438 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3439 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3440 &token);
3441 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3442 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3443
51fab693
LB		 3444 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3445 inode->i_atime.tv_sec, &token);
3446 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3447 inode->i_atime.tv_nsec, &token);
5f39d397 3448
51fab693
LB		 3449 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3450 inode->i_mtime.tv_sec, &token);
3451 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3452 inode->i_mtime.tv_nsec, &token);
5f39d397 3453
51fab693
LB		 3454 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3455 inode->i_ctime.tv_sec, &token);
3456 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3457 inode->i_ctime.tv_nsec, &token);
5f39d397 3458
51fab693
LB		 3459 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3460 &token);
3461 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3462 &token);
3463 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3464 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3465 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3466 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3467 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3468}
3469
d352ac68
CM		 3470/*
3471 * copy everything in the in-memory inode into the btree.
3472 */
2115133f 3473static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3474 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3475{
3476 struct btrfs_inode_item *inode_item;
3477 struct btrfs_path *path;
5f39d397 3478 struct extent_buffer *leaf;
39279cc3
CM		 3479 int ret;
3480
3481 path = btrfs_alloc_path();
16cdcec7
MX		 3482 if (!path)
3483 return -ENOMEM;
3484
b9473439 3485 path->leave_spinning = 1;
16cdcec7
MX		 3486 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3487 1);
39279cc3
CM		 3488 if (ret) {
3489 if (ret > 0)
3490 ret = -ENOENT;
3491 goto failed;
3492 }
3493
b4ce94de 3494 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3495 leaf = path->nodes[0];
3496 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3497 struct btrfs_inode_item);
39279cc3 3498
e02119d5 3499 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3500 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3501 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3502 ret = 0;
3503failed:
39279cc3
CM		 3504 btrfs_free_path(path);
3505 return ret;
3506}
3507
2115133f
CM		 3508/*
3509 * copy everything in the in-memory inode into the btree.
3510 */
3511noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root, struct inode *inode)
3513{
3514 int ret;
3515
3516 /*
3517 * If the inode is a free space inode, we can deadlock during commit
3518 * if we put it into the delayed code.
3519 *
3520 * The data relocation inode should also be directly updated
3521 * without delay
3522 */
83eea1f1 3523 if (!btrfs_is_free_space_inode(inode)
2115133f 3524 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3525 btrfs_update_root_times(trans, root);
3526
2115133f
CM		 3527 ret = btrfs_delayed_update_inode(trans, root, inode);
3528 if (!ret)
3529 btrfs_set_inode_last_trans(trans, inode);
3530 return ret;
3531 }
3532
3533 return btrfs_update_inode_item(trans, root, inode);
3534}
3535
be6aef60
JB		 3536noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3537 struct btrfs_root *root,
3538 struct inode *inode)
2115133f
CM		 3539{
3540 int ret;
3541
3542 ret = btrfs_update_inode(trans, root, inode);
3543 if (ret == -ENOSPC)
3544 return btrfs_update_inode_item(trans, root, inode);
3545 return ret;
3546}
3547
d352ac68
CM		 3548/*
3549 * unlink helper that gets used here in inode.c and in the tree logging
3550 * recovery code. It remove a link in a directory with a given name, and
3551 * also drops the back refs in the inode to the directory
3552 */
92986796
AV		 3553static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root,
3555 struct inode *dir, struct inode *inode,
3556 const char *name, int name_len)
39279cc3
CM		 3557{
3558 struct btrfs_path *path;
39279cc3 3559 int ret = 0;
5f39d397 3560 struct extent_buffer *leaf;
39279cc3 3561 struct btrfs_dir_item *di;
5f39d397 3562 struct btrfs_key key;
aec7477b 3563 u64 index;
33345d01
LZ		 3564 u64 ino = btrfs_ino(inode);
3565 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3566
3567 path = btrfs_alloc_path();
54aa1f4d
CM		 3568 if (!path) {
3569 ret = -ENOMEM;
554233a6 3570 goto out;
54aa1f4d
CM		 3571 }
3572
b9473439 3573 path->leave_spinning = 1;
33345d01 3574 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3575 name, name_len, -1);
3576 if (IS_ERR(di)) {
3577 ret = PTR_ERR(di);
3578 goto err;
3579 }
3580 if (!di) {
3581 ret = -ENOENT;
3582 goto err;
3583 }
5f39d397
CM		 3584 leaf = path->nodes[0];
3585 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3586 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3587 if (ret)
3588 goto err;
b3b4aa74 3589 btrfs_release_path(path);
39279cc3 3590
33345d01
LZ		 3591 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3592 dir_ino, &index);
aec7477b 3593 if (ret) {
c2cf52eb
SK		 3594 btrfs_info(root->fs_info,
3595 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3596 name_len, name, ino, dir_ino);
79787eaa 3597 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3598 goto err;
3599 }
3600
16cdcec7 3601 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3602 if (ret) {
3603 btrfs_abort_transaction(trans, root, ret);
39279cc3 3604 goto err;
79787eaa 3605 }
39279cc3 3606
e02119d5 3607 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3608 inode, dir_ino);
79787eaa
JM		 3609 if (ret != 0 && ret != -ENOENT) {
3610 btrfs_abort_transaction(trans, root, ret);
3611 goto err;
3612 }
e02119d5
CM		 3613
3614 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3615 dir, index);
6418c961
CM		 3616 if (ret == -ENOENT)
3617 ret = 0;
d4e3991b
ZB		 3618 else if (ret)
3619 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3620err:
3621 btrfs_free_path(path);
e02119d5
CM		 3622 if (ret)
3623 goto out;
3624
3625 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3626 inode_inc_iversion(inode);
3627 inode_inc_iversion(dir);
e02119d5 3628 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3629 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3630out:
39279cc3
CM		 3631 return ret;
3632}
3633
92986796
AV		 3634int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3635 struct btrfs_root *root,
3636 struct inode *dir, struct inode *inode,
3637 const char *name, int name_len)
3638{
3639 int ret;
3640 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3641 if (!ret) {
3642 btrfs_drop_nlink(inode);
3643 ret = btrfs_update_inode(trans, root, inode);
3644 }
3645 return ret;
3646}
39279cc3 3647
a22285a6
YZ		 3648/*
3649 * helper to start transaction for unlink and rmdir.
3650 *
d52be818
JB		 3651 * unlink and rmdir are special in btrfs, they do not always free space, so
3652 * if we cannot make our reservations the normal way try and see if there is
3653 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3654 * allow the unlink to occur.
a22285a6 3655 */
d52be818 3656static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3657{
39279cc3 3658 struct btrfs_trans_handle *trans;
a22285a6 3659 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3660 int ret;
3661
e70bea5f
JB		 3662 /*
3663 * 1 for the possible orphan item
3664 * 1 for the dir item
3665 * 1 for the dir index
3666 * 1 for the inode ref
e70bea5f
JB		 3667 * 1 for the inode
3668 */
6e137ed3 3669 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3670 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3671 return trans;
4df27c4d 3672
d52be818
JB		 3673 if (PTR_ERR(trans) == -ENOSPC) {
3674 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 3675
d52be818
JB		 3676 trans = btrfs_start_transaction(root, 0);
3677 if (IS_ERR(trans))
3678 return trans;
3679 ret = btrfs_cond_migrate_bytes(root->fs_info,
3680 &root->fs_info->trans_block_rsv,
3681 num_bytes, 5);
3682 if (ret) {
3683 btrfs_end_transaction(trans, root);
3684 return ERR_PTR(ret);
a22285a6 3685 }
5a77d76c 3686 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3687 trans->bytes_reserved = num_bytes;
a22285a6 3688 }
d52be818 3689 return trans;
a22285a6
YZ		 3690}
3691
3692static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3693{
3694 struct btrfs_root *root = BTRFS_I(dir)->root;
3695 struct btrfs_trans_handle *trans;
3696 struct inode *inode = dentry->d_inode;
3697 int ret;
a22285a6 3698
d52be818 3699 trans = __unlink_start_trans(dir);
a22285a6
YZ		 3700 if (IS_ERR(trans))
3701 return PTR_ERR(trans);
5f39d397 3702
12fcfd22
CM		 3703 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3704
e02119d5
CM		 3705 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3706 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3707 if (ret)
3708 goto out;
7b128766 3709
a22285a6 3710 if (inode->i_nlink == 0) {
7b128766 3711 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3712 if (ret)
3713 goto out;
a22285a6 3714 }
7b128766 3715
b532402e 3716out:
d52be818 3717 btrfs_end_transaction(trans, root);
b53d3f5d 3718 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3719 return ret;
3720}
3721
4df27c4d
YZ		 3722int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3723 struct btrfs_root *root,
3724 struct inode *dir, u64 objectid,
3725 const char *name, int name_len)
3726{
3727 struct btrfs_path *path;
3728 struct extent_buffer *leaf;
3729 struct btrfs_dir_item *di;
3730 struct btrfs_key key;
3731 u64 index;
3732 int ret;
33345d01 3733 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3734
3735 path = btrfs_alloc_path();
3736 if (!path)
3737 return -ENOMEM;
3738
33345d01 3739 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3740 name, name_len, -1);
79787eaa
JM		 3741 if (IS_ERR_OR_NULL(di)) {
3742 if (!di)
3743 ret = -ENOENT;
3744 else
3745 ret = PTR_ERR(di);
3746 goto out;
3747 }
4df27c4d
YZ		 3748
3749 leaf = path->nodes[0];
3750 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3751 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3752 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3753 if (ret) {
3754 btrfs_abort_transaction(trans, root, ret);
3755 goto out;
3756 }
b3b4aa74 3757 btrfs_release_path(path);
4df27c4d
YZ		 3758
3759 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3760 objectid, root->root_key.objectid,
33345d01 3761 dir_ino, &index, name, name_len);
4df27c4d 3762 if (ret < 0) {
79787eaa
JM		 3763 if (ret != -ENOENT) {
3764 btrfs_abort_transaction(trans, root, ret);
3765 goto out;
3766 }
33345d01 3767 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3768 name, name_len);
79787eaa
JM		 3769 if (IS_ERR_OR_NULL(di)) {
3770 if (!di)
3771 ret = -ENOENT;
3772 else
3773 ret = PTR_ERR(di);
3774 btrfs_abort_transaction(trans, root, ret);
3775 goto out;
3776 }
4df27c4d
YZ		 3777
3778 leaf = path->nodes[0];
3779 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3780 btrfs_release_path(path);
4df27c4d
YZ		 3781 index = key.offset;
3782 }
945d8962 3783 btrfs_release_path(path);
4df27c4d 3784
16cdcec7 3785 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3786 if (ret) {
3787 btrfs_abort_transaction(trans, root, ret);
3788 goto out;
3789 }
4df27c4d
YZ		 3790
3791 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3792 inode_inc_iversion(dir);
4df27c4d 3793 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3794 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3795 if (ret)
3796 btrfs_abort_transaction(trans, root, ret);
3797out:
71d7aed0 3798 btrfs_free_path(path);
79787eaa 3799 return ret;
4df27c4d
YZ		 3800}
3801
39279cc3
CM		 3802static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3803{
3804 struct inode *inode = dentry->d_inode;
1832a6d5 3805 int err = 0;
39279cc3 3806 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3807 struct btrfs_trans_handle *trans;
39279cc3 3808
b3ae244e 3809 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3810 return -ENOTEMPTY;
b3ae244e
DS		 3811 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3812 return -EPERM;
134d4512 3813
d52be818 3814 trans = __unlink_start_trans(dir);
a22285a6 3815 if (IS_ERR(trans))
5df6a9f6 3816 return PTR_ERR(trans);
5df6a9f6 3817
33345d01 3818 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3819 err = btrfs_unlink_subvol(trans, root, dir,
3820 BTRFS_I(inode)->location.objectid,
3821 dentry->d_name.name,
3822 dentry->d_name.len);
3823 goto out;
3824 }
3825
7b128766
JB		 3826 err = btrfs_orphan_add(trans, inode);
3827 if (err)
4df27c4d 3828 goto out;
7b128766 3829
39279cc3 3830 /* now the directory is empty */
e02119d5
CM		 3831 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3832 dentry->d_name.name, dentry->d_name.len);
d397712b 3833 if (!err)
dbe674a9 3834 btrfs_i_size_write(inode, 0);
4df27c4d 3835out:
d52be818 3836 btrfs_end_transaction(trans, root);
b53d3f5d 3837 btrfs_btree_balance_dirty(root);
3954401f 3838
39279cc3
CM		 3839 return err;
3840}
3841
39279cc3
CM		 3842/*
3843 * this can truncate away extent items, csum items and directory items.
3844 * It starts at a high offset and removes keys until it can't find
d352ac68 3845 * any higher than new_size
39279cc3
CM		 3846 *
3847 * csum items that cross the new i_size are truncated to the new size
3848 * as well.
7b128766
JB		 3849 *
3850 * min_type is the minimum key type to truncate down to. If set to 0, this
3851 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3852 */
8082510e
YZ		 3853int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3854 struct btrfs_root *root,
3855 struct inode *inode,
3856 u64 new_size, u32 min_type)
39279cc3 3857{
39279cc3 3858 struct btrfs_path *path;
5f39d397 3859 struct extent_buffer *leaf;
39279cc3 3860 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3861 struct btrfs_key key;
3862 struct btrfs_key found_key;
39279cc3 3863 u64 extent_start = 0;
db94535d 3864 u64 extent_num_bytes = 0;
5d4f98a2 3865 u64 extent_offset = 0;
39279cc3 3866 u64 item_end = 0;
7f4f6e0a 3867 u64 last_size = (u64)-1;
8082510e 3868 u32 found_type = (u8)-1;
39279cc3
CM		 3869 int found_extent;
3870 int del_item;
85e21bac
CM		 3871 int pending_del_nr = 0;
3872 int pending_del_slot = 0;
179e29e4 3873 int extent_type = -1;
8082510e
YZ		 3874 int ret;
3875 int err = 0;
33345d01 3876 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3877
3878 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3879
0eb0e19c
MF		 3880 path = btrfs_alloc_path();
3881 if (!path)
3882 return -ENOMEM;
3883 path->reada = -1;
3884
5dc562c5
JB		 3885 /*
3886 * We want to drop from the next block forward in case this new size is
3887 * not block aligned since we will be keeping the last block of the
3888 * extent just the way it is.
3889 */
0af3d00b 3890 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 3891 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3892 root->sectorsize), (u64)-1, 0);
8082510e 3893
16cdcec7
MX		 3894 /*
3895 * This function is also used to drop the items in the log tree before
3896 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3897 * it is used to drop the loged items. So we shouldn't kill the delayed
3898 * items.
3899 */
3900 if (min_type == 0 && root == BTRFS_I(inode)->root)
3901 btrfs_kill_delayed_inode_items(inode);
3902
33345d01 3903 key.objectid = ino;
39279cc3 3904 key.offset = (u64)-1;
5f39d397
CM		 3905 key.type = (u8)-1;
3906
85e21bac 3907search_again:
b9473439 3908 path->leave_spinning = 1;
85e21bac 3909 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3910 if (ret < 0) {
3911 err = ret;
3912 goto out;
3913 }
d397712b 3914
85e21bac 3915 if (ret > 0) {
e02119d5
CM		 3916 /* there are no items in the tree for us to truncate, we're
3917 * done
3918 */
8082510e
YZ		 3919 if (path->slots[0] == 0)
3920 goto out;
85e21bac
CM		 3921 path->slots[0]--;
3922 }
3923
d397712b 3924 while (1) {
39279cc3 3925 fi = NULL;
5f39d397
CM		 3926 leaf = path->nodes[0];
3927 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3928 found_type = btrfs_key_type(&found_key);
39279cc3 3929
33345d01 3930 if (found_key.objectid != ino)
39279cc3 3931 break;
5f39d397 3932
85e21bac 3933 if (found_type < min_type)
39279cc3
CM		 3934 break;
3935
5f39d397 3936 item_end = found_key.offset;
39279cc3 3937 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3938 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3939 struct btrfs_file_extent_item);
179e29e4
CM		 3940 extent_type = btrfs_file_extent_type(leaf, fi);
3941 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3942 item_end +=
db94535d 3943 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3944 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3945 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3946 fi);
39279cc3 3947 }
008630c1 3948 item_end--;
39279cc3 3949 }
8082510e
YZ		 3950 if (found_type > min_type) {
3951 del_item = 1;
3952 } else {
3953 if (item_end < new_size)
b888db2b 3954 break;
8082510e
YZ		 3955 if (found_key.offset >= new_size)
3956 del_item = 1;
3957 else
3958 del_item = 0;
39279cc3 3959 }
39279cc3 3960 found_extent = 0;
39279cc3 3961 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3962 if (found_type != BTRFS_EXTENT_DATA_KEY)
3963 goto delete;
3964
7f4f6e0a
JB		 3965 if (del_item)
3966 last_size = found_key.offset;
3967 else
3968 last_size = new_size;
3969
179e29e4 3970 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3971 u64 num_dec;
db94535d 3972 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3973 if (!del_item) {
db94535d
CM		 3974 u64 orig_num_bytes =
3975 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 3976 extent_num_bytes = ALIGN(new_size -
3977 found_key.offset,
3978 root->sectorsize);
db94535d
CM		 3979 btrfs_set_file_extent_num_bytes(leaf, fi,
3980 extent_num_bytes);
3981 num_dec = (orig_num_bytes -
9069218d 3982 extent_num_bytes);
e02119d5 3983 if (root->ref_cows && extent_start != 0)
a76a3cd4 3984 inode_sub_bytes(inode, num_dec);
5f39d397 3985 btrfs_mark_buffer_dirty(leaf);
39279cc3 3986 } else {
db94535d
CM		 3987 extent_num_bytes =
3988 btrfs_file_extent_disk_num_bytes(leaf,
3989 fi);
5d4f98a2
YZ		 3990 extent_offset = found_key.offset -
3991 btrfs_file_extent_offset(leaf, fi);
3992
39279cc3 3993 /* FIXME blocksize != 4096 */
9069218d 3994 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3995 if (extent_start != 0) {
3996 found_extent = 1;
e02119d5 3997 if (root->ref_cows)
a76a3cd4 3998 inode_sub_bytes(inode, num_dec);
e02119d5 3999 }
39279cc3 4000 }
9069218d 4001 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4002 /*
4003 * we can't truncate inline items that have had
4004 * special encodings
4005 */
4006 if (!del_item &&
4007 btrfs_file_extent_compression(leaf, fi) == 0 &&
4008 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4009 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4010 u32 size = new_size - found_key.offset;
4011
4012 if (root->ref_cows) {
a76a3cd4
YZ		 4013 inode_sub_bytes(inode, item_end + 1 -
4014 new_size);
e02119d5
CM		 4015 }
4016 size =
4017 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4018 btrfs_truncate_item(root, path, size, 1);
e02119d5 4019 } else if (root->ref_cows) {
a76a3cd4
YZ		 4020 inode_sub_bytes(inode, item_end + 1 -
4021 found_key.offset);
9069218d 4022 }
39279cc3 4023 }
179e29e4 4024delete:
39279cc3 4025 if (del_item) {
85e21bac
CM		 4026 if (!pending_del_nr) {
4027 /* no pending yet, add ourselves */
4028 pending_del_slot = path->slots[0];
4029 pending_del_nr = 1;
4030 } else if (pending_del_nr &&
4031 path->slots[0] + 1 == pending_del_slot) {
4032 /* hop on the pending chunk */
4033 pending_del_nr++;
4034 pending_del_slot = path->slots[0];
4035 } else {
d397712b 4036 BUG();
85e21bac 4037 }
39279cc3
CM		 4038 } else {
4039 break;
4040 }
0af3d00b
JB		 4041 if (found_extent && (root->ref_cows ||
4042 root == root->fs_info->tree_root)) {
b9473439 4043 btrfs_set_path_blocking(path);
39279cc3 4044 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4045 extent_num_bytes, 0,
4046 btrfs_header_owner(leaf),
66d7e7f0 4047 ino, extent_offset, 0);
39279cc3
CM		 4048 BUG_ON(ret);
4049 }
85e21bac 4050
8082510e
YZ		 4051 if (found_type == BTRFS_INODE_ITEM_KEY)
4052 break;
4053
4054 if (path->slots[0] == 0 ||
4055 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4056 if (pending_del_nr) {
4057 ret = btrfs_del_items(trans, root, path,
4058 pending_del_slot,
4059 pending_del_nr);
79787eaa
JM		 4060 if (ret) {
4061 btrfs_abort_transaction(trans,
4062 root, ret);
4063 goto error;
4064 }
8082510e
YZ		 4065 pending_del_nr = 0;
4066 }
b3b4aa74 4067 btrfs_release_path(path);
85e21bac 4068 goto search_again;
8082510e
YZ		 4069 } else {
4070 path->slots[0]--;
85e21bac 4071 }
39279cc3 4072 }
8082510e 4073out:
85e21bac
CM		 4074 if (pending_del_nr) {
4075 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4076 pending_del_nr);
79787eaa
JM		 4077 if (ret)
4078 btrfs_abort_transaction(trans, root, ret);
85e21bac 4079 }
79787eaa 4080error:
7f4f6e0a
JB		 4081 if (last_size != (u64)-1)
4082 btrfs_ordered_update_i_size(inode, last_size, NULL);
39279cc3 4083 btrfs_free_path(path);
8082510e 4084 return err;
39279cc3
CM		 4085}
4086
4087/*
2aaa6655
JB		 4088 * btrfs_truncate_page - read, zero a chunk and write a page
4089 * @inode - inode that we're zeroing
4090 * @from - the offset to start zeroing
4091 * @len - the length to zero, 0 to zero the entire range respective to the
4092 * offset
4093 * @front - zero up to the offset instead of from the offset on
4094 *
4095 * This will find the page for the "from" offset and cow the page and zero the
4096 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4097 */
2aaa6655
JB		 4098int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4099 int front)
39279cc3 4100{
2aaa6655 4101 struct address_space *mapping = inode->i_mapping;
db94535d 4102 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4103 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4104 struct btrfs_ordered_extent *ordered;
2ac55d41 4105 struct extent_state *cached_state = NULL;
e6dcd2dc 4106 char *kaddr;
db94535d 4107 u32 blocksize = root->sectorsize;
39279cc3
CM		 4108 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4109 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4110 struct page *page;
3b16a4e3 4111 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4112 int ret = 0;
a52d9a80 4113 u64 page_start;
e6dcd2dc 4114 u64 page_end;
39279cc3 4115
2aaa6655
JB		 4116 if ((offset & (blocksize - 1)) == 0 &&
4117 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4118 goto out;
0ca1f7ce 4119 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4120 if (ret)
4121 goto out;
39279cc3 4122
211c17f5 4123again:
3b16a4e3 4124 page = find_or_create_page(mapping, index, mask);
5d5e103a 4125 if (!page) {
0ca1f7ce 4126 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4127 ret = -ENOMEM;
39279cc3 4128 goto out;
5d5e103a 4129 }
e6dcd2dc
CM		 4130
4131 page_start = page_offset(page);
4132 page_end = page_start + PAGE_CACHE_SIZE - 1;
4133
39279cc3 4134 if (!PageUptodate(page)) {
9ebefb18 4135 ret = btrfs_readpage(NULL, page);
39279cc3 4136 lock_page(page);
211c17f5
CM		 4137 if (page->mapping != mapping) {
4138 unlock_page(page);
4139 page_cache_release(page);
4140 goto again;
4141 }
39279cc3
CM		 4142 if (!PageUptodate(page)) {
4143 ret = -EIO;
89642229 4144 goto out_unlock;
39279cc3
CM		 4145 }
4146 }
211c17f5 4147 wait_on_page_writeback(page);
e6dcd2dc 4148
d0082371 4149 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4150 set_page_extent_mapped(page);
4151
4152 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4153 if (ordered) {
2ac55d41
JB		 4154 unlock_extent_cached(io_tree, page_start, page_end,
4155 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4156 unlock_page(page);
4157 page_cache_release(page);
eb84ae03 4158 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4159 btrfs_put_ordered_extent(ordered);
4160 goto again;
4161 }
4162
2ac55d41 4163 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4164 EXTENT_DIRTY | EXTENT_DELALLOC |
4165 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4166 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4167
2ac55d41
JB		 4168 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4169 &cached_state);
9ed74f2d 4170 if (ret) {
2ac55d41
JB		 4171 unlock_extent_cached(io_tree, page_start, page_end,
4172 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4173 goto out_unlock;
4174 }
4175
e6dcd2dc 4176 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4177 if (!len)
4178 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4179 kaddr = kmap(page);
2aaa6655
JB		 4180 if (front)
4181 memset(kaddr, 0, offset);
4182 else
4183 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4184 flush_dcache_page(page);
4185 kunmap(page);
4186 }
247e743c 4187 ClearPageChecked(page);
e6dcd2dc 4188 set_page_dirty(page);
2ac55d41
JB		 4189 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4190 GFP_NOFS);
39279cc3 4191
89642229 4192out_unlock:
5d5e103a 4193 if (ret)
0ca1f7ce 4194 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4195 unlock_page(page);
4196 page_cache_release(page);
4197out:
4198 return ret;
4199}
4200
695a0d0d
JB		 4201/*
4202 * This function puts in dummy file extents for the area we're creating a hole
4203 * for. So if we are truncating this file to a larger size we need to insert
4204 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4205 * the range between oldsize and size
4206 */
a41ad394 4207int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4208{
9036c102
YZ		 4209 struct btrfs_trans_handle *trans;
4210 struct btrfs_root *root = BTRFS_I(inode)->root;
4211 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4212 struct extent_map *em = NULL;
2ac55d41 4213 struct extent_state *cached_state = NULL;
5dc562c5 4214 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4215 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4216 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4217 u64 last_byte;
4218 u64 cur_offset;
4219 u64 hole_size;
9ed74f2d 4220 int err = 0;
39279cc3 4221
a71754fc
JB		 4222 /*
4223 * If our size started in the middle of a page we need to zero out the
4224 * rest of the page before we expand the i_size, otherwise we could
4225 * expose stale data.
4226 */
4227 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4228 if (err)
4229 return err;
4230
9036c102
YZ		 4231 if (size <= hole_start)
4232 return 0;
4233
9036c102
YZ		 4234 while (1) {
4235 struct btrfs_ordered_extent *ordered;
4236 btrfs_wait_ordered_range(inode, hole_start,
4237 block_end - hole_start);
2ac55d41 4238 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4239 &cached_state);
9036c102
YZ		 4240 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4241 if (!ordered)
4242 break;
2ac55d41
JB		 4243 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4244 &cached_state, GFP_NOFS);
9036c102
YZ		 4245 btrfs_put_ordered_extent(ordered);
4246 }
39279cc3 4247
9036c102
YZ		 4248 cur_offset = hole_start;
4249 while (1) {
4250 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4251 block_end - cur_offset, 0);
79787eaa
JM		 4252 if (IS_ERR(em)) {
4253 err = PTR_ERR(em);
f2767956 4254 em = NULL;
79787eaa
JM		 4255 break;
4256 }
9036c102 4257 last_byte = min(extent_map_end(em), block_end);
fda2832f 4258 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4259 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4260 struct extent_map *hole_em;
9036c102 4261 hole_size = last_byte - cur_offset;
9ed74f2d 4262
3642320e 4263 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4264 if (IS_ERR(trans)) {
4265 err = PTR_ERR(trans);
9ed74f2d 4266 break;
a22285a6 4267 }
8082510e 4268
5dc562c5
JB		 4269 err = btrfs_drop_extents(trans, root, inode,
4270 cur_offset,
2671485d 4271 cur_offset + hole_size, 1);
5b397377 4272 if (err) {
79787eaa 4273 btrfs_abort_transaction(trans, root, err);
5b397377 4274 btrfs_end_transaction(trans, root);
3893e33b 4275 break;
5b397377 4276 }
8082510e 4277
9036c102 4278 err = btrfs_insert_file_extent(trans, root,
33345d01 4279 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4280 0, hole_size, 0, hole_size,
4281 0, 0, 0);
5b397377 4282 if (err) {
79787eaa 4283 btrfs_abort_transaction(trans, root, err);
5b397377 4284 btrfs_end_transaction(trans, root);
3893e33b 4285 break;
5b397377 4286 }
8082510e 4287
5dc562c5
JB		 4288 btrfs_drop_extent_cache(inode, cur_offset,
4289 cur_offset + hole_size - 1, 0);
4290 hole_em = alloc_extent_map();
4291 if (!hole_em) {
4292 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4293 &BTRFS_I(inode)->runtime_flags);
4294 goto next;
4295 }
4296 hole_em->start = cur_offset;
4297 hole_em->len = hole_size;
4298 hole_em->orig_start = cur_offset;
8082510e 4299
5dc562c5
JB		 4300 hole_em->block_start = EXTENT_MAP_HOLE;
4301 hole_em->block_len = 0;
b4939680 4302 hole_em->orig_block_len = 0;
cc95bef6 4303 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4304 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4305 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4306 hole_em->generation = trans->transid;
8082510e 4307
5dc562c5
JB		 4308 while (1) {
4309 write_lock(&em_tree->lock);
09a2a8f9 4310 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4311 write_unlock(&em_tree->lock);
4312 if (err != -EEXIST)
4313 break;
4314 btrfs_drop_extent_cache(inode, cur_offset,
4315 cur_offset +
4316 hole_size - 1, 0);
4317 }
4318 free_extent_map(hole_em);
4319next:
3642320e 4320 btrfs_update_inode(trans, root, inode);
8082510e 4321 btrfs_end_transaction(trans, root);
9036c102
YZ		 4322 }
4323 free_extent_map(em);
a22285a6 4324 em = NULL;
9036c102 4325 cur_offset = last_byte;
8082510e 4326 if (cur_offset >= block_end)
9036c102
YZ		 4327 break;
4328 }
1832a6d5 4329
a22285a6 4330 free_extent_map(em);
2ac55d41
JB		 4331 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4332 GFP_NOFS);
9036c102
YZ		 4333 return err;
4334}
39279cc3 4335
3972f260 4336static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4337{
f4a2f4c5
MX		 4338 struct btrfs_root *root = BTRFS_I(inode)->root;
4339 struct btrfs_trans_handle *trans;
a41ad394 4340 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4341 loff_t newsize = attr->ia_size;
4342 int mask = attr->ia_valid;
8082510e
YZ		 4343 int ret;
4344
3972f260
ES		 4345 /*
4346 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4347 * special case where we need to update the times despite not having
4348 * these flags set. For all other operations the VFS set these flags
4349 * explicitly if it wants a timestamp update.
4350 */
4351 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4352 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4353
a41ad394 4354 if (newsize > oldsize) {
7caef267 4355 truncate_pagecache(inode, newsize);
a41ad394 4356 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4357 if (ret)
8082510e 4358 return ret;
8082510e 4359
f4a2f4c5
MX		 4360 trans = btrfs_start_transaction(root, 1);
4361 if (IS_ERR(trans))
4362 return PTR_ERR(trans);
4363
4364 i_size_write(inode, newsize);
4365 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4366 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4367 btrfs_end_transaction(trans, root);
a41ad394 4368 } else {
8082510e 4369
a41ad394
JB		 4370 /*
4371 * We're truncating a file that used to have good data down to
4372 * zero. Make sure it gets into the ordered flush list so that
4373 * any new writes get down to disk quickly.
4374 */
4375 if (newsize == 0)
72ac3c0d
JB		 4376 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4377 &BTRFS_I(inode)->runtime_flags);
8082510e 4378
f3fe820c
JB		 4379 /*
4380 * 1 for the orphan item we're going to add
4381 * 1 for the orphan item deletion.
4382 */
4383 trans = btrfs_start_transaction(root, 2);
4384 if (IS_ERR(trans))
4385 return PTR_ERR(trans);
4386
4387 /*
4388 * We need to do this in case we fail at _any_ point during the
4389 * actual truncate. Once we do the truncate_setsize we could
4390 * invalidate pages which forces any outstanding ordered io to
4391 * be instantly completed which will give us extents that need
4392 * to be truncated. If we fail to get an orphan inode down we
4393 * could have left over extents that were never meant to live,
4394 * so we need to garuntee from this point on that everything
4395 * will be consistent.
4396 */
4397 ret = btrfs_orphan_add(trans, inode);
4398 btrfs_end_transaction(trans, root);
4399 if (ret)
4400 return ret;
4401
a41ad394
JB		 4402 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4403 truncate_setsize(inode, newsize);
2e60a51e
MX		 4404
4405 /* Disable nonlocked read DIO to avoid the end less truncate */
4406 btrfs_inode_block_unlocked_dio(inode);
4407 inode_dio_wait(inode);
4408 btrfs_inode_resume_unlocked_dio(inode);
4409
a41ad394 4410 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4411 if (ret && inode->i_nlink) {
4412 int err;
4413
4414 /*
4415 * failed to truncate, disk_i_size is only adjusted down
4416 * as we remove extents, so it should represent the true
4417 * size of the inode, so reset the in memory size and
4418 * delete our orphan entry.
4419 */
4420 trans = btrfs_join_transaction(root);
4421 if (IS_ERR(trans)) {
4422 btrfs_orphan_del(NULL, inode);
4423 return ret;
4424 }
4425 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4426 err = btrfs_orphan_del(trans, inode);
4427 if (err)
4428 btrfs_abort_transaction(trans, root, err);
4429 btrfs_end_transaction(trans, root);
4430 }
8082510e
YZ		 4431 }
4432
a41ad394 4433 return ret;
8082510e
YZ		 4434}
4435
9036c102
YZ		 4436static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4437{
4438 struct inode *inode = dentry->d_inode;
b83cc969 4439 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4440 int err;
39279cc3 4441
b83cc969
LZ		 4442 if (btrfs_root_readonly(root))
4443 return -EROFS;
4444
9036c102
YZ		 4445 err = inode_change_ok(inode, attr);
4446 if (err)
4447 return err;
2bf5a725 4448
5a3f23d5 4449 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4450 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4451 if (err)
4452 return err;
39279cc3 4453 }
9036c102 4454
1025774c
CH		 4455 if (attr->ia_valid) {
4456 setattr_copy(inode, attr);
0c4d2d95 4457 inode_inc_iversion(inode);
22c44fe6 4458 err = btrfs_dirty_inode(inode);
1025774c 4459
22c44fe6 4460 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4461 err = btrfs_acl_chmod(inode);
4462 }
33268eaf 4463
39279cc3
CM		 4464 return err;
4465}
61295eb8 4466
bd555975 4467void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4468{
4469 struct btrfs_trans_handle *trans;
4470 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4471 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4472 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4473 int ret;
4474
1abe9b8a 4475 trace_btrfs_inode_evict(inode);
4476
39279cc3 4477 truncate_inode_pages(&inode->i_data, 0);
69e9c6c6
SB		 4478 if (inode->i_nlink &&
4479 ((btrfs_root_refs(&root->root_item) != 0 &&
4480 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4481 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4482 goto no_delete;
4483
39279cc3 4484 if (is_bad_inode(inode)) {
7b128766 4485 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4486 goto no_delete;
4487 }
bd555975 4488 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4489 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4490
c71bf099 4491 if (root->fs_info->log_root_recovering) {
6bf02314 4492 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4493 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4494 goto no_delete;
4495 }
4496
76dda93c 4497 if (inode->i_nlink > 0) {
69e9c6c6
SB		 4498 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4499 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 4500 goto no_delete;
4501 }
4502
0e8c36a9
MX		 4503 ret = btrfs_commit_inode_delayed_inode(inode);
4504 if (ret) {
4505 btrfs_orphan_del(NULL, inode);
4506 goto no_delete;
4507 }
4508
66d8f3dd 4509 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4510 if (!rsv) {
4511 btrfs_orphan_del(NULL, inode);
4512 goto no_delete;
4513 }
4a338542 4514 rsv->size = min_size;
ca7e70f5 4515 rsv->failfast = 1;
726c35fa 4516 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4517
dbe674a9 4518 btrfs_i_size_write(inode, 0);
5f39d397 4519
4289a667 4520 /*
8407aa46
MX		 4521 * This is a bit simpler than btrfs_truncate since we've already
4522 * reserved our space for our orphan item in the unlink, so we just
4523 * need to reserve some slack space in case we add bytes and update
4524 * inode item when doing the truncate.
4289a667 4525 */
8082510e 4526 while (1) {
08e007d2
MX		 4527 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4528 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4529
4530 /*
4531 * Try and steal from the global reserve since we will
4532 * likely not use this space anyway, we want to try as
4533 * hard as possible to get this to work.
4534 */
4535 if (ret)
4536 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4537
d68fc57b 4538 if (ret) {
c2cf52eb
SK		 4539 btrfs_warn(root->fs_info,
4540 "Could not get space for a delete, will truncate on mount %d",
4541 ret);
4289a667
JB		 4542 btrfs_orphan_del(NULL, inode);
4543 btrfs_free_block_rsv(root, rsv);
4544 goto no_delete;
d68fc57b 4545 }
7b128766 4546
0e8c36a9 4547 trans = btrfs_join_transaction(root);
4289a667
JB		 4548 if (IS_ERR(trans)) {
4549 btrfs_orphan_del(NULL, inode);
4550 btrfs_free_block_rsv(root, rsv);
4551 goto no_delete;
d68fc57b 4552 }
7b128766 4553
4289a667
JB		 4554 trans->block_rsv = rsv;
4555
d68fc57b 4556 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4557 if (ret != -ENOSPC)
8082510e 4558 break;
85e21bac 4559
8407aa46 4560 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4561 btrfs_end_transaction(trans, root);
4562 trans = NULL;
b53d3f5d 4563 btrfs_btree_balance_dirty(root);
8082510e 4564 }
5f39d397 4565
4289a667
JB		 4566 btrfs_free_block_rsv(root, rsv);
4567
4ef31a45
JB		 4568 /*
4569 * Errors here aren't a big deal, it just means we leave orphan items
4570 * in the tree. They will be cleaned up on the next mount.
4571 */
8082510e 4572 if (ret == 0) {
4289a667 4573 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 4574 btrfs_orphan_del(trans, inode);
4575 } else {
4576 btrfs_orphan_del(NULL, inode);
8082510e 4577 }
54aa1f4d 4578
4289a667 4579 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4580 if (!(root == root->fs_info->tree_root ||
4581 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4582 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4583
54aa1f4d 4584 btrfs_end_transaction(trans, root);
b53d3f5d 4585 btrfs_btree_balance_dirty(root);
39279cc3 4586no_delete:
89042e5a 4587 btrfs_remove_delayed_node(inode);
dbd5768f 4588 clear_inode(inode);
8082510e 4589 return;
39279cc3
CM		 4590}
4591
4592/*
4593 * this returns the key found in the dir entry in the location pointer.
4594 * If no dir entries were found, location->objectid is 0.
4595 */
4596static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4597 struct btrfs_key *location)
4598{
4599 const char *name = dentry->d_name.name;
4600 int namelen = dentry->d_name.len;
4601 struct btrfs_dir_item *di;
4602 struct btrfs_path *path;
4603 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4604 int ret = 0;
39279cc3
CM		 4605
4606 path = btrfs_alloc_path();
d8926bb3
MF		 4607 if (!path)
4608 return -ENOMEM;
3954401f 4609
33345d01 4610 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4611 namelen, 0);
0d9f7f3e
Y		 4612 if (IS_ERR(di))
4613 ret = PTR_ERR(di);
d397712b 4614
c704005d 4615 if (IS_ERR_OR_NULL(di))
3954401f 4616 goto out_err;
d397712b 4617
5f39d397 4618 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4619out:
39279cc3
CM		 4620 btrfs_free_path(path);
4621 return ret;
3954401f
CM		 4622out_err:
4623 location->objectid = 0;
4624 goto out;
39279cc3
CM		 4625}
4626
4627/*
4628 * when we hit a tree root in a directory, the btrfs part of the inode
4629 * needs to be changed to reflect the root directory of the tree root. This
4630 * is kind of like crossing a mount point.
4631 */
4632static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4633 struct inode *dir,
4634 struct dentry *dentry,
4635 struct btrfs_key *location,
4636 struct btrfs_root **sub_root)
39279cc3 4637{
4df27c4d
YZ		 4638 struct btrfs_path *path;
4639 struct btrfs_root *new_root;
4640 struct btrfs_root_ref *ref;
4641 struct extent_buffer *leaf;
4642 int ret;
4643 int err = 0;
39279cc3 4644
4df27c4d
YZ		 4645 path = btrfs_alloc_path();
4646 if (!path) {
4647 err = -ENOMEM;
4648 goto out;
4649 }
39279cc3 4650
4df27c4d
YZ		 4651 err = -ENOENT;
4652 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4653 BTRFS_I(dir)->root->root_key.objectid,
4654 location->objectid);
4655 if (ret) {
4656 if (ret < 0)
4657 err = ret;
4658 goto out;
4659 }
39279cc3 4660
4df27c4d
YZ		 4661 leaf = path->nodes[0];
4662 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4663 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4664 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4665 goto out;
39279cc3 4666
4df27c4d
YZ		 4667 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4668 (unsigned long)(ref + 1),
4669 dentry->d_name.len);
4670 if (ret)
4671 goto out;
4672
b3b4aa74 4673 btrfs_release_path(path);
4df27c4d
YZ		 4674
4675 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4676 if (IS_ERR(new_root)) {
4677 err = PTR_ERR(new_root);
4678 goto out;
4679 }
4680
4df27c4d
YZ		 4681 *sub_root = new_root;
4682 location->objectid = btrfs_root_dirid(&new_root->root_item);
4683 location->type = BTRFS_INODE_ITEM_KEY;
4684 location->offset = 0;
4685 err = 0;
4686out:
4687 btrfs_free_path(path);
4688 return err;
39279cc3
CM		 4689}
4690
5d4f98a2
YZ		 4691static void inode_tree_add(struct inode *inode)
4692{
4693 struct btrfs_root *root = BTRFS_I(inode)->root;
4694 struct btrfs_inode *entry;
03e860bd
FNP		 4695 struct rb_node **p;
4696 struct rb_node *parent;
cef21937 4697 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 4698 u64 ino = btrfs_ino(inode);
5d4f98a2 4699
1d3382cb 4700 if (inode_unhashed(inode))
76dda93c 4701 return;
e1409cef 4702 parent = NULL;
5d4f98a2 4703 spin_lock(&root->inode_lock);
e1409cef 4704 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 4705 while (*p) {
4706 parent = *p;
4707 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4708
33345d01 4709 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4710 p = &parent->rb_left;
33345d01 4711 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4712 p = &parent->rb_right;
5d4f98a2
YZ		 4713 else {
4714 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4715 (I_WILL_FREE | I_FREEING)));
cef21937 4716 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 4717 RB_CLEAR_NODE(parent);
4718 spin_unlock(&root->inode_lock);
cef21937 4719 return;
5d4f98a2
YZ		 4720 }
4721 }
cef21937
FDBM		 4722 rb_link_node(new, parent, p);
4723 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 4724 spin_unlock(&root->inode_lock);
4725}
4726
4727static void inode_tree_del(struct inode *inode)
4728{
4729 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4730 int empty = 0;
5d4f98a2 4731
03e860bd 4732 spin_lock(&root->inode_lock);
5d4f98a2 4733 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4734 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4735 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4736 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4737 }
03e860bd 4738 spin_unlock(&root->inode_lock);
76dda93c 4739
69e9c6c6 4740 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 4741 synchronize_srcu(&root->fs_info->subvol_srcu);
4742 spin_lock(&root->inode_lock);
4743 empty = RB_EMPTY_ROOT(&root->inode_tree);
4744 spin_unlock(&root->inode_lock);
4745 if (empty)
4746 btrfs_add_dead_root(root);
4747 }
4748}
4749
143bede5 4750void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4751{
4752 struct rb_node *node;
4753 struct rb_node *prev;
4754 struct btrfs_inode *entry;
4755 struct inode *inode;
4756 u64 objectid = 0;
4757
4758 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4759
4760 spin_lock(&root->inode_lock);
4761again:
4762 node = root->inode_tree.rb_node;
4763 prev = NULL;
4764 while (node) {
4765 prev = node;
4766 entry = rb_entry(node, struct btrfs_inode, rb_node);
4767
33345d01 4768 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4769 node = node->rb_left;
33345d01 4770 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4771 node = node->rb_right;
4772 else
4773 break;
4774 }
4775 if (!node) {
4776 while (prev) {
4777 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4778 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4779 node = prev;
4780 break;
4781 }
4782 prev = rb_next(prev);
4783 }
4784 }
4785 while (node) {
4786 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4787 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4788 inode = igrab(&entry->vfs_inode);
4789 if (inode) {
4790 spin_unlock(&root->inode_lock);
4791 if (atomic_read(&inode->i_count) > 1)
4792 d_prune_aliases(inode);
4793 /*
45321ac5 4794 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4795 * the inode cache when its usage count
4796 * hits zero.
4797 */
4798 iput(inode);
4799 cond_resched();
4800 spin_lock(&root->inode_lock);
4801 goto again;
4802 }
4803
4804 if (cond_resched_lock(&root->inode_lock))
4805 goto again;
4806
4807 node = rb_next(node);
4808 }
4809 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4810}
4811
e02119d5
CM		 4812static int btrfs_init_locked_inode(struct inode *inode, void *p)
4813{
4814 struct btrfs_iget_args *args = p;
4815 inode->i_ino = args->ino;
e02119d5 4816 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4817 return 0;
4818}
4819
4820static int btrfs_find_actor(struct inode *inode, void *opaque)
4821{
4822 struct btrfs_iget_args *args = opaque;
33345d01 4823 return args->ino == btrfs_ino(inode) &&
d397712b 4824 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4825}
4826
5d4f98a2
YZ		 4827static struct inode *btrfs_iget_locked(struct super_block *s,
4828 u64 objectid,
4829 struct btrfs_root *root)
39279cc3
CM		 4830{
4831 struct inode *inode;
4832 struct btrfs_iget_args args;
4833 args.ino = objectid;
4834 args.root = root;
4835
4836 inode = iget5_locked(s, objectid, btrfs_find_actor,
4837 btrfs_init_locked_inode,
4838 (void *)&args);
4839 return inode;
4840}
4841
1a54ef8c
BR		 4842/* Get an inode object given its location and corresponding root.
4843 * Returns in *is_new if the inode was read from disk
4844 */
4845struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4846 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4847{
4848 struct inode *inode;
4849
4850 inode = btrfs_iget_locked(s, location->objectid, root);
4851 if (!inode)
5d4f98a2 4852 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4853
4854 if (inode->i_state & I_NEW) {
4855 BTRFS_I(inode)->root = root;
4856 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4857 btrfs_read_locked_inode(inode);
1748f843
MF		 4858 if (!is_bad_inode(inode)) {
4859 inode_tree_add(inode);
4860 unlock_new_inode(inode);
4861 if (new)
4862 *new = 1;
4863 } else {
e0b6d65b
ST		 4864 unlock_new_inode(inode);
4865 iput(inode);
4866 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4867 }
4868 }
4869
1a54ef8c
BR		 4870 return inode;
4871}
4872
4df27c4d
YZ		 4873static struct inode *new_simple_dir(struct super_block *s,
4874 struct btrfs_key *key,
4875 struct btrfs_root *root)
4876{
4877 struct inode *inode = new_inode(s);
4878
4879 if (!inode)
4880 return ERR_PTR(-ENOMEM);
4881
4df27c4d
YZ		 4882 BTRFS_I(inode)->root = root;
4883 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4884 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4885
4886 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4887 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4888 inode->i_fop = &simple_dir_operations;
4889 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4890 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4891
4892 return inode;
4893}
4894
3de4586c 4895struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4896{
d397712b 4897 struct inode *inode;
4df27c4d 4898 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4899 struct btrfs_root *sub_root = root;
4900 struct btrfs_key location;
76dda93c 4901 int index;
b4aff1f8 4902 int ret = 0;
39279cc3
CM		 4903
4904 if (dentry->d_name.len > BTRFS_NAME_LEN)
4905 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4906
39e3c955 4907 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 4908 if (ret < 0)
4909 return ERR_PTR(ret);
5f39d397 4910
4df27c4d
YZ		 4911 if (location.objectid == 0)
4912 return NULL;
4913
4914 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4915 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4916 return inode;
4917 }
4918
4919 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4920
76dda93c 4921 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4922 ret = fixup_tree_root_location(root, dir, dentry,
4923 &location, &sub_root);
4924 if (ret < 0) {
4925 if (ret != -ENOENT)
4926 inode = ERR_PTR(ret);
4927 else
4928 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4929 } else {
73f73415 4930 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4931 }
76dda93c
YZ		 4932 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4933
34d19bad 4934 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4935 down_read(&root->fs_info->cleanup_work_sem);
4936 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4937 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4938 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 4939 if (ret) {
4940 iput(inode);
66b4ffd1 4941 inode = ERR_PTR(ret);
01cd3367 4942 }
c71bf099
YZ		 4943 }
4944
3de4586c
CM		 4945 return inode;
4946}
4947
fe15ce44 4948static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4949{
4950 struct btrfs_root *root;
848cce0d 4951 struct inode *inode = dentry->d_inode;
76dda93c 4952
848cce0d
LZ		 4953 if (!inode && !IS_ROOT(dentry))
4954 inode = dentry->d_parent->d_inode;
76dda93c 4955
848cce0d
LZ		 4956 if (inode) {
4957 root = BTRFS_I(inode)->root;
efefb143
YZ		 4958 if (btrfs_root_refs(&root->root_item) == 0)
4959 return 1;
848cce0d
LZ		 4960
4961 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4962 return 1;
efefb143 4963 }
76dda93c
YZ		 4964 return 0;
4965}
4966
b4aff1f8
JB		 4967static void btrfs_dentry_release(struct dentry *dentry)
4968{
4969 if (dentry->d_fsdata)
4970 kfree(dentry->d_fsdata);
4971}
4972
3de4586c 4973static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 4974 unsigned int flags)
3de4586c 4975{
a66e7cc6
JB		 4976 struct dentry *ret;
4977
4978 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 4979 return ret;
39279cc3
CM		 4980}
4981
16cdcec7 4982unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4983 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4984};
4985
9cdda8d3 4986static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 4987{
9cdda8d3 4988 struct inode *inode = file_inode(file);
39279cc3
CM		 4989 struct btrfs_root *root = BTRFS_I(inode)->root;
4990 struct btrfs_item *item;
4991 struct btrfs_dir_item *di;
4992 struct btrfs_key key;
5f39d397 4993 struct btrfs_key found_key;
39279cc3 4994 struct btrfs_path *path;
16cdcec7
MX		 4995 struct list_head ins_list;
4996 struct list_head del_list;
39279cc3 4997 int ret;
5f39d397 4998 struct extent_buffer *leaf;
39279cc3 4999 int slot;
39279cc3
CM		 5000 unsigned char d_type;
5001 int over = 0;
5002 u32 di_cur;
5003 u32 di_total;
5004 u32 di_len;
5005 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5006 char tmp_name[32];
5007 char *name_ptr;
5008 int name_len;
9cdda8d3 5009 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5010
5011 /* FIXME, use a real flag for deciding about the key type */
5012 if (root->fs_info->tree_root == root)
5013 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5014
9cdda8d3
AV		 5015 if (!dir_emit_dots(file, ctx))
5016 return 0;
5017
49593bfa 5018 path = btrfs_alloc_path();
16cdcec7
MX		 5019 if (!path)
5020 return -ENOMEM;
ff5714cc 5021
026fd317 5022 path->reada = 1;
49593bfa 5023
16cdcec7
MX		 5024 if (key_type == BTRFS_DIR_INDEX_KEY) {
5025 INIT_LIST_HEAD(&ins_list);
5026 INIT_LIST_HEAD(&del_list);
5027 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5028 }
5029
39279cc3 5030 btrfs_set_key_type(&key, key_type);
9cdda8d3 5031 key.offset = ctx->pos;
33345d01 5032 key.objectid = btrfs_ino(inode);
5f39d397 5033
39279cc3
CM		 5034 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5035 if (ret < 0)
5036 goto err;
49593bfa
DW		 5037
5038 while (1) {
5f39d397 5039 leaf = path->nodes[0];
39279cc3 5040 slot = path->slots[0];
b9e03af0
LZ		 5041 if (slot >= btrfs_header_nritems(leaf)) {
5042 ret = btrfs_next_leaf(root, path);
5043 if (ret < 0)
5044 goto err;
5045 else if (ret > 0)
5046 break;
5047 continue;
39279cc3 5048 }
3de4586c 5049
dd3cc16b 5050 item = btrfs_item_nr(slot);
5f39d397
CM		 5051 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5052
5053 if (found_key.objectid != key.objectid)
39279cc3 5054 break;
5f39d397 5055 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5056 break;
9cdda8d3 5057 if (found_key.offset < ctx->pos)
b9e03af0 5058 goto next;
16cdcec7
MX		 5059 if (key_type == BTRFS_DIR_INDEX_KEY &&
5060 btrfs_should_delete_dir_index(&del_list,
5061 found_key.offset))
5062 goto next;
5f39d397 5063
9cdda8d3 5064 ctx->pos = found_key.offset;
16cdcec7 5065 is_curr = 1;
49593bfa 5066
39279cc3
CM		 5067 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5068 di_cur = 0;
5f39d397 5069 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5070
5071 while (di_cur < di_total) {
5f39d397
CM		 5072 struct btrfs_key location;
5073
22a94d44
JB		 5074 if (verify_dir_item(root, leaf, di))
5075 break;
5076
5f39d397 5077 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5078 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5079 name_ptr = tmp_name;
5080 } else {
5081 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5082 if (!name_ptr) {
5083 ret = -ENOMEM;
5084 goto err;
5085 }
5f39d397
CM		 5086 }
5087 read_extent_buffer(leaf, name_ptr,
5088 (unsigned long)(di + 1), name_len);
5089
5090 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5091 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5092
fede766f 5093
3de4586c 5094 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5095 * skip it.
5096 *
5097 * In contrast to old kernels, we insert the snapshot's
5098 * dir item and dir index after it has been created, so
5099 * we won't find a reference to our own snapshot. We
5100 * still keep the following code for backward
5101 * compatibility.
3de4586c
CM		 5102 */
5103 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5104 location.objectid == root->root_key.objectid) {
5105 over = 0;
5106 goto skip;
5107 }
9cdda8d3
AV		 5108 over = !dir_emit(ctx, name_ptr, name_len,
5109 location.objectid, d_type);
5f39d397 5110
3de4586c 5111skip:
5f39d397
CM		 5112 if (name_ptr != tmp_name)
5113 kfree(name_ptr);
5114
39279cc3
CM		 5115 if (over)
5116 goto nopos;
5103e947 5117 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5118 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5119 di_cur += di_len;
5120 di = (struct btrfs_dir_item *)((char *)di + di_len);
5121 }
b9e03af0
LZ		 5122next:
5123 path->slots[0]++;
39279cc3 5124 }
49593bfa 5125
16cdcec7
MX		 5126 if (key_type == BTRFS_DIR_INDEX_KEY) {
5127 if (is_curr)
9cdda8d3
AV		 5128 ctx->pos++;
5129 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5130 if (ret)
5131 goto nopos;
5132 }
5133
49593bfa 5134 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5135 ctx->pos++;
5136
5137 /*
5138 * Stop new entries from being returned after we return the last
5139 * entry.
5140 *
5141 * New directory entries are assigned a strictly increasing
5142 * offset. This means that new entries created during readdir
5143 * are *guaranteed* to be seen in the future by that readdir.
5144 * This has broken buggy programs which operate on names as
5145 * they're returned by readdir. Until we re-use freed offsets
5146 * we have this hack to stop new entries from being returned
5147 * under the assumption that they'll never reach this huge
5148 * offset.
5149 *
5150 * This is being careful not to overflow 32bit loff_t unless the
5151 * last entry requires it because doing so has broken 32bit apps
5152 * in the past.
5153 */
5154 if (key_type == BTRFS_DIR_INDEX_KEY) {
5155 if (ctx->pos >= INT_MAX)
5156 ctx->pos = LLONG_MAX;
5157 else
5158 ctx->pos = INT_MAX;
5159 }
39279cc3
CM		 5160nopos:
5161 ret = 0;
5162err:
16cdcec7
MX		 5163 if (key_type == BTRFS_DIR_INDEX_KEY)
5164 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5165 btrfs_free_path(path);
39279cc3
CM		 5166 return ret;
5167}
5168
a9185b41 5169int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5170{
5171 struct btrfs_root *root = BTRFS_I(inode)->root;
5172 struct btrfs_trans_handle *trans;
5173 int ret = 0;
0af3d00b 5174 bool nolock = false;
39279cc3 5175
72ac3c0d 5176 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5177 return 0;
5178
83eea1f1 5179 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5180 nolock = true;
0af3d00b 5181
a9185b41 5182 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5183 if (nolock)
7a7eaa40 5184 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5185 else
7a7eaa40 5186 trans = btrfs_join_transaction(root);
3612b495
TI		 5187 if (IS_ERR(trans))
5188 return PTR_ERR(trans);
a698d075 5189 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5190 }
5191 return ret;
5192}
5193
5194/*
54aa1f4d 5195 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5196 * inode changes. But, it is most likely to find the inode in cache.
5197 * FIXME, needs more benchmarking...there are no reasons other than performance
5198 * to keep or drop this code.
5199 */
48a3b636 5200static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5201{
5202 struct btrfs_root *root = BTRFS_I(inode)->root;
5203 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5204 int ret;
5205
72ac3c0d 5206 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5207 return 0;
39279cc3 5208
7a7eaa40 5209 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5210 if (IS_ERR(trans))
5211 return PTR_ERR(trans);
8929ecfa
YZ		 5212
5213 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5214 if (ret && ret == -ENOSPC) {
5215 /* whoops, lets try again with the full transaction */
5216 btrfs_end_transaction(trans, root);
5217 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5218 if (IS_ERR(trans))
5219 return PTR_ERR(trans);
8929ecfa 5220
94b60442 5221 ret = btrfs_update_inode(trans, root, inode);
94b60442 5222 }
39279cc3 5223 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5224 if (BTRFS_I(inode)->delayed_node)
5225 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5226
5227 return ret;
5228}
5229
5230/*
5231 * This is a copy of file_update_time. We need this so we can return error on
5232 * ENOSPC for updating the inode in the case of file write and mmap writes.
5233 */
e41f941a
JB		 5234static int btrfs_update_time(struct inode *inode, struct timespec *now,
5235 int flags)
22c44fe6 5236{
2bc55652
AB		 5237 struct btrfs_root *root = BTRFS_I(inode)->root;
5238
5239 if (btrfs_root_readonly(root))
5240 return -EROFS;
5241
e41f941a 5242 if (flags & S_VERSION)
22c44fe6 5243 inode_inc_iversion(inode);
e41f941a
JB		 5244 if (flags & S_CTIME)
5245 inode->i_ctime = *now;
5246 if (flags & S_MTIME)
5247 inode->i_mtime = *now;
5248 if (flags & S_ATIME)
5249 inode->i_atime = *now;
5250 return btrfs_dirty_inode(inode);
39279cc3
CM		 5251}
5252
d352ac68
CM		 5253/*
5254 * find the highest existing sequence number in a directory
5255 * and then set the in-memory index_cnt variable to reflect
5256 * free sequence numbers
5257 */
aec7477b
JB		 5258static int btrfs_set_inode_index_count(struct inode *inode)
5259{
5260 struct btrfs_root *root = BTRFS_I(inode)->root;
5261 struct btrfs_key key, found_key;
5262 struct btrfs_path *path;
5263 struct extent_buffer *leaf;
5264 int ret;
5265
33345d01 5266 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5267 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5268 key.offset = (u64)-1;
5269
5270 path = btrfs_alloc_path();
5271 if (!path)
5272 return -ENOMEM;
5273
5274 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5275 if (ret < 0)
5276 goto out;
5277 /* FIXME: we should be able to handle this */
5278 if (ret == 0)
5279 goto out;
5280 ret = 0;
5281
5282 /*
5283 * MAGIC NUMBER EXPLANATION:
5284 * since we search a directory based on f_pos we have to start at 2
5285 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5286 * else has to start at 2
5287 */
5288 if (path->slots[0] == 0) {
5289 BTRFS_I(inode)->index_cnt = 2;
5290 goto out;
5291 }
5292
5293 path->slots[0]--;
5294
5295 leaf = path->nodes[0];
5296 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5297
33345d01 5298 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5299 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5300 BTRFS_I(inode)->index_cnt = 2;
5301 goto out;
5302 }
5303
5304 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5305out:
5306 btrfs_free_path(path);
5307 return ret;
5308}
5309
d352ac68
CM		 5310/*
5311 * helper to find a free sequence number in a given directory. This current
5312 * code is very simple, later versions will do smarter things in the btree
5313 */
3de4586c 5314int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5315{
5316 int ret = 0;
5317
5318 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5319 ret = btrfs_inode_delayed_dir_index_count(dir);
5320 if (ret) {
5321 ret = btrfs_set_inode_index_count(dir);
5322 if (ret)
5323 return ret;
5324 }
aec7477b
JB		 5325 }
5326
00e4e6b3 5327 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5328 BTRFS_I(dir)->index_cnt++;
5329
5330 return ret;
5331}
5332
39279cc3
CM		 5333static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5334 struct btrfs_root *root,
aec7477b 5335 struct inode *dir,
9c58309d 5336 const char *name, int name_len,
175a4eb7
AV		 5337 u64 ref_objectid, u64 objectid,
5338 umode_t mode, u64 *index)
39279cc3
CM		 5339{
5340 struct inode *inode;
5f39d397 5341 struct btrfs_inode_item *inode_item;
39279cc3 5342 struct btrfs_key *location;
5f39d397 5343 struct btrfs_path *path;
9c58309d
CM		 5344 struct btrfs_inode_ref *ref;
5345 struct btrfs_key key[2];
5346 u32 sizes[2];
5347 unsigned long ptr;
39279cc3
CM		 5348 int ret;
5349 int owner;
5350
5f39d397 5351 path = btrfs_alloc_path();
d8926bb3
MF		 5352 if (!path)
5353 return ERR_PTR(-ENOMEM);
5f39d397 5354
39279cc3 5355 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5356 if (!inode) {
5357 btrfs_free_path(path);
39279cc3 5358 return ERR_PTR(-ENOMEM);
8fb27640 5359 }
39279cc3 5360
581bb050
LZ		 5361 /*
5362 * we have to initialize this early, so we can reclaim the inode
5363 * number if we fail afterwards in this function.
5364 */
5365 inode->i_ino = objectid;
5366
aec7477b 5367 if (dir) {
1abe9b8a 5368 trace_btrfs_inode_request(dir);
5369
3de4586c 5370 ret = btrfs_set_inode_index(dir, index);
09771430 5371 if (ret) {
8fb27640 5372 btrfs_free_path(path);
09771430 5373 iput(inode);
aec7477b 5374 return ERR_PTR(ret);
09771430 5375 }
aec7477b
JB		 5376 }
5377 /*
5378 * index_cnt is ignored for everything but a dir,
5379 * btrfs_get_inode_index_count has an explanation for the magic
5380 * number
5381 */
5382 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5383 BTRFS_I(inode)->root = root;
e02119d5 5384 BTRFS_I(inode)->generation = trans->transid;
76195853 5385 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5386
5dc562c5
JB		 5387 /*
5388 * We could have gotten an inode number from somebody who was fsynced
5389 * and then removed in this same transaction, so let's just set full
5390 * sync since it will be a full sync anyway and this will blow away the
5391 * old info in the log.
5392 */
5393 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5394
569254b0 5395 if (S_ISDIR(mode))
39279cc3
CM		 5396 owner = 0;
5397 else
5398 owner = 1;
9c58309d
CM		 5399
5400 key[0].objectid = objectid;
5401 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5402 key[0].offset = 0;
5403
f186373f
MF		 5404 /*
5405 * Start new inodes with an inode_ref. This is slightly more
5406 * efficient for small numbers of hard links since they will
5407 * be packed into one item. Extended refs will kick in if we
5408 * add more hard links than can fit in the ref item.
5409 */
9c58309d
CM		 5410 key[1].objectid = objectid;
5411 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5412 key[1].offset = ref_objectid;
5413
5414 sizes[0] = sizeof(struct btrfs_inode_item);
5415 sizes[1] = name_len + sizeof(*ref);
5416
b9473439 5417 path->leave_spinning = 1;
9c58309d
CM		 5418 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5419 if (ret != 0)
5f39d397
CM		 5420 goto fail;
5421
ecc11fab 5422 inode_init_owner(inode, dir, mode);
a76a3cd4 5423 inode_set_bytes(inode, 0);
39279cc3 5424 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5425 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5426 struct btrfs_inode_item);
293f7e07
LZ		 5427 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5428 sizeof(*inode_item));
e02119d5 5429 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5430
5431 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5432 struct btrfs_inode_ref);
5433 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5434 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5435 ptr = (unsigned long)(ref + 1);
5436 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5437
5f39d397
CM		 5438 btrfs_mark_buffer_dirty(path->nodes[0]);
5439 btrfs_free_path(path);
5440
39279cc3
CM		 5441 location = &BTRFS_I(inode)->location;
5442 location->objectid = objectid;
39279cc3
CM		 5443 location->offset = 0;
5444 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5445
6cbff00f
CH		 5446 btrfs_inherit_iflags(inode, dir);
5447
569254b0 5448 if (S_ISREG(mode)) {
94272164
CM		 5449 if (btrfs_test_opt(root, NODATASUM))
5450 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5451 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5452 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5453 BTRFS_INODE_NODATASUM;
94272164
CM		 5454 }
5455
39279cc3 5456 insert_inode_hash(inode);
5d4f98a2 5457 inode_tree_add(inode);
1abe9b8a 5458
5459 trace_btrfs_inode_new(inode);
1973f0fa 5460 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5461
8ea05e3a
AB		 5462 btrfs_update_root_times(trans, root);
5463
39279cc3 5464 return inode;
5f39d397 5465fail:
aec7477b
JB		 5466 if (dir)
5467 BTRFS_I(dir)->index_cnt--;
5f39d397 5468 btrfs_free_path(path);
09771430 5469 iput(inode);
5f39d397 5470 return ERR_PTR(ret);
39279cc3
CM		 5471}
5472
5473static inline u8 btrfs_inode_type(struct inode *inode)
5474{
5475 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5476}
5477
d352ac68
CM		 5478/*
5479 * utility function to add 'inode' into 'parent_inode' with
5480 * a give name and a given sequence number.
5481 * if 'add_backref' is true, also insert a backref from the
5482 * inode to the parent directory.
5483 */
e02119d5
CM		 5484int btrfs_add_link(struct btrfs_trans_handle *trans,
5485 struct inode *parent_inode, struct inode *inode,
5486 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5487{
4df27c4d 5488 int ret = 0;
39279cc3 5489 struct btrfs_key key;
e02119d5 5490 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5491 u64 ino = btrfs_ino(inode);
5492 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5493
33345d01 5494 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5495 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5496 } else {
33345d01 5497 key.objectid = ino;
4df27c4d
YZ		 5498 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5499 key.offset = 0;
5500 }
5501
33345d01 5502 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5503 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5504 key.objectid, root->root_key.objectid,
33345d01 5505 parent_ino, index, name, name_len);
4df27c4d 5506 } else if (add_backref) {
33345d01
LZ		 5507 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5508 parent_ino, index);
4df27c4d 5509 }
39279cc3 5510
79787eaa
JM		 5511 /* Nothing to clean up yet */
5512 if (ret)
5513 return ret;
4df27c4d 5514
79787eaa
JM		 5515 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5516 parent_inode, &key,
5517 btrfs_inode_type(inode), index);
9c52057c 5518 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5519 goto fail_dir_item;
5520 else if (ret) {
5521 btrfs_abort_transaction(trans, root, ret);
5522 return ret;
39279cc3 5523 }
79787eaa
JM		 5524
5525 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5526 name_len * 2);
0c4d2d95 5527 inode_inc_iversion(parent_inode);
79787eaa
JM		 5528 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5529 ret = btrfs_update_inode(trans, root, parent_inode);
5530 if (ret)
5531 btrfs_abort_transaction(trans, root, ret);
39279cc3 5532 return ret;
fe66a05a
CM		 5533
5534fail_dir_item:
5535 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5536 u64 local_index;
5537 int err;
5538 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5539 key.objectid, root->root_key.objectid,
5540 parent_ino, &local_index, name, name_len);
5541
5542 } else if (add_backref) {
5543 u64 local_index;
5544 int err;
5545
5546 err = btrfs_del_inode_ref(trans, root, name, name_len,
5547 ino, parent_ino, &local_index);
5548 }
5549 return ret;
39279cc3
CM		 5550}
5551
5552static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5553 struct inode *dir, struct dentry *dentry,
5554 struct inode *inode, int backref, u64 index)
39279cc3 5555{
a1b075d2
JB		 5556 int err = btrfs_add_link(trans, dir, inode,
5557 dentry->d_name.name, dentry->d_name.len,
5558 backref, index);
39279cc3
CM		 5559 if (err > 0)
5560 err = -EEXIST;
5561 return err;
5562}
5563
618e21d5 5564static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5565 umode_t mode, dev_t rdev)
618e21d5
JB		 5566{
5567 struct btrfs_trans_handle *trans;
5568 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5569 struct inode *inode = NULL;
618e21d5
JB		 5570 int err;
5571 int drop_inode = 0;
5572 u64 objectid;
00e4e6b3 5573 u64 index = 0;
618e21d5
JB		 5574
5575 if (!new_valid_dev(rdev))
5576 return -EINVAL;
5577
9ed74f2d
JB		 5578 /*
5579 * 2 for inode item and ref
5580 * 2 for dir items
5581 * 1 for xattr if selinux is on
5582 */
a22285a6
YZ		 5583 trans = btrfs_start_transaction(root, 5);
5584 if (IS_ERR(trans))
5585 return PTR_ERR(trans);
1832a6d5 5586
581bb050
LZ		 5587 err = btrfs_find_free_ino(root, &objectid);
5588 if (err)
5589 goto out_unlock;
5590
aec7477b 5591 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5592 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5593 mode, &index);
7cf96da3
TI		 5594 if (IS_ERR(inode)) {
5595 err = PTR_ERR(inode);
618e21d5 5596 goto out_unlock;
7cf96da3 5597 }
618e21d5 5598
2a7dba39 5599 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5600 if (err) {
5601 drop_inode = 1;
5602 goto out_unlock;
5603 }
5604
ad19db71
CS		 5605 /*
5606 * If the active LSM wants to access the inode during
5607 * d_instantiate it needs these. Smack checks to see
5608 * if the filesystem supports xattrs by looking at the
5609 * ops vector.
5610 */
5611
5612 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5613 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5614 if (err)
5615 drop_inode = 1;
5616 else {
618e21d5 5617 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5618 btrfs_update_inode(trans, root, inode);
08c422c2 5619 d_instantiate(dentry, inode);
618e21d5 5620 }
618e21d5 5621out_unlock:
7ad85bb7 5622 btrfs_end_transaction(trans, root);
b53d3f5d 5623 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5624 if (drop_inode) {
5625 inode_dec_link_count(inode);
5626 iput(inode);
5627 }
618e21d5
JB		 5628 return err;
5629}
5630
39279cc3 5631static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5632 umode_t mode, bool excl)
39279cc3
CM		 5633{
5634 struct btrfs_trans_handle *trans;
5635 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5636 struct inode *inode = NULL;
43baa579 5637 int drop_inode_on_err = 0;
a22285a6 5638 int err;
39279cc3 5639 u64 objectid;
00e4e6b3 5640 u64 index = 0;
39279cc3 5641
9ed74f2d
JB		 5642 /*
5643 * 2 for inode item and ref
5644 * 2 for dir items
5645 * 1 for xattr if selinux is on
5646 */
a22285a6
YZ		 5647 trans = btrfs_start_transaction(root, 5);
5648 if (IS_ERR(trans))
5649 return PTR_ERR(trans);
9ed74f2d 5650
581bb050
LZ		 5651 err = btrfs_find_free_ino(root, &objectid);
5652 if (err)
5653 goto out_unlock;
5654
aec7477b 5655 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5656 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5657 mode, &index);
7cf96da3
TI		 5658 if (IS_ERR(inode)) {
5659 err = PTR_ERR(inode);
39279cc3 5660 goto out_unlock;
7cf96da3 5661 }
43baa579 5662 drop_inode_on_err = 1;
39279cc3 5663
2a7dba39 5664 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5665 if (err)
33268eaf 5666 goto out_unlock;
33268eaf 5667
9185aa58
FB		 5668 err = btrfs_update_inode(trans, root, inode);
5669 if (err)
5670 goto out_unlock;
5671
ad19db71
CS		 5672 /*
5673 * If the active LSM wants to access the inode during
5674 * d_instantiate it needs these. Smack checks to see
5675 * if the filesystem supports xattrs by looking at the
5676 * ops vector.
5677 */
5678 inode->i_fop = &btrfs_file_operations;
5679 inode->i_op = &btrfs_file_inode_operations;
5680
a1b075d2 5681 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5682 if (err)
43baa579
FB		 5683 goto out_unlock;
5684
5685 inode->i_mapping->a_ops = &btrfs_aops;
5686 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5687 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5688 d_instantiate(dentry, inode);
5689
39279cc3 5690out_unlock:
7ad85bb7 5691 btrfs_end_transaction(trans, root);
43baa579 5692 if (err && drop_inode_on_err) {
39279cc3
CM		 5693 inode_dec_link_count(inode);
5694 iput(inode);
5695 }
b53d3f5d 5696 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5697 return err;
5698}
5699
5700static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5701 struct dentry *dentry)
5702{
5703 struct btrfs_trans_handle *trans;
5704 struct btrfs_root *root = BTRFS_I(dir)->root;
5705 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5706 u64 index;
39279cc3
CM		 5707 int err;
5708 int drop_inode = 0;
5709
4a8be425
TH		 5710 /* do not allow sys_link's with other subvols of the same device */
5711 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5712 return -EXDEV;
4a8be425 5713
f186373f 5714 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5715 return -EMLINK;
4a8be425 5716
3de4586c 5717 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5718 if (err)
5719 goto fail;
5720
a22285a6 5721 /*
7e6b6465 5722 * 2 items for inode and inode ref
a22285a6 5723 * 2 items for dir items
7e6b6465 5724 * 1 item for parent inode
a22285a6 5725 */
7e6b6465 5726 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5727 if (IS_ERR(trans)) {
5728 err = PTR_ERR(trans);
5729 goto fail;
5730 }
5f39d397 5731
3153495d 5732 btrfs_inc_nlink(inode);
0c4d2d95 5733 inode_inc_iversion(inode);
3153495d 5734 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5735 ihold(inode);
e9976151 5736 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5737
a1b075d2 5738 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5739
a5719521 5740 if (err) {
54aa1f4d 5741 drop_inode = 1;
a5719521 5742 } else {
10d9f309 5743 struct dentry *parent = dentry->d_parent;
a5719521 5744 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5745 if (err)
5746 goto fail;
08c422c2 5747 d_instantiate(dentry, inode);
6a912213 5748 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5749 }
39279cc3 5750
7ad85bb7 5751 btrfs_end_transaction(trans, root);
1832a6d5 5752fail:
39279cc3
CM		 5753 if (drop_inode) {
5754 inode_dec_link_count(inode);
5755 iput(inode);
5756 }
b53d3f5d 5757 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5758 return err;
5759}
5760
18bb1db3 5761static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5762{
b9d86667 5763 struct inode *inode = NULL;
39279cc3
CM		 5764 struct btrfs_trans_handle *trans;
5765 struct btrfs_root *root = BTRFS_I(dir)->root;
5766 int err = 0;
5767 int drop_on_err = 0;
b9d86667 5768 u64 objectid = 0;
00e4e6b3 5769 u64 index = 0;
39279cc3 5770
9ed74f2d
JB		 5771 /*
5772 * 2 items for inode and ref
5773 * 2 items for dir items
5774 * 1 for xattr if selinux is on
5775 */
a22285a6
YZ		 5776 trans = btrfs_start_transaction(root, 5);
5777 if (IS_ERR(trans))
5778 return PTR_ERR(trans);
39279cc3 5779
581bb050
LZ		 5780 err = btrfs_find_free_ino(root, &objectid);
5781 if (err)
5782 goto out_fail;
5783
aec7477b 5784 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5785 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5786 S_IFDIR | mode, &index);
39279cc3
CM		 5787 if (IS_ERR(inode)) {
5788 err = PTR_ERR(inode);
5789 goto out_fail;
5790 }
5f39d397 5791
39279cc3 5792 drop_on_err = 1;
33268eaf 5793
2a7dba39 5794 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5795 if (err)
5796 goto out_fail;
5797
39279cc3
CM		 5798 inode->i_op = &btrfs_dir_inode_operations;
5799 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5800
dbe674a9 5801 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5802 err = btrfs_update_inode(trans, root, inode);
5803 if (err)
5804 goto out_fail;
5f39d397 5805
a1b075d2
JB		 5806 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5807 dentry->d_name.len, 0, index);
39279cc3
CM		 5808 if (err)
5809 goto out_fail;
5f39d397 5810
39279cc3
CM		 5811 d_instantiate(dentry, inode);
5812 drop_on_err = 0;
39279cc3
CM		 5813
5814out_fail:
7ad85bb7 5815 btrfs_end_transaction(trans, root);
39279cc3
CM		 5816 if (drop_on_err)
5817 iput(inode);
b53d3f5d 5818 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5819 return err;
5820}
5821
d352ac68
CM		 5822/* helper for btfs_get_extent. Given an existing extent in the tree,
5823 * and an extent that you want to insert, deal with overlap and insert
5824 * the new extent into the tree.
5825 */
3b951516
CM		 5826static int merge_extent_mapping(struct extent_map_tree *em_tree,
5827 struct extent_map *existing,
e6dcd2dc
CM		 5828 struct extent_map *em,
5829 u64 map_start, u64 map_len)
3b951516
CM		 5830{
5831 u64 start_diff;
3b951516 5832
e6dcd2dc
CM		 5833 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5834 start_diff = map_start - em->start;
5835 em->start = map_start;
5836 em->len = map_len;
c8b97818
CM		 5837 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5838 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5839 em->block_start += start_diff;
c8b97818
CM		 5840 em->block_len -= start_diff;
5841 }
09a2a8f9 5842 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 5843}
5844
c8b97818
CM		 5845static noinline int uncompress_inline(struct btrfs_path *path,
5846 struct inode *inode, struct page *page,
5847 size_t pg_offset, u64 extent_offset,
5848 struct btrfs_file_extent_item *item)
5849{
5850 int ret;
5851 struct extent_buffer *leaf = path->nodes[0];
5852 char *tmp;
5853 size_t max_size;
5854 unsigned long inline_size;
5855 unsigned long ptr;
261507a0 5856 int compress_type;
c8b97818
CM		 5857
5858 WARN_ON(pg_offset != 0);
261507a0 5859 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5860 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5861 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 5862 btrfs_item_nr(path->slots[0]));
c8b97818 5863 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5864 if (!tmp)
5865 return -ENOMEM;
c8b97818
CM		 5866 ptr = btrfs_file_extent_inline_start(item);
5867
5868 read_extent_buffer(leaf, tmp, ptr, inline_size);
5869
5b050f04 5870 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5871 ret = btrfs_decompress(compress_type, tmp, page,
5872 extent_offset, inline_size, max_size);
c8b97818 5873 if (ret) {
7ac687d9 5874 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5875 unsigned long copy_size = min_t(u64,
5876 PAGE_CACHE_SIZE - pg_offset,
5877 max_size - extent_offset);
5878 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5879 kunmap_atomic(kaddr);
c8b97818
CM		 5880 }
5881 kfree(tmp);
5882 return 0;
5883}
5884
d352ac68
CM		 5885/*
5886 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5887 * the ugly parts come from merging extents from the disk with the in-ram
5888 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5889 * where the in-ram extents might be locked pending data=ordered completion.
5890 *
5891 * This also copies inline extents directly into the page.
5892 */
d397712b 5893
a52d9a80 5894struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5895 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5896 int create)
5897{
5898 int ret;
5899 int err = 0;
db94535d 5900 u64 bytenr;
a52d9a80
CM		 5901 u64 extent_start = 0;
5902 u64 extent_end = 0;
33345d01 5903 u64 objectid = btrfs_ino(inode);
a52d9a80 5904 u32 found_type;
f421950f 5905 struct btrfs_path *path = NULL;
a52d9a80
CM		 5906 struct btrfs_root *root = BTRFS_I(inode)->root;
5907 struct btrfs_file_extent_item *item;
5f39d397
CM		 5908 struct extent_buffer *leaf;
5909 struct btrfs_key found_key;
a52d9a80
CM		 5910 struct extent_map *em = NULL;
5911 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5912 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5913 struct btrfs_trans_handle *trans = NULL;
261507a0 5914 int compress_type;
a52d9a80 5915
a52d9a80 5916again:
890871be 5917 read_lock(&em_tree->lock);
d1310b2e 5918 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5919 if (em)
5920 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5921 read_unlock(&em_tree->lock);
d1310b2e 5922
a52d9a80 5923 if (em) {
e1c4b745
CM		 5924 if (em->start > start || em->start + em->len <= start)
5925 free_extent_map(em);
5926 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5927 free_extent_map(em);
5928 else
5929 goto out;
a52d9a80 5930 }
172ddd60 5931 em = alloc_extent_map();
a52d9a80 5932 if (!em) {
d1310b2e
CM		 5933 err = -ENOMEM;
5934 goto out;
a52d9a80 5935 }
e6dcd2dc 5936 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5937 em->start = EXTENT_MAP_HOLE;
445a6944 5938 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5939 em->len = (u64)-1;
c8b97818 5940 em->block_len = (u64)-1;
f421950f
CM		 5941
5942 if (!path) {
5943 path = btrfs_alloc_path();
026fd317
JB		 5944 if (!path) {
5945 err = -ENOMEM;
5946 goto out;
5947 }
5948 /*
5949 * Chances are we'll be called again, so go ahead and do
5950 * readahead
5951 */
5952 path->reada = 1;
f421950f
CM		 5953 }
5954
179e29e4
CM		 5955 ret = btrfs_lookup_file_extent(trans, root, path,
5956 objectid, start, trans != NULL);
a52d9a80
CM		 5957 if (ret < 0) {
5958 err = ret;
5959 goto out;
5960 }
5961
5962 if (ret != 0) {
5963 if (path->slots[0] == 0)
5964 goto not_found;
5965 path->slots[0]--;
5966 }
5967
5f39d397
CM		 5968 leaf = path->nodes[0];
5969 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5970 struct btrfs_file_extent_item);
a52d9a80 5971 /* are we inside the extent that was found? */
5f39d397
CM		 5972 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5973 found_type = btrfs_key_type(&found_key);
5974 if (found_key.objectid != objectid ||
a52d9a80
CM		 5975 found_type != BTRFS_EXTENT_DATA_KEY) {
5976 goto not_found;
5977 }
5978
5f39d397
CM		 5979 found_type = btrfs_file_extent_type(leaf, item);
5980 extent_start = found_key.offset;
261507a0 5981 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5982 if (found_type == BTRFS_FILE_EXTENT_REG ||
5983 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5984 extent_end = extent_start +
db94535d 5985 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5986 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5987 size_t size;
5988 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 5989 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102
YZ		 5990 }
5991
5992 if (start >= extent_end) {
5993 path->slots[0]++;
5994 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5995 ret = btrfs_next_leaf(root, path);
5996 if (ret < 0) {
5997 err = ret;
5998 goto out;
a52d9a80 5999 }
9036c102
YZ		 6000 if (ret > 0)
6001 goto not_found;
6002 leaf = path->nodes[0];
a52d9a80 6003 }
9036c102
YZ		 6004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6005 if (found_key.objectid != objectid ||
6006 found_key.type != BTRFS_EXTENT_DATA_KEY)
6007 goto not_found;
6008 if (start + len <= found_key.offset)
6009 goto not_found;
6010 em->start = start;
70c8a91c 6011 em->orig_start = start;
9036c102
YZ		 6012 em->len = found_key.offset - start;
6013 goto not_found_em;
6014 }
6015
cc95bef6 6016 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6017 if (found_type == BTRFS_FILE_EXTENT_REG ||
6018 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6019 em->start = extent_start;
6020 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6021 em->orig_start = extent_start -
6022 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6023 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6024 item);
db94535d
CM		 6025 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6026 if (bytenr == 0) {
5f39d397 6027 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6028 goto insert;
6029 }
261507a0 6030 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6031 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6032 em->compress_type = compress_type;
c8b97818 6033 em->block_start = bytenr;
b4939680 6034 em->block_len = em->orig_block_len;
c8b97818
CM		 6035 } else {
6036 bytenr += btrfs_file_extent_offset(leaf, item);
6037 em->block_start = bytenr;
6038 em->block_len = em->len;
d899e052
YZ		 6039 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6040 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6041 }
a52d9a80
CM		 6042 goto insert;
6043 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6044 unsigned long ptr;
a52d9a80 6045 char *map;
3326d1b0
CM		 6046 size_t size;
6047 size_t extent_offset;
6048 size_t copy_size;
a52d9a80 6049
689f9346 6050 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6051 if (!page || create) {
689f9346 6052 em->start = extent_start;
9036c102 6053 em->len = extent_end - extent_start;
689f9346
Y		 6054 goto out;
6055 }
5f39d397 6056
9036c102
YZ		 6057 size = btrfs_file_extent_inline_len(leaf, item);
6058 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6059 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6060 size - extent_offset);
3326d1b0 6061 em->start = extent_start + extent_offset;
fda2832f 6062 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6063 em->orig_block_len = em->len;
70c8a91c 6064 em->orig_start = em->start;
261507a0 6065 if (compress_type) {
c8b97818 6066 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6067 em->compress_type = compress_type;
6068 }
689f9346 6069 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6070 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6071 if (btrfs_file_extent_compression(leaf, item) !=
6072 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6073 ret = uncompress_inline(path, inode, page,
6074 pg_offset,
6075 extent_offset, item);
79787eaa 6076 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6077 } else {
6078 map = kmap(page);
6079 read_extent_buffer(leaf, map + pg_offset, ptr,
6080 copy_size);
93c82d57
CM		 6081 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6082 memset(map + pg_offset + copy_size, 0,
6083 PAGE_CACHE_SIZE - pg_offset -
6084 copy_size);
6085 }
c8b97818
CM		 6086 kunmap(page);
6087 }
179e29e4
CM		 6088 flush_dcache_page(page);
6089 } else if (create && PageUptodate(page)) {
6bf7e080 6090 BUG();
179e29e4
CM		 6091 if (!trans) {
6092 kunmap(page);
6093 free_extent_map(em);
6094 em = NULL;
ff5714cc 6095
b3b4aa74 6096 btrfs_release_path(path);
7a7eaa40 6097 trans = btrfs_join_transaction(root);
ff5714cc 6098
3612b495
TI		 6099 if (IS_ERR(trans))
6100 return ERR_CAST(trans);
179e29e4
CM		 6101 goto again;
6102 }
c8b97818 6103 map = kmap(page);
70dec807 6104 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6105 copy_size);
c8b97818 6106 kunmap(page);
179e29e4 6107 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6108 }
d1310b2e 6109 set_extent_uptodate(io_tree, em->start,
507903b8 6110 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6111 goto insert;
6112 } else {
31b1a2bd 6113 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6114 }
6115not_found:
6116 em->start = start;
70c8a91c 6117 em->orig_start = start;
d1310b2e 6118 em->len = len;
a52d9a80 6119not_found_em:
5f39d397 6120 em->block_start = EXTENT_MAP_HOLE;
9036c102 6121 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6122insert:
b3b4aa74 6123 btrfs_release_path(path);
d1310b2e 6124 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6125 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6126 em->start, em->len, start, len);
a52d9a80
CM		 6127 err = -EIO;
6128 goto out;
6129 }
d1310b2e
CM		 6130
6131 err = 0;
890871be 6132 write_lock(&em_tree->lock);
09a2a8f9 6133 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6134 /* it is possible that someone inserted the extent into the tree
6135 * while we had the lock dropped. It is also possible that
6136 * an overlapping map exists in the tree
6137 */
a52d9a80 6138 if (ret == -EEXIST) {
3b951516 6139 struct extent_map *existing;
e6dcd2dc
CM		 6140
6141 ret = 0;
6142
3b951516 6143 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6144 if (existing && (existing->start > start ||
6145 existing->start + existing->len <= start)) {
6146 free_extent_map(existing);
6147 existing = NULL;
6148 }
3b951516
CM		 6149 if (!existing) {
6150 existing = lookup_extent_mapping(em_tree, em->start,
6151 em->len);
6152 if (existing) {
6153 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6154 em, start,
6155 root->sectorsize);
3b951516
CM		 6156 free_extent_map(existing);
6157 if (err) {
6158 free_extent_map(em);
6159 em = NULL;
6160 }
6161 } else {
6162 err = -EIO;
3b951516
CM		 6163 free_extent_map(em);
6164 em = NULL;
6165 }
6166 } else {
6167 free_extent_map(em);
6168 em = existing;
e6dcd2dc 6169 err = 0;
a52d9a80 6170 }
a52d9a80 6171 }
890871be 6172 write_unlock(&em_tree->lock);
a52d9a80 6173out:
1abe9b8a 6174
f0bd95ea
TI		 6175 if (em)
6176 trace_btrfs_get_extent(root, em);
1abe9b8a 6177
f421950f
CM		 6178 if (path)
6179 btrfs_free_path(path);
a52d9a80
CM		 6180 if (trans) {
6181 ret = btrfs_end_transaction(trans, root);
d397712b 6182 if (!err)
a52d9a80
CM		 6183 err = ret;
6184 }
a52d9a80
CM		 6185 if (err) {
6186 free_extent_map(em);
a52d9a80
CM		 6187 return ERR_PTR(err);
6188 }
79787eaa 6189 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6190 return em;
6191}
6192
ec29ed5b
CM		 6193struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6194 size_t pg_offset, u64 start, u64 len,
6195 int create)
6196{
6197 struct extent_map *em;
6198 struct extent_map *hole_em = NULL;
6199 u64 range_start = start;
6200 u64 end;
6201 u64 found;
6202 u64 found_end;
6203 int err = 0;
6204
6205 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6206 if (IS_ERR(em))
6207 return em;
6208 if (em) {
6209 /*
f9e4fb53
LB		 6210 * if our em maps to
6211 * - a hole or
6212 * - a pre-alloc extent,
6213 * there might actually be delalloc bytes behind it.
ec29ed5b 6214 */
f9e4fb53
LB		 6215 if (em->block_start != EXTENT_MAP_HOLE &&
6216 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6217 return em;
6218 else
6219 hole_em = em;
6220 }
6221
6222 /* check to see if we've wrapped (len == -1 or similar) */
6223 end = start + len;
6224 if (end < start)
6225 end = (u64)-1;
6226 else
6227 end -= 1;
6228
6229 em = NULL;
6230
6231 /* ok, we didn't find anything, lets look for delalloc */
6232 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6233 end, len, EXTENT_DELALLOC, 1);
6234 found_end = range_start + found;
6235 if (found_end < range_start)
6236 found_end = (u64)-1;
6237
6238 /*
6239 * we didn't find anything useful, return
6240 * the original results from get_extent()
6241 */
6242 if (range_start > end || found_end <= start) {
6243 em = hole_em;
6244 hole_em = NULL;
6245 goto out;
6246 }
6247
6248 /* adjust the range_start to make sure it doesn't
6249 * go backwards from the start they passed in
6250 */
6251 range_start = max(start,range_start);
6252 found = found_end - range_start;
6253
6254 if (found > 0) {
6255 u64 hole_start = start;
6256 u64 hole_len = len;
6257
172ddd60 6258 em = alloc_extent_map();
ec29ed5b
CM		 6259 if (!em) {
6260 err = -ENOMEM;
6261 goto out;
6262 }
6263 /*
6264 * when btrfs_get_extent can't find anything it
6265 * returns one huge hole
6266 *
6267 * make sure what it found really fits our range, and
6268 * adjust to make sure it is based on the start from
6269 * the caller
6270 */
6271 if (hole_em) {
6272 u64 calc_end = extent_map_end(hole_em);
6273
6274 if (calc_end <= start || (hole_em->start > end)) {
6275 free_extent_map(hole_em);
6276 hole_em = NULL;
6277 } else {
6278 hole_start = max(hole_em->start, start);
6279 hole_len = calc_end - hole_start;
6280 }
6281 }
6282 em->bdev = NULL;
6283 if (hole_em && range_start > hole_start) {
6284 /* our hole starts before our delalloc, so we
6285 * have to return just the parts of the hole
6286 * that go until the delalloc starts
6287 */
6288 em->len = min(hole_len,
6289 range_start - hole_start);
6290 em->start = hole_start;
6291 em->orig_start = hole_start;
6292 /*
6293 * don't adjust block start at all,
6294 * it is fixed at EXTENT_MAP_HOLE
6295 */
6296 em->block_start = hole_em->block_start;
6297 em->block_len = hole_len;
f9e4fb53
LB		 6298 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6299 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6300 } else {
6301 em->start = range_start;
6302 em->len = found;
6303 em->orig_start = range_start;
6304 em->block_start = EXTENT_MAP_DELALLOC;
6305 em->block_len = found;
6306 }
6307 } else if (hole_em) {
6308 return hole_em;
6309 }
6310out:
6311
6312 free_extent_map(hole_em);
6313 if (err) {
6314 free_extent_map(em);
6315 return ERR_PTR(err);
6316 }
6317 return em;
6318}
6319
4b46fce2
JB		 6320static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6321 u64 start, u64 len)
6322{
6323 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 6324 struct extent_map *em;
4b46fce2
JB		 6325 struct btrfs_key ins;
6326 u64 alloc_hint;
6327 int ret;
4b46fce2 6328
4b46fce2 6329 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 6330 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
81c9ad23 6331 alloc_hint, &ins, 1);
00361589
JB		 6332 if (ret)
6333 return ERR_PTR(ret);
4b46fce2 6334
70c8a91c 6335 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6336 ins.offset, ins.offset, ins.offset, 0);
00361589
JB		 6337 if (IS_ERR(em)) {
6338 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6339 return em;
6340 }
4b46fce2
JB		 6341
6342 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6343 ins.offset, ins.offset, 0);
6344 if (ret) {
6345 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
00361589
JB		 6346 free_extent_map(em);
6347 return ERR_PTR(ret);
4b46fce2 6348 }
00361589 6349
4b46fce2
JB		 6350 return em;
6351}
6352
46bfbb5c
CM		 6353/*
6354 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6355 * block must be cow'd
6356 */
00361589 6357noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 6358 u64 *orig_start, u64 *orig_block_len,
6359 u64 *ram_bytes)
46bfbb5c 6360{
00361589 6361 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 6362 struct btrfs_path *path;
6363 int ret;
6364 struct extent_buffer *leaf;
6365 struct btrfs_root *root = BTRFS_I(inode)->root;
6366 struct btrfs_file_extent_item *fi;
6367 struct btrfs_key key;
6368 u64 disk_bytenr;
6369 u64 backref_offset;
6370 u64 extent_end;
6371 u64 num_bytes;
6372 int slot;
6373 int found_type;
7ee9e440 6374 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
46bfbb5c
CM		 6375 path = btrfs_alloc_path();
6376 if (!path)
6377 return -ENOMEM;
6378
00361589 6379 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6380 offset, 0);
6381 if (ret < 0)
6382 goto out;
6383
6384 slot = path->slots[0];
6385 if (ret == 1) {
6386 if (slot == 0) {
6387 /* can't find the item, must cow */
6388 ret = 0;
6389 goto out;
6390 }
6391 slot--;
6392 }
6393 ret = 0;
6394 leaf = path->nodes[0];
6395 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6396 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6397 key.type != BTRFS_EXTENT_DATA_KEY) {
6398 /* not our file or wrong item type, must cow */
6399 goto out;
6400 }
6401
6402 if (key.offset > offset) {
6403 /* Wrong offset, must cow */
6404 goto out;
6405 }
6406
6407 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6408 found_type = btrfs_file_extent_type(leaf, fi);
6409 if (found_type != BTRFS_FILE_EXTENT_REG &&
6410 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6411 /* not a regular extent, must cow */
6412 goto out;
6413 }
7ee9e440
JB		 6414
6415 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6416 goto out;
6417
46bfbb5c 6418 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 6419 if (disk_bytenr == 0)
6420 goto out;
6421
6422 if (btrfs_file_extent_compression(leaf, fi) ||
6423 btrfs_file_extent_encryption(leaf, fi) ||
6424 btrfs_file_extent_other_encoding(leaf, fi))
6425 goto out;
6426
46bfbb5c
CM		 6427 backref_offset = btrfs_file_extent_offset(leaf, fi);
6428
7ee9e440
JB		 6429 if (orig_start) {
6430 *orig_start = key.offset - backref_offset;
6431 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6432 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6433 }
eb384b55 6434
46bfbb5c 6435 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
46bfbb5c
CM		 6436
6437 if (btrfs_extent_readonly(root, disk_bytenr))
6438 goto out;
1bda19eb 6439 btrfs_release_path(path);
46bfbb5c
CM		 6440
6441 /*
6442 * look for other files referencing this extent, if we
6443 * find any we must cow
6444 */
00361589
JB		 6445 trans = btrfs_join_transaction(root);
6446 if (IS_ERR(trans)) {
6447 ret = 0;
46bfbb5c 6448 goto out;
00361589
JB		 6449 }
6450
6451 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6452 key.offset - backref_offset, disk_bytenr);
6453 btrfs_end_transaction(trans, root);
6454 if (ret) {
6455 ret = 0;
6456 goto out;
6457 }
46bfbb5c
CM		 6458
6459 /*
6460 * adjust disk_bytenr and num_bytes to cover just the bytes
6461 * in this extent we are about to write. If there
6462 * are any csums in that range we have to cow in order
6463 * to keep the csums correct
6464 */
6465 disk_bytenr += backref_offset;
6466 disk_bytenr += offset - key.offset;
eb384b55 6467 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6468 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6469 goto out;
6470 /*
6471 * all of the above have passed, it is safe to overwrite this extent
6472 * without cow
6473 */
eb384b55 6474 *len = num_bytes;
46bfbb5c
CM		 6475 ret = 1;
6476out:
6477 btrfs_free_path(path);
6478 return ret;
6479}
6480
eb838e73
JB		 6481static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6482 struct extent_state **cached_state, int writing)
6483{
6484 struct btrfs_ordered_extent *ordered;
6485 int ret = 0;
6486
6487 while (1) {
6488 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6489 0, cached_state);
6490 /*
6491 * We're concerned with the entire range that we're going to be
6492 * doing DIO to, so we need to make sure theres no ordered
6493 * extents in this range.
6494 */
6495 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6496 lockend - lockstart + 1);
6497
6498 /*
6499 * We need to make sure there are no buffered pages in this
6500 * range either, we could have raced between the invalidate in
6501 * generic_file_direct_write and locking the extent. The
6502 * invalidate needs to happen so that reads after a write do not
6503 * get stale data.
6504 */
6505 if (!ordered && (!writing ||
6506 !test_range_bit(&BTRFS_I(inode)->io_tree,
6507 lockstart, lockend, EXTENT_UPTODATE, 0,
6508 *cached_state)))
6509 break;
6510
6511 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6512 cached_state, GFP_NOFS);
6513
6514 if (ordered) {
6515 btrfs_start_ordered_extent(inode, ordered, 1);
6516 btrfs_put_ordered_extent(ordered);
6517 } else {
6518 /* Screw you mmap */
6519 ret = filemap_write_and_wait_range(inode->i_mapping,
6520 lockstart,
6521 lockend);
6522 if (ret)
6523 break;
6524
6525 /*
6526 * If we found a page that couldn't be invalidated just
6527 * fall back to buffered.
6528 */
6529 ret = invalidate_inode_pages2_range(inode->i_mapping,
6530 lockstart >> PAGE_CACHE_SHIFT,
6531 lockend >> PAGE_CACHE_SHIFT);
6532 if (ret)
6533 break;
6534 }
6535
6536 cond_resched();
6537 }
6538
6539 return ret;
6540}
6541
69ffb543
JB		 6542static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6543 u64 len, u64 orig_start,
6544 u64 block_start, u64 block_len,
cc95bef6
JB		 6545 u64 orig_block_len, u64 ram_bytes,
6546 int type)
69ffb543
JB		 6547{
6548 struct extent_map_tree *em_tree;
6549 struct extent_map *em;
6550 struct btrfs_root *root = BTRFS_I(inode)->root;
6551 int ret;
6552
6553 em_tree = &BTRFS_I(inode)->extent_tree;
6554 em = alloc_extent_map();
6555 if (!em)
6556 return ERR_PTR(-ENOMEM);
6557
6558 em->start = start;
6559 em->orig_start = orig_start;
2ab28f32
JB		 6560 em->mod_start = start;
6561 em->mod_len = len;
69ffb543
JB		 6562 em->len = len;
6563 em->block_len = block_len;
6564 em->block_start = block_start;
6565 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6566 em->orig_block_len = orig_block_len;
cc95bef6 6567 em->ram_bytes = ram_bytes;
70c8a91c 6568 em->generation = -1;
69ffb543
JB		 6569 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6570 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6571 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6572
6573 do {
6574 btrfs_drop_extent_cache(inode, em->start,
6575 em->start + em->len - 1, 0);
6576 write_lock(&em_tree->lock);
09a2a8f9 6577 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6578 write_unlock(&em_tree->lock);
6579 } while (ret == -EEXIST);
6580
6581 if (ret) {
6582 free_extent_map(em);
6583 return ERR_PTR(ret);
6584 }
6585
6586 return em;
6587}
6588
6589
4b46fce2
JB		 6590static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6591 struct buffer_head *bh_result, int create)
6592{
6593 struct extent_map *em;
6594 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6595 struct extent_state *cached_state = NULL;
4b46fce2 6596 u64 start = iblock << inode->i_blkbits;
eb838e73 6597 u64 lockstart, lockend;
4b46fce2 6598 u64 len = bh_result->b_size;
eb838e73 6599 int unlock_bits = EXTENT_LOCKED;
0934856d 6600 int ret = 0;
eb838e73 6601
172a5049 6602 if (create)
eb838e73 6603 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6604 else
c329861d 6605 len = min_t(u64, len, root->sectorsize);
eb838e73 6606
c329861d
JB		 6607 lockstart = start;
6608 lockend = start + len - 1;
6609
eb838e73
JB		 6610 /*
6611 * If this errors out it's because we couldn't invalidate pagecache for
6612 * this range and we need to fallback to buffered.
6613 */
6614 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6615 return -ENOTBLK;
6616
4b46fce2 6617 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6618 if (IS_ERR(em)) {
6619 ret = PTR_ERR(em);
6620 goto unlock_err;
6621 }
4b46fce2
JB		 6622
6623 /*
6624 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6625 * io. INLINE is special, and we could probably kludge it in here, but
6626 * it's still buffered so for safety lets just fall back to the generic
6627 * buffered path.
6628 *
6629 * For COMPRESSED we _have_ to read the entire extent in so we can
6630 * decompress it, so there will be buffering required no matter what we
6631 * do, so go ahead and fallback to buffered.
6632 *
6633 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6634 * to buffered IO. Don't blame me, this is the price we pay for using
6635 * the generic code.
6636 */
6637 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6638 em->block_start == EXTENT_MAP_INLINE) {
6639 free_extent_map(em);
eb838e73
JB		 6640 ret = -ENOTBLK;
6641 goto unlock_err;
4b46fce2
JB		 6642 }
6643
6644 /* Just a good old fashioned hole, return */
6645 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6646 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6647 free_extent_map(em);
eb838e73 6648 goto unlock_err;
4b46fce2
JB		 6649 }
6650
6651 /*
6652 * We don't allocate a new extent in the following cases
6653 *
6654 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6655 * existing extent.
6656 * 2) The extent is marked as PREALLOC. We're good to go here and can
6657 * just use the extent.
6658 *
6659 */
46bfbb5c 6660 if (!create) {
eb838e73
JB		 6661 len = min(len, em->len - (start - em->start));
6662 lockstart = start + len;
6663 goto unlock;
46bfbb5c 6664 }
4b46fce2
JB		 6665
6666 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6667 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6668 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6669 int type;
6670 int ret;
eb384b55 6671 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6672
6673 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6674 type = BTRFS_ORDERED_PREALLOC;
6675 else
6676 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6677 len = min(len, em->len - (start - em->start));
4b46fce2 6678 block_start = em->block_start + (start - em->start);
46bfbb5c 6679
00361589 6680 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 6681 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6682 if (type == BTRFS_ORDERED_PREALLOC) {
6683 free_extent_map(em);
6684 em = create_pinned_em(inode, start, len,
6685 orig_start,
b4939680 6686 block_start, len,
cc95bef6
JB		 6687 orig_block_len,
6688 ram_bytes, type);
00361589 6689 if (IS_ERR(em))
69ffb543 6690 goto unlock_err;
69ffb543
JB		 6691 }
6692
46bfbb5c
CM		 6693 ret = btrfs_add_ordered_extent_dio(inode, start,
6694 block_start, len, len, type);
46bfbb5c
CM		 6695 if (ret) {
6696 free_extent_map(em);
eb838e73 6697 goto unlock_err;
46bfbb5c
CM		 6698 }
6699 goto unlock;
4b46fce2 6700 }
4b46fce2 6701 }
00361589 6702
46bfbb5c
CM		 6703 /*
6704 * this will cow the extent, reset the len in case we changed
6705 * it above
6706 */
6707 len = bh_result->b_size;
70c8a91c
JB		 6708 free_extent_map(em);
6709 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6710 if (IS_ERR(em)) {
6711 ret = PTR_ERR(em);
6712 goto unlock_err;
6713 }
46bfbb5c
CM		 6714 len = min(len, em->len - (start - em->start));
6715unlock:
4b46fce2
JB		 6716 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6717 inode->i_blkbits;
46bfbb5c 6718 bh_result->b_size = len;
4b46fce2
JB		 6719 bh_result->b_bdev = em->bdev;
6720 set_buffer_mapped(bh_result);
c3473e83
JB		 6721 if (create) {
6722 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6723 set_buffer_new(bh_result);
6724
6725 /*
6726 * Need to update the i_size under the extent lock so buffered
6727 * readers will get the updated i_size when we unlock.
6728 */
6729 if (start + len > i_size_read(inode))
6730 i_size_write(inode, start + len);
0934856d 6731
172a5049
MX		 6732 spin_lock(&BTRFS_I(inode)->lock);
6733 BTRFS_I(inode)->outstanding_extents++;
6734 spin_unlock(&BTRFS_I(inode)->lock);
6735
0934856d
MX		 6736 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6737 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6738 &cached_state, GFP_NOFS);
6739 BUG_ON(ret);
c3473e83 6740 }
4b46fce2 6741
eb838e73
JB		 6742 /*
6743 * In the case of write we need to clear and unlock the entire range,
6744 * in the case of read we need to unlock only the end area that we
6745 * aren't using if there is any left over space.
6746 */
24c03fa5 6747 if (lockstart < lockend) {
0934856d
MX		 6748 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6749 lockend, unlock_bits, 1, 0,
6750 &cached_state, GFP_NOFS);
24c03fa5 6751 } else {
eb838e73 6752 free_extent_state(cached_state);
24c03fa5 6753 }
eb838e73 6754
4b46fce2
JB		 6755 free_extent_map(em);
6756
6757 return 0;
eb838e73
JB		 6758
6759unlock_err:
eb838e73
JB		 6760 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6761 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6762 return ret;
4b46fce2
JB		 6763}
6764
4b46fce2
JB		 6765static void btrfs_endio_direct_read(struct bio *bio, int err)
6766{
e65e1535 6767 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6768 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6769 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 6770 struct inode *inode = dip->inode;
6771 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 6772 struct bio *dio_bio;
facc8a22
MX		 6773 u32 *csums = (u32 *)dip->csum;
6774 int index = 0;
4b46fce2 6775 u64 start;
4b46fce2
JB		 6776
6777 start = dip->logical_offset;
6778 do {
6779 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6780 struct page *page = bvec->bv_page;
6781 char *kaddr;
6782 u32 csum = ~(u32)0;
6783 unsigned long flags;
6784
6785 local_irq_save(flags);
7ac687d9 6786 kaddr = kmap_atomic(page);
b0496686 6787 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6788 csum, bvec->bv_len);
6789 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6790 kunmap_atomic(kaddr);
4b46fce2
JB		 6791 local_irq_restore(flags);
6792
6793 flush_dcache_page(bvec->bv_page);
facc8a22
MX		 6794 if (csum != csums[index]) {
6795 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c
GU		 6796 btrfs_ino(inode), start, csum,
6797 csums[index]);
4b46fce2
JB		 6798 err = -EIO;
6799 }
6800 }
6801
6802 start += bvec->bv_len;
4b46fce2 6803 bvec++;
facc8a22 6804 index++;
4b46fce2
JB		 6805 } while (bvec <= bvec_end);
6806
6807 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6808 dip->logical_offset + dip->bytes - 1);
9be3395b 6809 dio_bio = dip->dio_bio;
4b46fce2 6810
4b46fce2 6811 kfree(dip);
c0da7aa1
JB		 6812
6813 /* If we had a csum failure make sure to clear the uptodate flag */
6814 if (err)
9be3395b
CM		 6815 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6816 dio_end_io(dio_bio, err);
6817 bio_put(bio);
4b46fce2
JB		 6818}
6819
6820static void btrfs_endio_direct_write(struct bio *bio, int err)
6821{
6822 struct btrfs_dio_private *dip = bio->bi_private;
6823 struct inode *inode = dip->inode;
6824 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6825 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 6826 u64 ordered_offset = dip->logical_offset;
6827 u64 ordered_bytes = dip->bytes;
9be3395b 6828 struct bio *dio_bio;
4b46fce2
JB		 6829 int ret;
6830
6831 if (err)
6832 goto out_done;
163cf09c
CM		 6833again:
6834 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6835 &ordered_offset,
5fd02043 6836 ordered_bytes, !err);
4b46fce2 6837 if (!ret)
163cf09c 6838 goto out_test;
4b46fce2 6839
5fd02043
JB		 6840 ordered->work.func = finish_ordered_fn;
6841 ordered->work.flags = 0;
6842 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6843 &ordered->work);
163cf09c
CM		 6844out_test:
6845 /*
6846 * our bio might span multiple ordered extents. If we haven't
6847 * completed the accounting for the whole dio, go back and try again
6848 */
6849 if (ordered_offset < dip->logical_offset + dip->bytes) {
6850 ordered_bytes = dip->logical_offset + dip->bytes -
6851 ordered_offset;
5fd02043 6852 ordered = NULL;
163cf09c
CM		 6853 goto again;
6854 }
4b46fce2 6855out_done:
9be3395b 6856 dio_bio = dip->dio_bio;
4b46fce2 6857
4b46fce2 6858 kfree(dip);
c0da7aa1
JB		 6859
6860 /* If we had an error make sure to clear the uptodate flag */
6861 if (err)
9be3395b
CM		 6862 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6863 dio_end_io(dio_bio, err);
6864 bio_put(bio);
4b46fce2
JB		 6865}
6866
eaf25d93
CM		 6867static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6868 struct bio *bio, int mirror_num,
6869 unsigned long bio_flags, u64 offset)
6870{
6871 int ret;
6872 struct btrfs_root *root = BTRFS_I(inode)->root;
6873 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6874 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6875 return 0;
6876}
6877
e65e1535
MX		 6878static void btrfs_end_dio_bio(struct bio *bio, int err)
6879{
6880 struct btrfs_dio_private *dip = bio->bi_private;
6881
6882 if (err) {
33345d01 6883 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6884 "sector %#Lx len %u err no %d\n",
c1c9ff7c 6885 btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6886 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6887 dip->errors = 1;
6888
6889 /*
6890 * before atomic variable goto zero, we must make sure
6891 * dip->errors is perceived to be set.
6892 */
6893 smp_mb__before_atomic_dec();
6894 }
6895
6896 /* if there are more bios still pending for this dio, just exit */
6897 if (!atomic_dec_and_test(&dip->pending_bios))
6898 goto out;
6899
9be3395b 6900 if (dip->errors) {
e65e1535 6901 bio_io_error(dip->orig_bio);
9be3395b
CM		 6902 } else {
6903 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 6904 bio_endio(dip->orig_bio, 0);
6905 }
6906out:
6907 bio_put(bio);
6908}
6909
6910static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6911 u64 first_sector, gfp_t gfp_flags)
6912{
6913 int nr_vecs = bio_get_nr_vecs(bdev);
6914 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6915}
6916
6917static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6918 int rw, u64 file_offset, int skip_sum,
c329861d 6919 int async_submit)
e65e1535 6920{
facc8a22 6921 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 6922 int write = rw & REQ_WRITE;
6923 struct btrfs_root *root = BTRFS_I(inode)->root;
6924 int ret;
6925
b812ce28
JB		 6926 if (async_submit)
6927 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6928
e65e1535 6929 bio_get(bio);
5fd02043
JB		 6930
6931 if (!write) {
6932 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6933 if (ret)
6934 goto err;
6935 }
e65e1535 6936
1ae39938
JB		 6937 if (skip_sum)
6938 goto map;
6939
6940 if (write && async_submit) {
e65e1535
MX		 6941 ret = btrfs_wq_submit_bio(root->fs_info,
6942 inode, rw, bio, 0, 0,
6943 file_offset,
6944 __btrfs_submit_bio_start_direct_io,
6945 __btrfs_submit_bio_done);
6946 goto err;
1ae39938
JB		 6947 } else if (write) {
6948 /*
6949 * If we aren't doing async submit, calculate the csum of the
6950 * bio now.
6951 */
6952 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6953 if (ret)
6954 goto err;
c2db1073 6955 } else if (!skip_sum) {
facc8a22
MX		 6956 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
6957 file_offset);
c2db1073
TI		 6958 if (ret)
6959 goto err;
6960 }
e65e1535 6961
1ae39938
JB		 6962map:
6963 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 6964err:
6965 bio_put(bio);
6966 return ret;
6967}
6968
6969static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6970 int skip_sum)
6971{
6972 struct inode *inode = dip->inode;
6973 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 6974 struct bio *bio;
6975 struct bio *orig_bio = dip->orig_bio;
6976 struct bio_vec *bvec = orig_bio->bi_io_vec;
6977 u64 start_sector = orig_bio->bi_sector;
6978 u64 file_offset = dip->logical_offset;
6979 u64 submit_len = 0;
6980 u64 map_length;
6981 int nr_pages = 0;
e65e1535 6982 int ret = 0;
1ae39938 6983 int async_submit = 0;
e65e1535 6984
e65e1535 6985 map_length = orig_bio->bi_size;
53b381b3 6986 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 6987 &map_length, NULL, 0);
6988 if (ret) {
64728bbb 6989 bio_put(orig_bio);
e65e1535
MX		 6990 return -EIO;
6991 }
facc8a22 6992
02f57c7a
JB		 6993 if (map_length >= orig_bio->bi_size) {
6994 bio = orig_bio;
6995 goto submit;
6996 }
6997
53b381b3
DW		 6998 /* async crcs make it difficult to collect full stripe writes. */
6999 if (btrfs_get_alloc_profile(root, 1) &
7000 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7001 async_submit = 0;
7002 else
7003 async_submit = 1;
7004
02f57c7a
JB		 7005 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7006 if (!bio)
7007 return -ENOMEM;
7008 bio->bi_private = dip;
7009 bio->bi_end_io = btrfs_end_dio_bio;
7010 atomic_inc(&dip->pending_bios);
7011
e65e1535
MX		 7012 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7013 if (unlikely(map_length < submit_len + bvec->bv_len ||
7014 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7015 bvec->bv_offset) < bvec->bv_len)) {
7016 /*
7017 * inc the count before we submit the bio so
7018 * we know the end IO handler won't happen before
7019 * we inc the count. Otherwise, the dip might get freed
7020 * before we're done setting it up
7021 */
7022 atomic_inc(&dip->pending_bios);
7023 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7024 file_offset, skip_sum,
c329861d 7025 async_submit);
e65e1535
MX		 7026 if (ret) {
7027 bio_put(bio);
7028 atomic_dec(&dip->pending_bios);
7029 goto out_err;
7030 }
7031
e65e1535
MX		 7032 start_sector += submit_len >> 9;
7033 file_offset += submit_len;
7034
7035 submit_len = 0;
7036 nr_pages = 0;
7037
7038 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7039 start_sector, GFP_NOFS);
7040 if (!bio)
7041 goto out_err;
7042 bio->bi_private = dip;
7043 bio->bi_end_io = btrfs_end_dio_bio;
7044
7045 map_length = orig_bio->bi_size;
53b381b3 7046 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7047 start_sector << 9,
e65e1535
MX		 7048 &map_length, NULL, 0);
7049 if (ret) {
7050 bio_put(bio);
7051 goto out_err;
7052 }
7053 } else {
7054 submit_len += bvec->bv_len;
7055 nr_pages ++;
7056 bvec++;
7057 }
7058 }
7059
02f57c7a 7060submit:
e65e1535 7061 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7062 async_submit);
e65e1535
MX		 7063 if (!ret)
7064 return 0;
7065
7066 bio_put(bio);
7067out_err:
7068 dip->errors = 1;
7069 /*
7070 * before atomic variable goto zero, we must
7071 * make sure dip->errors is perceived to be set.
7072 */
7073 smp_mb__before_atomic_dec();
7074 if (atomic_dec_and_test(&dip->pending_bios))
7075 bio_io_error(dip->orig_bio);
7076
7077 /* bio_end_io() will handle error, so we needn't return it */
7078 return 0;
7079}
7080
9be3395b
CM		 7081static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7082 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7083{
7084 struct btrfs_root *root = BTRFS_I(inode)->root;
7085 struct btrfs_dio_private *dip;
9be3395b 7086 struct bio *io_bio;
4b46fce2 7087 int skip_sum;
facc8a22 7088 int sum_len;
7b6d91da 7089 int write = rw & REQ_WRITE;
4b46fce2 7090 int ret = 0;
facc8a22 7091 u16 csum_size;
4b46fce2
JB		 7092
7093 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7094
9be3395b 7095 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 7096 if (!io_bio) {
7097 ret = -ENOMEM;
7098 goto free_ordered;
7099 }
7100
facc8a22
MX		 7101 if (!skip_sum && !write) {
7102 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7103 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7104 sum_len *= csum_size;
7105 } else {
7106 sum_len = 0;
7107 }
7108
7109 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
4b46fce2
JB		 7110 if (!dip) {
7111 ret = -ENOMEM;
9be3395b 7112 goto free_io_bio;
4b46fce2 7113 }
4b46fce2 7114
9be3395b 7115 dip->private = dio_bio->bi_private;
4b46fce2
JB		 7116 dip->inode = inode;
7117 dip->logical_offset = file_offset;
e6da5d2e 7118 dip->bytes = dio_bio->bi_size;
9be3395b
CM		 7119 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7120 io_bio->bi_private = dip;
e65e1535 7121 dip->errors = 0;
9be3395b
CM		 7122 dip->orig_bio = io_bio;
7123 dip->dio_bio = dio_bio;
e65e1535 7124 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7125
7126 if (write)
9be3395b 7127 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7128 else
9be3395b 7129 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7130
e65e1535
MX		 7131 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7132 if (!ret)
eaf25d93 7133 return;
9be3395b
CM		 7134
7135free_io_bio:
7136 bio_put(io_bio);
7137
4b46fce2
JB		 7138free_ordered:
7139 /*
7140 * If this is a write, we need to clean up the reserved space and kill
7141 * the ordered extent.
7142 */
7143 if (write) {
7144 struct btrfs_ordered_extent *ordered;
955256f2 7145 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7146 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7147 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7148 btrfs_free_reserved_extent(root, ordered->start,
7149 ordered->disk_len);
7150 btrfs_put_ordered_extent(ordered);
7151 btrfs_put_ordered_extent(ordered);
7152 }
9be3395b 7153 bio_endio(dio_bio, ret);
4b46fce2
JB		 7154}
7155
5a5f79b5
CM		 7156static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7157 const struct iovec *iov, loff_t offset,
7158 unsigned long nr_segs)
7159{
7160 int seg;
a1b75f7d 7161 int i;
5a5f79b5
CM		 7162 size_t size;
7163 unsigned long addr;
7164 unsigned blocksize_mask = root->sectorsize - 1;
7165 ssize_t retval = -EINVAL;
7166 loff_t end = offset;
7167
7168 if (offset & blocksize_mask)
7169 goto out;
7170
7171 /* Check the memory alignment. Blocks cannot straddle pages */
7172 for (seg = 0; seg < nr_segs; seg++) {
7173 addr = (unsigned long)iov[seg].iov_base;
7174 size = iov[seg].iov_len;
7175 end += size;
a1b75f7d 7176 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7177 goto out;
a1b75f7d
JB		 7178
7179 /* If this is a write we don't need to check anymore */
7180 if (rw & WRITE)
7181 continue;
7182
7183 /*
7184 * Check to make sure we don't have duplicate iov_base's in this
7185 * iovec, if so return EINVAL, otherwise we'll get csum errors
7186 * when reading back.
7187 */
7188 for (i = seg + 1; i < nr_segs; i++) {
7189 if (iov[seg].iov_base == iov[i].iov_base)
7190 goto out;
7191 }
5a5f79b5
CM		 7192 }
7193 retval = 0;
7194out:
7195 return retval;
7196}
eb838e73 7197
16432985
CM		 7198static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7199 const struct iovec *iov, loff_t offset,
7200 unsigned long nr_segs)
7201{
4b46fce2
JB		 7202 struct file *file = iocb->ki_filp;
7203 struct inode *inode = file->f_mapping->host;
0934856d 7204 size_t count = 0;
2e60a51e 7205 int flags = 0;
38851cc1
MX		 7206 bool wakeup = true;
7207 bool relock = false;
0934856d 7208 ssize_t ret;
4b46fce2 7209
5a5f79b5 7210 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7211 offset, nr_segs))
5a5f79b5 7212 return 0;
3f7c579c 7213
38851cc1
MX		 7214 atomic_inc(&inode->i_dio_count);
7215 smp_mb__after_atomic_inc();
7216
0e267c44
JB		 7217 /*
7218 * The generic stuff only does filemap_write_and_wait_range, which isn't
7219 * enough if we've written compressed pages to this area, so we need to
7220 * call btrfs_wait_ordered_range to make absolutely sure that any
7221 * outstanding dirty pages are on disk.
7222 */
7223 count = iov_length(iov, nr_segs);
7224 btrfs_wait_ordered_range(inode, offset, count);
7225
0934856d 7226 if (rw & WRITE) {
38851cc1
MX		 7227 /*
7228 * If the write DIO is beyond the EOF, we need update
7229 * the isize, but it is protected by i_mutex. So we can
7230 * not unlock the i_mutex at this case.
7231 */
7232 if (offset + count <= inode->i_size) {
7233 mutex_unlock(&inode->i_mutex);
7234 relock = true;
7235 }
0934856d
MX		 7236 ret = btrfs_delalloc_reserve_space(inode, count);
7237 if (ret)
38851cc1
MX		 7238 goto out;
7239 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7240 &BTRFS_I(inode)->runtime_flags))) {
7241 inode_dio_done(inode);
7242 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7243 wakeup = false;
0934856d
MX		 7244 }
7245
7246 ret = __blockdev_direct_IO(rw, iocb, inode,
7247 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7248 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7249 btrfs_submit_direct, flags);
0934856d
MX		 7250 if (rw & WRITE) {
7251 if (ret < 0 && ret != -EIOCBQUEUED)
7252 btrfs_delalloc_release_space(inode, count);
172a5049 7253 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7254 btrfs_delalloc_release_space(inode,
7255 count - (size_t)ret);
172a5049
MX		 7256 else
7257 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7258 }
38851cc1 7259out:
2e60a51e
MX		 7260 if (wakeup)
7261 inode_dio_done(inode);
38851cc1
MX		 7262 if (relock)
7263 mutex_lock(&inode->i_mutex);
0934856d
MX		 7264
7265 return ret;
16432985
CM		 7266}
7267
05dadc09
TI		 7268#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7269
1506fcc8
YS		 7270static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7271 __u64 start, __u64 len)
7272{
05dadc09
TI		 7273 int ret;
7274
7275 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7276 if (ret)
7277 return ret;
7278
ec29ed5b 7279 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7280}
7281
a52d9a80 7282int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7283{
d1310b2e
CM		 7284 struct extent_io_tree *tree;
7285 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7286 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7287}
1832a6d5 7288
a52d9a80 7289static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7290{
d1310b2e 7291 struct extent_io_tree *tree;
b888db2b
CM		 7292
7293
7294 if (current->flags & PF_MEMALLOC) {
7295 redirty_page_for_writepage(wbc, page);
7296 unlock_page(page);
7297 return 0;
7298 }
d1310b2e 7299 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7300 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7301}
7302
48a3b636
ES		 7303static int btrfs_writepages(struct address_space *mapping,
7304 struct writeback_control *wbc)
b293f02e 7305{
d1310b2e 7306 struct extent_io_tree *tree;
771ed689 7307
d1310b2e 7308 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7309 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7310}
7311
3ab2fb5a
CM		 7312static int
7313btrfs_readpages(struct file *file, struct address_space *mapping,
7314 struct list_head *pages, unsigned nr_pages)
7315{
d1310b2e
CM		 7316 struct extent_io_tree *tree;
7317 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7318 return extent_readpages(tree, mapping, pages, nr_pages,
7319 btrfs_get_extent);
7320}
e6dcd2dc 7321static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7322{
d1310b2e
CM		 7323 struct extent_io_tree *tree;
7324 struct extent_map_tree *map;
a52d9a80 7325 int ret;
8c2383c3 7326
d1310b2e
CM		 7327 tree = &BTRFS_I(page->mapping->host)->io_tree;
7328 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7329 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7330 if (ret == 1) {
7331 ClearPagePrivate(page);
7332 set_page_private(page, 0);
7333 page_cache_release(page);
39279cc3 7334 }
a52d9a80 7335 return ret;
39279cc3
CM		 7336}
7337
e6dcd2dc
CM		 7338static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7339{
98509cfc
CM		 7340 if (PageWriteback(page) || PageDirty(page))
7341 return 0;
b335b003 7342 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7343}
7344
d47992f8
LC		 7345static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7346 unsigned int length)
39279cc3 7347{
5fd02043 7348 struct inode *inode = page->mapping->host;
d1310b2e 7349 struct extent_io_tree *tree;
e6dcd2dc 7350 struct btrfs_ordered_extent *ordered;
2ac55d41 7351 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7352 u64 page_start = page_offset(page);
7353 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7354
8b62b72b
CM		 7355 /*
7356 * we have the page locked, so new writeback can't start,
7357 * and the dirty bit won't be cleared while we are here.
7358 *
7359 * Wait for IO on this page so that we can safely clear
7360 * the PagePrivate2 bit and do ordered accounting
7361 */
e6dcd2dc 7362 wait_on_page_writeback(page);
8b62b72b 7363
5fd02043 7364 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7365 if (offset) {
7366 btrfs_releasepage(page, GFP_NOFS);
7367 return;
7368 }
d0082371 7369 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7370 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7371 if (ordered) {
eb84ae03
CM		 7372 /*
7373 * IO on this page will never be started, so we need
7374 * to account for any ordered extents now
7375 */
e6dcd2dc
CM		 7376 clear_extent_bit(tree, page_start, page_end,
7377 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7378 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7379 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7380 /*
7381 * whoever cleared the private bit is responsible
7382 * for the finish_ordered_io
7383 */
77cef2ec
JB		 7384 if (TestClearPagePrivate2(page)) {
7385 struct btrfs_ordered_inode_tree *tree;
7386 u64 new_len;
7387
7388 tree = &BTRFS_I(inode)->ordered_tree;
7389
7390 spin_lock_irq(&tree->lock);
7391 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7392 new_len = page_start - ordered->file_offset;
7393 if (new_len < ordered->truncated_len)
7394 ordered->truncated_len = new_len;
7395 spin_unlock_irq(&tree->lock);
7396
7397 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7398 page_start,
7399 PAGE_CACHE_SIZE, 1))
7400 btrfs_finish_ordered_io(ordered);
8b62b72b 7401 }
e6dcd2dc 7402 btrfs_put_ordered_extent(ordered);
2ac55d41 7403 cached_state = NULL;
d0082371 7404 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7405 }
7406 clear_extent_bit(tree, page_start, page_end,
32c00aff 7407 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7408 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7409 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7410 __btrfs_releasepage(page, GFP_NOFS);
7411
4a096752 7412 ClearPageChecked(page);
9ad6b7bc 7413 if (PagePrivate(page)) {
9ad6b7bc
CM		 7414 ClearPagePrivate(page);
7415 set_page_private(page, 0);
7416 page_cache_release(page);
7417 }
39279cc3
CM		 7418}
7419
9ebefb18
CM		 7420/*
7421 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7422 * called from a page fault handler when a page is first dirtied. Hence we must
7423 * be careful to check for EOF conditions here. We set the page up correctly
7424 * for a written page which means we get ENOSPC checking when writing into
7425 * holes and correct delalloc and unwritten extent mapping on filesystems that
7426 * support these features.
7427 *
7428 * We are not allowed to take the i_mutex here so we have to play games to
7429 * protect against truncate races as the page could now be beyond EOF. Because
7430 * vmtruncate() writes the inode size before removing pages, once we have the
7431 * page lock we can determine safely if the page is beyond EOF. If it is not
7432 * beyond EOF, then the page is guaranteed safe against truncation until we
7433 * unlock the page.
7434 */
c2ec175c 7435int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7436{
c2ec175c 7437 struct page *page = vmf->page;
496ad9aa 7438 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7439 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7440 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7441 struct btrfs_ordered_extent *ordered;
2ac55d41 7442 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7443 char *kaddr;
7444 unsigned long zero_start;
9ebefb18 7445 loff_t size;
1832a6d5 7446 int ret;
9998eb70 7447 int reserved = 0;
a52d9a80 7448 u64 page_start;
e6dcd2dc 7449 u64 page_end;
9ebefb18 7450
b2b5ef5c 7451 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7452 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7453 if (!ret) {
e41f941a 7454 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7455 reserved = 1;
7456 }
56a76f82
NP		 7457 if (ret) {
7458 if (ret == -ENOMEM)
7459 ret = VM_FAULT_OOM;
7460 else /* -ENOSPC, -EIO, etc */
7461 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7462 if (reserved)
7463 goto out;
7464 goto out_noreserve;
56a76f82 7465 }
1832a6d5 7466
56a76f82 7467 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7468again:
9ebefb18 7469 lock_page(page);
9ebefb18 7470 size = i_size_read(inode);
e6dcd2dc
CM		 7471 page_start = page_offset(page);
7472 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7473
9ebefb18 7474 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7475 (page_start >= size)) {
9ebefb18
CM		 7476 /* page got truncated out from underneath us */
7477 goto out_unlock;
7478 }
e6dcd2dc
CM		 7479 wait_on_page_writeback(page);
7480
d0082371 7481 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7482 set_page_extent_mapped(page);
7483
eb84ae03
CM		 7484 /*
7485 * we can't set the delalloc bits if there are pending ordered
7486 * extents. Drop our locks and wait for them to finish
7487 */
e6dcd2dc
CM		 7488 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7489 if (ordered) {
2ac55d41
JB		 7490 unlock_extent_cached(io_tree, page_start, page_end,
7491 &cached_state, GFP_NOFS);
e6dcd2dc 7492 unlock_page(page);
eb84ae03 7493 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7494 btrfs_put_ordered_extent(ordered);
7495 goto again;
7496 }
7497
fbf19087
JB		 7498 /*
7499 * XXX - page_mkwrite gets called every time the page is dirtied, even
7500 * if it was already dirty, so for space accounting reasons we need to
7501 * clear any delalloc bits for the range we are fixing to save. There
7502 * is probably a better way to do this, but for now keep consistent with
7503 * prepare_pages in the normal write path.
7504 */
2ac55d41 7505 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7506 EXTENT_DIRTY | EXTENT_DELALLOC |
7507 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7508 0, 0, &cached_state, GFP_NOFS);
fbf19087 7509
2ac55d41
JB		 7510 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7511 &cached_state);
9ed74f2d 7512 if (ret) {
2ac55d41
JB		 7513 unlock_extent_cached(io_tree, page_start, page_end,
7514 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7515 ret = VM_FAULT_SIGBUS;
7516 goto out_unlock;
7517 }
e6dcd2dc 7518 ret = 0;
9ebefb18
CM		 7519
7520 /* page is wholly or partially inside EOF */
a52d9a80 7521 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7522 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7523 else
e6dcd2dc 7524 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7525
e6dcd2dc
CM		 7526 if (zero_start != PAGE_CACHE_SIZE) {
7527 kaddr = kmap(page);
7528 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7529 flush_dcache_page(page);
7530 kunmap(page);
7531 }
247e743c 7532 ClearPageChecked(page);
e6dcd2dc 7533 set_page_dirty(page);
50a9b214 7534 SetPageUptodate(page);
5a3f23d5 7535
257c62e1
CM		 7536 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7537 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7538 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7539
2ac55d41 7540 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7541
7542out_unlock:
b2b5ef5c
JK		 7543 if (!ret) {
7544 sb_end_pagefault(inode->i_sb);
50a9b214 7545 return VM_FAULT_LOCKED;
b2b5ef5c 7546 }
9ebefb18 7547 unlock_page(page);
1832a6d5 7548out:
ec39e180 7549 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7550out_noreserve:
b2b5ef5c 7551 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7552 return ret;
7553}
7554
a41ad394 7555static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7556{
7557 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7558 struct btrfs_block_rsv *rsv;
a71754fc 7559 int ret = 0;
3893e33b 7560 int err = 0;
39279cc3 7561 struct btrfs_trans_handle *trans;
dbe674a9 7562 u64 mask = root->sectorsize - 1;
07127184 7563 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7564
4a096752 7565 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 7566
fcb80c2a
JB		 7567 /*
7568 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7569 * 3 things going on here
7570 *
7571 * 1) We need to reserve space for our orphan item and the space to
7572 * delete our orphan item. Lord knows we don't want to have a dangling
7573 * orphan item because we didn't reserve space to remove it.
7574 *
7575 * 2) We need to reserve space to update our inode.
7576 *
7577 * 3) We need to have something to cache all the space that is going to
7578 * be free'd up by the truncate operation, but also have some slack
7579 * space reserved in case it uses space during the truncate (thank you
7580 * very much snapshotting).
7581 *
7582 * And we need these to all be seperate. The fact is we can use alot of
7583 * space doing the truncate, and we have no earthly idea how much space
7584 * we will use, so we need the truncate reservation to be seperate so it
7585 * doesn't end up using space reserved for updating the inode or
7586 * removing the orphan item. We also need to be able to stop the
7587 * transaction and start a new one, which means we need to be able to
7588 * update the inode several times, and we have no idea of knowing how
7589 * many times that will be, so we can't just reserve 1 item for the
7590 * entirety of the opration, so that has to be done seperately as well.
7591 * Then there is the orphan item, which does indeed need to be held on
7592 * to for the whole operation, and we need nobody to touch this reserved
7593 * space except the orphan code.
7594 *
7595 * So that leaves us with
7596 *
7597 * 1) root->orphan_block_rsv - for the orphan deletion.
7598 * 2) rsv - for the truncate reservation, which we will steal from the
7599 * transaction reservation.
7600 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7601 * updating the inode.
7602 */
66d8f3dd 7603 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 7604 if (!rsv)
7605 return -ENOMEM;
4a338542 7606 rsv->size = min_size;
ca7e70f5 7607 rsv->failfast = 1;
f0cd846e 7608
907cbceb 7609 /*
07127184 7610 * 1 for the truncate slack space
907cbceb
JB		 7611 * 1 for updating the inode.
7612 */
f3fe820c 7613 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 7614 if (IS_ERR(trans)) {
7615 err = PTR_ERR(trans);
7616 goto out;
7617 }
f0cd846e 7618
907cbceb
JB		 7619 /* Migrate the slack space for the truncate to our reserve */
7620 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7621 min_size);
fcb80c2a 7622 BUG_ON(ret);
f0cd846e 7623
5a3f23d5
CM		 7624 /*
7625 * setattr is responsible for setting the ordered_data_close flag,
7626 * but that is only tested during the last file release. That
7627 * could happen well after the next commit, leaving a great big
7628 * window where new writes may get lost if someone chooses to write
7629 * to this file after truncating to zero
7630 *
7631 * The inode doesn't have any dirty data here, and so if we commit
7632 * this is a noop. If someone immediately starts writing to the inode
7633 * it is very likely we'll catch some of their writes in this
7634 * transaction, and the commit will find this file on the ordered
7635 * data list with good things to send down.
7636 *
7637 * This is a best effort solution, there is still a window where
7638 * using truncate to replace the contents of the file will
7639 * end up with a zero length file after a crash.
7640 */
72ac3c0d
JB		 7641 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7642 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 7643 btrfs_add_ordered_operation(trans, root, inode);
7644
5dc562c5
JB		 7645 /*
7646 * So if we truncate and then write and fsync we normally would just
7647 * write the extents that changed, which is a problem if we need to
7648 * first truncate that entire inode. So set this flag so we write out
7649 * all of the extents in the inode to the sync log so we're completely
7650 * safe.
7651 */
7652 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7653 trans->block_rsv = rsv;
907cbceb 7654
8082510e
YZ		 7655 while (1) {
7656 ret = btrfs_truncate_inode_items(trans, root, inode,
7657 inode->i_size,
7658 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7659 if (ret != -ENOSPC) {
3893e33b 7660 err = ret;
8082510e 7661 break;
3893e33b 7662 }
39279cc3 7663
fcb80c2a 7664 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7665 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 7666 if (ret) {
7667 err = ret;
7668 break;
7669 }
ca7e70f5 7670
8082510e 7671 btrfs_end_transaction(trans, root);
b53d3f5d 7672 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 7673
7674 trans = btrfs_start_transaction(root, 2);
7675 if (IS_ERR(trans)) {
7676 ret = err = PTR_ERR(trans);
7677 trans = NULL;
7678 break;
7679 }
7680
7681 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7682 rsv, min_size);
7683 BUG_ON(ret); /* shouldn't happen */
7684 trans->block_rsv = rsv;
8082510e
YZ		 7685 }
7686
7687 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7688 trans->block_rsv = root->orphan_block_rsv;
8082510e 7689 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7690 if (ret)
7691 err = ret;
8082510e
YZ		 7692 }
7693
917c16b2
CM		 7694 if (trans) {
7695 trans->block_rsv = &root->fs_info->trans_block_rsv;
7696 ret = btrfs_update_inode(trans, root, inode);
7697 if (ret && !err)
7698 err = ret;
7b128766 7699
7ad85bb7 7700 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7701 btrfs_btree_balance_dirty(root);
917c16b2 7702 }
fcb80c2a
JB		 7703
7704out:
7705 btrfs_free_block_rsv(root, rsv);
7706
3893e33b
JB		 7707 if (ret && !err)
7708 err = ret;
a41ad394 7709
3893e33b 7710 return err;
39279cc3
CM		 7711}
7712
d352ac68
CM		 7713/*
7714 * create a new subvolume directory/inode (helper for the ioctl).
7715 */
d2fb3437 7716int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7717 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7718{
39279cc3 7719 struct inode *inode;
76dda93c 7720 int err;
00e4e6b3 7721 u64 index = 0;
39279cc3 7722
12fc9d09
FA		 7723 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7724 new_dirid, new_dirid,
7725 S_IFDIR | (~current_umask() & S_IRWXUGO),
7726 &index);
54aa1f4d 7727 if (IS_ERR(inode))
f46b5a66 7728 return PTR_ERR(inode);
39279cc3
CM		 7729 inode->i_op = &btrfs_dir_inode_operations;
7730 inode->i_fop = &btrfs_dir_file_operations;
7731
bfe86848 7732 set_nlink(inode, 1);
dbe674a9 7733 btrfs_i_size_write(inode, 0);
3b96362c 7734
76dda93c 7735 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7736
76dda93c 7737 iput(inode);
ce598979 7738 return err;
39279cc3
CM		 7739}
7740
39279cc3
CM		 7741struct inode *btrfs_alloc_inode(struct super_block *sb)
7742{
7743 struct btrfs_inode *ei;
2ead6ae7 7744 struct inode *inode;
39279cc3
CM		 7745
7746 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7747 if (!ei)
7748 return NULL;
2ead6ae7
YZ		 7749
7750 ei->root = NULL;
2ead6ae7 7751 ei->generation = 0;
15ee9bc7 7752 ei->last_trans = 0;
257c62e1 7753 ei->last_sub_trans = 0;
e02119d5 7754 ei->logged_trans = 0;
2ead6ae7 7755 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7756 ei->disk_i_size = 0;
7757 ei->flags = 0;
7709cde3 7758 ei->csum_bytes = 0;
2ead6ae7
YZ		 7759 ei->index_cnt = (u64)-1;
7760 ei->last_unlink_trans = 0;
46d8bc34 7761 ei->last_log_commit = 0;
2ead6ae7 7762
9e0baf60
JB		 7763 spin_lock_init(&ei->lock);
7764 ei->outstanding_extents = 0;
7765 ei->reserved_extents = 0;
2ead6ae7 7766
72ac3c0d 7767 ei->runtime_flags = 0;
261507a0 7768 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7769
16cdcec7
MX		 7770 ei->delayed_node = NULL;
7771
2ead6ae7 7772 inode = &ei->vfs_inode;
a8067e02 7773 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7774 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7775 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7776 ei->io_tree.track_uptodate = 1;
7777 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7778 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7779 mutex_init(&ei->log_mutex);
f248679e 7780 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7781 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7782 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7783 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7784 RB_CLEAR_NODE(&ei->rb_node);
7785
7786 return inode;
39279cc3
CM		 7787}
7788
fa0d7e3d
NP		 7789static void btrfs_i_callback(struct rcu_head *head)
7790{
7791 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7792 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7793}
7794
39279cc3
CM		 7795void btrfs_destroy_inode(struct inode *inode)
7796{
e6dcd2dc 7797 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7798 struct btrfs_root *root = BTRFS_I(inode)->root;
7799
b3d9b7a3 7800 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7801 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7802 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7803 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7804 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7805 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7806
a6dbd429
JB		 7807 /*
7808 * This can happen where we create an inode, but somebody else also
7809 * created the same inode and we need to destroy the one we already
7810 * created.
7811 */
7812 if (!root)
7813 goto free;
7814
5a3f23d5
CM		 7815 /*
7816 * Make sure we're properly removed from the ordered operation
7817 * lists.
7818 */
7819 smp_mb();
7820 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
199c2a9c 7821 spin_lock(&root->fs_info->ordered_root_lock);
5a3f23d5 7822 list_del_init(&BTRFS_I(inode)->ordered_operations);
199c2a9c 7823 spin_unlock(&root->fs_info->ordered_root_lock);
5a3f23d5
CM		 7824 }
7825
8a35d95f
JB		 7826 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7827 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 7828 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 7829 btrfs_ino(inode));
8a35d95f 7830 atomic_dec(&root->orphan_inodes);
7b128766 7831 }
7b128766 7832
d397712b 7833 while (1) {
e6dcd2dc
CM		 7834 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7835 if (!ordered)
7836 break;
7837 else {
c2cf52eb 7838 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 7839 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 7840 btrfs_remove_ordered_extent(inode, ordered);
7841 btrfs_put_ordered_extent(ordered);
7842 btrfs_put_ordered_extent(ordered);
7843 }
7844 }
5d4f98a2 7845 inode_tree_del(inode);
5b21f2ed 7846 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7847free:
fa0d7e3d 7848 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7849}
7850
45321ac5 7851int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7852{
7853 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7854
6379ef9f
NA		 7855 if (root == NULL)
7856 return 1;
7857
fa6ac876 7858 /* the snap/subvol tree is on deleting */
69e9c6c6 7859 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 7860 return 1;
76dda93c 7861 else
45321ac5 7862 return generic_drop_inode(inode);
76dda93c
YZ		 7863}
7864
0ee0fda0 7865static void init_once(void *foo)
39279cc3
CM		 7866{
7867 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7868
7869 inode_init_once(&ei->vfs_inode);
7870}
7871
7872void btrfs_destroy_cachep(void)
7873{
8c0a8537
KS		 7874 /*
7875 * Make sure all delayed rcu free inodes are flushed before we
7876 * destroy cache.
7877 */
7878 rcu_barrier();
39279cc3
CM		 7879 if (btrfs_inode_cachep)
7880 kmem_cache_destroy(btrfs_inode_cachep);
7881 if (btrfs_trans_handle_cachep)
7882 kmem_cache_destroy(btrfs_trans_handle_cachep);
7883 if (btrfs_transaction_cachep)
7884 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7885 if (btrfs_path_cachep)
7886 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7887 if (btrfs_free_space_cachep)
7888 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 7889 if (btrfs_delalloc_work_cachep)
7890 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 7891}
7892
7893int btrfs_init_cachep(void)
7894{
837e1972 7895 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 7896 sizeof(struct btrfs_inode), 0,
7897 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7898 if (!btrfs_inode_cachep)
7899 goto fail;
9601e3f6 7900
837e1972 7901 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 7902 sizeof(struct btrfs_trans_handle), 0,
7903 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7904 if (!btrfs_trans_handle_cachep)
7905 goto fail;
9601e3f6 7906
837e1972 7907 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 7908 sizeof(struct btrfs_transaction), 0,
7909 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7910 if (!btrfs_transaction_cachep)
7911 goto fail;
9601e3f6 7912
837e1972 7913 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 7914 sizeof(struct btrfs_path), 0,
7915 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7916 if (!btrfs_path_cachep)
7917 goto fail;
9601e3f6 7918
837e1972 7919 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 7920 sizeof(struct btrfs_free_space), 0,
7921 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7922 if (!btrfs_free_space_cachep)
7923 goto fail;
7924
8ccf6f19
MX		 7925 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7926 sizeof(struct btrfs_delalloc_work), 0,
7927 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7928 NULL);
7929 if (!btrfs_delalloc_work_cachep)
7930 goto fail;
7931
39279cc3
CM		 7932 return 0;
7933fail:
7934 btrfs_destroy_cachep();
7935 return -ENOMEM;
7936}
7937
7938static int btrfs_getattr(struct vfsmount *mnt,
7939 struct dentry *dentry, struct kstat *stat)
7940{
df0af1a5 7941 u64 delalloc_bytes;
39279cc3 7942 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7943 u32 blocksize = inode->i_sb->s_blocksize;
7944
39279cc3 7945 generic_fillattr(inode, stat);
0ee5dc67 7946 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7947 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 7948
7949 spin_lock(&BTRFS_I(inode)->lock);
7950 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7951 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 7952 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 7953 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7954 return 0;
7955}
7956
d397712b
CM		 7957static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7958 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7959{
7960 struct btrfs_trans_handle *trans;
7961 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7962 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7963 struct inode *new_inode = new_dentry->d_inode;
7964 struct inode *old_inode = old_dentry->d_inode;
7965 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7966 u64 index = 0;
4df27c4d 7967 u64 root_objectid;
39279cc3 7968 int ret;
33345d01 7969 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7970
33345d01 7971 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 7972 return -EPERM;
7973
4df27c4d 7974 /* we only allow rename subvolume link between subvolumes */
33345d01 7975 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 7976 return -EXDEV;
7977
33345d01
LZ		 7978 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7979 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 7980 return -ENOTEMPTY;
5f39d397 7981
4df27c4d
YZ		 7982 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7983 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7984 return -ENOTEMPTY;
9c52057c
CM		 7985
7986
7987 /* check for collisions, even if the name isn't there */
4871c158 7988 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 7989 new_dentry->d_name.name,
7990 new_dentry->d_name.len);
7991
7992 if (ret) {
7993 if (ret == -EEXIST) {
7994 /* we shouldn't get
7995 * eexist without a new_inode */
7996 if (!new_inode) {
7997 WARN_ON(1);
7998 return ret;
7999 }
8000 } else {
8001 /* maybe -EOVERFLOW */
8002 return ret;
8003 }
8004 }
8005 ret = 0;
8006
5a3f23d5
CM		 8007 /*
8008 * we're using rename to replace one file with another.
8009 * and the replacement file is large. Start IO on it now so
8010 * we don't add too much work to the end of the transaction
8011 */
4baf8c92 8012 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8013 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8014 filemap_flush(old_inode->i_mapping);
8015
76dda93c 8016 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8017 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8018 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8019 /*
8020 * We want to reserve the absolute worst case amount of items. So if
8021 * both inodes are subvols and we need to unlink them then that would
8022 * require 4 item modifications, but if they are both normal inodes it
8023 * would require 5 item modifications, so we'll assume their normal
8024 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8025 * should cover the worst case number of items we'll modify.
8026 */
6e137ed3 8027 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8028 if (IS_ERR(trans)) {
8029 ret = PTR_ERR(trans);
8030 goto out_notrans;
8031 }
76dda93c 8032
4df27c4d
YZ		 8033 if (dest != root)
8034 btrfs_record_root_in_trans(trans, dest);
5f39d397 8035
a5719521
YZ		 8036 ret = btrfs_set_inode_index(new_dir, &index);
8037 if (ret)
8038 goto out_fail;
5a3f23d5 8039
33345d01 8040 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8041 /* force full log commit if subvolume involved. */
8042 root->fs_info->last_trans_log_full_commit = trans->transid;
8043 } else {
a5719521
YZ		 8044 ret = btrfs_insert_inode_ref(trans, dest,
8045 new_dentry->d_name.name,
8046 new_dentry->d_name.len,
33345d01
LZ		 8047 old_ino,
8048 btrfs_ino(new_dir), index);
a5719521
YZ		 8049 if (ret)
8050 goto out_fail;
4df27c4d
YZ		 8051 /*
8052 * this is an ugly little race, but the rename is required
8053 * to make sure that if we crash, the inode is either at the
8054 * old name or the new one. pinning the log transaction lets
8055 * us make sure we don't allow a log commit to come in after
8056 * we unlink the name but before we add the new name back in.
8057 */
8058 btrfs_pin_log_trans(root);
8059 }
5a3f23d5
CM		 8060 /*
8061 * make sure the inode gets flushed if it is replacing
8062 * something.
8063 */
33345d01 8064 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8065 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8066
0c4d2d95
JB		 8067 inode_inc_iversion(old_dir);
8068 inode_inc_iversion(new_dir);
8069 inode_inc_iversion(old_inode);
39279cc3
CM		 8070 old_dir->i_ctime = old_dir->i_mtime = ctime;
8071 new_dir->i_ctime = new_dir->i_mtime = ctime;
8072 old_inode->i_ctime = ctime;
5f39d397 8073
12fcfd22
CM		 8074 if (old_dentry->d_parent != new_dentry->d_parent)
8075 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8076
33345d01 8077 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8078 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8079 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8080 old_dentry->d_name.name,
8081 old_dentry->d_name.len);
8082 } else {
92986796
AV		 8083 ret = __btrfs_unlink_inode(trans, root, old_dir,
8084 old_dentry->d_inode,
8085 old_dentry->d_name.name,
8086 old_dentry->d_name.len);
8087 if (!ret)
8088 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8089 }
79787eaa
JM		 8090 if (ret) {
8091 btrfs_abort_transaction(trans, root, ret);
8092 goto out_fail;
8093 }
39279cc3
CM		 8094
8095 if (new_inode) {
0c4d2d95 8096 inode_inc_iversion(new_inode);
39279cc3 8097 new_inode->i_ctime = CURRENT_TIME;
33345d01 8098 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8099 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8100 root_objectid = BTRFS_I(new_inode)->location.objectid;
8101 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8102 root_objectid,
8103 new_dentry->d_name.name,
8104 new_dentry->d_name.len);
8105 BUG_ON(new_inode->i_nlink == 0);
8106 } else {
8107 ret = btrfs_unlink_inode(trans, dest, new_dir,
8108 new_dentry->d_inode,
8109 new_dentry->d_name.name,
8110 new_dentry->d_name.len);
8111 }
4ef31a45 8112 if (!ret && new_inode->i_nlink == 0)
e02119d5 8113 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
79787eaa
JM		 8114 if (ret) {
8115 btrfs_abort_transaction(trans, root, ret);
8116 goto out_fail;
8117 }
39279cc3 8118 }
aec7477b 8119
4df27c4d
YZ		 8120 ret = btrfs_add_link(trans, new_dir, old_inode,
8121 new_dentry->d_name.name,
a5719521 8122 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8123 if (ret) {
8124 btrfs_abort_transaction(trans, root, ret);
8125 goto out_fail;
8126 }
39279cc3 8127
33345d01 8128 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8129 struct dentry *parent = new_dentry->d_parent;
6a912213 8130 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8131 btrfs_end_log_trans(root);
8132 }
39279cc3 8133out_fail:
7ad85bb7 8134 btrfs_end_transaction(trans, root);
b44c59a8 8135out_notrans:
33345d01 8136 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8137 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8138
39279cc3
CM		 8139 return ret;
8140}
8141
8ccf6f19
MX		 8142static void btrfs_run_delalloc_work(struct btrfs_work *work)
8143{
8144 struct btrfs_delalloc_work *delalloc_work;
8145
8146 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8147 work);
8148 if (delalloc_work->wait)
8149 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8150 else
8151 filemap_flush(delalloc_work->inode->i_mapping);
8152
8153 if (delalloc_work->delay_iput)
8154 btrfs_add_delayed_iput(delalloc_work->inode);
8155 else
8156 iput(delalloc_work->inode);
8157 complete(&delalloc_work->completion);
8158}
8159
8160struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8161 int wait, int delay_iput)
8162{
8163 struct btrfs_delalloc_work *work;
8164
8165 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8166 if (!work)
8167 return NULL;
8168
8169 init_completion(&work->completion);
8170 INIT_LIST_HEAD(&work->list);
8171 work->inode = inode;
8172 work->wait = wait;
8173 work->delay_iput = delay_iput;
8174 work->work.func = btrfs_run_delalloc_work;
8175
8176 return work;
8177}
8178
8179void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8180{
8181 wait_for_completion(&work->completion);
8182 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8183}
8184
d352ac68
CM		 8185/*
8186 * some fairly slow code that needs optimization. This walks the list
8187 * of all the inodes with pending delalloc and forces them to disk.
8188 */
eb73c1b7 8189static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8190{
ea8c2819 8191 struct btrfs_inode *binode;
5b21f2ed 8192 struct inode *inode;
8ccf6f19
MX		 8193 struct btrfs_delalloc_work *work, *next;
8194 struct list_head works;
1eafa6c7 8195 struct list_head splice;
8ccf6f19 8196 int ret = 0;
ea8c2819 8197
8ccf6f19 8198 INIT_LIST_HEAD(&works);
1eafa6c7 8199 INIT_LIST_HEAD(&splice);
63607cc8 8200
eb73c1b7
MX		 8201 spin_lock(&root->delalloc_lock);
8202 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8203 while (!list_empty(&splice)) {
8204 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8205 delalloc_inodes);
1eafa6c7 8206
eb73c1b7
MX		 8207 list_move_tail(&binode->delalloc_inodes,
8208 &root->delalloc_inodes);
5b21f2ed 8209 inode = igrab(&binode->vfs_inode);
df0af1a5 8210 if (!inode) {
eb73c1b7 8211 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8212 continue;
df0af1a5 8213 }
eb73c1b7 8214 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8215
8216 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8217 if (unlikely(!work)) {
f4ab9ea7
JB		 8218 if (delay_iput)
8219 btrfs_add_delayed_iput(inode);
8220 else
8221 iput(inode);
1eafa6c7
MX		 8222 ret = -ENOMEM;
8223 goto out;
5b21f2ed 8224 }
1eafa6c7
MX		 8225 list_add_tail(&work->list, &works);
8226 btrfs_queue_worker(&root->fs_info->flush_workers,
8227 &work->work);
8228
5b21f2ed 8229 cond_resched();
eb73c1b7 8230 spin_lock(&root->delalloc_lock);
ea8c2819 8231 }
eb73c1b7 8232 spin_unlock(&root->delalloc_lock);
8c8bee1d 8233
1eafa6c7
MX		 8234 list_for_each_entry_safe(work, next, &works, list) {
8235 list_del_init(&work->list);
8236 btrfs_wait_and_free_delalloc_work(work);
8237 }
eb73c1b7
MX		 8238 return 0;
8239out:
8240 list_for_each_entry_safe(work, next, &works, list) {
8241 list_del_init(&work->list);
8242 btrfs_wait_and_free_delalloc_work(work);
8243 }
8244
8245 if (!list_empty_careful(&splice)) {
8246 spin_lock(&root->delalloc_lock);
8247 list_splice_tail(&splice, &root->delalloc_inodes);
8248 spin_unlock(&root->delalloc_lock);
8249 }
8250 return ret;
8251}
1eafa6c7 8252
eb73c1b7
MX		 8253int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8254{
8255 int ret;
1eafa6c7 8256
eb73c1b7
MX		 8257 if (root->fs_info->sb->s_flags & MS_RDONLY)
8258 return -EROFS;
8259
8260 ret = __start_delalloc_inodes(root, delay_iput);
8261 /*
8262 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8263 * we have to make sure the IO is actually started and that
8264 * ordered extents get created before we return
8265 */
8266 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8267 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8268 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8269 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8270 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8271 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8272 }
8273 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8274 return ret;
8275}
8276
8277int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8278 int delay_iput)
8279{
8280 struct btrfs_root *root;
8281 struct list_head splice;
8282 int ret;
8283
8284 if (fs_info->sb->s_flags & MS_RDONLY)
8285 return -EROFS;
8286
8287 INIT_LIST_HEAD(&splice);
8288
8289 spin_lock(&fs_info->delalloc_root_lock);
8290 list_splice_init(&fs_info->delalloc_roots, &splice);
8291 while (!list_empty(&splice)) {
8292 root = list_first_entry(&splice, struct btrfs_root,
8293 delalloc_root);
8294 root = btrfs_grab_fs_root(root);
8295 BUG_ON(!root);
8296 list_move_tail(&root->delalloc_root,
8297 &fs_info->delalloc_roots);
8298 spin_unlock(&fs_info->delalloc_root_lock);
8299
8300 ret = __start_delalloc_inodes(root, delay_iput);
8301 btrfs_put_fs_root(root);
8302 if (ret)
8303 goto out;
8304
8305 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8306 }
eb73c1b7 8307 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8308
eb73c1b7
MX		 8309 atomic_inc(&fs_info->async_submit_draining);
8310 while (atomic_read(&fs_info->nr_async_submits) ||
8311 atomic_read(&fs_info->async_delalloc_pages)) {
8312 wait_event(fs_info->async_submit_wait,
8313 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8314 atomic_read(&fs_info->async_delalloc_pages) == 0));
8315 }
8316 atomic_dec(&fs_info->async_submit_draining);
8317 return 0;
8318out:
1eafa6c7 8319 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8320 spin_lock(&fs_info->delalloc_root_lock);
8321 list_splice_tail(&splice, &fs_info->delalloc_roots);
8322 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8323 }
8ccf6f19 8324 return ret;
ea8c2819
CM		 8325}
8326
39279cc3
CM		 8327static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8328 const char *symname)
8329{
8330 struct btrfs_trans_handle *trans;
8331 struct btrfs_root *root = BTRFS_I(dir)->root;
8332 struct btrfs_path *path;
8333 struct btrfs_key key;
1832a6d5 8334 struct inode *inode = NULL;
39279cc3
CM		 8335 int err;
8336 int drop_inode = 0;
8337 u64 objectid;
00e4e6b3 8338 u64 index = 0 ;
39279cc3
CM		 8339 int name_len;
8340 int datasize;
5f39d397 8341 unsigned long ptr;
39279cc3 8342 struct btrfs_file_extent_item *ei;
5f39d397 8343 struct extent_buffer *leaf;
39279cc3 8344
f06becc4 8345 name_len = strlen(symname);
39279cc3
CM		 8346 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8347 return -ENAMETOOLONG;
1832a6d5 8348
9ed74f2d
JB		 8349 /*
8350 * 2 items for inode item and ref
8351 * 2 items for dir items
8352 * 1 item for xattr if selinux is on
8353 */
a22285a6
YZ		 8354 trans = btrfs_start_transaction(root, 5);
8355 if (IS_ERR(trans))
8356 return PTR_ERR(trans);
1832a6d5 8357
581bb050
LZ		 8358 err = btrfs_find_free_ino(root, &objectid);
8359 if (err)
8360 goto out_unlock;
8361
aec7477b 8362 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8363 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8364 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8365 if (IS_ERR(inode)) {
8366 err = PTR_ERR(inode);
39279cc3 8367 goto out_unlock;
7cf96da3 8368 }
39279cc3 8369
2a7dba39 8370 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8371 if (err) {
8372 drop_inode = 1;
8373 goto out_unlock;
8374 }
8375
ad19db71
CS		 8376 /*
8377 * If the active LSM wants to access the inode during
8378 * d_instantiate it needs these. Smack checks to see
8379 * if the filesystem supports xattrs by looking at the
8380 * ops vector.
8381 */
8382 inode->i_fop = &btrfs_file_operations;
8383 inode->i_op = &btrfs_file_inode_operations;
8384
a1b075d2 8385 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8386 if (err)
8387 drop_inode = 1;
8388 else {
8389 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8390 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8391 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8392 }
39279cc3
CM		 8393 if (drop_inode)
8394 goto out_unlock;
8395
8396 path = btrfs_alloc_path();
d8926bb3
MF		 8397 if (!path) {
8398 err = -ENOMEM;
8399 drop_inode = 1;
8400 goto out_unlock;
8401 }
33345d01 8402 key.objectid = btrfs_ino(inode);
39279cc3 8403 key.offset = 0;
39279cc3
CM		 8404 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8405 datasize = btrfs_file_extent_calc_inline_size(name_len);
8406 err = btrfs_insert_empty_item(trans, root, path, &key,
8407 datasize);
54aa1f4d
CM		 8408 if (err) {
8409 drop_inode = 1;
b0839166 8410 btrfs_free_path(path);
54aa1f4d
CM		 8411 goto out_unlock;
8412 }
5f39d397
CM		 8413 leaf = path->nodes[0];
8414 ei = btrfs_item_ptr(leaf, path->slots[0],
8415 struct btrfs_file_extent_item);
8416 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8417 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8418 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8419 btrfs_set_file_extent_encryption(leaf, ei, 0);
8420 btrfs_set_file_extent_compression(leaf, ei, 0);
8421 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8422 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8423
39279cc3 8424 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8425 write_extent_buffer(leaf, symname, ptr, name_len);
8426 btrfs_mark_buffer_dirty(leaf);
39279cc3 8427 btrfs_free_path(path);
5f39d397 8428
39279cc3
CM		 8429 inode->i_op = &btrfs_symlink_inode_operations;
8430 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8431 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8432 inode_set_bytes(inode, name_len);
f06becc4 8433 btrfs_i_size_write(inode, name_len);
54aa1f4d
CM		 8434 err = btrfs_update_inode(trans, root, inode);
8435 if (err)
8436 drop_inode = 1;
39279cc3
CM		 8437
8438out_unlock:
08c422c2
AV		 8439 if (!err)
8440 d_instantiate(dentry, inode);
7ad85bb7 8441 btrfs_end_transaction(trans, root);
39279cc3
CM		 8442 if (drop_inode) {
8443 inode_dec_link_count(inode);
8444 iput(inode);
8445 }
b53d3f5d 8446 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8447 return err;
8448}
16432985 8449
0af3d00b
JB		 8450static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8451 u64 start, u64 num_bytes, u64 min_size,
8452 loff_t actual_len, u64 *alloc_hint,
8453 struct btrfs_trans_handle *trans)
d899e052 8454{
5dc562c5
JB		 8455 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8456 struct extent_map *em;
d899e052
YZ		 8457 struct btrfs_root *root = BTRFS_I(inode)->root;
8458 struct btrfs_key ins;
d899e052 8459 u64 cur_offset = start;
55a61d1d 8460 u64 i_size;
154ea289 8461 u64 cur_bytes;
d899e052 8462 int ret = 0;
0af3d00b 8463 bool own_trans = true;
d899e052 8464
0af3d00b
JB		 8465 if (trans)
8466 own_trans = false;
d899e052 8467 while (num_bytes > 0) {
0af3d00b
JB		 8468 if (own_trans) {
8469 trans = btrfs_start_transaction(root, 3);
8470 if (IS_ERR(trans)) {
8471 ret = PTR_ERR(trans);
8472 break;
8473 }
5a303d5d
YZ		 8474 }
8475
154ea289
CM		 8476 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8477 cur_bytes = max(cur_bytes, min_size);
00361589
JB		 8478 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8479 *alloc_hint, &ins, 1);
5a303d5d 8480 if (ret) {
0af3d00b
JB		 8481 if (own_trans)
8482 btrfs_end_transaction(trans, root);
a22285a6 8483 break;
d899e052 8484 }
5a303d5d 8485
d899e052
YZ		 8486 ret = insert_reserved_file_extent(trans, inode,
8487 cur_offset, ins.objectid,
8488 ins.offset, ins.offset,
920bbbfb 8489 ins.offset, 0, 0, 0,
d899e052 8490 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 8491 if (ret) {
8492 btrfs_abort_transaction(trans, root, ret);
8493 if (own_trans)
8494 btrfs_end_transaction(trans, root);
8495 break;
8496 }
a1ed835e
CM		 8497 btrfs_drop_extent_cache(inode, cur_offset,
8498 cur_offset + ins.offset -1, 0);
5a303d5d 8499
5dc562c5
JB		 8500 em = alloc_extent_map();
8501 if (!em) {
8502 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8503 &BTRFS_I(inode)->runtime_flags);
8504 goto next;
8505 }
8506
8507 em->start = cur_offset;
8508 em->orig_start = cur_offset;
8509 em->len = ins.offset;
8510 em->block_start = ins.objectid;
8511 em->block_len = ins.offset;
b4939680 8512 em->orig_block_len = ins.offset;
cc95bef6 8513 em->ram_bytes = ins.offset;
5dc562c5
JB		 8514 em->bdev = root->fs_info->fs_devices->latest_bdev;
8515 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8516 em->generation = trans->transid;
8517
8518 while (1) {
8519 write_lock(&em_tree->lock);
09a2a8f9 8520 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8521 write_unlock(&em_tree->lock);
8522 if (ret != -EEXIST)
8523 break;
8524 btrfs_drop_extent_cache(inode, cur_offset,
8525 cur_offset + ins.offset - 1,
8526 0);
8527 }
8528 free_extent_map(em);
8529next:
d899e052
YZ		 8530 num_bytes -= ins.offset;
8531 cur_offset += ins.offset;
efa56464 8532 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8533
0c4d2d95 8534 inode_inc_iversion(inode);
d899e052 8535 inode->i_ctime = CURRENT_TIME;
6cbff00f 8536 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8537 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8538 (actual_len > inode->i_size) &&
8539 (cur_offset > inode->i_size)) {
d1ea6a61 8540 if (cur_offset > actual_len)
55a61d1d 8541 i_size = actual_len;
d1ea6a61 8542 else
55a61d1d
JB		 8543 i_size = cur_offset;
8544 i_size_write(inode, i_size);
8545 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8546 }
8547
d899e052 8548 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8549
8550 if (ret) {
8551 btrfs_abort_transaction(trans, root, ret);
8552 if (own_trans)
8553 btrfs_end_transaction(trans, root);
8554 break;
8555 }
d899e052 8556
0af3d00b
JB		 8557 if (own_trans)
8558 btrfs_end_transaction(trans, root);
5a303d5d 8559 }
d899e052
YZ		 8560 return ret;
8561}
8562
0af3d00b
JB		 8563int btrfs_prealloc_file_range(struct inode *inode, int mode,
8564 u64 start, u64 num_bytes, u64 min_size,
8565 loff_t actual_len, u64 *alloc_hint)
8566{
8567 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8568 min_size, actual_len, alloc_hint,
8569 NULL);
8570}
8571
8572int btrfs_prealloc_file_range_trans(struct inode *inode,
8573 struct btrfs_trans_handle *trans, int mode,
8574 u64 start, u64 num_bytes, u64 min_size,
8575 loff_t actual_len, u64 *alloc_hint)
8576{
8577 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8578 min_size, actual_len, alloc_hint, trans);
8579}
8580
e6dcd2dc
CM		 8581static int btrfs_set_page_dirty(struct page *page)
8582{
e6dcd2dc
CM		 8583 return __set_page_dirty_nobuffers(page);
8584}
8585
10556cb2 8586static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8587{
b83cc969 8588 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8589 umode_t mode = inode->i_mode;
b83cc969 8590
cb6db4e5
JM		 8591 if (mask & MAY_WRITE &&
8592 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8593 if (btrfs_root_readonly(root))
8594 return -EROFS;
8595 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8596 return -EACCES;
8597 }
2830ba7f 8598 return generic_permission(inode, mask);
fdebe2bd 8599}
39279cc3 8600
6e1d5dcc 8601static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8602 .getattr = btrfs_getattr,
39279cc3
CM		 8603 .lookup = btrfs_lookup,
8604 .create = btrfs_create,
8605 .unlink = btrfs_unlink,
8606 .link = btrfs_link,
8607 .mkdir = btrfs_mkdir,
8608 .rmdir = btrfs_rmdir,
8609 .rename = btrfs_rename,
8610 .symlink = btrfs_symlink,
8611 .setattr = btrfs_setattr,
618e21d5 8612 .mknod = btrfs_mknod,
95819c05
CH		 8613 .setxattr = btrfs_setxattr,
8614 .getxattr = btrfs_getxattr,
5103e947 8615 .listxattr = btrfs_listxattr,
95819c05 8616 .removexattr = btrfs_removexattr,
fdebe2bd 8617 .permission = btrfs_permission,
4e34e719 8618 .get_acl = btrfs_get_acl,
93fd63c2 8619 .update_time = btrfs_update_time,
39279cc3 8620};
6e1d5dcc 8621static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8622 .lookup = btrfs_lookup,
fdebe2bd 8623 .permission = btrfs_permission,
4e34e719 8624 .get_acl = btrfs_get_acl,
93fd63c2 8625 .update_time = btrfs_update_time,
39279cc3 8626};
76dda93c 8627
828c0950 8628static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8629 .llseek = generic_file_llseek,
8630 .read = generic_read_dir,
9cdda8d3 8631 .iterate = btrfs_real_readdir,
34287aa3 8632 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8633#ifdef CONFIG_COMPAT
34287aa3 8634 .compat_ioctl = btrfs_ioctl,
39279cc3 8635#endif
6bf13c0c 8636 .release = btrfs_release_file,
e02119d5 8637 .fsync = btrfs_sync_file,
39279cc3
CM		 8638};
8639
d1310b2e 8640static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8641 .fill_delalloc = run_delalloc_range,
065631f6 8642 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8643 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8644 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8645 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8646 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8647 .set_bit_hook = btrfs_set_bit_hook,
8648 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8649 .merge_extent_hook = btrfs_merge_extent_hook,
8650 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8651};
8652
35054394
CM		 8653/*
8654 * btrfs doesn't support the bmap operation because swapfiles
8655 * use bmap to make a mapping of extents in the file. They assume
8656 * these extents won't change over the life of the file and they
8657 * use the bmap result to do IO directly to the drive.
8658 *
8659 * the btrfs bmap call would return logical addresses that aren't
8660 * suitable for IO and they also will change frequently as COW
8661 * operations happen. So, swapfile + btrfs == corruption.
8662 *
8663 * For now we're avoiding this by dropping bmap.
8664 */
7f09410b 8665static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8666 .readpage = btrfs_readpage,
8667 .writepage = btrfs_writepage,
b293f02e 8668 .writepages = btrfs_writepages,
3ab2fb5a 8669 .readpages = btrfs_readpages,
16432985 8670 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8671 .invalidatepage = btrfs_invalidatepage,
8672 .releasepage = btrfs_releasepage,
e6dcd2dc 8673 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8674 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8675};
8676
7f09410b 8677static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8678 .readpage = btrfs_readpage,
8679 .writepage = btrfs_writepage,
2bf5a725
CM		 8680 .invalidatepage = btrfs_invalidatepage,
8681 .releasepage = btrfs_releasepage,
39279cc3
CM		 8682};
8683
6e1d5dcc 8684static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8685 .getattr = btrfs_getattr,
8686 .setattr = btrfs_setattr,
95819c05
CH		 8687 .setxattr = btrfs_setxattr,
8688 .getxattr = btrfs_getxattr,
5103e947 8689 .listxattr = btrfs_listxattr,
95819c05 8690 .removexattr = btrfs_removexattr,
fdebe2bd 8691 .permission = btrfs_permission,
1506fcc8 8692 .fiemap = btrfs_fiemap,
4e34e719 8693 .get_acl = btrfs_get_acl,
e41f941a 8694 .update_time = btrfs_update_time,
39279cc3 8695};
6e1d5dcc 8696static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8697 .getattr = btrfs_getattr,
8698 .setattr = btrfs_setattr,
fdebe2bd 8699 .permission = btrfs_permission,
95819c05
CH		 8700 .setxattr = btrfs_setxattr,
8701 .getxattr = btrfs_getxattr,
33268eaf 8702 .listxattr = btrfs_listxattr,
95819c05 8703 .removexattr = btrfs_removexattr,
4e34e719 8704 .get_acl = btrfs_get_acl,
e41f941a 8705 .update_time = btrfs_update_time,
618e21d5 8706};
6e1d5dcc 8707static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8708 .readlink = generic_readlink,
8709 .follow_link = page_follow_link_light,
8710 .put_link = page_put_link,
f209561a 8711 .getattr = btrfs_getattr,
22c44fe6 8712 .setattr = btrfs_setattr,
fdebe2bd 8713 .permission = btrfs_permission,
0279b4cd
JO		 8714 .setxattr = btrfs_setxattr,
8715 .getxattr = btrfs_getxattr,
8716 .listxattr = btrfs_listxattr,
8717 .removexattr = btrfs_removexattr,
4e34e719 8718 .get_acl = btrfs_get_acl,
e41f941a 8719 .update_time = btrfs_update_time,
39279cc3 8720};
76dda93c 8721
82d339d9 8722const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8723 .d_delete = btrfs_dentry_delete,
b4aff1f8 8724 .d_release = btrfs_dentry_release,
76dda93c 8725};
Linux 6.x block layer and io_uring tree(s)