projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Replace kmalloc with kmalloc_array
[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>
39279cc3
CM		 46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
39279cc3 50#include "print-tree.h"
e6dcd2dc 51#include "ordered-data.h"
95819c05 52#include "xattr.h"
e02119d5 53#include "tree-log.h"
4a54c8c1 54#include "volumes.h"
c8b97818 55#include "compression.h"
b4ce94de 56#include "locking.h"
dc89e982 57#include "free-space-cache.h"
581bb050 58#include "inode-map.h"
38c227d8 59#include "backref.h"
f23b5a59 60#include "hash.h"
39279cc3
CM		 61
62struct btrfs_iget_args {
63 u64 ino;
64 struct btrfs_root *root;
65};
66
6e1d5dcc
AD		 67static const struct inode_operations btrfs_dir_inode_operations;
68static const struct inode_operations btrfs_symlink_inode_operations;
69static const struct inode_operations btrfs_dir_ro_inode_operations;
70static const struct inode_operations btrfs_special_inode_operations;
71static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 72static const struct address_space_operations btrfs_aops;
73static const struct address_space_operations btrfs_symlink_aops;
828c0950 74static const struct file_operations btrfs_dir_file_operations;
d1310b2e 75static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 76
77static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 78static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 79struct kmem_cache *btrfs_trans_handle_cachep;
80struct kmem_cache *btrfs_transaction_cachep;
39279cc3 81struct kmem_cache *btrfs_path_cachep;
dc89e982 82struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 83
84#define S_SHIFT 12
85static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
86 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
87 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
88 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
89 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
90 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
91 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
92 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
93};
94
3972f260 95static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 96static int btrfs_truncate(struct inode *inode);
5fd02043 97static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 98static noinline int cow_file_range(struct inode *inode,
99 struct page *locked_page,
100 u64 start, u64 end, int *page_started,
101 unsigned long *nr_written, int unlock);
70c8a91c
JB		 102static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
103 u64 len, u64 orig_start,
104 u64 block_start, u64 block_len,
cc95bef6
JB		 105 u64 orig_block_len, u64 ram_bytes,
106 int type);
7b128766 107
48a3b636 108static int btrfs_dirty_inode(struct inode *inode);
7b128766 109
f34f57a3 110static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 111 struct inode *inode, struct inode *dir,
112 const struct qstr *qstr)
0279b4cd
JO		 113{
114 int err;
115
f34f57a3 116 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 117 if (!err)
2a7dba39 118 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 119 return err;
120}
121
c8b97818
CM		 122/*
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
126 */
d397712b 127static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
fe3f566c 130 int compress_type,
c8b97818
CM		 131 struct page **compressed_pages)
132{
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
c8b97818 145
fe3f566c 146 if (compressed_size && compressed_pages)
c8b97818 147 cur_size = compressed_size;
c8b97818 148
d397712b
CM		 149 path = btrfs_alloc_path();
150 if (!path)
c8b97818
CM		 151 return -ENOMEM;
152
b9473439 153 path->leave_spinning = 1;
c8b97818 154
33345d01 155 key.objectid = btrfs_ino(inode);
c8b97818
CM		 156 key.offset = start;
157 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159
160 inode_add_bytes(inode, size);
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 datasize);
c8b97818
CM		 163 if (ret) {
164 err = ret;
c8b97818
CM		 165 goto fail;
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
261507a0 177 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 178 struct page *cpage;
179 int i = 0;
d397712b 180 while (compressed_size > 0) {
c8b97818 181 cpage = compressed_pages[i];
5b050f04 182 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 183 PAGE_CACHE_SIZE);
184
7ac687d9 185 kaddr = kmap_atomic(cpage);
c8b97818 186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 187 kunmap_atomic(kaddr);
c8b97818
CM		 188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
261507a0 194 compress_type);
c8b97818
CM		 195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 199 kaddr = kmap_atomic(page);
c8b97818
CM		 200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 202 kunmap_atomic(kaddr);
c8b97818
CM		 203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
207
c2167754
YZ		 208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
c8b97818 217 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 218 ret = btrfs_update_inode(trans, root, inode);
c2167754 219
79787eaa 220 return ret;
c8b97818
CM		 221fail:
222 btrfs_free_path(path);
223 return err;
224}
225
226
227/*
228 * conditionally insert an inline extent into the file. This
229 * does the checks required to make sure the data is small enough
230 * to fit as an inline extent.
231 */
00361589
JB		 232static noinline int cow_file_range_inline(struct btrfs_root *root,
233 struct inode *inode, u64 start,
234 u64 end, size_t compressed_size,
235 int compress_type,
236 struct page **compressed_pages)
c8b97818 237{
00361589 238 struct btrfs_trans_handle *trans;
c8b97818
CM		 239 u64 isize = i_size_read(inode);
240 u64 actual_end = min(end + 1, isize);
241 u64 inline_len = actual_end - start;
fda2832f 242 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 243 u64 data_len = inline_len;
244 int ret;
245
246 if (compressed_size)
247 data_len = compressed_size;
248
249 if (start > 0 ||
70b99e69 250 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 251 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
252 (!compressed_size &&
253 (actual_end & (root->sectorsize - 1)) == 0) ||
254 end + 1 < isize ||
255 data_len > root->fs_info->max_inline) {
256 return 1;
257 }
258
00361589
JB		 259 trans = btrfs_join_transaction(root);
260 if (IS_ERR(trans))
261 return PTR_ERR(trans);
262 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
263
2671485d 264 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
00361589
JB		 265 if (ret) {
266 btrfs_abort_transaction(trans, root, ret);
267 goto out;
268 }
c8b97818
CM		 269
270 if (isize > actual_end)
271 inline_len = min_t(u64, isize, actual_end);
272 ret = insert_inline_extent(trans, root, inode, start,
273 inline_len, compressed_size,
fe3f566c 274 compress_type, compressed_pages);
2adcac1a 275 if (ret && ret != -ENOSPC) {
79787eaa 276 btrfs_abort_transaction(trans, root, ret);
00361589 277 goto out;
2adcac1a 278 } else if (ret == -ENOSPC) {
00361589
JB		 279 ret = 1;
280 goto out;
79787eaa 281 }
2adcac1a 282
bdc20e67 283 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 284 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 285 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589
JB		 286out:
287 btrfs_end_transaction(trans, root);
288 return ret;
c8b97818
CM		 289}
290
771ed689
CM		 291struct async_extent {
292 u64 start;
293 u64 ram_size;
294 u64 compressed_size;
295 struct page **pages;
296 unsigned long nr_pages;
261507a0 297 int compress_type;
771ed689
CM		 298 struct list_head list;
299};
300
301struct async_cow {
302 struct inode *inode;
303 struct btrfs_root *root;
304 struct page *locked_page;
305 u64 start;
306 u64 end;
307 struct list_head extents;
308 struct btrfs_work work;
309};
310
311static noinline int add_async_extent(struct async_cow *cow,
312 u64 start, u64 ram_size,
313 u64 compressed_size,
314 struct page **pages,
261507a0
LZ		 315 unsigned long nr_pages,
316 int compress_type)
771ed689
CM		 317{
318 struct async_extent *async_extent;
319
320 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 321 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 322 async_extent->start = start;
323 async_extent->ram_size = ram_size;
324 async_extent->compressed_size = compressed_size;
325 async_extent->pages = pages;
326 async_extent->nr_pages = nr_pages;
261507a0 327 async_extent->compress_type = compress_type;
771ed689
CM		 328 list_add_tail(&async_extent->list, &cow->extents);
329 return 0;
330}
331
d352ac68 332/*
771ed689
CM		 333 * we create compressed extents in two phases. The first
334 * phase compresses a range of pages that have already been
335 * locked (both pages and state bits are locked).
c8b97818 336 *
771ed689
CM		 337 * This is done inside an ordered work queue, and the compression
338 * is spread across many cpus. The actual IO submission is step
339 * two, and the ordered work queue takes care of making sure that
340 * happens in the same order things were put onto the queue by
341 * writepages and friends.
c8b97818 342 *
771ed689
CM		 343 * If this code finds it can't get good compression, it puts an
344 * entry onto the work queue to write the uncompressed bytes. This
345 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 346 * are written in the same order that the flusher thread sent them
347 * down.
d352ac68 348 */
771ed689
CM		 349static noinline int compress_file_range(struct inode *inode,
350 struct page *locked_page,
351 u64 start, u64 end,
352 struct async_cow *async_cow,
353 int *num_added)
b888db2b
CM		 354{
355 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 356 u64 num_bytes;
db94535d 357 u64 blocksize = root->sectorsize;
c8b97818 358 u64 actual_end;
42dc7bab 359 u64 isize = i_size_read(inode);
e6dcd2dc 360 int ret = 0;
c8b97818
CM		 361 struct page **pages = NULL;
362 unsigned long nr_pages;
363 unsigned long nr_pages_ret = 0;
364 unsigned long total_compressed = 0;
365 unsigned long total_in = 0;
366 unsigned long max_compressed = 128 * 1024;
771ed689 367 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 368 int i;
369 int will_compress;
261507a0 370 int compress_type = root->fs_info->compress_type;
4adaa611 371 int redirty = 0;
b888db2b 372
4cb13e5d
LB		 373 /* if this is a small write inside eof, kick off a defrag */
374 if ((end - start + 1) < 16 * 1024 &&
375 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 376 btrfs_add_inode_defrag(NULL, inode);
377
42dc7bab 378 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 379again:
380 will_compress = 0;
381 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
382 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 383
f03d9301
CM		 384 /*
385 * we don't want to send crud past the end of i_size through
386 * compression, that's just a waste of CPU time. So, if the
387 * end of the file is before the start of our current
388 * requested range of bytes, we bail out to the uncompressed
389 * cleanup code that can deal with all of this.
390 *
391 * It isn't really the fastest way to fix things, but this is a
392 * very uncommon corner.
393 */
394 if (actual_end <= start)
395 goto cleanup_and_bail_uncompressed;
396
c8b97818
CM		 397 total_compressed = actual_end - start;
398
399 /* we want to make sure that amount of ram required to uncompress
400 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 401 * of a compressed extent to 128k. This is a crucial number
402 * because it also controls how easily we can spread reads across
403 * cpus for decompression.
404 *
405 * We also want to make sure the amount of IO required to do
406 * a random read is reasonably small, so we limit the size of
407 * a compressed extent to 128k.
c8b97818
CM		 408 */
409 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 410 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 411 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 412 total_in = 0;
413 ret = 0;
db94535d 414
771ed689
CM		 415 /*
416 * we do compression for mount -o compress and when the
417 * inode has not been flagged as nocompress. This flag can
418 * change at any time if we discover bad compression ratios.
c8b97818 419 */
6cbff00f 420 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 421 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 422 (BTRFS_I(inode)->force_compress) ||
423 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 424 WARN_ON(pages);
cfbc246e 425 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 426 if (!pages) {
427 /* just bail out to the uncompressed code */
428 goto cont;
429 }
c8b97818 430
261507a0
LZ		 431 if (BTRFS_I(inode)->force_compress)
432 compress_type = BTRFS_I(inode)->force_compress;
433
4adaa611
CM		 434 /*
435 * we need to call clear_page_dirty_for_io on each
436 * page in the range. Otherwise applications with the file
437 * mmap'd can wander in and change the page contents while
438 * we are compressing them.
439 *
440 * If the compression fails for any reason, we set the pages
441 * dirty again later on.
442 */
443 extent_range_clear_dirty_for_io(inode, start, end);
444 redirty = 1;
261507a0
LZ		 445 ret = btrfs_compress_pages(compress_type,
446 inode->i_mapping, start,
447 total_compressed, pages,
448 nr_pages, &nr_pages_ret,
449 &total_in,
450 &total_compressed,
451 max_compressed);
c8b97818
CM		 452
453 if (!ret) {
454 unsigned long offset = total_compressed &
455 (PAGE_CACHE_SIZE - 1);
456 struct page *page = pages[nr_pages_ret - 1];
457 char *kaddr;
458
459 /* zero the tail end of the last page, we might be
460 * sending it down to disk
461 */
462 if (offset) {
7ac687d9 463 kaddr = kmap_atomic(page);
c8b97818
CM		 464 memset(kaddr + offset, 0,
465 PAGE_CACHE_SIZE - offset);
7ac687d9 466 kunmap_atomic(kaddr);
c8b97818
CM		 467 }
468 will_compress = 1;
469 }
470 }
560f7d75 471cont:
c8b97818
CM		 472 if (start == 0) {
473 /* lets try to make an inline extent */
771ed689 474 if (ret || total_in < (actual_end - start)) {
c8b97818 475 /* we didn't compress the entire range, try
771ed689 476 * to make an uncompressed inline extent.
c8b97818 477 */
00361589
JB		 478 ret = cow_file_range_inline(root, inode, start, end,
479 0, 0, NULL);
c8b97818 480 } else {
771ed689 481 /* try making a compressed inline extent */
00361589 482 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 483 total_compressed,
484 compress_type, pages);
c8b97818 485 }
79787eaa 486 if (ret <= 0) {
151a41bc
JB		 487 unsigned long clear_flags = EXTENT_DELALLOC |
488 EXTENT_DEFRAG;
489 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
490
771ed689 491 /*
79787eaa
JM		 492 * inline extent creation worked or returned error,
493 * we don't need to create any more async work items.
494 * Unlock and free up our temp pages.
771ed689 495 */
c2790a2e 496 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 497 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 498 PAGE_CLEAR_DIRTY |
499 PAGE_SET_WRITEBACK |
500 PAGE_END_WRITEBACK);
c8b97818
CM		 501 goto free_pages_out;
502 }
503 }
504
505 if (will_compress) {
506 /*
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
509 * things
510 */
fda2832f 511 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 512
513 /*
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
516 */
fda2832f 517 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 518 if (total_compressed >= total_in) {
519 will_compress = 0;
520 } else {
c8b97818
CM		 521 num_bytes = total_in;
522 }
523 }
524 if (!will_compress && pages) {
525 /*
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
528 */
529 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 530 WARN_ON(pages[i]->mapping);
c8b97818
CM		 531 page_cache_release(pages[i]);
532 }
533 kfree(pages);
534 pages = NULL;
535 total_compressed = 0;
536 nr_pages_ret = 0;
537
538 /* flag the file so we don't compress in the future */
1e701a32
CM		 539 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
540 !(BTRFS_I(inode)->force_compress)) {
a555f810 541 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 542 }
c8b97818 543 }
771ed689
CM		 544 if (will_compress) {
545 *num_added += 1;
c8b97818 546
771ed689
CM		 547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
550 */
551 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 552 total_compressed, pages, nr_pages_ret,
553 compress_type);
179e29e4 554
24ae6365 555 if (start + num_bytes < end) {
771ed689
CM		 556 start += num_bytes;
557 pages = NULL;
558 cond_resched();
559 goto again;
560 }
561 } else {
f03d9301 562cleanup_and_bail_uncompressed:
771ed689
CM		 563 /*
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
569 */
570 if (page_offset(locked_page) >= start &&
571 page_offset(locked_page) <= end) {
572 __set_page_dirty_nobuffers(locked_page);
573 /* unlocked later on in the async handlers */
574 }
4adaa611
CM		 575 if (redirty)
576 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 577 add_async_extent(async_cow, start, end - start + 1,
578 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 579 *num_added += 1;
580 }
3b951516 581
771ed689 582out:
79787eaa 583 return ret;
771ed689
CM		 584
585free_pages_out:
586 for (i = 0; i < nr_pages_ret; i++) {
587 WARN_ON(pages[i]->mapping);
588 page_cache_release(pages[i]);
589 }
d397712b 590 kfree(pages);
771ed689
CM		 591
592 goto out;
593}
594
595/*
596 * phase two of compressed writeback. This is the ordered portion
597 * of the code, which only gets called in the order the work was
598 * queued. We walk all the async extents created by compress_file_range
599 * and send them down to the disk.
600 */
601static noinline int submit_compressed_extents(struct inode *inode,
602 struct async_cow *async_cow)
603{
604 struct async_extent *async_extent;
605 u64 alloc_hint = 0;
771ed689
CM		 606 struct btrfs_key ins;
607 struct extent_map *em;
608 struct btrfs_root *root = BTRFS_I(inode)->root;
609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
610 struct extent_io_tree *io_tree;
f5a84ee3 611 int ret = 0;
771ed689
CM		 612
613 if (list_empty(&async_cow->extents))
614 return 0;
615
3e04e7f1 616again:
d397712b 617 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 618 async_extent = list_entry(async_cow->extents.next,
619 struct async_extent, list);
620 list_del(&async_extent->list);
c8b97818 621
771ed689
CM		 622 io_tree = &BTRFS_I(inode)->io_tree;
623
f5a84ee3 624retry:
771ed689
CM		 625 /* did the compression code fall back to uncompressed IO? */
626 if (!async_extent->pages) {
627 int page_started = 0;
628 unsigned long nr_written = 0;
629
630 lock_extent(io_tree, async_extent->start,
2ac55d41 631 async_extent->start +
d0082371 632 async_extent->ram_size - 1);
771ed689
CM		 633
634 /* allocate blocks */
f5a84ee3
JB		 635 ret = cow_file_range(inode, async_cow->locked_page,
636 async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1,
639 &page_started, &nr_written, 0);
771ed689 640
79787eaa
JM		 641 /* JDM XXX */
642
771ed689
CM		 643 /*
644 * if page_started, cow_file_range inserted an
645 * inline extent and took care of all the unlocking
646 * and IO for us. Otherwise, we need to submit
647 * all those pages down to the drive.
648 */
f5a84ee3 649 if (!page_started && !ret)
771ed689
CM		 650 extent_write_locked_range(io_tree,
651 inode, async_extent->start,
d397712b 652 async_extent->start +
771ed689
CM		 653 async_extent->ram_size - 1,
654 btrfs_get_extent,
655 WB_SYNC_ALL);
3e04e7f1
JB		 656 else if (ret)
657 unlock_page(async_cow->locked_page);
771ed689
CM		 658 kfree(async_extent);
659 cond_resched();
660 continue;
661 }
662
663 lock_extent(io_tree, async_extent->start,
d0082371 664 async_extent->start + async_extent->ram_size - 1);
771ed689 665
00361589 666 ret = btrfs_reserve_extent(root,
771ed689
CM		 667 async_extent->compressed_size,
668 async_extent->compressed_size,
81c9ad23 669 0, alloc_hint, &ins, 1);
f5a84ee3
JB		 670 if (ret) {
671 int i;
3e04e7f1 672
f5a84ee3
JB		 673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
676 }
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
3e04e7f1 680
fdf8e2ea
JB		 681 if (ret == -ENOSPC) {
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1);
79787eaa 685 goto retry;
fdf8e2ea 686 }
3e04e7f1 687 goto out_free;
f5a84ee3
JB		 688 }
689
c2167754
YZ		 690 /*
691 * here we're doing allocation and writeback of the
692 * compressed pages
693 */
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
697
172ddd60 698 em = alloc_extent_map();
b9aa55be
LB		 699 if (!em) {
700 ret = -ENOMEM;
3e04e7f1 701 goto out_free_reserve;
b9aa55be 702 }
771ed689
CM		 703 em->start = async_extent->start;
704 em->len = async_extent->ram_size;
445a6944 705 em->orig_start = em->start;
2ab28f32
JB		 706 em->mod_start = em->start;
707 em->mod_len = em->len;
c8b97818 708
771ed689
CM		 709 em->block_start = ins.objectid;
710 em->block_len = ins.offset;
b4939680 711 em->orig_block_len = ins.offset;
cc95bef6 712 em->ram_bytes = async_extent->ram_size;
771ed689 713 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 714 em->compress_type = async_extent->compress_type;
771ed689
CM		 715 set_bit(EXTENT_FLAG_PINNED, &em->flags);
716 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 717 em->generation = -1;
771ed689 718
d397712b 719 while (1) {
890871be 720 write_lock(&em_tree->lock);
09a2a8f9 721 ret = add_extent_mapping(em_tree, em, 1);
890871be 722 write_unlock(&em_tree->lock);
771ed689
CM		 723 if (ret != -EEXIST) {
724 free_extent_map(em);
725 break;
726 }
727 btrfs_drop_extent_cache(inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1, 0);
730 }
731
3e04e7f1
JB		 732 if (ret)
733 goto out_free_reserve;
734
261507a0
LZ		 735 ret = btrfs_add_ordered_extent_compress(inode,
736 async_extent->start,
737 ins.objectid,
738 async_extent->ram_size,
739 ins.offset,
740 BTRFS_ORDERED_COMPRESSED,
741 async_extent->compress_type);
3e04e7f1
JB		 742 if (ret)
743 goto out_free_reserve;
771ed689 744
771ed689
CM		 745 /*
746 * clear dirty, set writeback and unlock the pages.
747 */
c2790a2e 748 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 749 async_extent->start +
750 async_extent->ram_size - 1,
151a41bc
JB		 751 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
752 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 753 PAGE_SET_WRITEBACK);
771ed689 754 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 755 async_extent->start,
756 async_extent->ram_size,
757 ins.objectid,
758 ins.offset, async_extent->pages,
759 async_extent->nr_pages);
771ed689
CM		 760 alloc_hint = ins.objectid + ins.offset;
761 kfree(async_extent);
3e04e7f1
JB		 762 if (ret)
763 goto out;
771ed689
CM		 764 cond_resched();
765 }
79787eaa
JM		 766 ret = 0;
767out:
768 return ret;
3e04e7f1
JB		 769out_free_reserve:
770 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 771out_free:
c2790a2e 772 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 773 async_extent->start +
774 async_extent->ram_size - 1,
c2790a2e 775 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 776 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
777 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
778 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 779 kfree(async_extent);
3e04e7f1 780 goto again;
771ed689
CM		 781}
782
4b46fce2
JB		 783static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
784 u64 num_bytes)
785{
786 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
787 struct extent_map *em;
788 u64 alloc_hint = 0;
789
790 read_lock(&em_tree->lock);
791 em = search_extent_mapping(em_tree, start, num_bytes);
792 if (em) {
793 /*
794 * if block start isn't an actual block number then find the
795 * first block in this inode and use that as a hint. If that
796 * block is also bogus then just don't worry about it.
797 */
798 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
799 free_extent_map(em);
800 em = search_extent_mapping(em_tree, 0, 0);
801 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
802 alloc_hint = em->block_start;
803 if (em)
804 free_extent_map(em);
805 } else {
806 alloc_hint = em->block_start;
807 free_extent_map(em);
808 }
809 }
810 read_unlock(&em_tree->lock);
811
812 return alloc_hint;
813}
814
771ed689
CM		 815/*
816 * when extent_io.c finds a delayed allocation range in the file,
817 * the call backs end up in this code. The basic idea is to
818 * allocate extents on disk for the range, and create ordered data structs
819 * in ram to track those extents.
820 *
821 * locked_page is the page that writepage had locked already. We use
822 * it to make sure we don't do extra locks or unlocks.
823 *
824 * *page_started is set to one if we unlock locked_page and do everything
825 * required to start IO on it. It may be clean and already done with
826 * IO when we return.
827 */
00361589
JB		 828static noinline int cow_file_range(struct inode *inode,
829 struct page *locked_page,
830 u64 start, u64 end, int *page_started,
831 unsigned long *nr_written,
832 int unlock)
771ed689 833{
00361589 834 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 835 u64 alloc_hint = 0;
836 u64 num_bytes;
837 unsigned long ram_size;
838 u64 disk_num_bytes;
839 u64 cur_alloc_size;
840 u64 blocksize = root->sectorsize;
771ed689
CM		 841 struct btrfs_key ins;
842 struct extent_map *em;
843 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
844 int ret = 0;
845
02ecd2c2
JB		 846 if (btrfs_is_free_space_inode(inode)) {
847 WARN_ON_ONCE(1);
848 return -EINVAL;
849 }
771ed689 850
fda2832f 851 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 852 num_bytes = max(blocksize, num_bytes);
853 disk_num_bytes = num_bytes;
771ed689 854
4cb5300b 855 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 856 if (num_bytes < 64 * 1024 &&
857 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 858 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 859
771ed689
CM		 860 if (start == 0) {
861 /* lets try to make an inline extent */
00361589
JB		 862 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
863 NULL);
771ed689 864 if (ret == 0) {
c2790a2e
JB		 865 extent_clear_unlock_delalloc(inode, start, end, NULL,
866 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 867 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 868 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
869 PAGE_END_WRITEBACK);
c2167754 870
771ed689
CM		 871 *nr_written = *nr_written +
872 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 *page_started = 1;
771ed689 874 goto out;
79787eaa 875 } else if (ret < 0) {
79787eaa 876 goto out_unlock;
771ed689
CM		 877 }
878 }
879
880 BUG_ON(disk_num_bytes >
6c41761f 881 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 882
4b46fce2 883 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 884 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
885
d397712b 886 while (disk_num_bytes > 0) {
a791e35e
CM		 887 unsigned long op;
888
287a0ab9 889 cur_alloc_size = disk_num_bytes;
00361589 890 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 891 root->sectorsize, 0, alloc_hint,
81c9ad23 892 &ins, 1);
00361589 893 if (ret < 0)
79787eaa 894 goto out_unlock;
d397712b 895
172ddd60 896 em = alloc_extent_map();
b9aa55be
LB		 897 if (!em) {
898 ret = -ENOMEM;
ace68bac 899 goto out_reserve;
b9aa55be 900 }
e6dcd2dc 901 em->start = start;
445a6944 902 em->orig_start = em->start;
771ed689
CM		 903 ram_size = ins.offset;
904 em->len = ins.offset;
2ab28f32
JB		 905 em->mod_start = em->start;
906 em->mod_len = em->len;
c8b97818 907
e6dcd2dc 908 em->block_start = ins.objectid;
c8b97818 909 em->block_len = ins.offset;
b4939680 910 em->orig_block_len = ins.offset;
cc95bef6 911 em->ram_bytes = ram_size;
e6dcd2dc 912 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 913 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 914 em->generation = -1;
c8b97818 915
d397712b 916 while (1) {
890871be 917 write_lock(&em_tree->lock);
09a2a8f9 918 ret = add_extent_mapping(em_tree, em, 1);
890871be 919 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 920 if (ret != -EEXIST) {
921 free_extent_map(em);
922 break;
923 }
924 btrfs_drop_extent_cache(inode, start,
c8b97818 925 start + ram_size - 1, 0);
e6dcd2dc 926 }
ace68bac
LB		 927 if (ret)
928 goto out_reserve;
e6dcd2dc 929
98d20f67 930 cur_alloc_size = ins.offset;
e6dcd2dc 931 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 932 ram_size, cur_alloc_size, 0);
ace68bac
LB		 933 if (ret)
934 goto out_reserve;
c8b97818 935
17d217fe
YZ		 936 if (root->root_key.objectid ==
937 BTRFS_DATA_RELOC_TREE_OBJECTID) {
938 ret = btrfs_reloc_clone_csums(inode, start,
939 cur_alloc_size);
00361589 940 if (ret)
ace68bac 941 goto out_reserve;
17d217fe
YZ		 942 }
943
d397712b 944 if (disk_num_bytes < cur_alloc_size)
3b951516 945 break;
d397712b 946
c8b97818
CM		 947 /* we're not doing compressed IO, don't unlock the first
948 * page (which the caller expects to stay locked), don't
949 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 950 *
951 * Do set the Private2 bit so we know this page was properly
952 * setup for writepage
c8b97818 953 */
c2790a2e
JB		 954 op = unlock ? PAGE_UNLOCK : 0;
955 op |= PAGE_SET_PRIVATE2;
a791e35e 956
c2790a2e
JB		 957 extent_clear_unlock_delalloc(inode, start,
958 start + ram_size - 1, locked_page,
959 EXTENT_LOCKED | EXTENT_DELALLOC,
960 op);
c8b97818 961 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 962 num_bytes -= cur_alloc_size;
963 alloc_hint = ins.objectid + ins.offset;
964 start += cur_alloc_size;
b888db2b 965 }
79787eaa 966out:
be20aa9d 967 return ret;
b7d5b0a8 968
ace68bac
LB		 969out_reserve:
970 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 971out_unlock:
c2790a2e 972 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 973 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
974 EXTENT_DELALLOC | EXTENT_DEFRAG,
975 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
976 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 977 goto out;
771ed689 978}
c8b97818 979
771ed689
CM		 980/*
981 * work queue call back to started compression on a file and pages
982 */
983static noinline void async_cow_start(struct btrfs_work *work)
984{
985 struct async_cow *async_cow;
986 int num_added = 0;
987 async_cow = container_of(work, struct async_cow, work);
988
989 compress_file_range(async_cow->inode, async_cow->locked_page,
990 async_cow->start, async_cow->end, async_cow,
991 &num_added);
8180ef88 992 if (num_added == 0) {
cb77fcd8 993 btrfs_add_delayed_iput(async_cow->inode);
771ed689 994 async_cow->inode = NULL;
8180ef88 995 }
771ed689
CM		 996}
997
998/*
999 * work queue call back to submit previously compressed pages
1000 */
1001static noinline void async_cow_submit(struct btrfs_work *work)
1002{
1003 struct async_cow *async_cow;
1004 struct btrfs_root *root;
1005 unsigned long nr_pages;
1006
1007 async_cow = container_of(work, struct async_cow, work);
1008
1009 root = async_cow->root;
1010 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1011 PAGE_CACHE_SHIFT;
1012
66657b31 1013 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1014 5 * 1024 * 1024 &&
771ed689
CM		 1015 waitqueue_active(&root->fs_info->async_submit_wait))
1016 wake_up(&root->fs_info->async_submit_wait);
1017
d397712b 1018 if (async_cow->inode)
771ed689 1019 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1020}
c8b97818 1021
771ed689
CM		 1022static noinline void async_cow_free(struct btrfs_work *work)
1023{
1024 struct async_cow *async_cow;
1025 async_cow = container_of(work, struct async_cow, work);
8180ef88 1026 if (async_cow->inode)
cb77fcd8 1027 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1028 kfree(async_cow);
1029}
1030
1031static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032 u64 start, u64 end, int *page_started,
1033 unsigned long *nr_written)
1034{
1035 struct async_cow *async_cow;
1036 struct btrfs_root *root = BTRFS_I(inode)->root;
1037 unsigned long nr_pages;
1038 u64 cur_end;
287082b0 1039 int limit = 10 * 1024 * 1024;
771ed689 1040
a3429ab7
CM		 1041 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042 1, 0, NULL, GFP_NOFS);
d397712b 1043 while (start < end) {
771ed689 1044 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1045 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1046 async_cow->inode = igrab(inode);
771ed689
CM		 1047 async_cow->root = root;
1048 async_cow->locked_page = locked_page;
1049 async_cow->start = start;
1050
6cbff00f 1051 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1052 cur_end = end;
1053 else
1054 cur_end = min(end, start + 512 * 1024 - 1);
1055
1056 async_cow->end = cur_end;
1057 INIT_LIST_HEAD(&async_cow->extents);
1058
1059 async_cow->work.func = async_cow_start;
1060 async_cow->work.ordered_func = async_cow_submit;
1061 async_cow->work.ordered_free = async_cow_free;
1062 async_cow->work.flags = 0;
1063
771ed689
CM		 1064 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065 PAGE_CACHE_SHIFT;
1066 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067
1068 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069 &async_cow->work);
1070
1071 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072 wait_event(root->fs_info->async_submit_wait,
1073 (atomic_read(&root->fs_info->async_delalloc_pages) <
1074 limit));
1075 }
1076
d397712b 1077 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1078 atomic_read(&root->fs_info->async_delalloc_pages)) {
1079 wait_event(root->fs_info->async_submit_wait,
1080 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1081 0));
1082 }
1083
1084 *nr_written += nr_pages;
1085 start = cur_end + 1;
1086 }
1087 *page_started = 1;
1088 return 0;
be20aa9d
CM		 1089}
1090
d397712b 1091static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1092 u64 bytenr, u64 num_bytes)
1093{
1094 int ret;
1095 struct btrfs_ordered_sum *sums;
1096 LIST_HEAD(list);
1097
07d400a6 1098 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1099 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1100 if (ret == 0 && list_empty(&list))
1101 return 0;
1102
1103 while (!list_empty(&list)) {
1104 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105 list_del(&sums->list);
1106 kfree(sums);
1107 }
1108 return 1;
1109}
1110
d352ac68
CM		 1111/*
1112 * when nowcow writeback call back. This checks for snapshots or COW copies
1113 * of the extents that exist in the file, and COWs the file as required.
1114 *
1115 * If no cow copies or snapshots exist, we write directly to the existing
1116 * blocks on disk
1117 */
7f366cfe
CM		 1118static noinline int run_delalloc_nocow(struct inode *inode,
1119 struct page *locked_page,
771ed689
CM		 1120 u64 start, u64 end, int *page_started, int force,
1121 unsigned long *nr_written)
be20aa9d 1122{
be20aa9d 1123 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1124 struct btrfs_trans_handle *trans;
be20aa9d 1125 struct extent_buffer *leaf;
be20aa9d 1126 struct btrfs_path *path;
80ff3856 1127 struct btrfs_file_extent_item *fi;
be20aa9d 1128 struct btrfs_key found_key;
80ff3856
YZ		 1129 u64 cow_start;
1130 u64 cur_offset;
1131 u64 extent_end;
5d4f98a2 1132 u64 extent_offset;
80ff3856
YZ		 1133 u64 disk_bytenr;
1134 u64 num_bytes;
b4939680 1135 u64 disk_num_bytes;
cc95bef6 1136 u64 ram_bytes;
80ff3856 1137 int extent_type;
79787eaa 1138 int ret, err;
d899e052 1139 int type;
80ff3856
YZ		 1140 int nocow;
1141 int check_prev = 1;
82d5902d 1142 bool nolock;
33345d01 1143 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1144
1145 path = btrfs_alloc_path();
17ca04af 1146 if (!path) {
c2790a2e
JB		 1147 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1148 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1149 EXTENT_DO_ACCOUNTING |
1150 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1151 PAGE_CLEAR_DIRTY |
1152 PAGE_SET_WRITEBACK |
1153 PAGE_END_WRITEBACK);
d8926bb3 1154 return -ENOMEM;
17ca04af 1155 }
82d5902d 1156
83eea1f1 1157 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1158
1159 if (nolock)
7a7eaa40 1160 trans = btrfs_join_transaction_nolock(root);
82d5902d 1161 else
7a7eaa40 1162 trans = btrfs_join_transaction(root);
ff5714cc 1163
79787eaa 1164 if (IS_ERR(trans)) {
c2790a2e
JB		 1165 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1167 EXTENT_DO_ACCOUNTING |
1168 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1169 PAGE_CLEAR_DIRTY |
1170 PAGE_SET_WRITEBACK |
1171 PAGE_END_WRITEBACK);
79787eaa
JM		 1172 btrfs_free_path(path);
1173 return PTR_ERR(trans);
1174 }
1175
74b21075 1176 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1177
80ff3856
YZ		 1178 cow_start = (u64)-1;
1179 cur_offset = start;
1180 while (1) {
33345d01 1181 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1182 cur_offset, 0);
d788a349 1183 if (ret < 0)
79787eaa 1184 goto error;
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);
d788a349 1198 if (ret < 0)
79787eaa 1199 goto error;
80ff3856
YZ		 1200 if (ret > 0)
1201 break;
1202 leaf = path->nodes[0];
1203 }
be20aa9d 1204
80ff3856
YZ		 1205 nocow = 0;
1206 disk_bytenr = 0;
17d217fe 1207 num_bytes = 0;
80ff3856
YZ		 1208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1209
33345d01 1210 if (found_key.objectid > ino ||
80ff3856
YZ		 1211 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1212 found_key.offset > end)
1213 break;
1214
1215 if (found_key.offset > cur_offset) {
1216 extent_end = found_key.offset;
e9061e21 1217 extent_type = 0;
80ff3856
YZ		 1218 goto out_check;
1219 }
1220
1221 fi = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_file_extent_item);
1223 extent_type = btrfs_file_extent_type(leaf, fi);
1224
cc95bef6 1225 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1226 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1227 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1228 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1229 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1230 extent_end = found_key.offset +
1231 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1232 disk_num_bytes =
1233 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1234 if (extent_end <= start) {
1235 path->slots[0]++;
1236 goto next_slot;
1237 }
17d217fe
YZ		 1238 if (disk_bytenr == 0)
1239 goto out_check;
80ff3856
YZ		 1240 if (btrfs_file_extent_compression(leaf, fi) ||
1241 btrfs_file_extent_encryption(leaf, fi) ||
1242 btrfs_file_extent_other_encoding(leaf, fi))
1243 goto out_check;
d899e052
YZ		 1244 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1245 goto out_check;
d2fb3437 1246 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1247 goto out_check;
33345d01 1248 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1249 found_key.offset -
1250 extent_offset, disk_bytenr))
17d217fe 1251 goto out_check;
5d4f98a2 1252 disk_bytenr += extent_offset;
17d217fe
YZ		 1253 disk_bytenr += cur_offset - found_key.offset;
1254 num_bytes = min(end + 1, extent_end) - cur_offset;
1255 /*
1256 * force cow if csum exists in the range.
1257 * this ensure that csum for a given extent are
1258 * either valid or do not exist.
1259 */
1260 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1261 goto out_check;
80ff3856
YZ		 1262 nocow = 1;
1263 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1264 extent_end = found_key.offset +
1265 btrfs_file_extent_inline_len(leaf, fi);
1266 extent_end = ALIGN(extent_end, root->sectorsize);
1267 } else {
1268 BUG_ON(1);
1269 }
1270out_check:
1271 if (extent_end <= start) {
1272 path->slots[0]++;
1273 goto next_slot;
1274 }
1275 if (!nocow) {
1276 if (cow_start == (u64)-1)
1277 cow_start = cur_offset;
1278 cur_offset = extent_end;
1279 if (cur_offset > end)
1280 break;
1281 path->slots[0]++;
1282 goto next_slot;
7ea394f1
YZ		 1283 }
1284
b3b4aa74 1285 btrfs_release_path(path);
80ff3856 1286 if (cow_start != (u64)-1) {
00361589
JB		 1287 ret = cow_file_range(inode, locked_page,
1288 cow_start, found_key.offset - 1,
1289 page_started, nr_written, 1);
d788a349 1290 if (ret)
79787eaa 1291 goto error;
80ff3856 1292 cow_start = (u64)-1;
7ea394f1 1293 }
80ff3856 1294
d899e052
YZ		 1295 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296 struct extent_map *em;
1297 struct extent_map_tree *em_tree;
1298 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1299 em = alloc_extent_map();
79787eaa 1300 BUG_ON(!em); /* -ENOMEM */
d899e052 1301 em->start = cur_offset;
70c8a91c 1302 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1303 em->len = num_bytes;
1304 em->block_len = num_bytes;
1305 em->block_start = disk_bytenr;
b4939680 1306 em->orig_block_len = disk_num_bytes;
cc95bef6 1307 em->ram_bytes = ram_bytes;
d899e052 1308 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1309 em->mod_start = em->start;
1310 em->mod_len = em->len;
d899e052 1311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1312 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1313 em->generation = -1;
d899e052 1314 while (1) {
890871be 1315 write_lock(&em_tree->lock);
09a2a8f9 1316 ret = add_extent_mapping(em_tree, em, 1);
890871be 1317 write_unlock(&em_tree->lock);
d899e052
YZ		 1318 if (ret != -EEXIST) {
1319 free_extent_map(em);
1320 break;
1321 }
1322 btrfs_drop_extent_cache(inode, em->start,
1323 em->start + em->len - 1, 0);
1324 }
1325 type = BTRFS_ORDERED_PREALLOC;
1326 } else {
1327 type = BTRFS_ORDERED_NOCOW;
1328 }
80ff3856
YZ		 1329
1330 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1331 num_bytes, num_bytes, type);
79787eaa 1332 BUG_ON(ret); /* -ENOMEM */
771ed689 1333
efa56464
YZ		 1334 if (root->root_key.objectid ==
1335 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1336 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1337 num_bytes);
d788a349 1338 if (ret)
79787eaa 1339 goto error;
efa56464
YZ		 1340 }
1341
c2790a2e
JB		 1342 extent_clear_unlock_delalloc(inode, cur_offset,
1343 cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_LOCKED |
1345 EXTENT_DELALLOC, PAGE_UNLOCK |
1346 PAGE_SET_PRIVATE2);
80ff3856
YZ		 1347 cur_offset = extent_end;
1348 if (cur_offset > end)
1349 break;
be20aa9d 1350 }
b3b4aa74 1351 btrfs_release_path(path);
80ff3856 1352
17ca04af 1353 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1354 cow_start = cur_offset;
17ca04af
JB		 1355 cur_offset = end;
1356 }
1357
80ff3856 1358 if (cow_start != (u64)-1) {
00361589
JB		 1359 ret = cow_file_range(inode, locked_page, cow_start, end,
1360 page_started, nr_written, 1);
d788a349 1361 if (ret)
79787eaa 1362 goto error;
80ff3856
YZ		 1363 }
1364
79787eaa 1365error:
a698d075 1366 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1367 if (!ret)
1368 ret = err;
1369
17ca04af 1370 if (ret && cur_offset < end)
c2790a2e
JB		 1371 extent_clear_unlock_delalloc(inode, cur_offset, end,
1372 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1373 EXTENT_DELALLOC | EXTENT_DEFRAG |
1374 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1375 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1376 PAGE_SET_WRITEBACK |
1377 PAGE_END_WRITEBACK);
7ea394f1 1378 btrfs_free_path(path);
79787eaa 1379 return ret;
be20aa9d
CM		 1380}
1381
d352ac68
CM		 1382/*
1383 * extent_io.c call back to do delayed allocation processing
1384 */
c8b97818 1385static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1386 u64 start, u64 end, int *page_started,
1387 unsigned long *nr_written)
be20aa9d 1388{
be20aa9d 1389 int ret;
7f366cfe 1390 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1391
7ddf5a42 1392 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1393 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1394 page_started, 1, nr_written);
7ddf5a42 1395 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1396 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1397 page_started, 0, nr_written);
7ddf5a42
JB		 1398 } else if (!btrfs_test_opt(root, COMPRESS) &&
1399 !(BTRFS_I(inode)->force_compress) &&
1400 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1401 ret = cow_file_range(inode, locked_page, start, end,
1402 page_started, nr_written, 1);
7ddf5a42
JB		 1403 } else {
1404 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1405 &BTRFS_I(inode)->runtime_flags);
771ed689 1406 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1407 page_started, nr_written);
7ddf5a42 1408 }
b888db2b
CM		 1409 return ret;
1410}
1411
1bf85046
JM		 1412static void btrfs_split_extent_hook(struct inode *inode,
1413 struct extent_state *orig, u64 split)
9ed74f2d 1414{
0ca1f7ce 1415 /* not delalloc, ignore it */
9ed74f2d 1416 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1417 return;
9ed74f2d 1418
9e0baf60
JB		 1419 spin_lock(&BTRFS_I(inode)->lock);
1420 BTRFS_I(inode)->outstanding_extents++;
1421 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1422}
1423
1424/*
1425 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1426 * extents so we can keep track of new extents that are just merged onto old
1427 * extents, such as when we are doing sequential writes, so we can properly
1428 * account for the metadata space we'll need.
1429 */
1bf85046
JM		 1430static void btrfs_merge_extent_hook(struct inode *inode,
1431 struct extent_state *new,
1432 struct extent_state *other)
9ed74f2d 1433{
9ed74f2d
JB		 1434 /* not delalloc, ignore it */
1435 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1436 return;
9ed74f2d 1437
9e0baf60
JB		 1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents--;
1440 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1441}
1442
eb73c1b7
MX		 1443static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1444 struct inode *inode)
1445{
1446 spin_lock(&root->delalloc_lock);
1447 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1448 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1449 &root->delalloc_inodes);
1450 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1451 &BTRFS_I(inode)->runtime_flags);
1452 root->nr_delalloc_inodes++;
1453 if (root->nr_delalloc_inodes == 1) {
1454 spin_lock(&root->fs_info->delalloc_root_lock);
1455 BUG_ON(!list_empty(&root->delalloc_root));
1456 list_add_tail(&root->delalloc_root,
1457 &root->fs_info->delalloc_roots);
1458 spin_unlock(&root->fs_info->delalloc_root_lock);
1459 }
1460 }
1461 spin_unlock(&root->delalloc_lock);
1462}
1463
1464static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1465 struct inode *inode)
1466{
1467 spin_lock(&root->delalloc_lock);
1468 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1470 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471 &BTRFS_I(inode)->runtime_flags);
1472 root->nr_delalloc_inodes--;
1473 if (!root->nr_delalloc_inodes) {
1474 spin_lock(&root->fs_info->delalloc_root_lock);
1475 BUG_ON(list_empty(&root->delalloc_root));
1476 list_del_init(&root->delalloc_root);
1477 spin_unlock(&root->fs_info->delalloc_root_lock);
1478 }
1479 }
1480 spin_unlock(&root->delalloc_lock);
1481}
1482
d352ac68
CM		 1483/*
1484 * extent_io.c set_bit_hook, used to track delayed allocation
1485 * bytes in this file, and to maintain the list of inodes that
1486 * have pending delalloc work to be done.
1487 */
1bf85046 1488static void btrfs_set_bit_hook(struct inode *inode,
41074888 1489 struct extent_state *state, unsigned long *bits)
291d673e 1490{
9ed74f2d 1491
75eff68e
CM		 1492 /*
1493 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1494 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1495 * bit, which is only set or cleared with irqs on
1496 */
0ca1f7ce 1497 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1498 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1499 u64 len = state->end + 1 - state->start;
83eea1f1 1500 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1501
9e0baf60 1502 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1503 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1504 } else {
1505 spin_lock(&BTRFS_I(inode)->lock);
1506 BTRFS_I(inode)->outstanding_extents++;
1507 spin_unlock(&BTRFS_I(inode)->lock);
1508 }
287a0ab9 1509
963d678b
MX		 1510 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1511 root->fs_info->delalloc_batch);
df0af1a5 1512 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1513 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5 1514 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1515 &BTRFS_I(inode)->runtime_flags))
1516 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1517 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1518 }
291d673e
CM		 1519}
1520
d352ac68
CM		 1521/*
1522 * extent_io.c clear_bit_hook, see set_bit_hook for why
1523 */
1bf85046 1524static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1525 struct extent_state *state,
1526 unsigned long *bits)
291d673e 1527{
75eff68e
CM		 1528 /*
1529 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1530 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1531 * bit, which is only set or cleared with irqs on
1532 */
0ca1f7ce 1533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1534 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1535 u64 len = state->end + 1 - state->start;
83eea1f1 1536 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1537
9e0baf60 1538 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1539 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1544 }
0ca1f7ce 1545
b6d08f06
JB		 1546 /*
1547 * We don't reserve metadata space for space cache inodes so we
1548 * don't need to call dellalloc_release_metadata if there is an
1549 * error.
1550 */
1551 if (*bits & EXTENT_DO_ACCOUNTING &&
1552 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1553 btrfs_delalloc_release_metadata(inode, len);
1554
0cb59c99 1555 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1556 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1557 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1558
963d678b
MX		 1559 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1560 root->fs_info->delalloc_batch);
df0af1a5 1561 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1562 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1563 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1564 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1565 &BTRFS_I(inode)->runtime_flags))
1566 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1567 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1568 }
291d673e
CM		 1569}
1570
d352ac68
CM		 1571/*
1572 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573 * we don't create bios that span stripes or chunks
1574 */
64a16701 1575int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1576 size_t size, struct bio *bio,
1577 unsigned long bio_flags)
239b14b3
CM		 1578{
1579 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1580 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1581 u64 length = 0;
1582 u64 map_length;
239b14b3
CM		 1583 int ret;
1584
771ed689
CM		 1585 if (bio_flags & EXTENT_BIO_COMPRESSED)
1586 return 0;
1587
f2d8d74d 1588 length = bio->bi_size;
239b14b3 1589 map_length = length;
64a16701 1590 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1591 &map_length, NULL, 0);
3ec706c8 1592 /* Will always return 0 with map_multi == NULL */
3444a972 1593 BUG_ON(ret < 0);
d397712b 1594 if (map_length < length + size)
239b14b3 1595 return 1;
3444a972 1596 return 0;
239b14b3
CM		 1597}
1598
d352ac68
CM		 1599/*
1600 * in order to insert checksums into the metadata in large chunks,
1601 * we wait until bio submission time. All the pages in the bio are
1602 * checksummed and sums are attached onto the ordered extent record.
1603 *
1604 * At IO completion time the cums attached on the ordered extent record
1605 * are inserted into the btree
1606 */
d397712b
CM		 1607static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608 struct bio *bio, int mirror_num,
eaf25d93
CM		 1609 unsigned long bio_flags,
1610 u64 bio_offset)
065631f6 1611{
065631f6 1612 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1613 int ret = 0;
e015640f 1614
d20f7043 1615 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1616 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1617 return 0;
1618}
e015640f 1619
4a69a410
CM		 1620/*
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1624 *
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1627 */
b2950863 1628static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1629 int mirror_num, unsigned long bio_flags,
1630 u64 bio_offset)
4a69a410
CM		 1631{
1632 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1633 int ret;
1634
1635 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1636 if (ret)
1637 bio_endio(bio, ret);
1638 return ret;
44b8bd7e
CM		 1639}
1640
d352ac68 1641/*
cad321ad
CM		 1642 * extent_io.c submission hook. This does the right thing for csum calculation
1643 * on write, or reading the csums from the tree before a read
d352ac68 1644 */
b2950863 1645static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1646 int mirror_num, unsigned long bio_flags,
1647 u64 bio_offset)
44b8bd7e
CM		 1648{
1649 struct btrfs_root *root = BTRFS_I(inode)->root;
1650 int ret = 0;
19b9bdb0 1651 int skip_sum;
0417341e 1652 int metadata = 0;
b812ce28 1653 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1654
6cbff00f 1655 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1656
83eea1f1 1657 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1658 metadata = 2;
1659
7b6d91da 1660 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1661 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1662 if (ret)
61891923 1663 goto out;
5fd02043 1664
d20f7043 1665 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1666 ret = btrfs_submit_compressed_read(inode, bio,
1667 mirror_num,
1668 bio_flags);
1669 goto out;
c2db1073
TI		 1670 } else if (!skip_sum) {
1671 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1672 if (ret)
61891923 1673 goto out;
c2db1073 1674 }
4d1b5fb4 1675 goto mapit;
b812ce28 1676 } else if (async && !skip_sum) {
17d217fe
YZ		 1677 /* csum items have already been cloned */
1678 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1679 goto mapit;
19b9bdb0 1680 /* we're doing a write, do the async checksumming */
61891923 1681 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1682 inode, rw, bio, mirror_num,
eaf25d93
CM		 1683 bio_flags, bio_offset,
1684 __btrfs_submit_bio_start,
4a69a410 1685 __btrfs_submit_bio_done);
61891923 1686 goto out;
b812ce28
JB		 1687 } else if (!skip_sum) {
1688 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1689 if (ret)
1690 goto out;
19b9bdb0
CM		 1691 }
1692
0b86a832 1693mapit:
61891923
SB		 1694 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1695
1696out:
1697 if (ret < 0)
1698 bio_endio(bio, ret);
1699 return ret;
065631f6 1700}
6885f308 1701
d352ac68
CM		 1702/*
1703 * given a list of ordered sums record them in the inode. This happens
1704 * at IO completion time based on sums calculated at bio submission time.
1705 */
ba1da2f4 1706static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1707 struct inode *inode, u64 file_offset,
1708 struct list_head *list)
1709{
e6dcd2dc
CM		 1710 struct btrfs_ordered_sum *sum;
1711
c6e30871 1712 list_for_each_entry(sum, list, list) {
39847c4d 1713 trans->adding_csums = 1;
d20f7043
CM		 1714 btrfs_csum_file_blocks(trans,
1715 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1716 trans->adding_csums = 0;
e6dcd2dc
CM		 1717 }
1718 return 0;
1719}
1720
2ac55d41
JB		 1721int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722 struct extent_state **cached_state)
ea8c2819 1723{
6c1500f2 1724 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1725 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1726 cached_state, GFP_NOFS);
ea8c2819
CM		 1727}
1728
d352ac68 1729/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1730struct btrfs_writepage_fixup {
1731 struct page *page;
1732 struct btrfs_work work;
1733};
1734
b2950863 1735static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1736{
1737 struct btrfs_writepage_fixup *fixup;
1738 struct btrfs_ordered_extent *ordered;
2ac55d41 1739 struct extent_state *cached_state = NULL;
247e743c
CM		 1740 struct page *page;
1741 struct inode *inode;
1742 u64 page_start;
1743 u64 page_end;
87826df0 1744 int ret;
247e743c
CM		 1745
1746 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1747 page = fixup->page;
4a096752 1748again:
247e743c
CM		 1749 lock_page(page);
1750 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751 ClearPageChecked(page);
1752 goto out_page;
1753 }
1754
1755 inode = page->mapping->host;
1756 page_start = page_offset(page);
1757 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1758
2ac55d41 1759 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1760 &cached_state);
4a096752
CM		 1761
1762 /* already ordered? We're done */
8b62b72b 1763 if (PagePrivate2(page))
247e743c 1764 goto out;
4a096752
CM		 1765
1766 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1767 if (ordered) {
2ac55d41
JB		 1768 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1770 unlock_page(page);
1771 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1772 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1773 goto again;
1774 }
247e743c 1775
87826df0
JM		 1776 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1777 if (ret) {
1778 mapping_set_error(page->mapping, ret);
1779 end_extent_writepage(page, ret, page_start, page_end);
1780 ClearPageChecked(page);
1781 goto out;
1782 }
1783
2ac55d41 1784 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1785 ClearPageChecked(page);
87826df0 1786 set_page_dirty(page);
247e743c 1787out:
2ac55d41
JB		 1788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789 &cached_state, GFP_NOFS);
247e743c
CM		 1790out_page:
1791 unlock_page(page);
1792 page_cache_release(page);
b897abec 1793 kfree(fixup);
247e743c
CM		 1794}
1795
1796/*
1797 * There are a few paths in the higher layers of the kernel that directly
1798 * set the page dirty bit without asking the filesystem if it is a
1799 * good idea. This causes problems because we want to make sure COW
1800 * properly happens and the data=ordered rules are followed.
1801 *
c8b97818 1802 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1803 * hasn't been properly setup for IO. We kick off an async process
1804 * to fix it up. The async helper will wait for ordered extents, set
1805 * the delalloc bit and make it safe to write the page.
1806 */
b2950863 1807static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1808{
1809 struct inode *inode = page->mapping->host;
1810 struct btrfs_writepage_fixup *fixup;
1811 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1812
8b62b72b
CM		 1813 /* this page is properly in the ordered list */
1814 if (TestClearPagePrivate2(page))
247e743c
CM		 1815 return 0;
1816
1817 if (PageChecked(page))
1818 return -EAGAIN;
1819
1820 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1821 if (!fixup)
1822 return -EAGAIN;
f421950f 1823
247e743c
CM		 1824 SetPageChecked(page);
1825 page_cache_get(page);
1826 fixup->work.func = btrfs_writepage_fixup_worker;
1827 fixup->page = page;
1828 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1829 return -EBUSY;
247e743c
CM		 1830}
1831
d899e052
YZ		 1832static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833 struct inode *inode, u64 file_pos,
1834 u64 disk_bytenr, u64 disk_num_bytes,
1835 u64 num_bytes, u64 ram_bytes,
1836 u8 compression, u8 encryption,
1837 u16 other_encoding, int extent_type)
1838{
1839 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 struct btrfs_file_extent_item *fi;
1841 struct btrfs_path *path;
1842 struct extent_buffer *leaf;
1843 struct btrfs_key ins;
d899e052
YZ		 1844 int ret;
1845
1846 path = btrfs_alloc_path();
d8926bb3
MF		 1847 if (!path)
1848 return -ENOMEM;
d899e052 1849
b9473439 1850 path->leave_spinning = 1;
a1ed835e
CM		 1851
1852 /*
1853 * we may be replacing one extent in the tree with another.
1854 * The new extent is pinned in the extent map, and we don't want
1855 * to drop it from the cache until it is completely in the btree.
1856 *
1857 * So, tell btrfs_drop_extents to leave this extent in the cache.
1858 * the caller is expected to unpin it and allow it to be merged
1859 * with the others.
1860 */
5dc562c5 1861 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1862 file_pos + num_bytes, 0);
79787eaa
JM		 1863 if (ret)
1864 goto out;
d899e052 1865
33345d01 1866 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1867 ins.offset = file_pos;
1868 ins.type = BTRFS_EXTENT_DATA_KEY;
1869 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1870 if (ret)
1871 goto out;
d899e052
YZ		 1872 leaf = path->nodes[0];
1873 fi = btrfs_item_ptr(leaf, path->slots[0],
1874 struct btrfs_file_extent_item);
1875 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876 btrfs_set_file_extent_type(leaf, fi, extent_type);
1877 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879 btrfs_set_file_extent_offset(leaf, fi, 0);
1880 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882 btrfs_set_file_extent_compression(leaf, fi, compression);
1883 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1885
d899e052 1886 btrfs_mark_buffer_dirty(leaf);
ce195332 1887 btrfs_release_path(path);
d899e052
YZ		 1888
1889 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1890
1891 ins.objectid = disk_bytenr;
1892 ins.offset = disk_num_bytes;
1893 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1894 ret = btrfs_alloc_reserved_file_extent(trans, root,
1895 root->root_key.objectid,
33345d01 1896 btrfs_ino(inode), file_pos, &ins);
79787eaa 1897out:
d899e052 1898 btrfs_free_path(path);
b9473439 1899
79787eaa 1900 return ret;
d899e052
YZ		 1901}
1902
38c227d8
LB		 1903/* snapshot-aware defrag */
1904struct sa_defrag_extent_backref {
1905 struct rb_node node;
1906 struct old_sa_defrag_extent *old;
1907 u64 root_id;
1908 u64 inum;
1909 u64 file_pos;
1910 u64 extent_offset;
1911 u64 num_bytes;
1912 u64 generation;
1913};
1914
1915struct old_sa_defrag_extent {
1916 struct list_head list;
1917 struct new_sa_defrag_extent *new;
1918
1919 u64 extent_offset;
1920 u64 bytenr;
1921 u64 offset;
1922 u64 len;
1923 int count;
1924};
1925
1926struct new_sa_defrag_extent {
1927 struct rb_root root;
1928 struct list_head head;
1929 struct btrfs_path *path;
1930 struct inode *inode;
1931 u64 file_pos;
1932 u64 len;
1933 u64 bytenr;
1934 u64 disk_len;
1935 u8 compress_type;
1936};
1937
1938static int backref_comp(struct sa_defrag_extent_backref *b1,
1939 struct sa_defrag_extent_backref *b2)
1940{
1941 if (b1->root_id < b2->root_id)
1942 return -1;
1943 else if (b1->root_id > b2->root_id)
1944 return 1;
1945
1946 if (b1->inum < b2->inum)
1947 return -1;
1948 else if (b1->inum > b2->inum)
1949 return 1;
1950
1951 if (b1->file_pos < b2->file_pos)
1952 return -1;
1953 else if (b1->file_pos > b2->file_pos)
1954 return 1;
1955
1956 /*
1957 * [------------------------------] ===> (a range of space)
1958 * |<--->| |<---->| =============> (fs/file tree A)
1959 * |<---------------------------->| ===> (fs/file tree B)
1960 *
1961 * A range of space can refer to two file extents in one tree while
1962 * refer to only one file extent in another tree.
1963 *
1964 * So we may process a disk offset more than one time(two extents in A)
1965 * and locate at the same extent(one extent in B), then insert two same
1966 * backrefs(both refer to the extent in B).
1967 */
1968 return 0;
1969}
1970
1971static void backref_insert(struct rb_root *root,
1972 struct sa_defrag_extent_backref *backref)
1973{
1974 struct rb_node **p = &root->rb_node;
1975 struct rb_node *parent = NULL;
1976 struct sa_defrag_extent_backref *entry;
1977 int ret;
1978
1979 while (*p) {
1980 parent = *p;
1981 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1982
1983 ret = backref_comp(backref, entry);
1984 if (ret < 0)
1985 p = &(*p)->rb_left;
1986 else
1987 p = &(*p)->rb_right;
1988 }
1989
1990 rb_link_node(&backref->node, parent, p);
1991 rb_insert_color(&backref->node, root);
1992}
1993
1994/*
1995 * Note the backref might has changed, and in this case we just return 0.
1996 */
1997static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
1998 void *ctx)
1999{
2000 struct btrfs_file_extent_item *extent;
2001 struct btrfs_fs_info *fs_info;
2002 struct old_sa_defrag_extent *old = ctx;
2003 struct new_sa_defrag_extent *new = old->new;
2004 struct btrfs_path *path = new->path;
2005 struct btrfs_key key;
2006 struct btrfs_root *root;
2007 struct sa_defrag_extent_backref *backref;
2008 struct extent_buffer *leaf;
2009 struct inode *inode = new->inode;
2010 int slot;
2011 int ret;
2012 u64 extent_offset;
2013 u64 num_bytes;
2014
2015 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2016 inum == btrfs_ino(inode))
2017 return 0;
2018
2019 key.objectid = root_id;
2020 key.type = BTRFS_ROOT_ITEM_KEY;
2021 key.offset = (u64)-1;
2022
2023 fs_info = BTRFS_I(inode)->root->fs_info;
2024 root = btrfs_read_fs_root_no_name(fs_info, &key);
2025 if (IS_ERR(root)) {
2026 if (PTR_ERR(root) == -ENOENT)
2027 return 0;
2028 WARN_ON(1);
2029 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2030 inum, offset, root_id);
2031 return PTR_ERR(root);
2032 }
2033
2034 key.objectid = inum;
2035 key.type = BTRFS_EXTENT_DATA_KEY;
2036 if (offset > (u64)-1 << 32)
2037 key.offset = 0;
2038 else
2039 key.offset = offset;
2040
2041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2042 if (WARN_ON(ret < 0))
38c227d8 2043 return ret;
50f1319c 2044 ret = 0;
38c227d8
LB		 2045
2046 while (1) {
2047 cond_resched();
2048
2049 leaf = path->nodes[0];
2050 slot = path->slots[0];
2051
2052 if (slot >= btrfs_header_nritems(leaf)) {
2053 ret = btrfs_next_leaf(root, path);
2054 if (ret < 0) {
2055 goto out;
2056 } else if (ret > 0) {
2057 ret = 0;
2058 goto out;
2059 }
2060 continue;
2061 }
2062
2063 path->slots[0]++;
2064
2065 btrfs_item_key_to_cpu(leaf, &key, slot);
2066
2067 if (key.objectid > inum)
2068 goto out;
2069
2070 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2071 continue;
2072
2073 extent = btrfs_item_ptr(leaf, slot,
2074 struct btrfs_file_extent_item);
2075
2076 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2077 continue;
2078
e68afa49
LB		 2079 /*
2080 * 'offset' refers to the exact key.offset,
2081 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2082 * (key.offset - extent_offset).
2083 */
2084 if (key.offset != offset)
38c227d8
LB		 2085 continue;
2086
e68afa49 2087 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2088 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2089
38c227d8
LB		 2090 if (extent_offset >= old->extent_offset + old->offset +
2091 old->len || extent_offset + num_bytes <=
2092 old->extent_offset + old->offset)
2093 continue;
38c227d8
LB		 2094 break;
2095 }
2096
2097 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2098 if (!backref) {
2099 ret = -ENOENT;
2100 goto out;
2101 }
2102
2103 backref->root_id = root_id;
2104 backref->inum = inum;
e68afa49 2105 backref->file_pos = offset;
38c227d8
LB		 2106 backref->num_bytes = num_bytes;
2107 backref->extent_offset = extent_offset;
2108 backref->generation = btrfs_file_extent_generation(leaf, extent);
2109 backref->old = old;
2110 backref_insert(&new->root, backref);
2111 old->count++;
2112out:
2113 btrfs_release_path(path);
2114 WARN_ON(ret);
2115 return ret;
2116}
2117
2118static noinline bool record_extent_backrefs(struct btrfs_path *path,
2119 struct new_sa_defrag_extent *new)
2120{
2121 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2122 struct old_sa_defrag_extent *old, *tmp;
2123 int ret;
2124
2125 new->path = path;
2126
2127 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2128 ret = iterate_inodes_from_logical(old->bytenr +
2129 old->extent_offset, fs_info,
38c227d8
LB		 2130 path, record_one_backref,
2131 old);
2132 BUG_ON(ret < 0 && ret != -ENOENT);
2133
2134 /* no backref to be processed for this extent */
2135 if (!old->count) {
2136 list_del(&old->list);
2137 kfree(old);
2138 }
2139 }
2140
2141 if (list_empty(&new->head))
2142 return false;
2143
2144 return true;
2145}
2146
2147static int relink_is_mergable(struct extent_buffer *leaf,
2148 struct btrfs_file_extent_item *fi,
116e0024 2149 struct new_sa_defrag_extent *new)
38c227d8 2150{
116e0024 2151 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2152 return 0;
2153
2154 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2155 return 0;
2156
116e0024
LB		 2157 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2158 return 0;
2159
2160 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2161 btrfs_file_extent_other_encoding(leaf, fi))
2162 return 0;
2163
2164 return 1;
2165}
2166
2167/*
2168 * Note the backref might has changed, and in this case we just return 0.
2169 */
2170static noinline int relink_extent_backref(struct btrfs_path *path,
2171 struct sa_defrag_extent_backref *prev,
2172 struct sa_defrag_extent_backref *backref)
2173{
2174 struct btrfs_file_extent_item *extent;
2175 struct btrfs_file_extent_item *item;
2176 struct btrfs_ordered_extent *ordered;
2177 struct btrfs_trans_handle *trans;
2178 struct btrfs_fs_info *fs_info;
2179 struct btrfs_root *root;
2180 struct btrfs_key key;
2181 struct extent_buffer *leaf;
2182 struct old_sa_defrag_extent *old = backref->old;
2183 struct new_sa_defrag_extent *new = old->new;
2184 struct inode *src_inode = new->inode;
2185 struct inode *inode;
2186 struct extent_state *cached = NULL;
2187 int ret = 0;
2188 u64 start;
2189 u64 len;
2190 u64 lock_start;
2191 u64 lock_end;
2192 bool merge = false;
2193 int index;
2194
2195 if (prev && prev->root_id == backref->root_id &&
2196 prev->inum == backref->inum &&
2197 prev->file_pos + prev->num_bytes == backref->file_pos)
2198 merge = true;
2199
2200 /* step 1: get root */
2201 key.objectid = backref->root_id;
2202 key.type = BTRFS_ROOT_ITEM_KEY;
2203 key.offset = (u64)-1;
2204
2205 fs_info = BTRFS_I(src_inode)->root->fs_info;
2206 index = srcu_read_lock(&fs_info->subvol_srcu);
2207
2208 root = btrfs_read_fs_root_no_name(fs_info, &key);
2209 if (IS_ERR(root)) {
2210 srcu_read_unlock(&fs_info->subvol_srcu, index);
2211 if (PTR_ERR(root) == -ENOENT)
2212 return 0;
2213 return PTR_ERR(root);
2214 }
38c227d8
LB		 2215
2216 /* step 2: get inode */
2217 key.objectid = backref->inum;
2218 key.type = BTRFS_INODE_ITEM_KEY;
2219 key.offset = 0;
2220
2221 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2222 if (IS_ERR(inode)) {
2223 srcu_read_unlock(&fs_info->subvol_srcu, index);
2224 return 0;
2225 }
2226
2227 srcu_read_unlock(&fs_info->subvol_srcu, index);
2228
2229 /* step 3: relink backref */
2230 lock_start = backref->file_pos;
2231 lock_end = backref->file_pos + backref->num_bytes - 1;
2232 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2233 0, &cached);
2234
2235 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2236 if (ordered) {
2237 btrfs_put_ordered_extent(ordered);
2238 goto out_unlock;
2239 }
2240
2241 trans = btrfs_join_transaction(root);
2242 if (IS_ERR(trans)) {
2243 ret = PTR_ERR(trans);
2244 goto out_unlock;
2245 }
2246
2247 key.objectid = backref->inum;
2248 key.type = BTRFS_EXTENT_DATA_KEY;
2249 key.offset = backref->file_pos;
2250
2251 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2252 if (ret < 0) {
2253 goto out_free_path;
2254 } else if (ret > 0) {
2255 ret = 0;
2256 goto out_free_path;
2257 }
2258
2259 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2260 struct btrfs_file_extent_item);
2261
2262 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2263 backref->generation)
2264 goto out_free_path;
2265
2266 btrfs_release_path(path);
2267
2268 start = backref->file_pos;
2269 if (backref->extent_offset < old->extent_offset + old->offset)
2270 start += old->extent_offset + old->offset -
2271 backref->extent_offset;
2272
2273 len = min(backref->extent_offset + backref->num_bytes,
2274 old->extent_offset + old->offset + old->len);
2275 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2276
2277 ret = btrfs_drop_extents(trans, root, inode, start,
2278 start + len, 1);
2279 if (ret)
2280 goto out_free_path;
2281again:
2282 key.objectid = btrfs_ino(inode);
2283 key.type = BTRFS_EXTENT_DATA_KEY;
2284 key.offset = start;
2285
a09a0a70 2286 path->leave_spinning = 1;
38c227d8
LB		 2287 if (merge) {
2288 struct btrfs_file_extent_item *fi;
2289 u64 extent_len;
2290 struct btrfs_key found_key;
2291
2292 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2293 if (ret < 0)
2294 goto out_free_path;
2295
2296 path->slots[0]--;
2297 leaf = path->nodes[0];
2298 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2299
2300 fi = btrfs_item_ptr(leaf, path->slots[0],
2301 struct btrfs_file_extent_item);
2302 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2303
116e0024
LB		 2304 if (extent_len + found_key.offset == start &&
2305 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2306 btrfs_set_file_extent_num_bytes(leaf, fi,
2307 extent_len + len);
2308 btrfs_mark_buffer_dirty(leaf);
2309 inode_add_bytes(inode, len);
2310
2311 ret = 1;
2312 goto out_free_path;
2313 } else {
2314 merge = false;
2315 btrfs_release_path(path);
2316 goto again;
2317 }
2318 }
2319
2320 ret = btrfs_insert_empty_item(trans, root, path, &key,
2321 sizeof(*extent));
2322 if (ret) {
2323 btrfs_abort_transaction(trans, root, ret);
2324 goto out_free_path;
2325 }
2326
2327 leaf = path->nodes[0];
2328 item = btrfs_item_ptr(leaf, path->slots[0],
2329 struct btrfs_file_extent_item);
2330 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2331 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2332 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2333 btrfs_set_file_extent_num_bytes(leaf, item, len);
2334 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2335 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2336 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2337 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2338 btrfs_set_file_extent_encryption(leaf, item, 0);
2339 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2340
2341 btrfs_mark_buffer_dirty(leaf);
2342 inode_add_bytes(inode, len);
a09a0a70 2343 btrfs_release_path(path);
38c227d8
LB		 2344
2345 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2346 new->disk_len, 0,
2347 backref->root_id, backref->inum,
2348 new->file_pos, 0); /* start - extent_offset */
2349 if (ret) {
2350 btrfs_abort_transaction(trans, root, ret);
2351 goto out_free_path;
2352 }
2353
2354 ret = 1;
2355out_free_path:
2356 btrfs_release_path(path);
a09a0a70 2357 path->leave_spinning = 0;
38c227d8
LB		 2358 btrfs_end_transaction(trans, root);
2359out_unlock:
2360 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2361 &cached, GFP_NOFS);
2362 iput(inode);
2363 return ret;
2364}
2365
6f519564
LB		 2366static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2367{
2368 struct old_sa_defrag_extent *old, *tmp;
2369
2370 if (!new)
2371 return;
2372
2373 list_for_each_entry_safe(old, tmp, &new->head, list) {
2374 list_del(&old->list);
2375 kfree(old);
2376 }
2377 kfree(new);
2378}
2379
38c227d8
LB		 2380static void relink_file_extents(struct new_sa_defrag_extent *new)
2381{
2382 struct btrfs_path *path;
38c227d8
LB		 2383 struct sa_defrag_extent_backref *backref;
2384 struct sa_defrag_extent_backref *prev = NULL;
2385 struct inode *inode;
2386 struct btrfs_root *root;
2387 struct rb_node *node;
2388 int ret;
2389
2390 inode = new->inode;
2391 root = BTRFS_I(inode)->root;
2392
2393 path = btrfs_alloc_path();
2394 if (!path)
2395 return;
2396
2397 if (!record_extent_backrefs(path, new)) {
2398 btrfs_free_path(path);
2399 goto out;
2400 }
2401 btrfs_release_path(path);
2402
2403 while (1) {
2404 node = rb_first(&new->root);
2405 if (!node)
2406 break;
2407 rb_erase(node, &new->root);
2408
2409 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2410
2411 ret = relink_extent_backref(path, prev, backref);
2412 WARN_ON(ret < 0);
2413
2414 kfree(prev);
2415
2416 if (ret == 1)
2417 prev = backref;
2418 else
2419 prev = NULL;
2420 cond_resched();
2421 }
2422 kfree(prev);
2423
2424 btrfs_free_path(path);
38c227d8 2425out:
6f519564
LB		 2426 free_sa_defrag_extent(new);
2427
38c227d8
LB		 2428 atomic_dec(&root->fs_info->defrag_running);
2429 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2430}
2431
2432static struct new_sa_defrag_extent *
2433record_old_file_extents(struct inode *inode,
2434 struct btrfs_ordered_extent *ordered)
2435{
2436 struct btrfs_root *root = BTRFS_I(inode)->root;
2437 struct btrfs_path *path;
2438 struct btrfs_key key;
6f519564 2439 struct old_sa_defrag_extent *old;
38c227d8
LB		 2440 struct new_sa_defrag_extent *new;
2441 int ret;
2442
2443 new = kmalloc(sizeof(*new), GFP_NOFS);
2444 if (!new)
2445 return NULL;
2446
2447 new->inode = inode;
2448 new->file_pos = ordered->file_offset;
2449 new->len = ordered->len;
2450 new->bytenr = ordered->start;
2451 new->disk_len = ordered->disk_len;
2452 new->compress_type = ordered->compress_type;
2453 new->root = RB_ROOT;
2454 INIT_LIST_HEAD(&new->head);
2455
2456 path = btrfs_alloc_path();
2457 if (!path)
2458 goto out_kfree;
2459
2460 key.objectid = btrfs_ino(inode);
2461 key.type = BTRFS_EXTENT_DATA_KEY;
2462 key.offset = new->file_pos;
2463
2464 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2465 if (ret < 0)
2466 goto out_free_path;
2467 if (ret > 0 && path->slots[0] > 0)
2468 path->slots[0]--;
2469
2470 /* find out all the old extents for the file range */
2471 while (1) {
2472 struct btrfs_file_extent_item *extent;
2473 struct extent_buffer *l;
2474 int slot;
2475 u64 num_bytes;
2476 u64 offset;
2477 u64 end;
2478 u64 disk_bytenr;
2479 u64 extent_offset;
2480
2481 l = path->nodes[0];
2482 slot = path->slots[0];
2483
2484 if (slot >= btrfs_header_nritems(l)) {
2485 ret = btrfs_next_leaf(root, path);
2486 if (ret < 0)
6f519564 2487 goto out_free_path;
38c227d8
LB		 2488 else if (ret > 0)
2489 break;
2490 continue;
2491 }
2492
2493 btrfs_item_key_to_cpu(l, &key, slot);
2494
2495 if (key.objectid != btrfs_ino(inode))
2496 break;
2497 if (key.type != BTRFS_EXTENT_DATA_KEY)
2498 break;
2499 if (key.offset >= new->file_pos + new->len)
2500 break;
2501
2502 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2503
2504 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2505 if (key.offset + num_bytes < new->file_pos)
2506 goto next;
2507
2508 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2509 if (!disk_bytenr)
2510 goto next;
2511
2512 extent_offset = btrfs_file_extent_offset(l, extent);
2513
2514 old = kmalloc(sizeof(*old), GFP_NOFS);
2515 if (!old)
6f519564 2516 goto out_free_path;
38c227d8
LB		 2517
2518 offset = max(new->file_pos, key.offset);
2519 end = min(new->file_pos + new->len, key.offset + num_bytes);
2520
2521 old->bytenr = disk_bytenr;
2522 old->extent_offset = extent_offset;
2523 old->offset = offset - key.offset;
2524 old->len = end - offset;
2525 old->new = new;
2526 old->count = 0;
2527 list_add_tail(&old->list, &new->head);
2528next:
2529 path->slots[0]++;
2530 cond_resched();
2531 }
2532
2533 btrfs_free_path(path);
2534 atomic_inc(&root->fs_info->defrag_running);
2535
2536 return new;
2537
38c227d8
LB		 2538out_free_path:
2539 btrfs_free_path(path);
2540out_kfree:
6f519564 2541 free_sa_defrag_extent(new);
38c227d8
LB		 2542 return NULL;
2543}
2544
5d13a98f
CM		 2545/*
2546 * helper function for btrfs_finish_ordered_io, this
2547 * just reads in some of the csum leaves to prime them into ram
2548 * before we start the transaction. It limits the amount of btree
2549 * reads required while inside the transaction.
2550 */
d352ac68
CM		 2551/* as ordered data IO finishes, this gets called so we can finish
2552 * an ordered extent if the range of bytes in the file it covers are
2553 * fully written.
2554 */
5fd02043 2555static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2556{
5fd02043 2557 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2558 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2559 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2560 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2561 struct extent_state *cached_state = NULL;
38c227d8 2562 struct new_sa_defrag_extent *new = NULL;
261507a0 2563 int compress_type = 0;
77cef2ec
JB		 2564 int ret = 0;
2565 u64 logical_len = ordered_extent->len;
82d5902d 2566 bool nolock;
77cef2ec 2567 bool truncated = false;
e6dcd2dc 2568
83eea1f1 2569 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2570
5fd02043
JB		 2571 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2572 ret = -EIO;
2573 goto out;
2574 }
2575
77cef2ec
JB		 2576 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2577 truncated = true;
2578 logical_len = ordered_extent->truncated_len;
2579 /* Truncated the entire extent, don't bother adding */
2580 if (!logical_len)
2581 goto out;
2582 }
2583
c2167754 2584 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2585 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2586 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2587 if (nolock)
2588 trans = btrfs_join_transaction_nolock(root);
2589 else
2590 trans = btrfs_join_transaction(root);
2591 if (IS_ERR(trans)) {
2592 ret = PTR_ERR(trans);
2593 trans = NULL;
2594 goto out;
c2167754 2595 }
6c760c07
JB		 2596 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2597 ret = btrfs_update_inode_fallback(trans, root, inode);
2598 if (ret) /* -ENOMEM or corruption */
2599 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2600 goto out;
2601 }
e6dcd2dc 2602
2ac55d41
JB		 2603 lock_extent_bits(io_tree, ordered_extent->file_offset,
2604 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2605 0, &cached_state);
e6dcd2dc 2606
38c227d8
LB		 2607 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2608 ordered_extent->file_offset + ordered_extent->len - 1,
2609 EXTENT_DEFRAG, 1, cached_state);
2610 if (ret) {
2611 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2612 if (last_snapshot >= BTRFS_I(inode)->generation)
2613 /* the inode is shared */
2614 new = record_old_file_extents(inode, ordered_extent);
2615
2616 clear_extent_bit(io_tree, ordered_extent->file_offset,
2617 ordered_extent->file_offset + ordered_extent->len - 1,
2618 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2619 }
2620
0cb59c99 2621 if (nolock)
7a7eaa40 2622 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2623 else
7a7eaa40 2624 trans = btrfs_join_transaction(root);
79787eaa
JM		 2625 if (IS_ERR(trans)) {
2626 ret = PTR_ERR(trans);
2627 trans = NULL;
2628 goto out_unlock;
2629 }
0ca1f7ce 2630 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2631
c8b97818 2632 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2633 compress_type = ordered_extent->compress_type;
d899e052 2634 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2635 BUG_ON(compress_type);
920bbbfb 2636 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2637 ordered_extent->file_offset,
2638 ordered_extent->file_offset +
77cef2ec 2639 logical_len);
d899e052 2640 } else {
0af3d00b 2641 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2642 ret = insert_reserved_file_extent(trans, inode,
2643 ordered_extent->file_offset,
2644 ordered_extent->start,
2645 ordered_extent->disk_len,
77cef2ec 2646 logical_len, logical_len,
261507a0 2647 compress_type, 0, 0,
d899e052 2648 BTRFS_FILE_EXTENT_REG);
d899e052 2649 }
5dc562c5
JB		 2650 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2651 ordered_extent->file_offset, ordered_extent->len,
2652 trans->transid);
79787eaa
JM		 2653 if (ret < 0) {
2654 btrfs_abort_transaction(trans, root, ret);
5fd02043 2655 goto out_unlock;
79787eaa 2656 }
2ac55d41 2657
e6dcd2dc
CM		 2658 add_pending_csums(trans, inode, ordered_extent->file_offset,
2659 &ordered_extent->list);
2660
6c760c07
JB		 2661 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2662 ret = btrfs_update_inode_fallback(trans, root, inode);
2663 if (ret) { /* -ENOMEM or corruption */
2664 btrfs_abort_transaction(trans, root, ret);
2665 goto out_unlock;
1ef30be1
JB		 2666 }
2667 ret = 0;
5fd02043
JB		 2668out_unlock:
2669 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2670 ordered_extent->file_offset +
2671 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2672out:
5b0e95bf 2673 if (root != root->fs_info->tree_root)
0cb59c99 2674 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2675 if (trans)
2676 btrfs_end_transaction(trans, root);
0cb59c99 2677
77cef2ec
JB		 2678 if (ret || truncated) {
2679 u64 start, end;
2680
2681 if (truncated)
2682 start = ordered_extent->file_offset + logical_len;
2683 else
2684 start = ordered_extent->file_offset;
2685 end = ordered_extent->file_offset + ordered_extent->len - 1;
2686 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2687
2688 /* Drop the cache for the part of the extent we didn't write. */
2689 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2690
0bec9ef5
JB		 2691 /*
2692 * If the ordered extent had an IOERR or something else went
2693 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2694 * back to the allocator. We only free the extent in the
2695 * truncated case if we didn't write out the extent at all.
0bec9ef5 2696 */
77cef2ec
JB		 2697 if ((ret || !logical_len) &&
2698 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2699 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2700 btrfs_free_reserved_extent(root, ordered_extent->start,
2701 ordered_extent->disk_len);
2702 }
2703
2704
5fd02043 2705 /*
8bad3c02
LB		 2706 * This needs to be done to make sure anybody waiting knows we are done
2707 * updating everything for this ordered extent.
5fd02043
JB		 2708 */
2709 btrfs_remove_ordered_extent(inode, ordered_extent);
2710
38c227d8 2711 /* for snapshot-aware defrag */
6f519564
LB		 2712 if (new) {
2713 if (ret) {
2714 free_sa_defrag_extent(new);
2715 atomic_dec(&root->fs_info->defrag_running);
2716 } else {
2717 relink_file_extents(new);
2718 }
2719 }
38c227d8 2720
e6dcd2dc
CM		 2721 /* once for us */
2722 btrfs_put_ordered_extent(ordered_extent);
2723 /* once for the tree */
2724 btrfs_put_ordered_extent(ordered_extent);
2725
5fd02043
JB		 2726 return ret;
2727}
2728
2729static void finish_ordered_fn(struct btrfs_work *work)
2730{
2731 struct btrfs_ordered_extent *ordered_extent;
2732 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2733 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2734}
2735
b2950863 2736static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2737 struct extent_state *state, int uptodate)
2738{
5fd02043
JB		 2739 struct inode *inode = page->mapping->host;
2740 struct btrfs_root *root = BTRFS_I(inode)->root;
2741 struct btrfs_ordered_extent *ordered_extent = NULL;
2742 struct btrfs_workers *workers;
2743
1abe9b8a 2744 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2745
8b62b72b 2746 ClearPagePrivate2(page);
5fd02043
JB		 2747 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2748 end - start + 1, uptodate))
2749 return 0;
2750
2751 ordered_extent->work.func = finish_ordered_fn;
2752 ordered_extent->work.flags = 0;
2753
83eea1f1 2754 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2755 workers = &root->fs_info->endio_freespace_worker;
2756 else
2757 workers = &root->fs_info->endio_write_workers;
2758 btrfs_queue_worker(workers, &ordered_extent->work);
2759
2760 return 0;
211f90e6
CM		 2761}
2762
d352ac68
CM		 2763/*
2764 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2765 * if there's a match, we allow the bio to finish. If not, the code in
2766 * extent_io.c will try to find good copies for us.
d352ac68 2767 */
facc8a22
MX		 2768static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2769 u64 phy_offset, struct page *page,
2770 u64 start, u64 end, int mirror)
07157aac 2771{
4eee4fa4 2772 size_t offset = start - page_offset(page);
07157aac 2773 struct inode *inode = page->mapping->host;
d1310b2e 2774 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2775 char *kaddr;
ff79f819 2776 struct btrfs_root *root = BTRFS_I(inode)->root;
facc8a22 2777 u32 csum_expected;
ff79f819 2778 u32 csum = ~(u32)0;
c2cf52eb
SK		 2779 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2780 DEFAULT_RATELIMIT_BURST);
d1310b2e 2781
d20f7043
CM		 2782 if (PageChecked(page)) {
2783 ClearPageChecked(page);
2784 goto good;
2785 }
6cbff00f
CH		 2786
2787 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2788 goto good;
17d217fe
YZ		 2789
2790 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2791 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2792 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2793 GFP_NOFS);
b6cda9bc 2794 return 0;
17d217fe 2795 }
d20f7043 2796
facc8a22
MX		 2797 phy_offset >>= inode->i_sb->s_blocksize_bits;
2798 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
d397712b 2799
facc8a22 2800 kaddr = kmap_atomic(page);
b0496686 2801 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2802 btrfs_csum_final(csum, (char *)&csum);
facc8a22 2803 if (csum != csum_expected)
07157aac 2804 goto zeroit;
d397712b 2805
7ac687d9 2806 kunmap_atomic(kaddr);
d20f7043 2807good:
07157aac
CM		 2808 return 0;
2809
2810zeroit:
c2cf52eb 2811 if (__ratelimit(&_rs))
facc8a22 2812 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c 2813 btrfs_ino(page->mapping->host), start, csum, csum_expected);
db94535d
CM		 2814 memset(kaddr + offset, 1, end - start + 1);
2815 flush_dcache_page(page);
7ac687d9 2816 kunmap_atomic(kaddr);
facc8a22 2817 if (csum_expected == 0)
3b951516 2818 return 0;
7e38326f 2819 return -EIO;
07157aac 2820}
b888db2b 2821
24bbcf04
YZ		 2822struct delayed_iput {
2823 struct list_head list;
2824 struct inode *inode;
2825};
2826
79787eaa
JM		 2827/* JDM: If this is fs-wide, why can't we add a pointer to
2828 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2829void btrfs_add_delayed_iput(struct inode *inode)
2830{
2831 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2832 struct delayed_iput *delayed;
2833
2834 if (atomic_add_unless(&inode->i_count, -1, 1))
2835 return;
2836
2837 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2838 delayed->inode = inode;
2839
2840 spin_lock(&fs_info->delayed_iput_lock);
2841 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2842 spin_unlock(&fs_info->delayed_iput_lock);
2843}
2844
2845void btrfs_run_delayed_iputs(struct btrfs_root *root)
2846{
2847 LIST_HEAD(list);
2848 struct btrfs_fs_info *fs_info = root->fs_info;
2849 struct delayed_iput *delayed;
2850 int empty;
2851
2852 spin_lock(&fs_info->delayed_iput_lock);
2853 empty = list_empty(&fs_info->delayed_iputs);
2854 spin_unlock(&fs_info->delayed_iput_lock);
2855 if (empty)
2856 return;
2857
24bbcf04
YZ		 2858 spin_lock(&fs_info->delayed_iput_lock);
2859 list_splice_init(&fs_info->delayed_iputs, &list);
2860 spin_unlock(&fs_info->delayed_iput_lock);
2861
2862 while (!list_empty(&list)) {
2863 delayed = list_entry(list.next, struct delayed_iput, list);
2864 list_del(&delayed->list);
2865 iput(delayed->inode);
2866 kfree(delayed);
2867 }
24bbcf04
YZ		 2868}
2869
d68fc57b 2870/*
42b2aa86 2871 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2872 * files in the subvolume, it removes orphan item and frees block_rsv
2873 * structure.
2874 */
2875void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2876 struct btrfs_root *root)
2877{
90290e19 2878 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2879 int ret;
2880
8a35d95f 2881 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2882 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2883 return;
2884
90290e19 2885 spin_lock(&root->orphan_lock);
8a35d95f 2886 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2887 spin_unlock(&root->orphan_lock);
2888 return;
2889 }
2890
2891 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2892 spin_unlock(&root->orphan_lock);
2893 return;
2894 }
2895
2896 block_rsv = root->orphan_block_rsv;
2897 root->orphan_block_rsv = NULL;
2898 spin_unlock(&root->orphan_lock);
2899
d68fc57b
YZ		 2900 if (root->orphan_item_inserted &&
2901 btrfs_root_refs(&root->root_item) > 0) {
2902 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2903 root->root_key.objectid);
4ef31a45
JB		 2904 if (ret)
2905 btrfs_abort_transaction(trans, root, ret);
2906 else
2907 root->orphan_item_inserted = 0;
d68fc57b
YZ		 2908 }
2909
90290e19
JB		 2910 if (block_rsv) {
2911 WARN_ON(block_rsv->size > 0);
2912 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2913 }
2914}
2915
7b128766
JB		 2916/*
2917 * This creates an orphan entry for the given inode in case something goes
2918 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2919 *
2920 * NOTE: caller of this function should reserve 5 units of metadata for
2921 * this function.
7b128766
JB		 2922 */
2923int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2924{
2925 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2926 struct btrfs_block_rsv *block_rsv = NULL;
2927 int reserve = 0;
2928 int insert = 0;
2929 int ret;
7b128766 2930
d68fc57b 2931 if (!root->orphan_block_rsv) {
66d8f3dd 2932 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2933 if (!block_rsv)
2934 return -ENOMEM;
d68fc57b 2935 }
7b128766 2936
d68fc57b
YZ		 2937 spin_lock(&root->orphan_lock);
2938 if (!root->orphan_block_rsv) {
2939 root->orphan_block_rsv = block_rsv;
2940 } else if (block_rsv) {
2941 btrfs_free_block_rsv(root, block_rsv);
2942 block_rsv = NULL;
7b128766 2943 }
7b128766 2944
8a35d95f
JB		 2945 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2946 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 2947#if 0
2948 /*
2949 * For proper ENOSPC handling, we should do orphan
2950 * cleanup when mounting. But this introduces backward
2951 * compatibility issue.
2952 */
2953 if (!xchg(&root->orphan_item_inserted, 1))
2954 insert = 2;
2955 else
2956 insert = 1;
2957#endif
2958 insert = 1;
321f0e70 2959 atomic_inc(&root->orphan_inodes);
7b128766
JB		 2960 }
2961
72ac3c0d
JB		 2962 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2963 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2964 reserve = 1;
d68fc57b 2965 spin_unlock(&root->orphan_lock);
7b128766 2966
d68fc57b
YZ		 2967 /* grab metadata reservation from transaction handle */
2968 if (reserve) {
2969 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 2970 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 2971 }
7b128766 2972
d68fc57b
YZ		 2973 /* insert an orphan item to track this unlinked/truncated file */
2974 if (insert >= 1) {
33345d01 2975 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 2976 if (ret) {
703c88e0 2977 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 2978 if (reserve) {
2979 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2980 &BTRFS_I(inode)->runtime_flags);
2981 btrfs_orphan_release_metadata(inode);
2982 }
2983 if (ret != -EEXIST) {
e8e7cff6
JB		 2984 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2985 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 2986 btrfs_abort_transaction(trans, root, ret);
2987 return ret;
2988 }
79787eaa
JM		 2989 }
2990 ret = 0;
d68fc57b
YZ		 2991 }
2992
2993 /* insert an orphan item to track subvolume contains orphan files */
2994 if (insert >= 2) {
2995 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2996 root->root_key.objectid);
79787eaa
JM		 2997 if (ret && ret != -EEXIST) {
2998 btrfs_abort_transaction(trans, root, ret);
2999 return ret;
3000 }
d68fc57b
YZ		 3001 }
3002 return 0;
7b128766
JB		 3003}
3004
3005/*
3006 * We have done the truncate/delete so we can go ahead and remove the orphan
3007 * item for this particular inode.
3008 */
48a3b636
ES		 3009static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3010 struct inode *inode)
7b128766
JB		 3011{
3012 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3013 int delete_item = 0;
3014 int release_rsv = 0;
7b128766
JB		 3015 int ret = 0;
3016
d68fc57b 3017 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3018 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3019 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3020 delete_item = 1;
7b128766 3021
72ac3c0d
JB		 3022 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3023 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3024 release_rsv = 1;
d68fc57b 3025 spin_unlock(&root->orphan_lock);
7b128766 3026
703c88e0 3027 if (delete_item) {
8a35d95f 3028 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3029 if (trans)
3030 ret = btrfs_del_orphan_item(trans, root,
3031 btrfs_ino(inode));
8a35d95f 3032 }
7b128766 3033
703c88e0
FDBM		 3034 if (release_rsv)
3035 btrfs_orphan_release_metadata(inode);
3036
4ef31a45 3037 return ret;
7b128766
JB		 3038}
3039
3040/*
3041 * this cleans up any orphans that may be left on the list from the last use
3042 * of this root.
3043 */
66b4ffd1 3044int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3045{
3046 struct btrfs_path *path;
3047 struct extent_buffer *leaf;
7b128766
JB		 3048 struct btrfs_key key, found_key;
3049 struct btrfs_trans_handle *trans;
3050 struct inode *inode;
8f6d7f4f 3051 u64 last_objectid = 0;
7b128766
JB		 3052 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3053
d68fc57b 3054 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3055 return 0;
c71bf099
YZ		 3056
3057 path = btrfs_alloc_path();
66b4ffd1
JB		 3058 if (!path) {
3059 ret = -ENOMEM;
3060 goto out;
3061 }
7b128766
JB		 3062 path->reada = -1;
3063
3064 key.objectid = BTRFS_ORPHAN_OBJECTID;
3065 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3066 key.offset = (u64)-1;
3067
7b128766
JB		 3068 while (1) {
3069 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3070 if (ret < 0)
3071 goto out;
7b128766
JB		 3072
3073 /*
3074 * if ret == 0 means we found what we were searching for, which
25985edc 3075 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3076 * find the key and see if we have stuff that matches
3077 */
3078 if (ret > 0) {
66b4ffd1 3079 ret = 0;
7b128766
JB		 3080 if (path->slots[0] == 0)
3081 break;
3082 path->slots[0]--;
3083 }
3084
3085 /* pull out the item */
3086 leaf = path->nodes[0];
7b128766
JB		 3087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3088
3089 /* make sure the item matches what we want */
3090 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3091 break;
3092 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3093 break;
3094
3095 /* release the path since we're done with it */
b3b4aa74 3096 btrfs_release_path(path);
7b128766
JB		 3097
3098 /*
3099 * this is where we are basically btrfs_lookup, without the
3100 * crossing root thing. we store the inode number in the
3101 * offset of the orphan item.
3102 */
8f6d7f4f
JB		 3103
3104 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3105 btrfs_err(root->fs_info,
3106 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3107 ret = -EINVAL;
3108 goto out;
3109 }
3110
3111 last_objectid = found_key.offset;
3112
5d4f98a2
YZ		 3113 found_key.objectid = found_key.offset;
3114 found_key.type = BTRFS_INODE_ITEM_KEY;
3115 found_key.offset = 0;
73f73415 3116 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3117 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3118 if (ret && ret != -ESTALE)
66b4ffd1 3119 goto out;
7b128766 3120
f8e9e0b0
AJ		 3121 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3122 struct btrfs_root *dead_root;
3123 struct btrfs_fs_info *fs_info = root->fs_info;
3124 int is_dead_root = 0;
3125
3126 /*
3127 * this is an orphan in the tree root. Currently these
3128 * could come from 2 sources:
3129 * a) a snapshot deletion in progress
3130 * b) a free space cache inode
3131 * We need to distinguish those two, as the snapshot
3132 * orphan must not get deleted.
3133 * find_dead_roots already ran before us, so if this
3134 * is a snapshot deletion, we should find the root
3135 * in the dead_roots list
3136 */
3137 spin_lock(&fs_info->trans_lock);
3138 list_for_each_entry(dead_root, &fs_info->dead_roots,
3139 root_list) {
3140 if (dead_root->root_key.objectid ==
3141 found_key.objectid) {
3142 is_dead_root = 1;
3143 break;
3144 }
3145 }
3146 spin_unlock(&fs_info->trans_lock);
3147 if (is_dead_root) {
3148 /* prevent this orphan from being found again */
3149 key.offset = found_key.objectid - 1;
3150 continue;
3151 }
3152 }
7b128766 3153 /*
a8c9e576
JB		 3154 * Inode is already gone but the orphan item is still there,
3155 * kill the orphan item.
7b128766 3156 */
a8c9e576
JB		 3157 if (ret == -ESTALE) {
3158 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3159 if (IS_ERR(trans)) {
3160 ret = PTR_ERR(trans);
3161 goto out;
3162 }
c2cf52eb
SK		 3163 btrfs_debug(root->fs_info, "auto deleting %Lu",
3164 found_key.objectid);
a8c9e576
JB		 3165 ret = btrfs_del_orphan_item(trans, root,
3166 found_key.objectid);
5b21f2ed 3167 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3168 if (ret)
3169 goto out;
7b128766
JB		 3170 continue;
3171 }
3172
a8c9e576
JB		 3173 /*
3174 * add this inode to the orphan list so btrfs_orphan_del does
3175 * the proper thing when we hit it
3176 */
8a35d95f
JB		 3177 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3178 &BTRFS_I(inode)->runtime_flags);
925396ec 3179 atomic_inc(&root->orphan_inodes);
a8c9e576 3180
7b128766
JB		 3181 /* if we have links, this was a truncate, lets do that */
3182 if (inode->i_nlink) {
fae7f21c 3183 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3184 iput(inode);
3185 continue;
3186 }
7b128766 3187 nr_truncate++;
f3fe820c
JB		 3188
3189 /* 1 for the orphan item deletion. */
3190 trans = btrfs_start_transaction(root, 1);
3191 if (IS_ERR(trans)) {
c69b26b0 3192 iput(inode);
f3fe820c
JB		 3193 ret = PTR_ERR(trans);
3194 goto out;
3195 }
3196 ret = btrfs_orphan_add(trans, inode);
3197 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3198 if (ret) {
3199 iput(inode);
f3fe820c 3200 goto out;
c69b26b0 3201 }
f3fe820c 3202
66b4ffd1 3203 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3204 if (ret)
3205 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3206 } else {
3207 nr_unlink++;
3208 }
3209
3210 /* this will do delete_inode and everything for us */
3211 iput(inode);
66b4ffd1
JB		 3212 if (ret)
3213 goto out;
7b128766 3214 }
3254c876
MX		 3215 /* release the path since we're done with it */
3216 btrfs_release_path(path);
3217
d68fc57b
YZ		 3218 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3219
3220 if (root->orphan_block_rsv)
3221 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3222 (u64)-1);
3223
3224 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3225 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3226 if (!IS_ERR(trans))
3227 btrfs_end_transaction(trans, root);
d68fc57b 3228 }
7b128766
JB		 3229
3230 if (nr_unlink)
4884b476 3231 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3232 if (nr_truncate)
4884b476 3233 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3234
3235out:
3236 if (ret)
c2cf52eb
SK		 3237 btrfs_crit(root->fs_info,
3238 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3239 btrfs_free_path(path);
3240 return ret;
7b128766
JB		 3241}
3242
46a53cca
CM		 3243/*
3244 * very simple check to peek ahead in the leaf looking for xattrs. If we
3245 * don't find any xattrs, we know there can't be any acls.
3246 *
3247 * slot is the slot the inode is in, objectid is the objectid of the inode
3248 */
3249static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3250 int slot, u64 objectid)
3251{
3252 u32 nritems = btrfs_header_nritems(leaf);
3253 struct btrfs_key found_key;
f23b5a59
JB		 3254 static u64 xattr_access = 0;
3255 static u64 xattr_default = 0;
46a53cca
CM		 3256 int scanned = 0;
3257
f23b5a59
JB		 3258 if (!xattr_access) {
3259 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3260 strlen(POSIX_ACL_XATTR_ACCESS));
3261 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3262 strlen(POSIX_ACL_XATTR_DEFAULT));
3263 }
3264
46a53cca
CM		 3265 slot++;
3266 while (slot < nritems) {
3267 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3268
3269 /* we found a different objectid, there must not be acls */
3270 if (found_key.objectid != objectid)
3271 return 0;
3272
3273 /* we found an xattr, assume we've got an acl */
f23b5a59
JB		 3274 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3275 if (found_key.offset == xattr_access ||
3276 found_key.offset == xattr_default)
3277 return 1;
3278 }
46a53cca
CM		 3279
3280 /*
3281 * we found a key greater than an xattr key, there can't
3282 * be any acls later on
3283 */
3284 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3285 return 0;
3286
3287 slot++;
3288 scanned++;
3289
3290 /*
3291 * it goes inode, inode backrefs, xattrs, extents,
3292 * so if there are a ton of hard links to an inode there can
3293 * be a lot of backrefs. Don't waste time searching too hard,
3294 * this is just an optimization
3295 */
3296 if (scanned >= 8)
3297 break;
3298 }
3299 /* we hit the end of the leaf before we found an xattr or
3300 * something larger than an xattr. We have to assume the inode
3301 * has acls
3302 */
3303 return 1;
3304}
3305
d352ac68
CM		 3306/*
3307 * read an inode from the btree into the in-memory inode
3308 */
5d4f98a2 3309static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3310{
3311 struct btrfs_path *path;
5f39d397 3312 struct extent_buffer *leaf;
39279cc3 3313 struct btrfs_inode_item *inode_item;
0b86a832 3314 struct btrfs_timespec *tspec;
39279cc3
CM		 3315 struct btrfs_root *root = BTRFS_I(inode)->root;
3316 struct btrfs_key location;
46a53cca 3317 int maybe_acls;
618e21d5 3318 u32 rdev;
39279cc3 3319 int ret;
2f7e33d4
MX		 3320 bool filled = false;
3321
3322 ret = btrfs_fill_inode(inode, &rdev);
3323 if (!ret)
3324 filled = true;
39279cc3
CM		 3325
3326 path = btrfs_alloc_path();
1748f843
MF		 3327 if (!path)
3328 goto make_bad;
3329
d90c7321 3330 path->leave_spinning = 1;
39279cc3 3331 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3332
39279cc3 3333 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3334 if (ret)
39279cc3 3335 goto make_bad;
39279cc3 3336
5f39d397 3337 leaf = path->nodes[0];
2f7e33d4
MX		 3338
3339 if (filled)
3340 goto cache_acl;
3341
5f39d397
CM		 3342 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3343 struct btrfs_inode_item);
5f39d397 3344 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3345 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3346 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3347 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3348 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3349
3350 tspec = btrfs_inode_atime(inode_item);
3351 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3352 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3353
3354 tspec = btrfs_inode_mtime(inode_item);
3355 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3356 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3357
3358 tspec = btrfs_inode_ctime(inode_item);
3359 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3360 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3361
a76a3cd4 3362 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3363 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3364 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3365
3366 /*
3367 * If we were modified in the current generation and evicted from memory
3368 * and then re-read we need to do a full sync since we don't have any
3369 * idea about which extents were modified before we were evicted from
3370 * cache.
3371 */
3372 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3373 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3374 &BTRFS_I(inode)->runtime_flags);
3375
0c4d2d95 3376 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3377 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3378 inode->i_rdev = 0;
5f39d397
CM		 3379 rdev = btrfs_inode_rdev(leaf, inode_item);
3380
aec7477b 3381 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3382 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3383cache_acl:
46a53cca
CM		 3384 /*
3385 * try to precache a NULL acl entry for files that don't have
3386 * any xattrs or acls
3387 */
33345d01
LZ		 3388 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3389 btrfs_ino(inode));
72c04902
AV		 3390 if (!maybe_acls)
3391 cache_no_acl(inode);
46a53cca 3392
39279cc3 3393 btrfs_free_path(path);
39279cc3 3394
39279cc3 3395 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3396 case S_IFREG:
3397 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3398 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3399 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3400 inode->i_fop = &btrfs_file_operations;
3401 inode->i_op = &btrfs_file_inode_operations;
3402 break;
3403 case S_IFDIR:
3404 inode->i_fop = &btrfs_dir_file_operations;
3405 if (root == root->fs_info->tree_root)
3406 inode->i_op = &btrfs_dir_ro_inode_operations;
3407 else
3408 inode->i_op = &btrfs_dir_inode_operations;
3409 break;
3410 case S_IFLNK:
3411 inode->i_op = &btrfs_symlink_inode_operations;
3412 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3413 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3414 break;
618e21d5 3415 default:
0279b4cd 3416 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3417 init_special_inode(inode, inode->i_mode, rdev);
3418 break;
39279cc3 3419 }
6cbff00f
CH		 3420
3421 btrfs_update_iflags(inode);
39279cc3
CM		 3422 return;
3423
3424make_bad:
39279cc3 3425 btrfs_free_path(path);
39279cc3
CM		 3426 make_bad_inode(inode);
3427}
3428
d352ac68
CM		 3429/*
3430 * given a leaf and an inode, copy the inode fields into the leaf
3431 */
e02119d5
CM		 3432static void fill_inode_item(struct btrfs_trans_handle *trans,
3433 struct extent_buffer *leaf,
5f39d397 3434 struct btrfs_inode_item *item,
39279cc3
CM		 3435 struct inode *inode)
3436{
51fab693
LB		 3437 struct btrfs_map_token token;
3438
3439 btrfs_init_map_token(&token);
5f39d397 3440
51fab693
LB		 3441 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3442 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3443 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3444 &token);
3445 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3446 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3447
51fab693
LB		 3448 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3449 inode->i_atime.tv_sec, &token);
3450 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3451 inode->i_atime.tv_nsec, &token);
5f39d397 3452
51fab693
LB		 3453 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3454 inode->i_mtime.tv_sec, &token);
3455 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3456 inode->i_mtime.tv_nsec, &token);
5f39d397 3457
51fab693
LB		 3458 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3459 inode->i_ctime.tv_sec, &token);
3460 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3461 inode->i_ctime.tv_nsec, &token);
5f39d397 3462
51fab693
LB		 3463 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3464 &token);
3465 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3466 &token);
3467 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3468 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3469 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3470 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3471 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3472}
3473
d352ac68
CM		 3474/*
3475 * copy everything in the in-memory inode into the btree.
3476 */
2115133f 3477static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3478 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3479{
3480 struct btrfs_inode_item *inode_item;
3481 struct btrfs_path *path;
5f39d397 3482 struct extent_buffer *leaf;
39279cc3
CM		 3483 int ret;
3484
3485 path = btrfs_alloc_path();
16cdcec7
MX		 3486 if (!path)
3487 return -ENOMEM;
3488
b9473439 3489 path->leave_spinning = 1;
16cdcec7
MX		 3490 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3491 1);
39279cc3
CM		 3492 if (ret) {
3493 if (ret > 0)
3494 ret = -ENOENT;
3495 goto failed;
3496 }
3497
b4ce94de 3498 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3499 leaf = path->nodes[0];
3500 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3501 struct btrfs_inode_item);
39279cc3 3502
e02119d5 3503 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3504 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3505 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3506 ret = 0;
3507failed:
39279cc3
CM		 3508 btrfs_free_path(path);
3509 return ret;
3510}
3511
2115133f
CM		 3512/*
3513 * copy everything in the in-memory inode into the btree.
3514 */
3515noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3516 struct btrfs_root *root, struct inode *inode)
3517{
3518 int ret;
3519
3520 /*
3521 * If the inode is a free space inode, we can deadlock during commit
3522 * if we put it into the delayed code.
3523 *
3524 * The data relocation inode should also be directly updated
3525 * without delay
3526 */
83eea1f1 3527 if (!btrfs_is_free_space_inode(inode)
2115133f 3528 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3529 btrfs_update_root_times(trans, root);
3530
2115133f
CM		 3531 ret = btrfs_delayed_update_inode(trans, root, inode);
3532 if (!ret)
3533 btrfs_set_inode_last_trans(trans, inode);
3534 return ret;
3535 }
3536
3537 return btrfs_update_inode_item(trans, root, inode);
3538}
3539
be6aef60
JB		 3540noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root,
3542 struct inode *inode)
2115133f
CM		 3543{
3544 int ret;
3545
3546 ret = btrfs_update_inode(trans, root, inode);
3547 if (ret == -ENOSPC)
3548 return btrfs_update_inode_item(trans, root, inode);
3549 return ret;
3550}
3551
d352ac68
CM		 3552/*
3553 * unlink helper that gets used here in inode.c and in the tree logging
3554 * recovery code. It remove a link in a directory with a given name, and
3555 * also drops the back refs in the inode to the directory
3556 */
92986796
AV		 3557static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3558 struct btrfs_root *root,
3559 struct inode *dir, struct inode *inode,
3560 const char *name, int name_len)
39279cc3
CM		 3561{
3562 struct btrfs_path *path;
39279cc3 3563 int ret = 0;
5f39d397 3564 struct extent_buffer *leaf;
39279cc3 3565 struct btrfs_dir_item *di;
5f39d397 3566 struct btrfs_key key;
aec7477b 3567 u64 index;
33345d01
LZ		 3568 u64 ino = btrfs_ino(inode);
3569 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3570
3571 path = btrfs_alloc_path();
54aa1f4d
CM		 3572 if (!path) {
3573 ret = -ENOMEM;
554233a6 3574 goto out;
54aa1f4d
CM		 3575 }
3576
b9473439 3577 path->leave_spinning = 1;
33345d01 3578 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3579 name, name_len, -1);
3580 if (IS_ERR(di)) {
3581 ret = PTR_ERR(di);
3582 goto err;
3583 }
3584 if (!di) {
3585 ret = -ENOENT;
3586 goto err;
3587 }
5f39d397
CM		 3588 leaf = path->nodes[0];
3589 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3590 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3591 if (ret)
3592 goto err;
b3b4aa74 3593 btrfs_release_path(path);
39279cc3 3594
33345d01
LZ		 3595 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3596 dir_ino, &index);
aec7477b 3597 if (ret) {
c2cf52eb
SK		 3598 btrfs_info(root->fs_info,
3599 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3600 name_len, name, ino, dir_ino);
79787eaa 3601 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3602 goto err;
3603 }
3604
16cdcec7 3605 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3606 if (ret) {
3607 btrfs_abort_transaction(trans, root, ret);
39279cc3 3608 goto err;
79787eaa 3609 }
39279cc3 3610
e02119d5 3611 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3612 inode, dir_ino);
79787eaa
JM		 3613 if (ret != 0 && ret != -ENOENT) {
3614 btrfs_abort_transaction(trans, root, ret);
3615 goto err;
3616 }
e02119d5
CM		 3617
3618 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3619 dir, index);
6418c961
CM		 3620 if (ret == -ENOENT)
3621 ret = 0;
d4e3991b
ZB		 3622 else if (ret)
3623 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3624err:
3625 btrfs_free_path(path);
e02119d5
CM		 3626 if (ret)
3627 goto out;
3628
3629 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3630 inode_inc_iversion(inode);
3631 inode_inc_iversion(dir);
e02119d5 3632 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3633 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3634out:
39279cc3
CM		 3635 return ret;
3636}
3637
92986796
AV		 3638int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3639 struct btrfs_root *root,
3640 struct inode *dir, struct inode *inode,
3641 const char *name, int name_len)
3642{
3643 int ret;
3644 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3645 if (!ret) {
8b558c5f 3646 drop_nlink(inode);
92986796
AV		 3647 ret = btrfs_update_inode(trans, root, inode);
3648 }
3649 return ret;
3650}
39279cc3 3651
a22285a6
YZ		 3652/*
3653 * helper to start transaction for unlink and rmdir.
3654 *
d52be818
JB		 3655 * unlink and rmdir are special in btrfs, they do not always free space, so
3656 * if we cannot make our reservations the normal way try and see if there is
3657 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3658 * allow the unlink to occur.
a22285a6 3659 */
d52be818 3660static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3661{
39279cc3 3662 struct btrfs_trans_handle *trans;
a22285a6 3663 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3664 int ret;
3665
e70bea5f
JB		 3666 /*
3667 * 1 for the possible orphan item
3668 * 1 for the dir item
3669 * 1 for the dir index
3670 * 1 for the inode ref
e70bea5f
JB		 3671 * 1 for the inode
3672 */
6e137ed3 3673 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3674 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3675 return trans;
4df27c4d 3676
d52be818
JB		 3677 if (PTR_ERR(trans) == -ENOSPC) {
3678 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 3679
d52be818
JB		 3680 trans = btrfs_start_transaction(root, 0);
3681 if (IS_ERR(trans))
3682 return trans;
3683 ret = btrfs_cond_migrate_bytes(root->fs_info,
3684 &root->fs_info->trans_block_rsv,
3685 num_bytes, 5);
3686 if (ret) {
3687 btrfs_end_transaction(trans, root);
3688 return ERR_PTR(ret);
a22285a6 3689 }
5a77d76c 3690 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3691 trans->bytes_reserved = num_bytes;
a22285a6 3692 }
d52be818 3693 return trans;
a22285a6
YZ		 3694}
3695
3696static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3697{
3698 struct btrfs_root *root = BTRFS_I(dir)->root;
3699 struct btrfs_trans_handle *trans;
3700 struct inode *inode = dentry->d_inode;
3701 int ret;
a22285a6 3702
d52be818 3703 trans = __unlink_start_trans(dir);
a22285a6
YZ		 3704 if (IS_ERR(trans))
3705 return PTR_ERR(trans);
5f39d397 3706
12fcfd22
CM		 3707 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3708
e02119d5
CM		 3709 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3710 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3711 if (ret)
3712 goto out;
7b128766 3713
a22285a6 3714 if (inode->i_nlink == 0) {
7b128766 3715 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3716 if (ret)
3717 goto out;
a22285a6 3718 }
7b128766 3719
b532402e 3720out:
d52be818 3721 btrfs_end_transaction(trans, root);
b53d3f5d 3722 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3723 return ret;
3724}
3725
4df27c4d
YZ		 3726int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3727 struct btrfs_root *root,
3728 struct inode *dir, u64 objectid,
3729 const char *name, int name_len)
3730{
3731 struct btrfs_path *path;
3732 struct extent_buffer *leaf;
3733 struct btrfs_dir_item *di;
3734 struct btrfs_key key;
3735 u64 index;
3736 int ret;
33345d01 3737 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3738
3739 path = btrfs_alloc_path();
3740 if (!path)
3741 return -ENOMEM;
3742
33345d01 3743 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3744 name, name_len, -1);
79787eaa
JM		 3745 if (IS_ERR_OR_NULL(di)) {
3746 if (!di)
3747 ret = -ENOENT;
3748 else
3749 ret = PTR_ERR(di);
3750 goto out;
3751 }
4df27c4d
YZ		 3752
3753 leaf = path->nodes[0];
3754 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3755 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3756 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3757 if (ret) {
3758 btrfs_abort_transaction(trans, root, ret);
3759 goto out;
3760 }
b3b4aa74 3761 btrfs_release_path(path);
4df27c4d
YZ		 3762
3763 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3764 objectid, root->root_key.objectid,
33345d01 3765 dir_ino, &index, name, name_len);
4df27c4d 3766 if (ret < 0) {
79787eaa
JM		 3767 if (ret != -ENOENT) {
3768 btrfs_abort_transaction(trans, root, ret);
3769 goto out;
3770 }
33345d01 3771 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3772 name, name_len);
79787eaa
JM		 3773 if (IS_ERR_OR_NULL(di)) {
3774 if (!di)
3775 ret = -ENOENT;
3776 else
3777 ret = PTR_ERR(di);
3778 btrfs_abort_transaction(trans, root, ret);
3779 goto out;
3780 }
4df27c4d
YZ		 3781
3782 leaf = path->nodes[0];
3783 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3784 btrfs_release_path(path);
4df27c4d
YZ		 3785 index = key.offset;
3786 }
945d8962 3787 btrfs_release_path(path);
4df27c4d 3788
16cdcec7 3789 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3790 if (ret) {
3791 btrfs_abort_transaction(trans, root, ret);
3792 goto out;
3793 }
4df27c4d
YZ		 3794
3795 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3796 inode_inc_iversion(dir);
4df27c4d 3797 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3798 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3799 if (ret)
3800 btrfs_abort_transaction(trans, root, ret);
3801out:
71d7aed0 3802 btrfs_free_path(path);
79787eaa 3803 return ret;
4df27c4d
YZ		 3804}
3805
39279cc3
CM		 3806static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3807{
3808 struct inode *inode = dentry->d_inode;
1832a6d5 3809 int err = 0;
39279cc3 3810 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3811 struct btrfs_trans_handle *trans;
39279cc3 3812
b3ae244e 3813 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3814 return -ENOTEMPTY;
b3ae244e
DS		 3815 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3816 return -EPERM;
134d4512 3817
d52be818 3818 trans = __unlink_start_trans(dir);
a22285a6 3819 if (IS_ERR(trans))
5df6a9f6 3820 return PTR_ERR(trans);
5df6a9f6 3821
33345d01 3822 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3823 err = btrfs_unlink_subvol(trans, root, dir,
3824 BTRFS_I(inode)->location.objectid,
3825 dentry->d_name.name,
3826 dentry->d_name.len);
3827 goto out;
3828 }
3829
7b128766
JB		 3830 err = btrfs_orphan_add(trans, inode);
3831 if (err)
4df27c4d 3832 goto out;
7b128766 3833
39279cc3 3834 /* now the directory is empty */
e02119d5
CM		 3835 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3836 dentry->d_name.name, dentry->d_name.len);
d397712b 3837 if (!err)
dbe674a9 3838 btrfs_i_size_write(inode, 0);
4df27c4d 3839out:
d52be818 3840 btrfs_end_transaction(trans, root);
b53d3f5d 3841 btrfs_btree_balance_dirty(root);
3954401f 3842
39279cc3
CM		 3843 return err;
3844}
3845
39279cc3
CM		 3846/*
3847 * this can truncate away extent items, csum items and directory items.
3848 * It starts at a high offset and removes keys until it can't find
d352ac68 3849 * any higher than new_size
39279cc3
CM		 3850 *
3851 * csum items that cross the new i_size are truncated to the new size
3852 * as well.
7b128766
JB		 3853 *
3854 * min_type is the minimum key type to truncate down to. If set to 0, this
3855 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3856 */
8082510e
YZ		 3857int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3858 struct btrfs_root *root,
3859 struct inode *inode,
3860 u64 new_size, u32 min_type)
39279cc3 3861{
39279cc3 3862 struct btrfs_path *path;
5f39d397 3863 struct extent_buffer *leaf;
39279cc3 3864 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3865 struct btrfs_key key;
3866 struct btrfs_key found_key;
39279cc3 3867 u64 extent_start = 0;
db94535d 3868 u64 extent_num_bytes = 0;
5d4f98a2 3869 u64 extent_offset = 0;
39279cc3 3870 u64 item_end = 0;
7f4f6e0a 3871 u64 last_size = (u64)-1;
8082510e 3872 u32 found_type = (u8)-1;
39279cc3
CM		 3873 int found_extent;
3874 int del_item;
85e21bac
CM		 3875 int pending_del_nr = 0;
3876 int pending_del_slot = 0;
179e29e4 3877 int extent_type = -1;
8082510e
YZ		 3878 int ret;
3879 int err = 0;
33345d01 3880 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3881
3882 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3883
0eb0e19c
MF		 3884 path = btrfs_alloc_path();
3885 if (!path)
3886 return -ENOMEM;
3887 path->reada = -1;
3888
5dc562c5
JB		 3889 /*
3890 * We want to drop from the next block forward in case this new size is
3891 * not block aligned since we will be keeping the last block of the
3892 * extent just the way it is.
3893 */
0af3d00b 3894 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 3895 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3896 root->sectorsize), (u64)-1, 0);
8082510e 3897
16cdcec7
MX		 3898 /*
3899 * This function is also used to drop the items in the log tree before
3900 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3901 * it is used to drop the loged items. So we shouldn't kill the delayed
3902 * items.
3903 */
3904 if (min_type == 0 && root == BTRFS_I(inode)->root)
3905 btrfs_kill_delayed_inode_items(inode);
3906
33345d01 3907 key.objectid = ino;
39279cc3 3908 key.offset = (u64)-1;
5f39d397
CM		 3909 key.type = (u8)-1;
3910
85e21bac 3911search_again:
b9473439 3912 path->leave_spinning = 1;
85e21bac 3913 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3914 if (ret < 0) {
3915 err = ret;
3916 goto out;
3917 }
d397712b 3918
85e21bac 3919 if (ret > 0) {
e02119d5
CM		 3920 /* there are no items in the tree for us to truncate, we're
3921 * done
3922 */
8082510e
YZ		 3923 if (path->slots[0] == 0)
3924 goto out;
85e21bac
CM		 3925 path->slots[0]--;
3926 }
3927
d397712b 3928 while (1) {
39279cc3 3929 fi = NULL;
5f39d397
CM		 3930 leaf = path->nodes[0];
3931 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3932 found_type = btrfs_key_type(&found_key);
39279cc3 3933
33345d01 3934 if (found_key.objectid != ino)
39279cc3 3935 break;
5f39d397 3936
85e21bac 3937 if (found_type < min_type)
39279cc3
CM		 3938 break;
3939
5f39d397 3940 item_end = found_key.offset;
39279cc3 3941 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3942 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3943 struct btrfs_file_extent_item);
179e29e4
CM		 3944 extent_type = btrfs_file_extent_type(leaf, fi);
3945 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3946 item_end +=
db94535d 3947 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3948 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3949 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3950 fi);
39279cc3 3951 }
008630c1 3952 item_end--;
39279cc3 3953 }
8082510e
YZ		 3954 if (found_type > min_type) {
3955 del_item = 1;
3956 } else {
3957 if (item_end < new_size)
b888db2b 3958 break;
8082510e
YZ		 3959 if (found_key.offset >= new_size)
3960 del_item = 1;
3961 else
3962 del_item = 0;
39279cc3 3963 }
39279cc3 3964 found_extent = 0;
39279cc3 3965 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3966 if (found_type != BTRFS_EXTENT_DATA_KEY)
3967 goto delete;
3968
7f4f6e0a
JB		 3969 if (del_item)
3970 last_size = found_key.offset;
3971 else
3972 last_size = new_size;
3973
179e29e4 3974 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3975 u64 num_dec;
db94535d 3976 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3977 if (!del_item) {
db94535d
CM		 3978 u64 orig_num_bytes =
3979 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 3980 extent_num_bytes = ALIGN(new_size -
3981 found_key.offset,
3982 root->sectorsize);
db94535d
CM		 3983 btrfs_set_file_extent_num_bytes(leaf, fi,
3984 extent_num_bytes);
3985 num_dec = (orig_num_bytes -
9069218d 3986 extent_num_bytes);
e02119d5 3987 if (root->ref_cows && extent_start != 0)
a76a3cd4 3988 inode_sub_bytes(inode, num_dec);
5f39d397 3989 btrfs_mark_buffer_dirty(leaf);
39279cc3 3990 } else {
db94535d
CM		 3991 extent_num_bytes =
3992 btrfs_file_extent_disk_num_bytes(leaf,
3993 fi);
5d4f98a2
YZ		 3994 extent_offset = found_key.offset -
3995 btrfs_file_extent_offset(leaf, fi);
3996
39279cc3 3997 /* FIXME blocksize != 4096 */
9069218d 3998 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3999 if (extent_start != 0) {
4000 found_extent = 1;
e02119d5 4001 if (root->ref_cows)
a76a3cd4 4002 inode_sub_bytes(inode, num_dec);
e02119d5 4003 }
39279cc3 4004 }
9069218d 4005 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4006 /*
4007 * we can't truncate inline items that have had
4008 * special encodings
4009 */
4010 if (!del_item &&
4011 btrfs_file_extent_compression(leaf, fi) == 0 &&
4012 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4013 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4014 u32 size = new_size - found_key.offset;
4015
4016 if (root->ref_cows) {
a76a3cd4
YZ		 4017 inode_sub_bytes(inode, item_end + 1 -
4018 new_size);
e02119d5
CM		 4019 }
4020 size =
4021 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4022 btrfs_truncate_item(root, path, size, 1);
e02119d5 4023 } else if (root->ref_cows) {
a76a3cd4
YZ		 4024 inode_sub_bytes(inode, item_end + 1 -
4025 found_key.offset);
9069218d 4026 }
39279cc3 4027 }
179e29e4 4028delete:
39279cc3 4029 if (del_item) {
85e21bac
CM		 4030 if (!pending_del_nr) {
4031 /* no pending yet, add ourselves */
4032 pending_del_slot = path->slots[0];
4033 pending_del_nr = 1;
4034 } else if (pending_del_nr &&
4035 path->slots[0] + 1 == pending_del_slot) {
4036 /* hop on the pending chunk */
4037 pending_del_nr++;
4038 pending_del_slot = path->slots[0];
4039 } else {
d397712b 4040 BUG();
85e21bac 4041 }
39279cc3
CM		 4042 } else {
4043 break;
4044 }
0af3d00b
JB		 4045 if (found_extent && (root->ref_cows ||
4046 root == root->fs_info->tree_root)) {
b9473439 4047 btrfs_set_path_blocking(path);
39279cc3 4048 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4049 extent_num_bytes, 0,
4050 btrfs_header_owner(leaf),
66d7e7f0 4051 ino, extent_offset, 0);
39279cc3
CM		 4052 BUG_ON(ret);
4053 }
85e21bac 4054
8082510e
YZ		 4055 if (found_type == BTRFS_INODE_ITEM_KEY)
4056 break;
4057
4058 if (path->slots[0] == 0 ||
4059 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4060 if (pending_del_nr) {
4061 ret = btrfs_del_items(trans, root, path,
4062 pending_del_slot,
4063 pending_del_nr);
79787eaa
JM		 4064 if (ret) {
4065 btrfs_abort_transaction(trans,
4066 root, ret);
4067 goto error;
4068 }
8082510e
YZ		 4069 pending_del_nr = 0;
4070 }
b3b4aa74 4071 btrfs_release_path(path);
85e21bac 4072 goto search_again;
8082510e
YZ		 4073 } else {
4074 path->slots[0]--;
85e21bac 4075 }
39279cc3 4076 }
8082510e 4077out:
85e21bac
CM		 4078 if (pending_del_nr) {
4079 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4080 pending_del_nr);
79787eaa
JM		 4081 if (ret)
4082 btrfs_abort_transaction(trans, root, ret);
85e21bac 4083 }
79787eaa 4084error:
7f4f6e0a
JB		 4085 if (last_size != (u64)-1)
4086 btrfs_ordered_update_i_size(inode, last_size, NULL);
39279cc3 4087 btrfs_free_path(path);
8082510e 4088 return err;
39279cc3
CM		 4089}
4090
4091/*
2aaa6655
JB		 4092 * btrfs_truncate_page - read, zero a chunk and write a page
4093 * @inode - inode that we're zeroing
4094 * @from - the offset to start zeroing
4095 * @len - the length to zero, 0 to zero the entire range respective to the
4096 * offset
4097 * @front - zero up to the offset instead of from the offset on
4098 *
4099 * This will find the page for the "from" offset and cow the page and zero the
4100 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4101 */
2aaa6655
JB		 4102int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4103 int front)
39279cc3 4104{
2aaa6655 4105 struct address_space *mapping = inode->i_mapping;
db94535d 4106 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4107 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4108 struct btrfs_ordered_extent *ordered;
2ac55d41 4109 struct extent_state *cached_state = NULL;
e6dcd2dc 4110 char *kaddr;
db94535d 4111 u32 blocksize = root->sectorsize;
39279cc3
CM		 4112 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4113 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4114 struct page *page;
3b16a4e3 4115 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4116 int ret = 0;
a52d9a80 4117 u64 page_start;
e6dcd2dc 4118 u64 page_end;
39279cc3 4119
2aaa6655
JB		 4120 if ((offset & (blocksize - 1)) == 0 &&
4121 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4122 goto out;
0ca1f7ce 4123 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4124 if (ret)
4125 goto out;
39279cc3 4126
211c17f5 4127again:
3b16a4e3 4128 page = find_or_create_page(mapping, index, mask);
5d5e103a 4129 if (!page) {
0ca1f7ce 4130 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4131 ret = -ENOMEM;
39279cc3 4132 goto out;
5d5e103a 4133 }
e6dcd2dc
CM		 4134
4135 page_start = page_offset(page);
4136 page_end = page_start + PAGE_CACHE_SIZE - 1;
4137
39279cc3 4138 if (!PageUptodate(page)) {
9ebefb18 4139 ret = btrfs_readpage(NULL, page);
39279cc3 4140 lock_page(page);
211c17f5
CM		 4141 if (page->mapping != mapping) {
4142 unlock_page(page);
4143 page_cache_release(page);
4144 goto again;
4145 }
39279cc3
CM		 4146 if (!PageUptodate(page)) {
4147 ret = -EIO;
89642229 4148 goto out_unlock;
39279cc3
CM		 4149 }
4150 }
211c17f5 4151 wait_on_page_writeback(page);
e6dcd2dc 4152
d0082371 4153 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4154 set_page_extent_mapped(page);
4155
4156 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4157 if (ordered) {
2ac55d41
JB		 4158 unlock_extent_cached(io_tree, page_start, page_end,
4159 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4160 unlock_page(page);
4161 page_cache_release(page);
eb84ae03 4162 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4163 btrfs_put_ordered_extent(ordered);
4164 goto again;
4165 }
4166
2ac55d41 4167 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4168 EXTENT_DIRTY | EXTENT_DELALLOC |
4169 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4170 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4171
2ac55d41
JB		 4172 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4173 &cached_state);
9ed74f2d 4174 if (ret) {
2ac55d41
JB		 4175 unlock_extent_cached(io_tree, page_start, page_end,
4176 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4177 goto out_unlock;
4178 }
4179
e6dcd2dc 4180 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4181 if (!len)
4182 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4183 kaddr = kmap(page);
2aaa6655
JB		 4184 if (front)
4185 memset(kaddr, 0, offset);
4186 else
4187 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4188 flush_dcache_page(page);
4189 kunmap(page);
4190 }
247e743c 4191 ClearPageChecked(page);
e6dcd2dc 4192 set_page_dirty(page);
2ac55d41
JB		 4193 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4194 GFP_NOFS);
39279cc3 4195
89642229 4196out_unlock:
5d5e103a 4197 if (ret)
0ca1f7ce 4198 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4199 unlock_page(page);
4200 page_cache_release(page);
4201out:
4202 return ret;
4203}
4204
695a0d0d
JB		 4205/*
4206 * This function puts in dummy file extents for the area we're creating a hole
4207 * for. So if we are truncating this file to a larger size we need to insert
4208 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4209 * the range between oldsize and size
4210 */
a41ad394 4211int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4212{
9036c102
YZ		 4213 struct btrfs_trans_handle *trans;
4214 struct btrfs_root *root = BTRFS_I(inode)->root;
4215 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4216 struct extent_map *em = NULL;
2ac55d41 4217 struct extent_state *cached_state = NULL;
5dc562c5 4218 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4219 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4220 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4221 u64 last_byte;
4222 u64 cur_offset;
4223 u64 hole_size;
9ed74f2d 4224 int err = 0;
39279cc3 4225
a71754fc
JB		 4226 /*
4227 * If our size started in the middle of a page we need to zero out the
4228 * rest of the page before we expand the i_size, otherwise we could
4229 * expose stale data.
4230 */
4231 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4232 if (err)
4233 return err;
4234
9036c102
YZ		 4235 if (size <= hole_start)
4236 return 0;
4237
9036c102
YZ		 4238 while (1) {
4239 struct btrfs_ordered_extent *ordered;
fa7c1494 4240
2ac55d41 4241 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4242 &cached_state);
fa7c1494
MX		 4243 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4244 block_end - hole_start);
9036c102
YZ		 4245 if (!ordered)
4246 break;
2ac55d41
JB		 4247 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4248 &cached_state, GFP_NOFS);
fa7c1494 4249 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4250 btrfs_put_ordered_extent(ordered);
4251 }
39279cc3 4252
9036c102
YZ		 4253 cur_offset = hole_start;
4254 while (1) {
4255 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4256 block_end - cur_offset, 0);
79787eaa
JM		 4257 if (IS_ERR(em)) {
4258 err = PTR_ERR(em);
f2767956 4259 em = NULL;
79787eaa
JM		 4260 break;
4261 }
9036c102 4262 last_byte = min(extent_map_end(em), block_end);
fda2832f 4263 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4264 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4265 struct extent_map *hole_em;
9036c102 4266 hole_size = last_byte - cur_offset;
9ed74f2d 4267
3642320e 4268 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4269 if (IS_ERR(trans)) {
4270 err = PTR_ERR(trans);
9ed74f2d 4271 break;
a22285a6 4272 }
8082510e 4273
5dc562c5
JB		 4274 err = btrfs_drop_extents(trans, root, inode,
4275 cur_offset,
2671485d 4276 cur_offset + hole_size, 1);
5b397377 4277 if (err) {
79787eaa 4278 btrfs_abort_transaction(trans, root, err);
5b397377 4279 btrfs_end_transaction(trans, root);
3893e33b 4280 break;
5b397377 4281 }
8082510e 4282
9036c102 4283 err = btrfs_insert_file_extent(trans, root,
33345d01 4284 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4285 0, hole_size, 0, hole_size,
4286 0, 0, 0);
5b397377 4287 if (err) {
79787eaa 4288 btrfs_abort_transaction(trans, root, err);
5b397377 4289 btrfs_end_transaction(trans, root);
3893e33b 4290 break;
5b397377 4291 }
8082510e 4292
5dc562c5
JB		 4293 btrfs_drop_extent_cache(inode, cur_offset,
4294 cur_offset + hole_size - 1, 0);
4295 hole_em = alloc_extent_map();
4296 if (!hole_em) {
4297 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4298 &BTRFS_I(inode)->runtime_flags);
4299 goto next;
4300 }
4301 hole_em->start = cur_offset;
4302 hole_em->len = hole_size;
4303 hole_em->orig_start = cur_offset;
8082510e 4304
5dc562c5
JB		 4305 hole_em->block_start = EXTENT_MAP_HOLE;
4306 hole_em->block_len = 0;
b4939680 4307 hole_em->orig_block_len = 0;
cc95bef6 4308 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4309 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4310 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4311 hole_em->generation = trans->transid;
8082510e 4312
5dc562c5
JB		 4313 while (1) {
4314 write_lock(&em_tree->lock);
09a2a8f9 4315 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4316 write_unlock(&em_tree->lock);
4317 if (err != -EEXIST)
4318 break;
4319 btrfs_drop_extent_cache(inode, cur_offset,
4320 cur_offset +
4321 hole_size - 1, 0);
4322 }
4323 free_extent_map(hole_em);
4324next:
3642320e 4325 btrfs_update_inode(trans, root, inode);
8082510e 4326 btrfs_end_transaction(trans, root);
9036c102
YZ		 4327 }
4328 free_extent_map(em);
a22285a6 4329 em = NULL;
9036c102 4330 cur_offset = last_byte;
8082510e 4331 if (cur_offset >= block_end)
9036c102
YZ		 4332 break;
4333 }
1832a6d5 4334
a22285a6 4335 free_extent_map(em);
2ac55d41
JB		 4336 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4337 GFP_NOFS);
9036c102
YZ		 4338 return err;
4339}
39279cc3 4340
3972f260 4341static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4342{
f4a2f4c5
MX		 4343 struct btrfs_root *root = BTRFS_I(inode)->root;
4344 struct btrfs_trans_handle *trans;
a41ad394 4345 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4346 loff_t newsize = attr->ia_size;
4347 int mask = attr->ia_valid;
8082510e
YZ		 4348 int ret;
4349
3972f260
ES		 4350 /*
4351 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4352 * special case where we need to update the times despite not having
4353 * these flags set. For all other operations the VFS set these flags
4354 * explicitly if it wants a timestamp update.
4355 */
4356 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4357 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4358
a41ad394 4359 if (newsize > oldsize) {
7caef267 4360 truncate_pagecache(inode, newsize);
a41ad394 4361 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4362 if (ret)
8082510e 4363 return ret;
8082510e 4364
f4a2f4c5
MX		 4365 trans = btrfs_start_transaction(root, 1);
4366 if (IS_ERR(trans))
4367 return PTR_ERR(trans);
4368
4369 i_size_write(inode, newsize);
4370 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4371 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4372 btrfs_end_transaction(trans, root);
a41ad394 4373 } else {
8082510e 4374
a41ad394
JB		 4375 /*
4376 * We're truncating a file that used to have good data down to
4377 * zero. Make sure it gets into the ordered flush list so that
4378 * any new writes get down to disk quickly.
4379 */
4380 if (newsize == 0)
72ac3c0d
JB		 4381 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4382 &BTRFS_I(inode)->runtime_flags);
8082510e 4383
f3fe820c
JB		 4384 /*
4385 * 1 for the orphan item we're going to add
4386 * 1 for the orphan item deletion.
4387 */
4388 trans = btrfs_start_transaction(root, 2);
4389 if (IS_ERR(trans))
4390 return PTR_ERR(trans);
4391
4392 /*
4393 * We need to do this in case we fail at _any_ point during the
4394 * actual truncate. Once we do the truncate_setsize we could
4395 * invalidate pages which forces any outstanding ordered io to
4396 * be instantly completed which will give us extents that need
4397 * to be truncated. If we fail to get an orphan inode down we
4398 * could have left over extents that were never meant to live,
4399 * so we need to garuntee from this point on that everything
4400 * will be consistent.
4401 */
4402 ret = btrfs_orphan_add(trans, inode);
4403 btrfs_end_transaction(trans, root);
4404 if (ret)
4405 return ret;
4406
a41ad394
JB		 4407 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4408 truncate_setsize(inode, newsize);
2e60a51e
MX		 4409
4410 /* Disable nonlocked read DIO to avoid the end less truncate */
4411 btrfs_inode_block_unlocked_dio(inode);
4412 inode_dio_wait(inode);
4413 btrfs_inode_resume_unlocked_dio(inode);
4414
a41ad394 4415 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4416 if (ret && inode->i_nlink) {
4417 int err;
4418
4419 /*
4420 * failed to truncate, disk_i_size is only adjusted down
4421 * as we remove extents, so it should represent the true
4422 * size of the inode, so reset the in memory size and
4423 * delete our orphan entry.
4424 */
4425 trans = btrfs_join_transaction(root);
4426 if (IS_ERR(trans)) {
4427 btrfs_orphan_del(NULL, inode);
4428 return ret;
4429 }
4430 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4431 err = btrfs_orphan_del(trans, inode);
4432 if (err)
4433 btrfs_abort_transaction(trans, root, err);
4434 btrfs_end_transaction(trans, root);
4435 }
8082510e
YZ		 4436 }
4437
a41ad394 4438 return ret;
8082510e
YZ		 4439}
4440
9036c102
YZ		 4441static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4442{
4443 struct inode *inode = dentry->d_inode;
b83cc969 4444 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4445 int err;
39279cc3 4446
b83cc969
LZ		 4447 if (btrfs_root_readonly(root))
4448 return -EROFS;
4449
9036c102
YZ		 4450 err = inode_change_ok(inode, attr);
4451 if (err)
4452 return err;
2bf5a725 4453
5a3f23d5 4454 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4455 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4456 if (err)
4457 return err;
39279cc3 4458 }
9036c102 4459
1025774c
CH		 4460 if (attr->ia_valid) {
4461 setattr_copy(inode, attr);
0c4d2d95 4462 inode_inc_iversion(inode);
22c44fe6 4463 err = btrfs_dirty_inode(inode);
1025774c 4464
22c44fe6 4465 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4466 err = btrfs_acl_chmod(inode);
4467 }
33268eaf 4468
39279cc3
CM		 4469 return err;
4470}
61295eb8 4471
bd555975 4472void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4473{
4474 struct btrfs_trans_handle *trans;
4475 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4476 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4477 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4478 int ret;
4479
1abe9b8a 4480 trace_btrfs_inode_evict(inode);
4481
39279cc3 4482 truncate_inode_pages(&inode->i_data, 0);
69e9c6c6
SB		 4483 if (inode->i_nlink &&
4484 ((btrfs_root_refs(&root->root_item) != 0 &&
4485 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4486 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4487 goto no_delete;
4488
39279cc3 4489 if (is_bad_inode(inode)) {
7b128766 4490 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4491 goto no_delete;
4492 }
bd555975 4493 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4494 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4495
c71bf099 4496 if (root->fs_info->log_root_recovering) {
6bf02314 4497 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4498 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4499 goto no_delete;
4500 }
4501
76dda93c 4502 if (inode->i_nlink > 0) {
69e9c6c6
SB		 4503 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4504 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 4505 goto no_delete;
4506 }
4507
0e8c36a9
MX		 4508 ret = btrfs_commit_inode_delayed_inode(inode);
4509 if (ret) {
4510 btrfs_orphan_del(NULL, inode);
4511 goto no_delete;
4512 }
4513
66d8f3dd 4514 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4515 if (!rsv) {
4516 btrfs_orphan_del(NULL, inode);
4517 goto no_delete;
4518 }
4a338542 4519 rsv->size = min_size;
ca7e70f5 4520 rsv->failfast = 1;
726c35fa 4521 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4522
dbe674a9 4523 btrfs_i_size_write(inode, 0);
5f39d397 4524
4289a667 4525 /*
8407aa46
MX		 4526 * This is a bit simpler than btrfs_truncate since we've already
4527 * reserved our space for our orphan item in the unlink, so we just
4528 * need to reserve some slack space in case we add bytes and update
4529 * inode item when doing the truncate.
4289a667 4530 */
8082510e 4531 while (1) {
08e007d2
MX		 4532 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4533 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4534
4535 /*
4536 * Try and steal from the global reserve since we will
4537 * likely not use this space anyway, we want to try as
4538 * hard as possible to get this to work.
4539 */
4540 if (ret)
4541 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4542
d68fc57b 4543 if (ret) {
c2cf52eb
SK		 4544 btrfs_warn(root->fs_info,
4545 "Could not get space for a delete, will truncate on mount %d",
4546 ret);
4289a667
JB		 4547 btrfs_orphan_del(NULL, inode);
4548 btrfs_free_block_rsv(root, rsv);
4549 goto no_delete;
d68fc57b 4550 }
7b128766 4551
0e8c36a9 4552 trans = btrfs_join_transaction(root);
4289a667
JB		 4553 if (IS_ERR(trans)) {
4554 btrfs_orphan_del(NULL, inode);
4555 btrfs_free_block_rsv(root, rsv);
4556 goto no_delete;
d68fc57b 4557 }
7b128766 4558
4289a667
JB		 4559 trans->block_rsv = rsv;
4560
d68fc57b 4561 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4562 if (ret != -ENOSPC)
8082510e 4563 break;
85e21bac 4564
8407aa46 4565 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4566 btrfs_end_transaction(trans, root);
4567 trans = NULL;
b53d3f5d 4568 btrfs_btree_balance_dirty(root);
8082510e 4569 }
5f39d397 4570
4289a667
JB		 4571 btrfs_free_block_rsv(root, rsv);
4572
4ef31a45
JB		 4573 /*
4574 * Errors here aren't a big deal, it just means we leave orphan items
4575 * in the tree. They will be cleaned up on the next mount.
4576 */
8082510e 4577 if (ret == 0) {
4289a667 4578 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 4579 btrfs_orphan_del(trans, inode);
4580 } else {
4581 btrfs_orphan_del(NULL, inode);
8082510e 4582 }
54aa1f4d 4583
4289a667 4584 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4585 if (!(root == root->fs_info->tree_root ||
4586 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4587 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4588
54aa1f4d 4589 btrfs_end_transaction(trans, root);
b53d3f5d 4590 btrfs_btree_balance_dirty(root);
39279cc3 4591no_delete:
89042e5a 4592 btrfs_remove_delayed_node(inode);
dbd5768f 4593 clear_inode(inode);
8082510e 4594 return;
39279cc3
CM		 4595}
4596
4597/*
4598 * this returns the key found in the dir entry in the location pointer.
4599 * If no dir entries were found, location->objectid is 0.
4600 */
4601static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4602 struct btrfs_key *location)
4603{
4604 const char *name = dentry->d_name.name;
4605 int namelen = dentry->d_name.len;
4606 struct btrfs_dir_item *di;
4607 struct btrfs_path *path;
4608 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4609 int ret = 0;
39279cc3
CM		 4610
4611 path = btrfs_alloc_path();
d8926bb3
MF		 4612 if (!path)
4613 return -ENOMEM;
3954401f 4614
33345d01 4615 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4616 namelen, 0);
0d9f7f3e
Y		 4617 if (IS_ERR(di))
4618 ret = PTR_ERR(di);
d397712b 4619
c704005d 4620 if (IS_ERR_OR_NULL(di))
3954401f 4621 goto out_err;
d397712b 4622
5f39d397 4623 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4624out:
39279cc3
CM		 4625 btrfs_free_path(path);
4626 return ret;
3954401f
CM		 4627out_err:
4628 location->objectid = 0;
4629 goto out;
39279cc3
CM		 4630}
4631
4632/*
4633 * when we hit a tree root in a directory, the btrfs part of the inode
4634 * needs to be changed to reflect the root directory of the tree root. This
4635 * is kind of like crossing a mount point.
4636 */
4637static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4638 struct inode *dir,
4639 struct dentry *dentry,
4640 struct btrfs_key *location,
4641 struct btrfs_root **sub_root)
39279cc3 4642{
4df27c4d
YZ		 4643 struct btrfs_path *path;
4644 struct btrfs_root *new_root;
4645 struct btrfs_root_ref *ref;
4646 struct extent_buffer *leaf;
4647 int ret;
4648 int err = 0;
39279cc3 4649
4df27c4d
YZ		 4650 path = btrfs_alloc_path();
4651 if (!path) {
4652 err = -ENOMEM;
4653 goto out;
4654 }
39279cc3 4655
4df27c4d
YZ		 4656 err = -ENOENT;
4657 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4658 BTRFS_I(dir)->root->root_key.objectid,
4659 location->objectid);
4660 if (ret) {
4661 if (ret < 0)
4662 err = ret;
4663 goto out;
4664 }
39279cc3 4665
4df27c4d
YZ		 4666 leaf = path->nodes[0];
4667 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4668 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4669 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4670 goto out;
39279cc3 4671
4df27c4d
YZ		 4672 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4673 (unsigned long)(ref + 1),
4674 dentry->d_name.len);
4675 if (ret)
4676 goto out;
4677
b3b4aa74 4678 btrfs_release_path(path);
4df27c4d
YZ		 4679
4680 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4681 if (IS_ERR(new_root)) {
4682 err = PTR_ERR(new_root);
4683 goto out;
4684 }
4685
4df27c4d
YZ		 4686 *sub_root = new_root;
4687 location->objectid = btrfs_root_dirid(&new_root->root_item);
4688 location->type = BTRFS_INODE_ITEM_KEY;
4689 location->offset = 0;
4690 err = 0;
4691out:
4692 btrfs_free_path(path);
4693 return err;
39279cc3
CM		 4694}
4695
5d4f98a2
YZ		 4696static void inode_tree_add(struct inode *inode)
4697{
4698 struct btrfs_root *root = BTRFS_I(inode)->root;
4699 struct btrfs_inode *entry;
03e860bd
FNP		 4700 struct rb_node **p;
4701 struct rb_node *parent;
cef21937 4702 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 4703 u64 ino = btrfs_ino(inode);
5d4f98a2 4704
1d3382cb 4705 if (inode_unhashed(inode))
76dda93c 4706 return;
e1409cef 4707 parent = NULL;
5d4f98a2 4708 spin_lock(&root->inode_lock);
e1409cef 4709 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 4710 while (*p) {
4711 parent = *p;
4712 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4713
33345d01 4714 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4715 p = &parent->rb_left;
33345d01 4716 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4717 p = &parent->rb_right;
5d4f98a2
YZ		 4718 else {
4719 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4720 (I_WILL_FREE | I_FREEING)));
cef21937 4721 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 4722 RB_CLEAR_NODE(parent);
4723 spin_unlock(&root->inode_lock);
cef21937 4724 return;
5d4f98a2
YZ		 4725 }
4726 }
cef21937
FDBM		 4727 rb_link_node(new, parent, p);
4728 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 4729 spin_unlock(&root->inode_lock);
4730}
4731
4732static void inode_tree_del(struct inode *inode)
4733{
4734 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4735 int empty = 0;
5d4f98a2 4736
03e860bd 4737 spin_lock(&root->inode_lock);
5d4f98a2 4738 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4739 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4740 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4741 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4742 }
03e860bd 4743 spin_unlock(&root->inode_lock);
76dda93c 4744
69e9c6c6 4745 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 4746 synchronize_srcu(&root->fs_info->subvol_srcu);
4747 spin_lock(&root->inode_lock);
4748 empty = RB_EMPTY_ROOT(&root->inode_tree);
4749 spin_unlock(&root->inode_lock);
4750 if (empty)
4751 btrfs_add_dead_root(root);
4752 }
4753}
4754
143bede5 4755void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4756{
4757 struct rb_node *node;
4758 struct rb_node *prev;
4759 struct btrfs_inode *entry;
4760 struct inode *inode;
4761 u64 objectid = 0;
4762
4763 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4764
4765 spin_lock(&root->inode_lock);
4766again:
4767 node = root->inode_tree.rb_node;
4768 prev = NULL;
4769 while (node) {
4770 prev = node;
4771 entry = rb_entry(node, struct btrfs_inode, rb_node);
4772
33345d01 4773 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4774 node = node->rb_left;
33345d01 4775 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4776 node = node->rb_right;
4777 else
4778 break;
4779 }
4780 if (!node) {
4781 while (prev) {
4782 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4783 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4784 node = prev;
4785 break;
4786 }
4787 prev = rb_next(prev);
4788 }
4789 }
4790 while (node) {
4791 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4792 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4793 inode = igrab(&entry->vfs_inode);
4794 if (inode) {
4795 spin_unlock(&root->inode_lock);
4796 if (atomic_read(&inode->i_count) > 1)
4797 d_prune_aliases(inode);
4798 /*
45321ac5 4799 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4800 * the inode cache when its usage count
4801 * hits zero.
4802 */
4803 iput(inode);
4804 cond_resched();
4805 spin_lock(&root->inode_lock);
4806 goto again;
4807 }
4808
4809 if (cond_resched_lock(&root->inode_lock))
4810 goto again;
4811
4812 node = rb_next(node);
4813 }
4814 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4815}
4816
e02119d5
CM		 4817static int btrfs_init_locked_inode(struct inode *inode, void *p)
4818{
4819 struct btrfs_iget_args *args = p;
4820 inode->i_ino = args->ino;
e02119d5 4821 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4822 return 0;
4823}
4824
4825static int btrfs_find_actor(struct inode *inode, void *opaque)
4826{
4827 struct btrfs_iget_args *args = opaque;
33345d01 4828 return args->ino == btrfs_ino(inode) &&
d397712b 4829 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4830}
4831
5d4f98a2
YZ		 4832static struct inode *btrfs_iget_locked(struct super_block *s,
4833 u64 objectid,
4834 struct btrfs_root *root)
39279cc3
CM		 4835{
4836 struct inode *inode;
4837 struct btrfs_iget_args args;
778ba82b
FDBM		 4838 unsigned long hashval = btrfs_inode_hash(objectid, root);
4839
39279cc3
CM		 4840 args.ino = objectid;
4841 args.root = root;
4842
778ba82b 4843 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 4844 btrfs_init_locked_inode,
4845 (void *)&args);
4846 return inode;
4847}
4848
1a54ef8c
BR		 4849/* Get an inode object given its location and corresponding root.
4850 * Returns in *is_new if the inode was read from disk
4851 */
4852struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4853 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4854{
4855 struct inode *inode;
4856
4857 inode = btrfs_iget_locked(s, location->objectid, root);
4858 if (!inode)
5d4f98a2 4859 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4860
4861 if (inode->i_state & I_NEW) {
4862 BTRFS_I(inode)->root = root;
4863 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4864 btrfs_read_locked_inode(inode);
1748f843
MF		 4865 if (!is_bad_inode(inode)) {
4866 inode_tree_add(inode);
4867 unlock_new_inode(inode);
4868 if (new)
4869 *new = 1;
4870 } else {
e0b6d65b
ST		 4871 unlock_new_inode(inode);
4872 iput(inode);
4873 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4874 }
4875 }
4876
1a54ef8c
BR		 4877 return inode;
4878}
4879
4df27c4d
YZ		 4880static struct inode *new_simple_dir(struct super_block *s,
4881 struct btrfs_key *key,
4882 struct btrfs_root *root)
4883{
4884 struct inode *inode = new_inode(s);
4885
4886 if (!inode)
4887 return ERR_PTR(-ENOMEM);
4888
4df27c4d
YZ		 4889 BTRFS_I(inode)->root = root;
4890 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4891 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4892
4893 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4894 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4895 inode->i_fop = &simple_dir_operations;
4896 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4897 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4898
4899 return inode;
4900}
4901
3de4586c 4902struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4903{
d397712b 4904 struct inode *inode;
4df27c4d 4905 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4906 struct btrfs_root *sub_root = root;
4907 struct btrfs_key location;
76dda93c 4908 int index;
b4aff1f8 4909 int ret = 0;
39279cc3
CM		 4910
4911 if (dentry->d_name.len > BTRFS_NAME_LEN)
4912 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4913
39e3c955 4914 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 4915 if (ret < 0)
4916 return ERR_PTR(ret);
5f39d397 4917
4df27c4d
YZ		 4918 if (location.objectid == 0)
4919 return NULL;
4920
4921 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4922 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4923 return inode;
4924 }
4925
4926 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4927
76dda93c 4928 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4929 ret = fixup_tree_root_location(root, dir, dentry,
4930 &location, &sub_root);
4931 if (ret < 0) {
4932 if (ret != -ENOENT)
4933 inode = ERR_PTR(ret);
4934 else
4935 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4936 } else {
73f73415 4937 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4938 }
76dda93c
YZ		 4939 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4940
34d19bad 4941 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4942 down_read(&root->fs_info->cleanup_work_sem);
4943 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4944 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4945 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 4946 if (ret) {
4947 iput(inode);
66b4ffd1 4948 inode = ERR_PTR(ret);
01cd3367 4949 }
c71bf099
YZ		 4950 }
4951
3de4586c
CM		 4952 return inode;
4953}
4954
fe15ce44 4955static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4956{
4957 struct btrfs_root *root;
848cce0d 4958 struct inode *inode = dentry->d_inode;
76dda93c 4959
848cce0d
LZ		 4960 if (!inode && !IS_ROOT(dentry))
4961 inode = dentry->d_parent->d_inode;
76dda93c 4962
848cce0d
LZ		 4963 if (inode) {
4964 root = BTRFS_I(inode)->root;
efefb143
YZ		 4965 if (btrfs_root_refs(&root->root_item) == 0)
4966 return 1;
848cce0d
LZ		 4967
4968 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4969 return 1;
efefb143 4970 }
76dda93c
YZ		 4971 return 0;
4972}
4973
b4aff1f8
JB		 4974static void btrfs_dentry_release(struct dentry *dentry)
4975{
4976 if (dentry->d_fsdata)
4977 kfree(dentry->d_fsdata);
4978}
4979
3de4586c 4980static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 4981 unsigned int flags)
3de4586c 4982{
a66e7cc6
JB		 4983 struct dentry *ret;
4984
4985 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 4986 return ret;
39279cc3
CM		 4987}
4988
16cdcec7 4989unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4990 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4991};
4992
9cdda8d3 4993static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 4994{
9cdda8d3 4995 struct inode *inode = file_inode(file);
39279cc3
CM		 4996 struct btrfs_root *root = BTRFS_I(inode)->root;
4997 struct btrfs_item *item;
4998 struct btrfs_dir_item *di;
4999 struct btrfs_key key;
5f39d397 5000 struct btrfs_key found_key;
39279cc3 5001 struct btrfs_path *path;
16cdcec7
MX		 5002 struct list_head ins_list;
5003 struct list_head del_list;
39279cc3 5004 int ret;
5f39d397 5005 struct extent_buffer *leaf;
39279cc3 5006 int slot;
39279cc3
CM		 5007 unsigned char d_type;
5008 int over = 0;
5009 u32 di_cur;
5010 u32 di_total;
5011 u32 di_len;
5012 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5013 char tmp_name[32];
5014 char *name_ptr;
5015 int name_len;
9cdda8d3 5016 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5017
5018 /* FIXME, use a real flag for deciding about the key type */
5019 if (root->fs_info->tree_root == root)
5020 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5021
9cdda8d3
AV		 5022 if (!dir_emit_dots(file, ctx))
5023 return 0;
5024
49593bfa 5025 path = btrfs_alloc_path();
16cdcec7
MX		 5026 if (!path)
5027 return -ENOMEM;
ff5714cc 5028
026fd317 5029 path->reada = 1;
49593bfa 5030
16cdcec7
MX		 5031 if (key_type == BTRFS_DIR_INDEX_KEY) {
5032 INIT_LIST_HEAD(&ins_list);
5033 INIT_LIST_HEAD(&del_list);
5034 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5035 }
5036
39279cc3 5037 btrfs_set_key_type(&key, key_type);
9cdda8d3 5038 key.offset = ctx->pos;
33345d01 5039 key.objectid = btrfs_ino(inode);
5f39d397 5040
39279cc3
CM		 5041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5042 if (ret < 0)
5043 goto err;
49593bfa
DW		 5044
5045 while (1) {
5f39d397 5046 leaf = path->nodes[0];
39279cc3 5047 slot = path->slots[0];
b9e03af0
LZ		 5048 if (slot >= btrfs_header_nritems(leaf)) {
5049 ret = btrfs_next_leaf(root, path);
5050 if (ret < 0)
5051 goto err;
5052 else if (ret > 0)
5053 break;
5054 continue;
39279cc3 5055 }
3de4586c 5056
dd3cc16b 5057 item = btrfs_item_nr(slot);
5f39d397
CM		 5058 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5059
5060 if (found_key.objectid != key.objectid)
39279cc3 5061 break;
5f39d397 5062 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5063 break;
9cdda8d3 5064 if (found_key.offset < ctx->pos)
b9e03af0 5065 goto next;
16cdcec7
MX		 5066 if (key_type == BTRFS_DIR_INDEX_KEY &&
5067 btrfs_should_delete_dir_index(&del_list,
5068 found_key.offset))
5069 goto next;
5f39d397 5070
9cdda8d3 5071 ctx->pos = found_key.offset;
16cdcec7 5072 is_curr = 1;
49593bfa 5073
39279cc3
CM		 5074 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5075 di_cur = 0;
5f39d397 5076 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5077
5078 while (di_cur < di_total) {
5f39d397
CM		 5079 struct btrfs_key location;
5080
22a94d44
JB		 5081 if (verify_dir_item(root, leaf, di))
5082 break;
5083
5f39d397 5084 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5085 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5086 name_ptr = tmp_name;
5087 } else {
5088 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5089 if (!name_ptr) {
5090 ret = -ENOMEM;
5091 goto err;
5092 }
5f39d397
CM		 5093 }
5094 read_extent_buffer(leaf, name_ptr,
5095 (unsigned long)(di + 1), name_len);
5096
5097 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5098 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5099
fede766f 5100
3de4586c 5101 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5102 * skip it.
5103 *
5104 * In contrast to old kernels, we insert the snapshot's
5105 * dir item and dir index after it has been created, so
5106 * we won't find a reference to our own snapshot. We
5107 * still keep the following code for backward
5108 * compatibility.
3de4586c
CM		 5109 */
5110 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5111 location.objectid == root->root_key.objectid) {
5112 over = 0;
5113 goto skip;
5114 }
9cdda8d3
AV		 5115 over = !dir_emit(ctx, name_ptr, name_len,
5116 location.objectid, d_type);
5f39d397 5117
3de4586c 5118skip:
5f39d397
CM		 5119 if (name_ptr != tmp_name)
5120 kfree(name_ptr);
5121
39279cc3
CM		 5122 if (over)
5123 goto nopos;
5103e947 5124 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5125 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5126 di_cur += di_len;
5127 di = (struct btrfs_dir_item *)((char *)di + di_len);
5128 }
b9e03af0
LZ		 5129next:
5130 path->slots[0]++;
39279cc3 5131 }
49593bfa 5132
16cdcec7
MX		 5133 if (key_type == BTRFS_DIR_INDEX_KEY) {
5134 if (is_curr)
9cdda8d3
AV		 5135 ctx->pos++;
5136 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5137 if (ret)
5138 goto nopos;
5139 }
5140
49593bfa 5141 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5142 ctx->pos++;
5143
5144 /*
5145 * Stop new entries from being returned after we return the last
5146 * entry.
5147 *
5148 * New directory entries are assigned a strictly increasing
5149 * offset. This means that new entries created during readdir
5150 * are *guaranteed* to be seen in the future by that readdir.
5151 * This has broken buggy programs which operate on names as
5152 * they're returned by readdir. Until we re-use freed offsets
5153 * we have this hack to stop new entries from being returned
5154 * under the assumption that they'll never reach this huge
5155 * offset.
5156 *
5157 * This is being careful not to overflow 32bit loff_t unless the
5158 * last entry requires it because doing so has broken 32bit apps
5159 * in the past.
5160 */
5161 if (key_type == BTRFS_DIR_INDEX_KEY) {
5162 if (ctx->pos >= INT_MAX)
5163 ctx->pos = LLONG_MAX;
5164 else
5165 ctx->pos = INT_MAX;
5166 }
39279cc3
CM		 5167nopos:
5168 ret = 0;
5169err:
16cdcec7
MX		 5170 if (key_type == BTRFS_DIR_INDEX_KEY)
5171 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5172 btrfs_free_path(path);
39279cc3
CM		 5173 return ret;
5174}
5175
a9185b41 5176int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5177{
5178 struct btrfs_root *root = BTRFS_I(inode)->root;
5179 struct btrfs_trans_handle *trans;
5180 int ret = 0;
0af3d00b 5181 bool nolock = false;
39279cc3 5182
72ac3c0d 5183 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5184 return 0;
5185
83eea1f1 5186 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5187 nolock = true;
0af3d00b 5188
a9185b41 5189 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5190 if (nolock)
7a7eaa40 5191 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5192 else
7a7eaa40 5193 trans = btrfs_join_transaction(root);
3612b495
TI		 5194 if (IS_ERR(trans))
5195 return PTR_ERR(trans);
a698d075 5196 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5197 }
5198 return ret;
5199}
5200
5201/*
54aa1f4d 5202 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5203 * inode changes. But, it is most likely to find the inode in cache.
5204 * FIXME, needs more benchmarking...there are no reasons other than performance
5205 * to keep or drop this code.
5206 */
48a3b636 5207static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5208{
5209 struct btrfs_root *root = BTRFS_I(inode)->root;
5210 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5211 int ret;
5212
72ac3c0d 5213 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5214 return 0;
39279cc3 5215
7a7eaa40 5216 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5217 if (IS_ERR(trans))
5218 return PTR_ERR(trans);
8929ecfa
YZ		 5219
5220 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5221 if (ret && ret == -ENOSPC) {
5222 /* whoops, lets try again with the full transaction */
5223 btrfs_end_transaction(trans, root);
5224 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5225 if (IS_ERR(trans))
5226 return PTR_ERR(trans);
8929ecfa 5227
94b60442 5228 ret = btrfs_update_inode(trans, root, inode);
94b60442 5229 }
39279cc3 5230 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5231 if (BTRFS_I(inode)->delayed_node)
5232 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5233
5234 return ret;
5235}
5236
5237/*
5238 * This is a copy of file_update_time. We need this so we can return error on
5239 * ENOSPC for updating the inode in the case of file write and mmap writes.
5240 */
e41f941a
JB		 5241static int btrfs_update_time(struct inode *inode, struct timespec *now,
5242 int flags)
22c44fe6 5243{
2bc55652
AB		 5244 struct btrfs_root *root = BTRFS_I(inode)->root;
5245
5246 if (btrfs_root_readonly(root))
5247 return -EROFS;
5248
e41f941a 5249 if (flags & S_VERSION)
22c44fe6 5250 inode_inc_iversion(inode);
e41f941a
JB		 5251 if (flags & S_CTIME)
5252 inode->i_ctime = *now;
5253 if (flags & S_MTIME)
5254 inode->i_mtime = *now;
5255 if (flags & S_ATIME)
5256 inode->i_atime = *now;
5257 return btrfs_dirty_inode(inode);
39279cc3
CM		 5258}
5259
d352ac68
CM		 5260/*
5261 * find the highest existing sequence number in a directory
5262 * and then set the in-memory index_cnt variable to reflect
5263 * free sequence numbers
5264 */
aec7477b
JB		 5265static int btrfs_set_inode_index_count(struct inode *inode)
5266{
5267 struct btrfs_root *root = BTRFS_I(inode)->root;
5268 struct btrfs_key key, found_key;
5269 struct btrfs_path *path;
5270 struct extent_buffer *leaf;
5271 int ret;
5272
33345d01 5273 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5274 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5275 key.offset = (u64)-1;
5276
5277 path = btrfs_alloc_path();
5278 if (!path)
5279 return -ENOMEM;
5280
5281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5282 if (ret < 0)
5283 goto out;
5284 /* FIXME: we should be able to handle this */
5285 if (ret == 0)
5286 goto out;
5287 ret = 0;
5288
5289 /*
5290 * MAGIC NUMBER EXPLANATION:
5291 * since we search a directory based on f_pos we have to start at 2
5292 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5293 * else has to start at 2
5294 */
5295 if (path->slots[0] == 0) {
5296 BTRFS_I(inode)->index_cnt = 2;
5297 goto out;
5298 }
5299
5300 path->slots[0]--;
5301
5302 leaf = path->nodes[0];
5303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5304
33345d01 5305 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5306 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5307 BTRFS_I(inode)->index_cnt = 2;
5308 goto out;
5309 }
5310
5311 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5312out:
5313 btrfs_free_path(path);
5314 return ret;
5315}
5316
d352ac68
CM		 5317/*
5318 * helper to find a free sequence number in a given directory. This current
5319 * code is very simple, later versions will do smarter things in the btree
5320 */
3de4586c 5321int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5322{
5323 int ret = 0;
5324
5325 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5326 ret = btrfs_inode_delayed_dir_index_count(dir);
5327 if (ret) {
5328 ret = btrfs_set_inode_index_count(dir);
5329 if (ret)
5330 return ret;
5331 }
aec7477b
JB		 5332 }
5333
00e4e6b3 5334 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5335 BTRFS_I(dir)->index_cnt++;
5336
5337 return ret;
5338}
5339
39279cc3
CM		 5340static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5341 struct btrfs_root *root,
aec7477b 5342 struct inode *dir,
9c58309d 5343 const char *name, int name_len,
175a4eb7
AV		 5344 u64 ref_objectid, u64 objectid,
5345 umode_t mode, u64 *index)
39279cc3
CM		 5346{
5347 struct inode *inode;
5f39d397 5348 struct btrfs_inode_item *inode_item;
39279cc3 5349 struct btrfs_key *location;
5f39d397 5350 struct btrfs_path *path;
9c58309d
CM		 5351 struct btrfs_inode_ref *ref;
5352 struct btrfs_key key[2];
5353 u32 sizes[2];
5354 unsigned long ptr;
39279cc3
CM		 5355 int ret;
5356 int owner;
5357
5f39d397 5358 path = btrfs_alloc_path();
d8926bb3
MF		 5359 if (!path)
5360 return ERR_PTR(-ENOMEM);
5f39d397 5361
39279cc3 5362 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5363 if (!inode) {
5364 btrfs_free_path(path);
39279cc3 5365 return ERR_PTR(-ENOMEM);
8fb27640 5366 }
39279cc3 5367
581bb050
LZ		 5368 /*
5369 * we have to initialize this early, so we can reclaim the inode
5370 * number if we fail afterwards in this function.
5371 */
5372 inode->i_ino = objectid;
5373
aec7477b 5374 if (dir) {
1abe9b8a 5375 trace_btrfs_inode_request(dir);
5376
3de4586c 5377 ret = btrfs_set_inode_index(dir, index);
09771430 5378 if (ret) {
8fb27640 5379 btrfs_free_path(path);
09771430 5380 iput(inode);
aec7477b 5381 return ERR_PTR(ret);
09771430 5382 }
aec7477b
JB		 5383 }
5384 /*
5385 * index_cnt is ignored for everything but a dir,
5386 * btrfs_get_inode_index_count has an explanation for the magic
5387 * number
5388 */
5389 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5390 BTRFS_I(inode)->root = root;
e02119d5 5391 BTRFS_I(inode)->generation = trans->transid;
76195853 5392 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5393
5dc562c5
JB		 5394 /*
5395 * We could have gotten an inode number from somebody who was fsynced
5396 * and then removed in this same transaction, so let's just set full
5397 * sync since it will be a full sync anyway and this will blow away the
5398 * old info in the log.
5399 */
5400 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5401
569254b0 5402 if (S_ISDIR(mode))
39279cc3
CM		 5403 owner = 0;
5404 else
5405 owner = 1;
9c58309d
CM		 5406
5407 key[0].objectid = objectid;
5408 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5409 key[0].offset = 0;
5410
f186373f
MF		 5411 /*
5412 * Start new inodes with an inode_ref. This is slightly more
5413 * efficient for small numbers of hard links since they will
5414 * be packed into one item. Extended refs will kick in if we
5415 * add more hard links than can fit in the ref item.
5416 */
9c58309d
CM		 5417 key[1].objectid = objectid;
5418 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5419 key[1].offset = ref_objectid;
5420
5421 sizes[0] = sizeof(struct btrfs_inode_item);
5422 sizes[1] = name_len + sizeof(*ref);
5423
b9473439 5424 path->leave_spinning = 1;
9c58309d
CM		 5425 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5426 if (ret != 0)
5f39d397
CM		 5427 goto fail;
5428
ecc11fab 5429 inode_init_owner(inode, dir, mode);
a76a3cd4 5430 inode_set_bytes(inode, 0);
39279cc3 5431 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5432 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5433 struct btrfs_inode_item);
293f7e07
LZ		 5434 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5435 sizeof(*inode_item));
e02119d5 5436 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5437
5438 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5439 struct btrfs_inode_ref);
5440 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5441 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5442 ptr = (unsigned long)(ref + 1);
5443 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5444
5f39d397
CM		 5445 btrfs_mark_buffer_dirty(path->nodes[0]);
5446 btrfs_free_path(path);
5447
39279cc3
CM		 5448 location = &BTRFS_I(inode)->location;
5449 location->objectid = objectid;
39279cc3
CM		 5450 location->offset = 0;
5451 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5452
6cbff00f
CH		 5453 btrfs_inherit_iflags(inode, dir);
5454
569254b0 5455 if (S_ISREG(mode)) {
94272164
CM		 5456 if (btrfs_test_opt(root, NODATASUM))
5457 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5458 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5459 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5460 BTRFS_INODE_NODATASUM;
94272164
CM		 5461 }
5462
778ba82b 5463 btrfs_insert_inode_hash(inode);
5d4f98a2 5464 inode_tree_add(inode);
1abe9b8a 5465
5466 trace_btrfs_inode_new(inode);
1973f0fa 5467 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5468
8ea05e3a
AB		 5469 btrfs_update_root_times(trans, root);
5470
39279cc3 5471 return inode;
5f39d397 5472fail:
aec7477b
JB		 5473 if (dir)
5474 BTRFS_I(dir)->index_cnt--;
5f39d397 5475 btrfs_free_path(path);
09771430 5476 iput(inode);
5f39d397 5477 return ERR_PTR(ret);
39279cc3
CM		 5478}
5479
5480static inline u8 btrfs_inode_type(struct inode *inode)
5481{
5482 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5483}
5484
d352ac68
CM		 5485/*
5486 * utility function to add 'inode' into 'parent_inode' with
5487 * a give name and a given sequence number.
5488 * if 'add_backref' is true, also insert a backref from the
5489 * inode to the parent directory.
5490 */
e02119d5
CM		 5491int btrfs_add_link(struct btrfs_trans_handle *trans,
5492 struct inode *parent_inode, struct inode *inode,
5493 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5494{
4df27c4d 5495 int ret = 0;
39279cc3 5496 struct btrfs_key key;
e02119d5 5497 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5498 u64 ino = btrfs_ino(inode);
5499 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5500
33345d01 5501 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5502 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5503 } else {
33345d01 5504 key.objectid = ino;
4df27c4d
YZ		 5505 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5506 key.offset = 0;
5507 }
5508
33345d01 5509 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5510 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5511 key.objectid, root->root_key.objectid,
33345d01 5512 parent_ino, index, name, name_len);
4df27c4d 5513 } else if (add_backref) {
33345d01
LZ		 5514 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5515 parent_ino, index);
4df27c4d 5516 }
39279cc3 5517
79787eaa
JM		 5518 /* Nothing to clean up yet */
5519 if (ret)
5520 return ret;
4df27c4d 5521
79787eaa
JM		 5522 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5523 parent_inode, &key,
5524 btrfs_inode_type(inode), index);
9c52057c 5525 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5526 goto fail_dir_item;
5527 else if (ret) {
5528 btrfs_abort_transaction(trans, root, ret);
5529 return ret;
39279cc3 5530 }
79787eaa
JM		 5531
5532 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5533 name_len * 2);
0c4d2d95 5534 inode_inc_iversion(parent_inode);
79787eaa
JM		 5535 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5536 ret = btrfs_update_inode(trans, root, parent_inode);
5537 if (ret)
5538 btrfs_abort_transaction(trans, root, ret);
39279cc3 5539 return ret;
fe66a05a
CM		 5540
5541fail_dir_item:
5542 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5543 u64 local_index;
5544 int err;
5545 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5546 key.objectid, root->root_key.objectid,
5547 parent_ino, &local_index, name, name_len);
5548
5549 } else if (add_backref) {
5550 u64 local_index;
5551 int err;
5552
5553 err = btrfs_del_inode_ref(trans, root, name, name_len,
5554 ino, parent_ino, &local_index);
5555 }
5556 return ret;
39279cc3
CM		 5557}
5558
5559static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5560 struct inode *dir, struct dentry *dentry,
5561 struct inode *inode, int backref, u64 index)
39279cc3 5562{
a1b075d2
JB		 5563 int err = btrfs_add_link(trans, dir, inode,
5564 dentry->d_name.name, dentry->d_name.len,
5565 backref, index);
39279cc3
CM		 5566 if (err > 0)
5567 err = -EEXIST;
5568 return err;
5569}
5570
618e21d5 5571static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5572 umode_t mode, dev_t rdev)
618e21d5
JB		 5573{
5574 struct btrfs_trans_handle *trans;
5575 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5576 struct inode *inode = NULL;
618e21d5
JB		 5577 int err;
5578 int drop_inode = 0;
5579 u64 objectid;
00e4e6b3 5580 u64 index = 0;
618e21d5
JB		 5581
5582 if (!new_valid_dev(rdev))
5583 return -EINVAL;
5584
9ed74f2d
JB		 5585 /*
5586 * 2 for inode item and ref
5587 * 2 for dir items
5588 * 1 for xattr if selinux is on
5589 */
a22285a6
YZ		 5590 trans = btrfs_start_transaction(root, 5);
5591 if (IS_ERR(trans))
5592 return PTR_ERR(trans);
1832a6d5 5593
581bb050
LZ		 5594 err = btrfs_find_free_ino(root, &objectid);
5595 if (err)
5596 goto out_unlock;
5597
aec7477b 5598 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5599 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5600 mode, &index);
7cf96da3
TI		 5601 if (IS_ERR(inode)) {
5602 err = PTR_ERR(inode);
618e21d5 5603 goto out_unlock;
7cf96da3 5604 }
618e21d5 5605
2a7dba39 5606 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5607 if (err) {
5608 drop_inode = 1;
5609 goto out_unlock;
5610 }
5611
ad19db71
CS		 5612 /*
5613 * If the active LSM wants to access the inode during
5614 * d_instantiate it needs these. Smack checks to see
5615 * if the filesystem supports xattrs by looking at the
5616 * ops vector.
5617 */
5618
5619 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5620 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5621 if (err)
5622 drop_inode = 1;
5623 else {
618e21d5 5624 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5625 btrfs_update_inode(trans, root, inode);
08c422c2 5626 d_instantiate(dentry, inode);
618e21d5 5627 }
618e21d5 5628out_unlock:
7ad85bb7 5629 btrfs_end_transaction(trans, root);
b53d3f5d 5630 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5631 if (drop_inode) {
5632 inode_dec_link_count(inode);
5633 iput(inode);
5634 }
618e21d5
JB		 5635 return err;
5636}
5637
39279cc3 5638static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5639 umode_t mode, bool excl)
39279cc3
CM		 5640{
5641 struct btrfs_trans_handle *trans;
5642 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5643 struct inode *inode = NULL;
43baa579 5644 int drop_inode_on_err = 0;
a22285a6 5645 int err;
39279cc3 5646 u64 objectid;
00e4e6b3 5647 u64 index = 0;
39279cc3 5648
9ed74f2d
JB		 5649 /*
5650 * 2 for inode item and ref
5651 * 2 for dir items
5652 * 1 for xattr if selinux is on
5653 */
a22285a6
YZ		 5654 trans = btrfs_start_transaction(root, 5);
5655 if (IS_ERR(trans))
5656 return PTR_ERR(trans);
9ed74f2d 5657
581bb050
LZ		 5658 err = btrfs_find_free_ino(root, &objectid);
5659 if (err)
5660 goto out_unlock;
5661
aec7477b 5662 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5663 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5664 mode, &index);
7cf96da3
TI		 5665 if (IS_ERR(inode)) {
5666 err = PTR_ERR(inode);
39279cc3 5667 goto out_unlock;
7cf96da3 5668 }
43baa579 5669 drop_inode_on_err = 1;
39279cc3 5670
2a7dba39 5671 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5672 if (err)
33268eaf 5673 goto out_unlock;
33268eaf 5674
9185aa58
FB		 5675 err = btrfs_update_inode(trans, root, inode);
5676 if (err)
5677 goto out_unlock;
5678
ad19db71
CS		 5679 /*
5680 * If the active LSM wants to access the inode during
5681 * d_instantiate it needs these. Smack checks to see
5682 * if the filesystem supports xattrs by looking at the
5683 * ops vector.
5684 */
5685 inode->i_fop = &btrfs_file_operations;
5686 inode->i_op = &btrfs_file_inode_operations;
5687
a1b075d2 5688 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5689 if (err)
43baa579
FB		 5690 goto out_unlock;
5691
5692 inode->i_mapping->a_ops = &btrfs_aops;
5693 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5694 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5695 d_instantiate(dentry, inode);
5696
39279cc3 5697out_unlock:
7ad85bb7 5698 btrfs_end_transaction(trans, root);
43baa579 5699 if (err && drop_inode_on_err) {
39279cc3
CM		 5700 inode_dec_link_count(inode);
5701 iput(inode);
5702 }
b53d3f5d 5703 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5704 return err;
5705}
5706
5707static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5708 struct dentry *dentry)
5709{
5710 struct btrfs_trans_handle *trans;
5711 struct btrfs_root *root = BTRFS_I(dir)->root;
5712 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5713 u64 index;
39279cc3
CM		 5714 int err;
5715 int drop_inode = 0;
5716
4a8be425
TH		 5717 /* do not allow sys_link's with other subvols of the same device */
5718 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5719 return -EXDEV;
4a8be425 5720
f186373f 5721 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5722 return -EMLINK;
4a8be425 5723
3de4586c 5724 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5725 if (err)
5726 goto fail;
5727
a22285a6 5728 /*
7e6b6465 5729 * 2 items for inode and inode ref
a22285a6 5730 * 2 items for dir items
7e6b6465 5731 * 1 item for parent inode
a22285a6 5732 */
7e6b6465 5733 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5734 if (IS_ERR(trans)) {
5735 err = PTR_ERR(trans);
5736 goto fail;
5737 }
5f39d397 5738
8b558c5f 5739 inc_nlink(inode);
0c4d2d95 5740 inode_inc_iversion(inode);
3153495d 5741 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5742 ihold(inode);
e9976151 5743 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5744
a1b075d2 5745 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5746
a5719521 5747 if (err) {
54aa1f4d 5748 drop_inode = 1;
a5719521 5749 } else {
10d9f309 5750 struct dentry *parent = dentry->d_parent;
a5719521 5751 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5752 if (err)
5753 goto fail;
08c422c2 5754 d_instantiate(dentry, inode);
6a912213 5755 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5756 }
39279cc3 5757
7ad85bb7 5758 btrfs_end_transaction(trans, root);
1832a6d5 5759fail:
39279cc3
CM		 5760 if (drop_inode) {
5761 inode_dec_link_count(inode);
5762 iput(inode);
5763 }
b53d3f5d 5764 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5765 return err;
5766}
5767
18bb1db3 5768static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5769{
b9d86667 5770 struct inode *inode = NULL;
39279cc3
CM		 5771 struct btrfs_trans_handle *trans;
5772 struct btrfs_root *root = BTRFS_I(dir)->root;
5773 int err = 0;
5774 int drop_on_err = 0;
b9d86667 5775 u64 objectid = 0;
00e4e6b3 5776 u64 index = 0;
39279cc3 5777
9ed74f2d
JB		 5778 /*
5779 * 2 items for inode and ref
5780 * 2 items for dir items
5781 * 1 for xattr if selinux is on
5782 */
a22285a6
YZ		 5783 trans = btrfs_start_transaction(root, 5);
5784 if (IS_ERR(trans))
5785 return PTR_ERR(trans);
39279cc3 5786
581bb050
LZ		 5787 err = btrfs_find_free_ino(root, &objectid);
5788 if (err)
5789 goto out_fail;
5790
aec7477b 5791 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5792 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5793 S_IFDIR | mode, &index);
39279cc3
CM		 5794 if (IS_ERR(inode)) {
5795 err = PTR_ERR(inode);
5796 goto out_fail;
5797 }
5f39d397 5798
39279cc3 5799 drop_on_err = 1;
33268eaf 5800
2a7dba39 5801 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5802 if (err)
5803 goto out_fail;
5804
39279cc3
CM		 5805 inode->i_op = &btrfs_dir_inode_operations;
5806 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5807
dbe674a9 5808 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5809 err = btrfs_update_inode(trans, root, inode);
5810 if (err)
5811 goto out_fail;
5f39d397 5812
a1b075d2
JB		 5813 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5814 dentry->d_name.len, 0, index);
39279cc3
CM		 5815 if (err)
5816 goto out_fail;
5f39d397 5817
39279cc3
CM		 5818 d_instantiate(dentry, inode);
5819 drop_on_err = 0;
39279cc3
CM		 5820
5821out_fail:
7ad85bb7 5822 btrfs_end_transaction(trans, root);
39279cc3
CM		 5823 if (drop_on_err)
5824 iput(inode);
b53d3f5d 5825 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5826 return err;
5827}
5828
d352ac68
CM		 5829/* helper for btfs_get_extent. Given an existing extent in the tree,
5830 * and an extent that you want to insert, deal with overlap and insert
5831 * the new extent into the tree.
5832 */
3b951516
CM		 5833static int merge_extent_mapping(struct extent_map_tree *em_tree,
5834 struct extent_map *existing,
e6dcd2dc
CM		 5835 struct extent_map *em,
5836 u64 map_start, u64 map_len)
3b951516
CM		 5837{
5838 u64 start_diff;
3b951516 5839
e6dcd2dc
CM		 5840 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5841 start_diff = map_start - em->start;
5842 em->start = map_start;
5843 em->len = map_len;
c8b97818
CM		 5844 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5845 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5846 em->block_start += start_diff;
c8b97818
CM		 5847 em->block_len -= start_diff;
5848 }
09a2a8f9 5849 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 5850}
5851
c8b97818
CM		 5852static noinline int uncompress_inline(struct btrfs_path *path,
5853 struct inode *inode, struct page *page,
5854 size_t pg_offset, u64 extent_offset,
5855 struct btrfs_file_extent_item *item)
5856{
5857 int ret;
5858 struct extent_buffer *leaf = path->nodes[0];
5859 char *tmp;
5860 size_t max_size;
5861 unsigned long inline_size;
5862 unsigned long ptr;
261507a0 5863 int compress_type;
c8b97818
CM		 5864
5865 WARN_ON(pg_offset != 0);
261507a0 5866 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5867 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5868 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 5869 btrfs_item_nr(path->slots[0]));
c8b97818 5870 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5871 if (!tmp)
5872 return -ENOMEM;
c8b97818
CM		 5873 ptr = btrfs_file_extent_inline_start(item);
5874
5875 read_extent_buffer(leaf, tmp, ptr, inline_size);
5876
5b050f04 5877 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5878 ret = btrfs_decompress(compress_type, tmp, page,
5879 extent_offset, inline_size, max_size);
c8b97818 5880 if (ret) {
7ac687d9 5881 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5882 unsigned long copy_size = min_t(u64,
5883 PAGE_CACHE_SIZE - pg_offset,
5884 max_size - extent_offset);
5885 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5886 kunmap_atomic(kaddr);
c8b97818
CM		 5887 }
5888 kfree(tmp);
5889 return 0;
5890}
5891
d352ac68
CM		 5892/*
5893 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5894 * the ugly parts come from merging extents from the disk with the in-ram
5895 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5896 * where the in-ram extents might be locked pending data=ordered completion.
5897 *
5898 * This also copies inline extents directly into the page.
5899 */
d397712b 5900
a52d9a80 5901struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5902 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5903 int create)
5904{
5905 int ret;
5906 int err = 0;
db94535d 5907 u64 bytenr;
a52d9a80
CM		 5908 u64 extent_start = 0;
5909 u64 extent_end = 0;
33345d01 5910 u64 objectid = btrfs_ino(inode);
a52d9a80 5911 u32 found_type;
f421950f 5912 struct btrfs_path *path = NULL;
a52d9a80
CM		 5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 struct btrfs_file_extent_item *item;
5f39d397
CM		 5915 struct extent_buffer *leaf;
5916 struct btrfs_key found_key;
a52d9a80
CM		 5917 struct extent_map *em = NULL;
5918 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5919 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5920 struct btrfs_trans_handle *trans = NULL;
261507a0 5921 int compress_type;
a52d9a80 5922
a52d9a80 5923again:
890871be 5924 read_lock(&em_tree->lock);
d1310b2e 5925 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5926 if (em)
5927 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5928 read_unlock(&em_tree->lock);
d1310b2e 5929
a52d9a80 5930 if (em) {
e1c4b745
CM		 5931 if (em->start > start || em->start + em->len <= start)
5932 free_extent_map(em);
5933 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5934 free_extent_map(em);
5935 else
5936 goto out;
a52d9a80 5937 }
172ddd60 5938 em = alloc_extent_map();
a52d9a80 5939 if (!em) {
d1310b2e
CM		 5940 err = -ENOMEM;
5941 goto out;
a52d9a80 5942 }
e6dcd2dc 5943 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5944 em->start = EXTENT_MAP_HOLE;
445a6944 5945 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5946 em->len = (u64)-1;
c8b97818 5947 em->block_len = (u64)-1;
f421950f
CM		 5948
5949 if (!path) {
5950 path = btrfs_alloc_path();
026fd317
JB		 5951 if (!path) {
5952 err = -ENOMEM;
5953 goto out;
5954 }
5955 /*
5956 * Chances are we'll be called again, so go ahead and do
5957 * readahead
5958 */
5959 path->reada = 1;
f421950f
CM		 5960 }
5961
179e29e4
CM		 5962 ret = btrfs_lookup_file_extent(trans, root, path,
5963 objectid, start, trans != NULL);
a52d9a80
CM		 5964 if (ret < 0) {
5965 err = ret;
5966 goto out;
5967 }
5968
5969 if (ret != 0) {
5970 if (path->slots[0] == 0)
5971 goto not_found;
5972 path->slots[0]--;
5973 }
5974
5f39d397
CM		 5975 leaf = path->nodes[0];
5976 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5977 struct btrfs_file_extent_item);
a52d9a80 5978 /* are we inside the extent that was found? */
5f39d397
CM		 5979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5980 found_type = btrfs_key_type(&found_key);
5981 if (found_key.objectid != objectid ||
a52d9a80 5982 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 5983 /*
5984 * If we backup past the first extent we want to move forward
5985 * and see if there is an extent in front of us, otherwise we'll
5986 * say there is a hole for our whole search range which can
5987 * cause problems.
5988 */
5989 extent_end = start;
5990 goto next;
a52d9a80
CM		 5991 }
5992
5f39d397
CM		 5993 found_type = btrfs_file_extent_type(leaf, item);
5994 extent_start = found_key.offset;
261507a0 5995 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5996 if (found_type == BTRFS_FILE_EXTENT_REG ||
5997 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5998 extent_end = extent_start +
db94535d 5999 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6000 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6001 size_t size;
6002 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 6003 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6004 }
25a50341 6005next:
9036c102
YZ		 6006 if (start >= extent_end) {
6007 path->slots[0]++;
6008 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6009 ret = btrfs_next_leaf(root, path);
6010 if (ret < 0) {
6011 err = ret;
6012 goto out;
a52d9a80 6013 }
9036c102
YZ		 6014 if (ret > 0)
6015 goto not_found;
6016 leaf = path->nodes[0];
a52d9a80 6017 }
9036c102
YZ		 6018 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6019 if (found_key.objectid != objectid ||
6020 found_key.type != BTRFS_EXTENT_DATA_KEY)
6021 goto not_found;
6022 if (start + len <= found_key.offset)
6023 goto not_found;
6024 em->start = start;
70c8a91c 6025 em->orig_start = start;
9036c102
YZ		 6026 em->len = found_key.offset - start;
6027 goto not_found_em;
6028 }
6029
cc95bef6 6030 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6031 if (found_type == BTRFS_FILE_EXTENT_REG ||
6032 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6033 em->start = extent_start;
6034 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6035 em->orig_start = extent_start -
6036 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6037 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6038 item);
db94535d
CM		 6039 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6040 if (bytenr == 0) {
5f39d397 6041 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6042 goto insert;
6043 }
261507a0 6044 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6045 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6046 em->compress_type = compress_type;
c8b97818 6047 em->block_start = bytenr;
b4939680 6048 em->block_len = em->orig_block_len;
c8b97818
CM		 6049 } else {
6050 bytenr += btrfs_file_extent_offset(leaf, item);
6051 em->block_start = bytenr;
6052 em->block_len = em->len;
d899e052
YZ		 6053 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6054 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6055 }
a52d9a80
CM		 6056 goto insert;
6057 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6058 unsigned long ptr;
a52d9a80 6059 char *map;
3326d1b0
CM		 6060 size_t size;
6061 size_t extent_offset;
6062 size_t copy_size;
a52d9a80 6063
689f9346 6064 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6065 if (!page || create) {
689f9346 6066 em->start = extent_start;
9036c102 6067 em->len = extent_end - extent_start;
689f9346
Y		 6068 goto out;
6069 }
5f39d397 6070
9036c102
YZ		 6071 size = btrfs_file_extent_inline_len(leaf, item);
6072 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6073 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6074 size - extent_offset);
3326d1b0 6075 em->start = extent_start + extent_offset;
fda2832f 6076 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6077 em->orig_block_len = em->len;
70c8a91c 6078 em->orig_start = em->start;
261507a0 6079 if (compress_type) {
c8b97818 6080 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6081 em->compress_type = compress_type;
6082 }
689f9346 6083 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6084 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6085 if (btrfs_file_extent_compression(leaf, item) !=
6086 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6087 ret = uncompress_inline(path, inode, page,
6088 pg_offset,
6089 extent_offset, item);
79787eaa 6090 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6091 } else {
6092 map = kmap(page);
6093 read_extent_buffer(leaf, map + pg_offset, ptr,
6094 copy_size);
93c82d57
CM		 6095 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6096 memset(map + pg_offset + copy_size, 0,
6097 PAGE_CACHE_SIZE - pg_offset -
6098 copy_size);
6099 }
c8b97818
CM		 6100 kunmap(page);
6101 }
179e29e4
CM		 6102 flush_dcache_page(page);
6103 } else if (create && PageUptodate(page)) {
6bf7e080 6104 BUG();
179e29e4
CM		 6105 if (!trans) {
6106 kunmap(page);
6107 free_extent_map(em);
6108 em = NULL;
ff5714cc 6109
b3b4aa74 6110 btrfs_release_path(path);
7a7eaa40 6111 trans = btrfs_join_transaction(root);
ff5714cc 6112
3612b495
TI		 6113 if (IS_ERR(trans))
6114 return ERR_CAST(trans);
179e29e4
CM		 6115 goto again;
6116 }
c8b97818 6117 map = kmap(page);
70dec807 6118 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6119 copy_size);
c8b97818 6120 kunmap(page);
179e29e4 6121 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6122 }
d1310b2e 6123 set_extent_uptodate(io_tree, em->start,
507903b8 6124 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6125 goto insert;
6126 } else {
31b1a2bd 6127 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6128 }
6129not_found:
6130 em->start = start;
70c8a91c 6131 em->orig_start = start;
d1310b2e 6132 em->len = len;
a52d9a80 6133not_found_em:
5f39d397 6134 em->block_start = EXTENT_MAP_HOLE;
9036c102 6135 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6136insert:
b3b4aa74 6137 btrfs_release_path(path);
d1310b2e 6138 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6139 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6140 em->start, em->len, start, len);
a52d9a80
CM		 6141 err = -EIO;
6142 goto out;
6143 }
d1310b2e
CM		 6144
6145 err = 0;
890871be 6146 write_lock(&em_tree->lock);
09a2a8f9 6147 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6148 /* it is possible that someone inserted the extent into the tree
6149 * while we had the lock dropped. It is also possible that
6150 * an overlapping map exists in the tree
6151 */
a52d9a80 6152 if (ret == -EEXIST) {
3b951516 6153 struct extent_map *existing;
e6dcd2dc
CM		 6154
6155 ret = 0;
6156
3b951516 6157 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6158 if (existing && (existing->start > start ||
6159 existing->start + existing->len <= start)) {
6160 free_extent_map(existing);
6161 existing = NULL;
6162 }
3b951516
CM		 6163 if (!existing) {
6164 existing = lookup_extent_mapping(em_tree, em->start,
6165 em->len);
6166 if (existing) {
6167 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6168 em, start,
6169 root->sectorsize);
3b951516
CM		 6170 free_extent_map(existing);
6171 if (err) {
6172 free_extent_map(em);
6173 em = NULL;
6174 }
6175 } else {
6176 err = -EIO;
3b951516
CM		 6177 free_extent_map(em);
6178 em = NULL;
6179 }
6180 } else {
6181 free_extent_map(em);
6182 em = existing;
e6dcd2dc 6183 err = 0;
a52d9a80 6184 }
a52d9a80 6185 }
890871be 6186 write_unlock(&em_tree->lock);
a52d9a80 6187out:
1abe9b8a 6188
f0bd95ea
TI		 6189 if (em)
6190 trace_btrfs_get_extent(root, em);
1abe9b8a 6191
f421950f
CM		 6192 if (path)
6193 btrfs_free_path(path);
a52d9a80
CM		 6194 if (trans) {
6195 ret = btrfs_end_transaction(trans, root);
d397712b 6196 if (!err)
a52d9a80
CM		 6197 err = ret;
6198 }
a52d9a80
CM		 6199 if (err) {
6200 free_extent_map(em);
a52d9a80
CM		 6201 return ERR_PTR(err);
6202 }
79787eaa 6203 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6204 return em;
6205}
6206
ec29ed5b
CM		 6207struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6208 size_t pg_offset, u64 start, u64 len,
6209 int create)
6210{
6211 struct extent_map *em;
6212 struct extent_map *hole_em = NULL;
6213 u64 range_start = start;
6214 u64 end;
6215 u64 found;
6216 u64 found_end;
6217 int err = 0;
6218
6219 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6220 if (IS_ERR(em))
6221 return em;
6222 if (em) {
6223 /*
f9e4fb53
LB		 6224 * if our em maps to
6225 * - a hole or
6226 * - a pre-alloc extent,
6227 * there might actually be delalloc bytes behind it.
ec29ed5b 6228 */
f9e4fb53
LB		 6229 if (em->block_start != EXTENT_MAP_HOLE &&
6230 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6231 return em;
6232 else
6233 hole_em = em;
6234 }
6235
6236 /* check to see if we've wrapped (len == -1 or similar) */
6237 end = start + len;
6238 if (end < start)
6239 end = (u64)-1;
6240 else
6241 end -= 1;
6242
6243 em = NULL;
6244
6245 /* ok, we didn't find anything, lets look for delalloc */
6246 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6247 end, len, EXTENT_DELALLOC, 1);
6248 found_end = range_start + found;
6249 if (found_end < range_start)
6250 found_end = (u64)-1;
6251
6252 /*
6253 * we didn't find anything useful, return
6254 * the original results from get_extent()
6255 */
6256 if (range_start > end || found_end <= start) {
6257 em = hole_em;
6258 hole_em = NULL;
6259 goto out;
6260 }
6261
6262 /* adjust the range_start to make sure it doesn't
6263 * go backwards from the start they passed in
6264 */
6265 range_start = max(start,range_start);
6266 found = found_end - range_start;
6267
6268 if (found > 0) {
6269 u64 hole_start = start;
6270 u64 hole_len = len;
6271
172ddd60 6272 em = alloc_extent_map();
ec29ed5b
CM		 6273 if (!em) {
6274 err = -ENOMEM;
6275 goto out;
6276 }
6277 /*
6278 * when btrfs_get_extent can't find anything it
6279 * returns one huge hole
6280 *
6281 * make sure what it found really fits our range, and
6282 * adjust to make sure it is based on the start from
6283 * the caller
6284 */
6285 if (hole_em) {
6286 u64 calc_end = extent_map_end(hole_em);
6287
6288 if (calc_end <= start || (hole_em->start > end)) {
6289 free_extent_map(hole_em);
6290 hole_em = NULL;
6291 } else {
6292 hole_start = max(hole_em->start, start);
6293 hole_len = calc_end - hole_start;
6294 }
6295 }
6296 em->bdev = NULL;
6297 if (hole_em && range_start > hole_start) {
6298 /* our hole starts before our delalloc, so we
6299 * have to return just the parts of the hole
6300 * that go until the delalloc starts
6301 */
6302 em->len = min(hole_len,
6303 range_start - hole_start);
6304 em->start = hole_start;
6305 em->orig_start = hole_start;
6306 /*
6307 * don't adjust block start at all,
6308 * it is fixed at EXTENT_MAP_HOLE
6309 */
6310 em->block_start = hole_em->block_start;
6311 em->block_len = hole_len;
f9e4fb53
LB		 6312 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6313 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6314 } else {
6315 em->start = range_start;
6316 em->len = found;
6317 em->orig_start = range_start;
6318 em->block_start = EXTENT_MAP_DELALLOC;
6319 em->block_len = found;
6320 }
6321 } else if (hole_em) {
6322 return hole_em;
6323 }
6324out:
6325
6326 free_extent_map(hole_em);
6327 if (err) {
6328 free_extent_map(em);
6329 return ERR_PTR(err);
6330 }
6331 return em;
6332}
6333
4b46fce2
JB		 6334static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6335 u64 start, u64 len)
6336{
6337 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 6338 struct extent_map *em;
4b46fce2
JB		 6339 struct btrfs_key ins;
6340 u64 alloc_hint;
6341 int ret;
4b46fce2 6342
4b46fce2 6343 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 6344 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
81c9ad23 6345 alloc_hint, &ins, 1);
00361589
JB		 6346 if (ret)
6347 return ERR_PTR(ret);
4b46fce2 6348
70c8a91c 6349 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6350 ins.offset, ins.offset, ins.offset, 0);
00361589
JB		 6351 if (IS_ERR(em)) {
6352 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6353 return em;
6354 }
4b46fce2
JB		 6355
6356 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6357 ins.offset, ins.offset, 0);
6358 if (ret) {
6359 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
00361589
JB		 6360 free_extent_map(em);
6361 return ERR_PTR(ret);
4b46fce2 6362 }
00361589 6363
4b46fce2
JB		 6364 return em;
6365}
6366
46bfbb5c
CM		 6367/*
6368 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6369 * block must be cow'd
6370 */
00361589 6371noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 6372 u64 *orig_start, u64 *orig_block_len,
6373 u64 *ram_bytes)
46bfbb5c 6374{
00361589 6375 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 6376 struct btrfs_path *path;
6377 int ret;
6378 struct extent_buffer *leaf;
6379 struct btrfs_root *root = BTRFS_I(inode)->root;
6380 struct btrfs_file_extent_item *fi;
6381 struct btrfs_key key;
6382 u64 disk_bytenr;
6383 u64 backref_offset;
6384 u64 extent_end;
6385 u64 num_bytes;
6386 int slot;
6387 int found_type;
7ee9e440 6388 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
46bfbb5c
CM		 6389 path = btrfs_alloc_path();
6390 if (!path)
6391 return -ENOMEM;
6392
00361589 6393 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6394 offset, 0);
6395 if (ret < 0)
6396 goto out;
6397
6398 slot = path->slots[0];
6399 if (ret == 1) {
6400 if (slot == 0) {
6401 /* can't find the item, must cow */
6402 ret = 0;
6403 goto out;
6404 }
6405 slot--;
6406 }
6407 ret = 0;
6408 leaf = path->nodes[0];
6409 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6410 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6411 key.type != BTRFS_EXTENT_DATA_KEY) {
6412 /* not our file or wrong item type, must cow */
6413 goto out;
6414 }
6415
6416 if (key.offset > offset) {
6417 /* Wrong offset, must cow */
6418 goto out;
6419 }
6420
6421 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6422 found_type = btrfs_file_extent_type(leaf, fi);
6423 if (found_type != BTRFS_FILE_EXTENT_REG &&
6424 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6425 /* not a regular extent, must cow */
6426 goto out;
6427 }
7ee9e440
JB		 6428
6429 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6430 goto out;
6431
46bfbb5c 6432 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 6433 if (disk_bytenr == 0)
6434 goto out;
6435
6436 if (btrfs_file_extent_compression(leaf, fi) ||
6437 btrfs_file_extent_encryption(leaf, fi) ||
6438 btrfs_file_extent_other_encoding(leaf, fi))
6439 goto out;
6440
46bfbb5c
CM		 6441 backref_offset = btrfs_file_extent_offset(leaf, fi);
6442
7ee9e440
JB		 6443 if (orig_start) {
6444 *orig_start = key.offset - backref_offset;
6445 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6446 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6447 }
eb384b55 6448
46bfbb5c 6449 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
46bfbb5c
CM		 6450
6451 if (btrfs_extent_readonly(root, disk_bytenr))
6452 goto out;
1bda19eb 6453 btrfs_release_path(path);
46bfbb5c
CM		 6454
6455 /*
6456 * look for other files referencing this extent, if we
6457 * find any we must cow
6458 */
00361589
JB		 6459 trans = btrfs_join_transaction(root);
6460 if (IS_ERR(trans)) {
6461 ret = 0;
46bfbb5c 6462 goto out;
00361589
JB		 6463 }
6464
6465 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6466 key.offset - backref_offset, disk_bytenr);
6467 btrfs_end_transaction(trans, root);
6468 if (ret) {
6469 ret = 0;
6470 goto out;
6471 }
46bfbb5c
CM		 6472
6473 /*
6474 * adjust disk_bytenr and num_bytes to cover just the bytes
6475 * in this extent we are about to write. If there
6476 * are any csums in that range we have to cow in order
6477 * to keep the csums correct
6478 */
6479 disk_bytenr += backref_offset;
6480 disk_bytenr += offset - key.offset;
eb384b55 6481 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6482 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6483 goto out;
6484 /*
6485 * all of the above have passed, it is safe to overwrite this extent
6486 * without cow
6487 */
eb384b55 6488 *len = num_bytes;
46bfbb5c
CM		 6489 ret = 1;
6490out:
6491 btrfs_free_path(path);
6492 return ret;
6493}
6494
eb838e73
JB		 6495static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6496 struct extent_state **cached_state, int writing)
6497{
6498 struct btrfs_ordered_extent *ordered;
6499 int ret = 0;
6500
6501 while (1) {
6502 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6503 0, cached_state);
6504 /*
6505 * We're concerned with the entire range that we're going to be
6506 * doing DIO to, so we need to make sure theres no ordered
6507 * extents in this range.
6508 */
6509 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6510 lockend - lockstart + 1);
6511
6512 /*
6513 * We need to make sure there are no buffered pages in this
6514 * range either, we could have raced between the invalidate in
6515 * generic_file_direct_write and locking the extent. The
6516 * invalidate needs to happen so that reads after a write do not
6517 * get stale data.
6518 */
6519 if (!ordered && (!writing ||
6520 !test_range_bit(&BTRFS_I(inode)->io_tree,
6521 lockstart, lockend, EXTENT_UPTODATE, 0,
6522 *cached_state)))
6523 break;
6524
6525 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6526 cached_state, GFP_NOFS);
6527
6528 if (ordered) {
6529 btrfs_start_ordered_extent(inode, ordered, 1);
6530 btrfs_put_ordered_extent(ordered);
6531 } else {
6532 /* Screw you mmap */
6533 ret = filemap_write_and_wait_range(inode->i_mapping,
6534 lockstart,
6535 lockend);
6536 if (ret)
6537 break;
6538
6539 /*
6540 * If we found a page that couldn't be invalidated just
6541 * fall back to buffered.
6542 */
6543 ret = invalidate_inode_pages2_range(inode->i_mapping,
6544 lockstart >> PAGE_CACHE_SHIFT,
6545 lockend >> PAGE_CACHE_SHIFT);
6546 if (ret)
6547 break;
6548 }
6549
6550 cond_resched();
6551 }
6552
6553 return ret;
6554}
6555
69ffb543
JB		 6556static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6557 u64 len, u64 orig_start,
6558 u64 block_start, u64 block_len,
cc95bef6
JB		 6559 u64 orig_block_len, u64 ram_bytes,
6560 int type)
69ffb543
JB		 6561{
6562 struct extent_map_tree *em_tree;
6563 struct extent_map *em;
6564 struct btrfs_root *root = BTRFS_I(inode)->root;
6565 int ret;
6566
6567 em_tree = &BTRFS_I(inode)->extent_tree;
6568 em = alloc_extent_map();
6569 if (!em)
6570 return ERR_PTR(-ENOMEM);
6571
6572 em->start = start;
6573 em->orig_start = orig_start;
2ab28f32
JB		 6574 em->mod_start = start;
6575 em->mod_len = len;
69ffb543
JB		 6576 em->len = len;
6577 em->block_len = block_len;
6578 em->block_start = block_start;
6579 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6580 em->orig_block_len = orig_block_len;
cc95bef6 6581 em->ram_bytes = ram_bytes;
70c8a91c 6582 em->generation = -1;
69ffb543
JB		 6583 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6584 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6585 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6586
6587 do {
6588 btrfs_drop_extent_cache(inode, em->start,
6589 em->start + em->len - 1, 0);
6590 write_lock(&em_tree->lock);
09a2a8f9 6591 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6592 write_unlock(&em_tree->lock);
6593 } while (ret == -EEXIST);
6594
6595 if (ret) {
6596 free_extent_map(em);
6597 return ERR_PTR(ret);
6598 }
6599
6600 return em;
6601}
6602
6603
4b46fce2
JB		 6604static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6605 struct buffer_head *bh_result, int create)
6606{
6607 struct extent_map *em;
6608 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6609 struct extent_state *cached_state = NULL;
4b46fce2 6610 u64 start = iblock << inode->i_blkbits;
eb838e73 6611 u64 lockstart, lockend;
4b46fce2 6612 u64 len = bh_result->b_size;
eb838e73 6613 int unlock_bits = EXTENT_LOCKED;
0934856d 6614 int ret = 0;
eb838e73 6615
172a5049 6616 if (create)
eb838e73 6617 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6618 else
c329861d 6619 len = min_t(u64, len, root->sectorsize);
eb838e73 6620
c329861d
JB		 6621 lockstart = start;
6622 lockend = start + len - 1;
6623
eb838e73
JB		 6624 /*
6625 * If this errors out it's because we couldn't invalidate pagecache for
6626 * this range and we need to fallback to buffered.
6627 */
6628 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6629 return -ENOTBLK;
6630
4b46fce2 6631 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6632 if (IS_ERR(em)) {
6633 ret = PTR_ERR(em);
6634 goto unlock_err;
6635 }
4b46fce2
JB		 6636
6637 /*
6638 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6639 * io. INLINE is special, and we could probably kludge it in here, but
6640 * it's still buffered so for safety lets just fall back to the generic
6641 * buffered path.
6642 *
6643 * For COMPRESSED we _have_ to read the entire extent in so we can
6644 * decompress it, so there will be buffering required no matter what we
6645 * do, so go ahead and fallback to buffered.
6646 *
6647 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6648 * to buffered IO. Don't blame me, this is the price we pay for using
6649 * the generic code.
6650 */
6651 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6652 em->block_start == EXTENT_MAP_INLINE) {
6653 free_extent_map(em);
eb838e73
JB		 6654 ret = -ENOTBLK;
6655 goto unlock_err;
4b46fce2
JB		 6656 }
6657
6658 /* Just a good old fashioned hole, return */
6659 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6660 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6661 free_extent_map(em);
eb838e73 6662 goto unlock_err;
4b46fce2
JB		 6663 }
6664
6665 /*
6666 * We don't allocate a new extent in the following cases
6667 *
6668 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6669 * existing extent.
6670 * 2) The extent is marked as PREALLOC. We're good to go here and can
6671 * just use the extent.
6672 *
6673 */
46bfbb5c 6674 if (!create) {
eb838e73
JB		 6675 len = min(len, em->len - (start - em->start));
6676 lockstart = start + len;
6677 goto unlock;
46bfbb5c 6678 }
4b46fce2
JB		 6679
6680 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6681 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6682 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6683 int type;
6684 int ret;
eb384b55 6685 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6686
6687 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6688 type = BTRFS_ORDERED_PREALLOC;
6689 else
6690 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6691 len = min(len, em->len - (start - em->start));
4b46fce2 6692 block_start = em->block_start + (start - em->start);
46bfbb5c 6693
00361589 6694 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 6695 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6696 if (type == BTRFS_ORDERED_PREALLOC) {
6697 free_extent_map(em);
6698 em = create_pinned_em(inode, start, len,
6699 orig_start,
b4939680 6700 block_start, len,
cc95bef6
JB		 6701 orig_block_len,
6702 ram_bytes, type);
00361589 6703 if (IS_ERR(em))
69ffb543 6704 goto unlock_err;
69ffb543
JB		 6705 }
6706
46bfbb5c
CM		 6707 ret = btrfs_add_ordered_extent_dio(inode, start,
6708 block_start, len, len, type);
46bfbb5c
CM		 6709 if (ret) {
6710 free_extent_map(em);
eb838e73 6711 goto unlock_err;
46bfbb5c
CM		 6712 }
6713 goto unlock;
4b46fce2 6714 }
4b46fce2 6715 }
00361589 6716
46bfbb5c
CM		 6717 /*
6718 * this will cow the extent, reset the len in case we changed
6719 * it above
6720 */
6721 len = bh_result->b_size;
70c8a91c
JB		 6722 free_extent_map(em);
6723 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6724 if (IS_ERR(em)) {
6725 ret = PTR_ERR(em);
6726 goto unlock_err;
6727 }
46bfbb5c
CM		 6728 len = min(len, em->len - (start - em->start));
6729unlock:
4b46fce2
JB		 6730 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6731 inode->i_blkbits;
46bfbb5c 6732 bh_result->b_size = len;
4b46fce2
JB		 6733 bh_result->b_bdev = em->bdev;
6734 set_buffer_mapped(bh_result);
c3473e83
JB		 6735 if (create) {
6736 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6737 set_buffer_new(bh_result);
6738
6739 /*
6740 * Need to update the i_size under the extent lock so buffered
6741 * readers will get the updated i_size when we unlock.
6742 */
6743 if (start + len > i_size_read(inode))
6744 i_size_write(inode, start + len);
0934856d 6745
172a5049
MX		 6746 spin_lock(&BTRFS_I(inode)->lock);
6747 BTRFS_I(inode)->outstanding_extents++;
6748 spin_unlock(&BTRFS_I(inode)->lock);
6749
0934856d
MX		 6750 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6751 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6752 &cached_state, GFP_NOFS);
6753 BUG_ON(ret);
c3473e83 6754 }
4b46fce2 6755
eb838e73
JB		 6756 /*
6757 * In the case of write we need to clear and unlock the entire range,
6758 * in the case of read we need to unlock only the end area that we
6759 * aren't using if there is any left over space.
6760 */
24c03fa5 6761 if (lockstart < lockend) {
0934856d
MX		 6762 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6763 lockend, unlock_bits, 1, 0,
6764 &cached_state, GFP_NOFS);
24c03fa5 6765 } else {
eb838e73 6766 free_extent_state(cached_state);
24c03fa5 6767 }
eb838e73 6768
4b46fce2
JB		 6769 free_extent_map(em);
6770
6771 return 0;
eb838e73
JB		 6772
6773unlock_err:
eb838e73
JB		 6774 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6775 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6776 return ret;
4b46fce2
JB		 6777}
6778
4b46fce2
JB		 6779static void btrfs_endio_direct_read(struct bio *bio, int err)
6780{
e65e1535 6781 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6782 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6783 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 6784 struct inode *inode = dip->inode;
6785 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 6786 struct bio *dio_bio;
facc8a22
MX		 6787 u32 *csums = (u32 *)dip->csum;
6788 int index = 0;
4b46fce2 6789 u64 start;
4b46fce2
JB		 6790
6791 start = dip->logical_offset;
6792 do {
6793 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6794 struct page *page = bvec->bv_page;
6795 char *kaddr;
6796 u32 csum = ~(u32)0;
6797 unsigned long flags;
6798
6799 local_irq_save(flags);
7ac687d9 6800 kaddr = kmap_atomic(page);
b0496686 6801 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6802 csum, bvec->bv_len);
6803 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6804 kunmap_atomic(kaddr);
4b46fce2
JB		 6805 local_irq_restore(flags);
6806
6807 flush_dcache_page(bvec->bv_page);
facc8a22
MX		 6808 if (csum != csums[index]) {
6809 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c
GU		 6810 btrfs_ino(inode), start, csum,
6811 csums[index]);
4b46fce2
JB		 6812 err = -EIO;
6813 }
6814 }
6815
6816 start += bvec->bv_len;
4b46fce2 6817 bvec++;
facc8a22 6818 index++;
4b46fce2
JB		 6819 } while (bvec <= bvec_end);
6820
6821 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6822 dip->logical_offset + dip->bytes - 1);
9be3395b 6823 dio_bio = dip->dio_bio;
4b46fce2 6824
4b46fce2 6825 kfree(dip);
c0da7aa1
JB		 6826
6827 /* If we had a csum failure make sure to clear the uptodate flag */
6828 if (err)
9be3395b
CM		 6829 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6830 dio_end_io(dio_bio, err);
6831 bio_put(bio);
4b46fce2
JB		 6832}
6833
6834static void btrfs_endio_direct_write(struct bio *bio, int err)
6835{
6836 struct btrfs_dio_private *dip = bio->bi_private;
6837 struct inode *inode = dip->inode;
6838 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6839 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 6840 u64 ordered_offset = dip->logical_offset;
6841 u64 ordered_bytes = dip->bytes;
9be3395b 6842 struct bio *dio_bio;
4b46fce2
JB		 6843 int ret;
6844
6845 if (err)
6846 goto out_done;
163cf09c
CM		 6847again:
6848 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6849 &ordered_offset,
5fd02043 6850 ordered_bytes, !err);
4b46fce2 6851 if (!ret)
163cf09c 6852 goto out_test;
4b46fce2 6853
5fd02043
JB		 6854 ordered->work.func = finish_ordered_fn;
6855 ordered->work.flags = 0;
6856 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6857 &ordered->work);
163cf09c
CM		 6858out_test:
6859 /*
6860 * our bio might span multiple ordered extents. If we haven't
6861 * completed the accounting for the whole dio, go back and try again
6862 */
6863 if (ordered_offset < dip->logical_offset + dip->bytes) {
6864 ordered_bytes = dip->logical_offset + dip->bytes -
6865 ordered_offset;
5fd02043 6866 ordered = NULL;
163cf09c
CM		 6867 goto again;
6868 }
4b46fce2 6869out_done:
9be3395b 6870 dio_bio = dip->dio_bio;
4b46fce2 6871
4b46fce2 6872 kfree(dip);
c0da7aa1
JB		 6873
6874 /* If we had an error make sure to clear the uptodate flag */
6875 if (err)
9be3395b
CM		 6876 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6877 dio_end_io(dio_bio, err);
6878 bio_put(bio);
4b46fce2
JB		 6879}
6880
eaf25d93
CM		 6881static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6882 struct bio *bio, int mirror_num,
6883 unsigned long bio_flags, u64 offset)
6884{
6885 int ret;
6886 struct btrfs_root *root = BTRFS_I(inode)->root;
6887 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6888 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6889 return 0;
6890}
6891
e65e1535
MX		 6892static void btrfs_end_dio_bio(struct bio *bio, int err)
6893{
6894 struct btrfs_dio_private *dip = bio->bi_private;
6895
6896 if (err) {
33345d01 6897 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6898 "sector %#Lx len %u err no %d\n",
c1c9ff7c 6899 btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6900 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6901 dip->errors = 1;
6902
6903 /*
6904 * before atomic variable goto zero, we must make sure
6905 * dip->errors is perceived to be set.
6906 */
6907 smp_mb__before_atomic_dec();
6908 }
6909
6910 /* if there are more bios still pending for this dio, just exit */
6911 if (!atomic_dec_and_test(&dip->pending_bios))
6912 goto out;
6913
9be3395b 6914 if (dip->errors) {
e65e1535 6915 bio_io_error(dip->orig_bio);
9be3395b
CM		 6916 } else {
6917 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 6918 bio_endio(dip->orig_bio, 0);
6919 }
6920out:
6921 bio_put(bio);
6922}
6923
6924static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6925 u64 first_sector, gfp_t gfp_flags)
6926{
6927 int nr_vecs = bio_get_nr_vecs(bdev);
6928 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6929}
6930
6931static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6932 int rw, u64 file_offset, int skip_sum,
c329861d 6933 int async_submit)
e65e1535 6934{
facc8a22 6935 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 6936 int write = rw & REQ_WRITE;
6937 struct btrfs_root *root = BTRFS_I(inode)->root;
6938 int ret;
6939
b812ce28
JB		 6940 if (async_submit)
6941 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6942
e65e1535 6943 bio_get(bio);
5fd02043
JB		 6944
6945 if (!write) {
6946 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6947 if (ret)
6948 goto err;
6949 }
e65e1535 6950
1ae39938
JB		 6951 if (skip_sum)
6952 goto map;
6953
6954 if (write && async_submit) {
e65e1535
MX		 6955 ret = btrfs_wq_submit_bio(root->fs_info,
6956 inode, rw, bio, 0, 0,
6957 file_offset,
6958 __btrfs_submit_bio_start_direct_io,
6959 __btrfs_submit_bio_done);
6960 goto err;
1ae39938
JB		 6961 } else if (write) {
6962 /*
6963 * If we aren't doing async submit, calculate the csum of the
6964 * bio now.
6965 */
6966 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6967 if (ret)
6968 goto err;
c2db1073 6969 } else if (!skip_sum) {
facc8a22
MX		 6970 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
6971 file_offset);
c2db1073
TI		 6972 if (ret)
6973 goto err;
6974 }
e65e1535 6975
1ae39938
JB		 6976map:
6977 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 6978err:
6979 bio_put(bio);
6980 return ret;
6981}
6982
6983static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6984 int skip_sum)
6985{
6986 struct inode *inode = dip->inode;
6987 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 6988 struct bio *bio;
6989 struct bio *orig_bio = dip->orig_bio;
6990 struct bio_vec *bvec = orig_bio->bi_io_vec;
6991 u64 start_sector = orig_bio->bi_sector;
6992 u64 file_offset = dip->logical_offset;
6993 u64 submit_len = 0;
6994 u64 map_length;
6995 int nr_pages = 0;
e65e1535 6996 int ret = 0;
1ae39938 6997 int async_submit = 0;
e65e1535 6998
e65e1535 6999 map_length = orig_bio->bi_size;
53b381b3 7000 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 7001 &map_length, NULL, 0);
7002 if (ret) {
64728bbb 7003 bio_put(orig_bio);
e65e1535
MX		 7004 return -EIO;
7005 }
facc8a22 7006
02f57c7a
JB		 7007 if (map_length >= orig_bio->bi_size) {
7008 bio = orig_bio;
7009 goto submit;
7010 }
7011
53b381b3
DW		 7012 /* async crcs make it difficult to collect full stripe writes. */
7013 if (btrfs_get_alloc_profile(root, 1) &
7014 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7015 async_submit = 0;
7016 else
7017 async_submit = 1;
7018
02f57c7a
JB		 7019 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7020 if (!bio)
7021 return -ENOMEM;
7022 bio->bi_private = dip;
7023 bio->bi_end_io = btrfs_end_dio_bio;
7024 atomic_inc(&dip->pending_bios);
7025
e65e1535
MX		 7026 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7027 if (unlikely(map_length < submit_len + bvec->bv_len ||
7028 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7029 bvec->bv_offset) < bvec->bv_len)) {
7030 /*
7031 * inc the count before we submit the bio so
7032 * we know the end IO handler won't happen before
7033 * we inc the count. Otherwise, the dip might get freed
7034 * before we're done setting it up
7035 */
7036 atomic_inc(&dip->pending_bios);
7037 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7038 file_offset, skip_sum,
c329861d 7039 async_submit);
e65e1535
MX		 7040 if (ret) {
7041 bio_put(bio);
7042 atomic_dec(&dip->pending_bios);
7043 goto out_err;
7044 }
7045
e65e1535
MX		 7046 start_sector += submit_len >> 9;
7047 file_offset += submit_len;
7048
7049 submit_len = 0;
7050 nr_pages = 0;
7051
7052 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7053 start_sector, GFP_NOFS);
7054 if (!bio)
7055 goto out_err;
7056 bio->bi_private = dip;
7057 bio->bi_end_io = btrfs_end_dio_bio;
7058
7059 map_length = orig_bio->bi_size;
53b381b3 7060 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7061 start_sector << 9,
e65e1535
MX		 7062 &map_length, NULL, 0);
7063 if (ret) {
7064 bio_put(bio);
7065 goto out_err;
7066 }
7067 } else {
7068 submit_len += bvec->bv_len;
7069 nr_pages ++;
7070 bvec++;
7071 }
7072 }
7073
02f57c7a 7074submit:
e65e1535 7075 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7076 async_submit);
e65e1535
MX		 7077 if (!ret)
7078 return 0;
7079
7080 bio_put(bio);
7081out_err:
7082 dip->errors = 1;
7083 /*
7084 * before atomic variable goto zero, we must
7085 * make sure dip->errors is perceived to be set.
7086 */
7087 smp_mb__before_atomic_dec();
7088 if (atomic_dec_and_test(&dip->pending_bios))
7089 bio_io_error(dip->orig_bio);
7090
7091 /* bio_end_io() will handle error, so we needn't return it */
7092 return 0;
7093}
7094
9be3395b
CM		 7095static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7096 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7097{
7098 struct btrfs_root *root = BTRFS_I(inode)->root;
7099 struct btrfs_dio_private *dip;
9be3395b 7100 struct bio *io_bio;
4b46fce2 7101 int skip_sum;
facc8a22 7102 int sum_len;
7b6d91da 7103 int write = rw & REQ_WRITE;
4b46fce2 7104 int ret = 0;
facc8a22 7105 u16 csum_size;
4b46fce2
JB		 7106
7107 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7108
9be3395b 7109 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 7110 if (!io_bio) {
7111 ret = -ENOMEM;
7112 goto free_ordered;
7113 }
7114
facc8a22
MX		 7115 if (!skip_sum && !write) {
7116 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7117 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7118 sum_len *= csum_size;
7119 } else {
7120 sum_len = 0;
7121 }
7122
7123 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
4b46fce2
JB		 7124 if (!dip) {
7125 ret = -ENOMEM;
9be3395b 7126 goto free_io_bio;
4b46fce2 7127 }
4b46fce2 7128
9be3395b 7129 dip->private = dio_bio->bi_private;
4b46fce2
JB		 7130 dip->inode = inode;
7131 dip->logical_offset = file_offset;
e6da5d2e 7132 dip->bytes = dio_bio->bi_size;
9be3395b
CM		 7133 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7134 io_bio->bi_private = dip;
e65e1535 7135 dip->errors = 0;
9be3395b
CM		 7136 dip->orig_bio = io_bio;
7137 dip->dio_bio = dio_bio;
e65e1535 7138 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7139
7140 if (write)
9be3395b 7141 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7142 else
9be3395b 7143 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7144
e65e1535
MX		 7145 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7146 if (!ret)
eaf25d93 7147 return;
9be3395b
CM		 7148
7149free_io_bio:
7150 bio_put(io_bio);
7151
4b46fce2
JB		 7152free_ordered:
7153 /*
7154 * If this is a write, we need to clean up the reserved space and kill
7155 * the ordered extent.
7156 */
7157 if (write) {
7158 struct btrfs_ordered_extent *ordered;
955256f2 7159 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7160 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7161 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7162 btrfs_free_reserved_extent(root, ordered->start,
7163 ordered->disk_len);
7164 btrfs_put_ordered_extent(ordered);
7165 btrfs_put_ordered_extent(ordered);
7166 }
9be3395b 7167 bio_endio(dio_bio, ret);
4b46fce2
JB		 7168}
7169
5a5f79b5
CM		 7170static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7171 const struct iovec *iov, loff_t offset,
7172 unsigned long nr_segs)
7173{
7174 int seg;
a1b75f7d 7175 int i;
5a5f79b5
CM		 7176 size_t size;
7177 unsigned long addr;
7178 unsigned blocksize_mask = root->sectorsize - 1;
7179 ssize_t retval = -EINVAL;
7180 loff_t end = offset;
7181
7182 if (offset & blocksize_mask)
7183 goto out;
7184
7185 /* Check the memory alignment. Blocks cannot straddle pages */
7186 for (seg = 0; seg < nr_segs; seg++) {
7187 addr = (unsigned long)iov[seg].iov_base;
7188 size = iov[seg].iov_len;
7189 end += size;
a1b75f7d 7190 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7191 goto out;
a1b75f7d
JB		 7192
7193 /* If this is a write we don't need to check anymore */
7194 if (rw & WRITE)
7195 continue;
7196
7197 /*
7198 * Check to make sure we don't have duplicate iov_base's in this
7199 * iovec, if so return EINVAL, otherwise we'll get csum errors
7200 * when reading back.
7201 */
7202 for (i = seg + 1; i < nr_segs; i++) {
7203 if (iov[seg].iov_base == iov[i].iov_base)
7204 goto out;
7205 }
5a5f79b5
CM		 7206 }
7207 retval = 0;
7208out:
7209 return retval;
7210}
eb838e73 7211
16432985
CM		 7212static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7213 const struct iovec *iov, loff_t offset,
7214 unsigned long nr_segs)
7215{
4b46fce2
JB		 7216 struct file *file = iocb->ki_filp;
7217 struct inode *inode = file->f_mapping->host;
0934856d 7218 size_t count = 0;
2e60a51e 7219 int flags = 0;
38851cc1
MX		 7220 bool wakeup = true;
7221 bool relock = false;
0934856d 7222 ssize_t ret;
4b46fce2 7223
5a5f79b5 7224 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7225 offset, nr_segs))
5a5f79b5 7226 return 0;
3f7c579c 7227
38851cc1
MX		 7228 atomic_inc(&inode->i_dio_count);
7229 smp_mb__after_atomic_inc();
7230
0e267c44
JB		 7231 /*
7232 * The generic stuff only does filemap_write_and_wait_range, which isn't
7233 * enough if we've written compressed pages to this area, so we need to
7234 * call btrfs_wait_ordered_range to make absolutely sure that any
7235 * outstanding dirty pages are on disk.
7236 */
7237 count = iov_length(iov, nr_segs);
0ef8b726
JB		 7238 ret = btrfs_wait_ordered_range(inode, offset, count);
7239 if (ret)
7240 return ret;
0e267c44 7241
0934856d 7242 if (rw & WRITE) {
38851cc1
MX		 7243 /*
7244 * If the write DIO is beyond the EOF, we need update
7245 * the isize, but it is protected by i_mutex. So we can
7246 * not unlock the i_mutex at this case.
7247 */
7248 if (offset + count <= inode->i_size) {
7249 mutex_unlock(&inode->i_mutex);
7250 relock = true;
7251 }
0934856d
MX		 7252 ret = btrfs_delalloc_reserve_space(inode, count);
7253 if (ret)
38851cc1
MX		 7254 goto out;
7255 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7256 &BTRFS_I(inode)->runtime_flags))) {
7257 inode_dio_done(inode);
7258 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7259 wakeup = false;
0934856d
MX		 7260 }
7261
7262 ret = __blockdev_direct_IO(rw, iocb, inode,
7263 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7264 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7265 btrfs_submit_direct, flags);
0934856d
MX		 7266 if (rw & WRITE) {
7267 if (ret < 0 && ret != -EIOCBQUEUED)
7268 btrfs_delalloc_release_space(inode, count);
172a5049 7269 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7270 btrfs_delalloc_release_space(inode,
7271 count - (size_t)ret);
172a5049
MX		 7272 else
7273 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7274 }
38851cc1 7275out:
2e60a51e
MX		 7276 if (wakeup)
7277 inode_dio_done(inode);
38851cc1
MX		 7278 if (relock)
7279 mutex_lock(&inode->i_mutex);
0934856d
MX		 7280
7281 return ret;
16432985
CM		 7282}
7283
05dadc09
TI		 7284#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7285
1506fcc8
YS		 7286static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7287 __u64 start, __u64 len)
7288{
05dadc09
TI		 7289 int ret;
7290
7291 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7292 if (ret)
7293 return ret;
7294
ec29ed5b 7295 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7296}
7297
a52d9a80 7298int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7299{
d1310b2e
CM		 7300 struct extent_io_tree *tree;
7301 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7302 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7303}
1832a6d5 7304
a52d9a80 7305static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7306{
d1310b2e 7307 struct extent_io_tree *tree;
b888db2b
CM		 7308
7309
7310 if (current->flags & PF_MEMALLOC) {
7311 redirty_page_for_writepage(wbc, page);
7312 unlock_page(page);
7313 return 0;
7314 }
d1310b2e 7315 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7316 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7317}
7318
48a3b636
ES		 7319static int btrfs_writepages(struct address_space *mapping,
7320 struct writeback_control *wbc)
b293f02e 7321{
d1310b2e 7322 struct extent_io_tree *tree;
771ed689 7323
d1310b2e 7324 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7325 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7326}
7327
3ab2fb5a
CM		 7328static int
7329btrfs_readpages(struct file *file, struct address_space *mapping,
7330 struct list_head *pages, unsigned nr_pages)
7331{
d1310b2e
CM		 7332 struct extent_io_tree *tree;
7333 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7334 return extent_readpages(tree, mapping, pages, nr_pages,
7335 btrfs_get_extent);
7336}
e6dcd2dc 7337static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7338{
d1310b2e
CM		 7339 struct extent_io_tree *tree;
7340 struct extent_map_tree *map;
a52d9a80 7341 int ret;
8c2383c3 7342
d1310b2e
CM		 7343 tree = &BTRFS_I(page->mapping->host)->io_tree;
7344 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7345 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7346 if (ret == 1) {
7347 ClearPagePrivate(page);
7348 set_page_private(page, 0);
7349 page_cache_release(page);
39279cc3 7350 }
a52d9a80 7351 return ret;
39279cc3
CM		 7352}
7353
e6dcd2dc
CM		 7354static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7355{
98509cfc
CM		 7356 if (PageWriteback(page) || PageDirty(page))
7357 return 0;
b335b003 7358 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7359}
7360
d47992f8
LC		 7361static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7362 unsigned int length)
39279cc3 7363{
5fd02043 7364 struct inode *inode = page->mapping->host;
d1310b2e 7365 struct extent_io_tree *tree;
e6dcd2dc 7366 struct btrfs_ordered_extent *ordered;
2ac55d41 7367 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7368 u64 page_start = page_offset(page);
7369 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7370
8b62b72b
CM		 7371 /*
7372 * we have the page locked, so new writeback can't start,
7373 * and the dirty bit won't be cleared while we are here.
7374 *
7375 * Wait for IO on this page so that we can safely clear
7376 * the PagePrivate2 bit and do ordered accounting
7377 */
e6dcd2dc 7378 wait_on_page_writeback(page);
8b62b72b 7379
5fd02043 7380 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7381 if (offset) {
7382 btrfs_releasepage(page, GFP_NOFS);
7383 return;
7384 }
d0082371 7385 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7386 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7387 if (ordered) {
eb84ae03
CM		 7388 /*
7389 * IO on this page will never be started, so we need
7390 * to account for any ordered extents now
7391 */
e6dcd2dc
CM		 7392 clear_extent_bit(tree, page_start, page_end,
7393 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7394 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7395 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7396 /*
7397 * whoever cleared the private bit is responsible
7398 * for the finish_ordered_io
7399 */
77cef2ec
JB		 7400 if (TestClearPagePrivate2(page)) {
7401 struct btrfs_ordered_inode_tree *tree;
7402 u64 new_len;
7403
7404 tree = &BTRFS_I(inode)->ordered_tree;
7405
7406 spin_lock_irq(&tree->lock);
7407 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7408 new_len = page_start - ordered->file_offset;
7409 if (new_len < ordered->truncated_len)
7410 ordered->truncated_len = new_len;
7411 spin_unlock_irq(&tree->lock);
7412
7413 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7414 page_start,
7415 PAGE_CACHE_SIZE, 1))
7416 btrfs_finish_ordered_io(ordered);
8b62b72b 7417 }
e6dcd2dc 7418 btrfs_put_ordered_extent(ordered);
2ac55d41 7419 cached_state = NULL;
d0082371 7420 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7421 }
7422 clear_extent_bit(tree, page_start, page_end,
32c00aff 7423 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7424 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7425 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7426 __btrfs_releasepage(page, GFP_NOFS);
7427
4a096752 7428 ClearPageChecked(page);
9ad6b7bc 7429 if (PagePrivate(page)) {
9ad6b7bc
CM		 7430 ClearPagePrivate(page);
7431 set_page_private(page, 0);
7432 page_cache_release(page);
7433 }
39279cc3
CM		 7434}
7435
9ebefb18
CM		 7436/*
7437 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7438 * called from a page fault handler when a page is first dirtied. Hence we must
7439 * be careful to check for EOF conditions here. We set the page up correctly
7440 * for a written page which means we get ENOSPC checking when writing into
7441 * holes and correct delalloc and unwritten extent mapping on filesystems that
7442 * support these features.
7443 *
7444 * We are not allowed to take the i_mutex here so we have to play games to
7445 * protect against truncate races as the page could now be beyond EOF. Because
7446 * vmtruncate() writes the inode size before removing pages, once we have the
7447 * page lock we can determine safely if the page is beyond EOF. If it is not
7448 * beyond EOF, then the page is guaranteed safe against truncation until we
7449 * unlock the page.
7450 */
c2ec175c 7451int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7452{
c2ec175c 7453 struct page *page = vmf->page;
496ad9aa 7454 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7455 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7456 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7457 struct btrfs_ordered_extent *ordered;
2ac55d41 7458 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7459 char *kaddr;
7460 unsigned long zero_start;
9ebefb18 7461 loff_t size;
1832a6d5 7462 int ret;
9998eb70 7463 int reserved = 0;
a52d9a80 7464 u64 page_start;
e6dcd2dc 7465 u64 page_end;
9ebefb18 7466
b2b5ef5c 7467 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7468 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7469 if (!ret) {
e41f941a 7470 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7471 reserved = 1;
7472 }
56a76f82
NP		 7473 if (ret) {
7474 if (ret == -ENOMEM)
7475 ret = VM_FAULT_OOM;
7476 else /* -ENOSPC, -EIO, etc */
7477 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7478 if (reserved)
7479 goto out;
7480 goto out_noreserve;
56a76f82 7481 }
1832a6d5 7482
56a76f82 7483 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7484again:
9ebefb18 7485 lock_page(page);
9ebefb18 7486 size = i_size_read(inode);
e6dcd2dc
CM		 7487 page_start = page_offset(page);
7488 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7489
9ebefb18 7490 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7491 (page_start >= size)) {
9ebefb18
CM		 7492 /* page got truncated out from underneath us */
7493 goto out_unlock;
7494 }
e6dcd2dc
CM		 7495 wait_on_page_writeback(page);
7496
d0082371 7497 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7498 set_page_extent_mapped(page);
7499
eb84ae03
CM		 7500 /*
7501 * we can't set the delalloc bits if there are pending ordered
7502 * extents. Drop our locks and wait for them to finish
7503 */
e6dcd2dc
CM		 7504 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7505 if (ordered) {
2ac55d41
JB		 7506 unlock_extent_cached(io_tree, page_start, page_end,
7507 &cached_state, GFP_NOFS);
e6dcd2dc 7508 unlock_page(page);
eb84ae03 7509 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7510 btrfs_put_ordered_extent(ordered);
7511 goto again;
7512 }
7513
fbf19087
JB		 7514 /*
7515 * XXX - page_mkwrite gets called every time the page is dirtied, even
7516 * if it was already dirty, so for space accounting reasons we need to
7517 * clear any delalloc bits for the range we are fixing to save. There
7518 * is probably a better way to do this, but for now keep consistent with
7519 * prepare_pages in the normal write path.
7520 */
2ac55d41 7521 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7522 EXTENT_DIRTY | EXTENT_DELALLOC |
7523 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7524 0, 0, &cached_state, GFP_NOFS);
fbf19087 7525
2ac55d41
JB		 7526 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7527 &cached_state);
9ed74f2d 7528 if (ret) {
2ac55d41
JB		 7529 unlock_extent_cached(io_tree, page_start, page_end,
7530 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7531 ret = VM_FAULT_SIGBUS;
7532 goto out_unlock;
7533 }
e6dcd2dc 7534 ret = 0;
9ebefb18
CM		 7535
7536 /* page is wholly or partially inside EOF */
a52d9a80 7537 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7538 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7539 else
e6dcd2dc 7540 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7541
e6dcd2dc
CM		 7542 if (zero_start != PAGE_CACHE_SIZE) {
7543 kaddr = kmap(page);
7544 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7545 flush_dcache_page(page);
7546 kunmap(page);
7547 }
247e743c 7548 ClearPageChecked(page);
e6dcd2dc 7549 set_page_dirty(page);
50a9b214 7550 SetPageUptodate(page);
5a3f23d5 7551
257c62e1
CM		 7552 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7553 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7554 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7555
2ac55d41 7556 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7557
7558out_unlock:
b2b5ef5c
JK		 7559 if (!ret) {
7560 sb_end_pagefault(inode->i_sb);
50a9b214 7561 return VM_FAULT_LOCKED;
b2b5ef5c 7562 }
9ebefb18 7563 unlock_page(page);
1832a6d5 7564out:
ec39e180 7565 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7566out_noreserve:
b2b5ef5c 7567 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7568 return ret;
7569}
7570
a41ad394 7571static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7572{
7573 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7574 struct btrfs_block_rsv *rsv;
a71754fc 7575 int ret = 0;
3893e33b 7576 int err = 0;
39279cc3 7577 struct btrfs_trans_handle *trans;
dbe674a9 7578 u64 mask = root->sectorsize - 1;
07127184 7579 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7580
0ef8b726
JB		 7581 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7582 (u64)-1);
7583 if (ret)
7584 return ret;
39279cc3 7585
fcb80c2a
JB		 7586 /*
7587 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7588 * 3 things going on here
7589 *
7590 * 1) We need to reserve space for our orphan item and the space to
7591 * delete our orphan item. Lord knows we don't want to have a dangling
7592 * orphan item because we didn't reserve space to remove it.
7593 *
7594 * 2) We need to reserve space to update our inode.
7595 *
7596 * 3) We need to have something to cache all the space that is going to
7597 * be free'd up by the truncate operation, but also have some slack
7598 * space reserved in case it uses space during the truncate (thank you
7599 * very much snapshotting).
7600 *
7601 * And we need these to all be seperate. The fact is we can use alot of
7602 * space doing the truncate, and we have no earthly idea how much space
7603 * we will use, so we need the truncate reservation to be seperate so it
7604 * doesn't end up using space reserved for updating the inode or
7605 * removing the orphan item. We also need to be able to stop the
7606 * transaction and start a new one, which means we need to be able to
7607 * update the inode several times, and we have no idea of knowing how
7608 * many times that will be, so we can't just reserve 1 item for the
7609 * entirety of the opration, so that has to be done seperately as well.
7610 * Then there is the orphan item, which does indeed need to be held on
7611 * to for the whole operation, and we need nobody to touch this reserved
7612 * space except the orphan code.
7613 *
7614 * So that leaves us with
7615 *
7616 * 1) root->orphan_block_rsv - for the orphan deletion.
7617 * 2) rsv - for the truncate reservation, which we will steal from the
7618 * transaction reservation.
7619 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7620 * updating the inode.
7621 */
66d8f3dd 7622 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 7623 if (!rsv)
7624 return -ENOMEM;
4a338542 7625 rsv->size = min_size;
ca7e70f5 7626 rsv->failfast = 1;
f0cd846e 7627
907cbceb 7628 /*
07127184 7629 * 1 for the truncate slack space
907cbceb
JB		 7630 * 1 for updating the inode.
7631 */
f3fe820c 7632 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 7633 if (IS_ERR(trans)) {
7634 err = PTR_ERR(trans);
7635 goto out;
7636 }
f0cd846e 7637
907cbceb
JB		 7638 /* Migrate the slack space for the truncate to our reserve */
7639 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7640 min_size);
fcb80c2a 7641 BUG_ON(ret);
f0cd846e 7642
5a3f23d5
CM		 7643 /*
7644 * setattr is responsible for setting the ordered_data_close flag,
7645 * but that is only tested during the last file release. That
7646 * could happen well after the next commit, leaving a great big
7647 * window where new writes may get lost if someone chooses to write
7648 * to this file after truncating to zero
7649 *
7650 * The inode doesn't have any dirty data here, and so if we commit
7651 * this is a noop. If someone immediately starts writing to the inode
7652 * it is very likely we'll catch some of their writes in this
7653 * transaction, and the commit will find this file on the ordered
7654 * data list with good things to send down.
7655 *
7656 * This is a best effort solution, there is still a window where
7657 * using truncate to replace the contents of the file will
7658 * end up with a zero length file after a crash.
7659 */
72ac3c0d
JB		 7660 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7661 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 7662 btrfs_add_ordered_operation(trans, root, inode);
7663
5dc562c5
JB		 7664 /*
7665 * So if we truncate and then write and fsync we normally would just
7666 * write the extents that changed, which is a problem if we need to
7667 * first truncate that entire inode. So set this flag so we write out
7668 * all of the extents in the inode to the sync log so we're completely
7669 * safe.
7670 */
7671 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7672 trans->block_rsv = rsv;
907cbceb 7673
8082510e
YZ		 7674 while (1) {
7675 ret = btrfs_truncate_inode_items(trans, root, inode,
7676 inode->i_size,
7677 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7678 if (ret != -ENOSPC) {
3893e33b 7679 err = ret;
8082510e 7680 break;
3893e33b 7681 }
39279cc3 7682
fcb80c2a 7683 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7684 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 7685 if (ret) {
7686 err = ret;
7687 break;
7688 }
ca7e70f5 7689
8082510e 7690 btrfs_end_transaction(trans, root);
b53d3f5d 7691 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 7692
7693 trans = btrfs_start_transaction(root, 2);
7694 if (IS_ERR(trans)) {
7695 ret = err = PTR_ERR(trans);
7696 trans = NULL;
7697 break;
7698 }
7699
7700 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7701 rsv, min_size);
7702 BUG_ON(ret); /* shouldn't happen */
7703 trans->block_rsv = rsv;
8082510e
YZ		 7704 }
7705
7706 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7707 trans->block_rsv = root->orphan_block_rsv;
8082510e 7708 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7709 if (ret)
7710 err = ret;
8082510e
YZ		 7711 }
7712
917c16b2
CM		 7713 if (trans) {
7714 trans->block_rsv = &root->fs_info->trans_block_rsv;
7715 ret = btrfs_update_inode(trans, root, inode);
7716 if (ret && !err)
7717 err = ret;
7b128766 7718
7ad85bb7 7719 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7720 btrfs_btree_balance_dirty(root);
917c16b2 7721 }
fcb80c2a
JB		 7722
7723out:
7724 btrfs_free_block_rsv(root, rsv);
7725
3893e33b
JB		 7726 if (ret && !err)
7727 err = ret;
a41ad394 7728
3893e33b 7729 return err;
39279cc3
CM		 7730}
7731
d352ac68
CM		 7732/*
7733 * create a new subvolume directory/inode (helper for the ioctl).
7734 */
d2fb3437 7735int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7736 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7737{
39279cc3 7738 struct inode *inode;
76dda93c 7739 int err;
00e4e6b3 7740 u64 index = 0;
39279cc3 7741
12fc9d09
FA		 7742 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7743 new_dirid, new_dirid,
7744 S_IFDIR | (~current_umask() & S_IRWXUGO),
7745 &index);
54aa1f4d 7746 if (IS_ERR(inode))
f46b5a66 7747 return PTR_ERR(inode);
39279cc3
CM		 7748 inode->i_op = &btrfs_dir_inode_operations;
7749 inode->i_fop = &btrfs_dir_file_operations;
7750
bfe86848 7751 set_nlink(inode, 1);
dbe674a9 7752 btrfs_i_size_write(inode, 0);
3b96362c 7753
76dda93c 7754 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7755
76dda93c 7756 iput(inode);
ce598979 7757 return err;
39279cc3
CM		 7758}
7759
39279cc3
CM		 7760struct inode *btrfs_alloc_inode(struct super_block *sb)
7761{
7762 struct btrfs_inode *ei;
2ead6ae7 7763 struct inode *inode;
39279cc3
CM		 7764
7765 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7766 if (!ei)
7767 return NULL;
2ead6ae7
YZ		 7768
7769 ei->root = NULL;
2ead6ae7 7770 ei->generation = 0;
15ee9bc7 7771 ei->last_trans = 0;
257c62e1 7772 ei->last_sub_trans = 0;
e02119d5 7773 ei->logged_trans = 0;
2ead6ae7 7774 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7775 ei->disk_i_size = 0;
7776 ei->flags = 0;
7709cde3 7777 ei->csum_bytes = 0;
2ead6ae7
YZ		 7778 ei->index_cnt = (u64)-1;
7779 ei->last_unlink_trans = 0;
46d8bc34 7780 ei->last_log_commit = 0;
2ead6ae7 7781
9e0baf60
JB		 7782 spin_lock_init(&ei->lock);
7783 ei->outstanding_extents = 0;
7784 ei->reserved_extents = 0;
2ead6ae7 7785
72ac3c0d 7786 ei->runtime_flags = 0;
261507a0 7787 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7788
16cdcec7
MX		 7789 ei->delayed_node = NULL;
7790
2ead6ae7 7791 inode = &ei->vfs_inode;
a8067e02 7792 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7793 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7794 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7795 ei->io_tree.track_uptodate = 1;
7796 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7797 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7798 mutex_init(&ei->log_mutex);
f248679e 7799 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7800 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7801 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7802 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7803 RB_CLEAR_NODE(&ei->rb_node);
7804
7805 return inode;
39279cc3
CM		 7806}
7807
aaedb55b
JB		 7808#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7809void btrfs_test_destroy_inode(struct inode *inode)
7810{
7811 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7812 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7813}
7814#endif
7815
fa0d7e3d
NP		 7816static void btrfs_i_callback(struct rcu_head *head)
7817{
7818 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7819 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7820}
7821
39279cc3
CM		 7822void btrfs_destroy_inode(struct inode *inode)
7823{
e6dcd2dc 7824 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7825 struct btrfs_root *root = BTRFS_I(inode)->root;
7826
b3d9b7a3 7827 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7828 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7829 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7830 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7831 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7832 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7833
a6dbd429
JB		 7834 /*
7835 * This can happen where we create an inode, but somebody else also
7836 * created the same inode and we need to destroy the one we already
7837 * created.
7838 */
7839 if (!root)
7840 goto free;
7841
5a3f23d5
CM		 7842 /*
7843 * Make sure we're properly removed from the ordered operation
7844 * lists.
7845 */
7846 smp_mb();
7847 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
199c2a9c 7848 spin_lock(&root->fs_info->ordered_root_lock);
5a3f23d5 7849 list_del_init(&BTRFS_I(inode)->ordered_operations);
199c2a9c 7850 spin_unlock(&root->fs_info->ordered_root_lock);
5a3f23d5
CM		 7851 }
7852
8a35d95f
JB		 7853 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7854 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 7855 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 7856 btrfs_ino(inode));
8a35d95f 7857 atomic_dec(&root->orphan_inodes);
7b128766 7858 }
7b128766 7859
d397712b 7860 while (1) {
e6dcd2dc
CM		 7861 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7862 if (!ordered)
7863 break;
7864 else {
c2cf52eb 7865 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 7866 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 7867 btrfs_remove_ordered_extent(inode, ordered);
7868 btrfs_put_ordered_extent(ordered);
7869 btrfs_put_ordered_extent(ordered);
7870 }
7871 }
5d4f98a2 7872 inode_tree_del(inode);
5b21f2ed 7873 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7874free:
fa0d7e3d 7875 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7876}
7877
45321ac5 7878int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7879{
7880 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7881
6379ef9f
NA		 7882 if (root == NULL)
7883 return 1;
7884
fa6ac876 7885 /* the snap/subvol tree is on deleting */
69e9c6c6 7886 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 7887 return 1;
76dda93c 7888 else
45321ac5 7889 return generic_drop_inode(inode);
76dda93c
YZ		 7890}
7891
0ee0fda0 7892static void init_once(void *foo)
39279cc3
CM		 7893{
7894 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7895
7896 inode_init_once(&ei->vfs_inode);
7897}
7898
7899void btrfs_destroy_cachep(void)
7900{
8c0a8537
KS		 7901 /*
7902 * Make sure all delayed rcu free inodes are flushed before we
7903 * destroy cache.
7904 */
7905 rcu_barrier();
39279cc3
CM		 7906 if (btrfs_inode_cachep)
7907 kmem_cache_destroy(btrfs_inode_cachep);
7908 if (btrfs_trans_handle_cachep)
7909 kmem_cache_destroy(btrfs_trans_handle_cachep);
7910 if (btrfs_transaction_cachep)
7911 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7912 if (btrfs_path_cachep)
7913 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7914 if (btrfs_free_space_cachep)
7915 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 7916 if (btrfs_delalloc_work_cachep)
7917 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 7918}
7919
7920int btrfs_init_cachep(void)
7921{
837e1972 7922 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 7923 sizeof(struct btrfs_inode), 0,
7924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7925 if (!btrfs_inode_cachep)
7926 goto fail;
9601e3f6 7927
837e1972 7928 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 7929 sizeof(struct btrfs_trans_handle), 0,
7930 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7931 if (!btrfs_trans_handle_cachep)
7932 goto fail;
9601e3f6 7933
837e1972 7934 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 7935 sizeof(struct btrfs_transaction), 0,
7936 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7937 if (!btrfs_transaction_cachep)
7938 goto fail;
9601e3f6 7939
837e1972 7940 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 7941 sizeof(struct btrfs_path), 0,
7942 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7943 if (!btrfs_path_cachep)
7944 goto fail;
9601e3f6 7945
837e1972 7946 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 7947 sizeof(struct btrfs_free_space), 0,
7948 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7949 if (!btrfs_free_space_cachep)
7950 goto fail;
7951
8ccf6f19
MX		 7952 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7953 sizeof(struct btrfs_delalloc_work), 0,
7954 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7955 NULL);
7956 if (!btrfs_delalloc_work_cachep)
7957 goto fail;
7958
39279cc3
CM		 7959 return 0;
7960fail:
7961 btrfs_destroy_cachep();
7962 return -ENOMEM;
7963}
7964
7965static int btrfs_getattr(struct vfsmount *mnt,
7966 struct dentry *dentry, struct kstat *stat)
7967{
df0af1a5 7968 u64 delalloc_bytes;
39279cc3 7969 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7970 u32 blocksize = inode->i_sb->s_blocksize;
7971
39279cc3 7972 generic_fillattr(inode, stat);
0ee5dc67 7973 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7974 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 7975
7976 spin_lock(&BTRFS_I(inode)->lock);
7977 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7978 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 7979 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 7980 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7981 return 0;
7982}
7983
d397712b
CM		 7984static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7985 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7986{
7987 struct btrfs_trans_handle *trans;
7988 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7989 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7990 struct inode *new_inode = new_dentry->d_inode;
7991 struct inode *old_inode = old_dentry->d_inode;
7992 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7993 u64 index = 0;
4df27c4d 7994 u64 root_objectid;
39279cc3 7995 int ret;
33345d01 7996 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7997
33345d01 7998 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 7999 return -EPERM;
8000
4df27c4d 8001 /* we only allow rename subvolume link between subvolumes */
33345d01 8002 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8003 return -EXDEV;
8004
33345d01
LZ		 8005 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8006 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8007 return -ENOTEMPTY;
5f39d397 8008
4df27c4d
YZ		 8009 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8010 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8011 return -ENOTEMPTY;
9c52057c
CM		 8012
8013
8014 /* check for collisions, even if the name isn't there */
4871c158 8015 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 8016 new_dentry->d_name.name,
8017 new_dentry->d_name.len);
8018
8019 if (ret) {
8020 if (ret == -EEXIST) {
8021 /* we shouldn't get
8022 * eexist without a new_inode */
fae7f21c 8023 if (WARN_ON(!new_inode)) {
9c52057c
CM		 8024 return ret;
8025 }
8026 } else {
8027 /* maybe -EOVERFLOW */
8028 return ret;
8029 }
8030 }
8031 ret = 0;
8032
5a3f23d5
CM		 8033 /*
8034 * we're using rename to replace one file with another.
8035 * and the replacement file is large. Start IO on it now so
8036 * we don't add too much work to the end of the transaction
8037 */
4baf8c92 8038 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8039 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8040 filemap_flush(old_inode->i_mapping);
8041
76dda93c 8042 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8043 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8044 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8045 /*
8046 * We want to reserve the absolute worst case amount of items. So if
8047 * both inodes are subvols and we need to unlink them then that would
8048 * require 4 item modifications, but if they are both normal inodes it
8049 * would require 5 item modifications, so we'll assume their normal
8050 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8051 * should cover the worst case number of items we'll modify.
8052 */
6e137ed3 8053 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8054 if (IS_ERR(trans)) {
8055 ret = PTR_ERR(trans);
8056 goto out_notrans;
8057 }
76dda93c 8058
4df27c4d
YZ		 8059 if (dest != root)
8060 btrfs_record_root_in_trans(trans, dest);
5f39d397 8061
a5719521
YZ		 8062 ret = btrfs_set_inode_index(new_dir, &index);
8063 if (ret)
8064 goto out_fail;
5a3f23d5 8065
33345d01 8066 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8067 /* force full log commit if subvolume involved. */
8068 root->fs_info->last_trans_log_full_commit = trans->transid;
8069 } else {
a5719521
YZ		 8070 ret = btrfs_insert_inode_ref(trans, dest,
8071 new_dentry->d_name.name,
8072 new_dentry->d_name.len,
33345d01
LZ		 8073 old_ino,
8074 btrfs_ino(new_dir), index);
a5719521
YZ		 8075 if (ret)
8076 goto out_fail;
4df27c4d
YZ		 8077 /*
8078 * this is an ugly little race, but the rename is required
8079 * to make sure that if we crash, the inode is either at the
8080 * old name or the new one. pinning the log transaction lets
8081 * us make sure we don't allow a log commit to come in after
8082 * we unlink the name but before we add the new name back in.
8083 */
8084 btrfs_pin_log_trans(root);
8085 }
5a3f23d5
CM		 8086 /*
8087 * make sure the inode gets flushed if it is replacing
8088 * something.
8089 */
33345d01 8090 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8091 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8092
0c4d2d95
JB		 8093 inode_inc_iversion(old_dir);
8094 inode_inc_iversion(new_dir);
8095 inode_inc_iversion(old_inode);
39279cc3
CM		 8096 old_dir->i_ctime = old_dir->i_mtime = ctime;
8097 new_dir->i_ctime = new_dir->i_mtime = ctime;
8098 old_inode->i_ctime = ctime;
5f39d397 8099
12fcfd22
CM		 8100 if (old_dentry->d_parent != new_dentry->d_parent)
8101 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8102
33345d01 8103 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8104 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8105 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8106 old_dentry->d_name.name,
8107 old_dentry->d_name.len);
8108 } else {
92986796
AV		 8109 ret = __btrfs_unlink_inode(trans, root, old_dir,
8110 old_dentry->d_inode,
8111 old_dentry->d_name.name,
8112 old_dentry->d_name.len);
8113 if (!ret)
8114 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8115 }
79787eaa
JM		 8116 if (ret) {
8117 btrfs_abort_transaction(trans, root, ret);
8118 goto out_fail;
8119 }
39279cc3
CM		 8120
8121 if (new_inode) {
0c4d2d95 8122 inode_inc_iversion(new_inode);
39279cc3 8123 new_inode->i_ctime = CURRENT_TIME;
33345d01 8124 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8125 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8126 root_objectid = BTRFS_I(new_inode)->location.objectid;
8127 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8128 root_objectid,
8129 new_dentry->d_name.name,
8130 new_dentry->d_name.len);
8131 BUG_ON(new_inode->i_nlink == 0);
8132 } else {
8133 ret = btrfs_unlink_inode(trans, dest, new_dir,
8134 new_dentry->d_inode,
8135 new_dentry->d_name.name,
8136 new_dentry->d_name.len);
8137 }
4ef31a45 8138 if (!ret && new_inode->i_nlink == 0)
e02119d5 8139 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
79787eaa
JM		 8140 if (ret) {
8141 btrfs_abort_transaction(trans, root, ret);
8142 goto out_fail;
8143 }
39279cc3 8144 }
aec7477b 8145
4df27c4d
YZ		 8146 ret = btrfs_add_link(trans, new_dir, old_inode,
8147 new_dentry->d_name.name,
a5719521 8148 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8149 if (ret) {
8150 btrfs_abort_transaction(trans, root, ret);
8151 goto out_fail;
8152 }
39279cc3 8153
33345d01 8154 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8155 struct dentry *parent = new_dentry->d_parent;
6a912213 8156 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8157 btrfs_end_log_trans(root);
8158 }
39279cc3 8159out_fail:
7ad85bb7 8160 btrfs_end_transaction(trans, root);
b44c59a8 8161out_notrans:
33345d01 8162 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8163 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8164
39279cc3
CM		 8165 return ret;
8166}
8167
8ccf6f19
MX		 8168static void btrfs_run_delalloc_work(struct btrfs_work *work)
8169{
8170 struct btrfs_delalloc_work *delalloc_work;
9f23e289 8171 struct inode *inode;
8ccf6f19
MX		 8172
8173 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8174 work);
9f23e289
JB		 8175 inode = delalloc_work->inode;
8176 if (delalloc_work->wait) {
8177 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8178 } else {
8179 filemap_flush(inode->i_mapping);
8180 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8181 &BTRFS_I(inode)->runtime_flags))
8182 filemap_flush(inode->i_mapping);
8183 }
8ccf6f19
MX		 8184
8185 if (delalloc_work->delay_iput)
9f23e289 8186 btrfs_add_delayed_iput(inode);
8ccf6f19 8187 else
9f23e289 8188 iput(inode);
8ccf6f19
MX		 8189 complete(&delalloc_work->completion);
8190}
8191
8192struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8193 int wait, int delay_iput)
8194{
8195 struct btrfs_delalloc_work *work;
8196
8197 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8198 if (!work)
8199 return NULL;
8200
8201 init_completion(&work->completion);
8202 INIT_LIST_HEAD(&work->list);
8203 work->inode = inode;
8204 work->wait = wait;
8205 work->delay_iput = delay_iput;
8206 work->work.func = btrfs_run_delalloc_work;
8207
8208 return work;
8209}
8210
8211void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8212{
8213 wait_for_completion(&work->completion);
8214 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8215}
8216
d352ac68
CM		 8217/*
8218 * some fairly slow code that needs optimization. This walks the list
8219 * of all the inodes with pending delalloc and forces them to disk.
8220 */
eb73c1b7 8221static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8222{
ea8c2819 8223 struct btrfs_inode *binode;
5b21f2ed 8224 struct inode *inode;
8ccf6f19
MX		 8225 struct btrfs_delalloc_work *work, *next;
8226 struct list_head works;
1eafa6c7 8227 struct list_head splice;
8ccf6f19 8228 int ret = 0;
ea8c2819 8229
8ccf6f19 8230 INIT_LIST_HEAD(&works);
1eafa6c7 8231 INIT_LIST_HEAD(&splice);
63607cc8 8232
eb73c1b7
MX		 8233 spin_lock(&root->delalloc_lock);
8234 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8235 while (!list_empty(&splice)) {
8236 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8237 delalloc_inodes);
1eafa6c7 8238
eb73c1b7
MX		 8239 list_move_tail(&binode->delalloc_inodes,
8240 &root->delalloc_inodes);
5b21f2ed 8241 inode = igrab(&binode->vfs_inode);
df0af1a5 8242 if (!inode) {
eb73c1b7 8243 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8244 continue;
df0af1a5 8245 }
eb73c1b7 8246 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8247
8248 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8249 if (unlikely(!work)) {
f4ab9ea7
JB		 8250 if (delay_iput)
8251 btrfs_add_delayed_iput(inode);
8252 else
8253 iput(inode);
1eafa6c7
MX		 8254 ret = -ENOMEM;
8255 goto out;
5b21f2ed 8256 }
1eafa6c7
MX		 8257 list_add_tail(&work->list, &works);
8258 btrfs_queue_worker(&root->fs_info->flush_workers,
8259 &work->work);
8260
5b21f2ed 8261 cond_resched();
eb73c1b7 8262 spin_lock(&root->delalloc_lock);
ea8c2819 8263 }
eb73c1b7 8264 spin_unlock(&root->delalloc_lock);
8c8bee1d 8265
1eafa6c7
MX		 8266 list_for_each_entry_safe(work, next, &works, list) {
8267 list_del_init(&work->list);
8268 btrfs_wait_and_free_delalloc_work(work);
8269 }
eb73c1b7
MX		 8270 return 0;
8271out:
8272 list_for_each_entry_safe(work, next, &works, list) {
8273 list_del_init(&work->list);
8274 btrfs_wait_and_free_delalloc_work(work);
8275 }
8276
8277 if (!list_empty_careful(&splice)) {
8278 spin_lock(&root->delalloc_lock);
8279 list_splice_tail(&splice, &root->delalloc_inodes);
8280 spin_unlock(&root->delalloc_lock);
8281 }
8282 return ret;
8283}
1eafa6c7 8284
eb73c1b7
MX		 8285int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8286{
8287 int ret;
1eafa6c7 8288
eb73c1b7
MX		 8289 if (root->fs_info->sb->s_flags & MS_RDONLY)
8290 return -EROFS;
8291
8292 ret = __start_delalloc_inodes(root, delay_iput);
8293 /*
8294 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8295 * we have to make sure the IO is actually started and that
8296 * ordered extents get created before we return
8297 */
8298 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8299 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8300 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8301 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8302 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8303 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8304 }
8305 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8306 return ret;
8307}
8308
8309int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8310 int delay_iput)
8311{
8312 struct btrfs_root *root;
8313 struct list_head splice;
8314 int ret;
8315
8316 if (fs_info->sb->s_flags & MS_RDONLY)
8317 return -EROFS;
8318
8319 INIT_LIST_HEAD(&splice);
8320
8321 spin_lock(&fs_info->delalloc_root_lock);
8322 list_splice_init(&fs_info->delalloc_roots, &splice);
8323 while (!list_empty(&splice)) {
8324 root = list_first_entry(&splice, struct btrfs_root,
8325 delalloc_root);
8326 root = btrfs_grab_fs_root(root);
8327 BUG_ON(!root);
8328 list_move_tail(&root->delalloc_root,
8329 &fs_info->delalloc_roots);
8330 spin_unlock(&fs_info->delalloc_root_lock);
8331
8332 ret = __start_delalloc_inodes(root, delay_iput);
8333 btrfs_put_fs_root(root);
8334 if (ret)
8335 goto out;
8336
8337 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8338 }
eb73c1b7 8339 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8340
eb73c1b7
MX		 8341 atomic_inc(&fs_info->async_submit_draining);
8342 while (atomic_read(&fs_info->nr_async_submits) ||
8343 atomic_read(&fs_info->async_delalloc_pages)) {
8344 wait_event(fs_info->async_submit_wait,
8345 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8346 atomic_read(&fs_info->async_delalloc_pages) == 0));
8347 }
8348 atomic_dec(&fs_info->async_submit_draining);
8349 return 0;
8350out:
1eafa6c7 8351 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8352 spin_lock(&fs_info->delalloc_root_lock);
8353 list_splice_tail(&splice, &fs_info->delalloc_roots);
8354 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8355 }
8ccf6f19 8356 return ret;
ea8c2819
CM		 8357}
8358
39279cc3
CM		 8359static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8360 const char *symname)
8361{
8362 struct btrfs_trans_handle *trans;
8363 struct btrfs_root *root = BTRFS_I(dir)->root;
8364 struct btrfs_path *path;
8365 struct btrfs_key key;
1832a6d5 8366 struct inode *inode = NULL;
39279cc3
CM		 8367 int err;
8368 int drop_inode = 0;
8369 u64 objectid;
00e4e6b3 8370 u64 index = 0 ;
39279cc3
CM		 8371 int name_len;
8372 int datasize;
5f39d397 8373 unsigned long ptr;
39279cc3 8374 struct btrfs_file_extent_item *ei;
5f39d397 8375 struct extent_buffer *leaf;
39279cc3 8376
f06becc4 8377 name_len = strlen(symname);
39279cc3
CM		 8378 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8379 return -ENAMETOOLONG;
1832a6d5 8380
9ed74f2d
JB		 8381 /*
8382 * 2 items for inode item and ref
8383 * 2 items for dir items
8384 * 1 item for xattr if selinux is on
8385 */
a22285a6
YZ		 8386 trans = btrfs_start_transaction(root, 5);
8387 if (IS_ERR(trans))
8388 return PTR_ERR(trans);
1832a6d5 8389
581bb050
LZ		 8390 err = btrfs_find_free_ino(root, &objectid);
8391 if (err)
8392 goto out_unlock;
8393
aec7477b 8394 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8395 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8396 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8397 if (IS_ERR(inode)) {
8398 err = PTR_ERR(inode);
39279cc3 8399 goto out_unlock;
7cf96da3 8400 }
39279cc3 8401
2a7dba39 8402 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8403 if (err) {
8404 drop_inode = 1;
8405 goto out_unlock;
8406 }
8407
ad19db71
CS		 8408 /*
8409 * If the active LSM wants to access the inode during
8410 * d_instantiate it needs these. Smack checks to see
8411 * if the filesystem supports xattrs by looking at the
8412 * ops vector.
8413 */
8414 inode->i_fop = &btrfs_file_operations;
8415 inode->i_op = &btrfs_file_inode_operations;
8416
a1b075d2 8417 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8418 if (err)
8419 drop_inode = 1;
8420 else {
8421 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8422 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8423 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8424 }
39279cc3
CM		 8425 if (drop_inode)
8426 goto out_unlock;
8427
8428 path = btrfs_alloc_path();
d8926bb3
MF		 8429 if (!path) {
8430 err = -ENOMEM;
8431 drop_inode = 1;
8432 goto out_unlock;
8433 }
33345d01 8434 key.objectid = btrfs_ino(inode);
39279cc3 8435 key.offset = 0;
39279cc3
CM		 8436 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8437 datasize = btrfs_file_extent_calc_inline_size(name_len);
8438 err = btrfs_insert_empty_item(trans, root, path, &key,
8439 datasize);
54aa1f4d
CM		 8440 if (err) {
8441 drop_inode = 1;
b0839166 8442 btrfs_free_path(path);
54aa1f4d
CM		 8443 goto out_unlock;
8444 }
5f39d397
CM		 8445 leaf = path->nodes[0];
8446 ei = btrfs_item_ptr(leaf, path->slots[0],
8447 struct btrfs_file_extent_item);
8448 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8449 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8450 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8451 btrfs_set_file_extent_encryption(leaf, ei, 0);
8452 btrfs_set_file_extent_compression(leaf, ei, 0);
8453 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8454 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8455
39279cc3 8456 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8457 write_extent_buffer(leaf, symname, ptr, name_len);
8458 btrfs_mark_buffer_dirty(leaf);
39279cc3 8459 btrfs_free_path(path);
5f39d397 8460
39279cc3
CM		 8461 inode->i_op = &btrfs_symlink_inode_operations;
8462 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8463 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8464 inode_set_bytes(inode, name_len);
f06becc4 8465 btrfs_i_size_write(inode, name_len);
54aa1f4d
CM		 8466 err = btrfs_update_inode(trans, root, inode);
8467 if (err)
8468 drop_inode = 1;
39279cc3
CM		 8469
8470out_unlock:
08c422c2
AV		 8471 if (!err)
8472 d_instantiate(dentry, inode);
7ad85bb7 8473 btrfs_end_transaction(trans, root);
39279cc3
CM		 8474 if (drop_inode) {
8475 inode_dec_link_count(inode);
8476 iput(inode);
8477 }
b53d3f5d 8478 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8479 return err;
8480}
16432985 8481
0af3d00b
JB		 8482static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8483 u64 start, u64 num_bytes, u64 min_size,
8484 loff_t actual_len, u64 *alloc_hint,
8485 struct btrfs_trans_handle *trans)
d899e052 8486{
5dc562c5
JB		 8487 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8488 struct extent_map *em;
d899e052
YZ		 8489 struct btrfs_root *root = BTRFS_I(inode)->root;
8490 struct btrfs_key ins;
d899e052 8491 u64 cur_offset = start;
55a61d1d 8492 u64 i_size;
154ea289 8493 u64 cur_bytes;
d899e052 8494 int ret = 0;
0af3d00b 8495 bool own_trans = true;
d899e052 8496
0af3d00b
JB		 8497 if (trans)
8498 own_trans = false;
d899e052 8499 while (num_bytes > 0) {
0af3d00b
JB		 8500 if (own_trans) {
8501 trans = btrfs_start_transaction(root, 3);
8502 if (IS_ERR(trans)) {
8503 ret = PTR_ERR(trans);
8504 break;
8505 }
5a303d5d
YZ		 8506 }
8507
154ea289
CM		 8508 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8509 cur_bytes = max(cur_bytes, min_size);
00361589
JB		 8510 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8511 *alloc_hint, &ins, 1);
5a303d5d 8512 if (ret) {
0af3d00b
JB		 8513 if (own_trans)
8514 btrfs_end_transaction(trans, root);
a22285a6 8515 break;
d899e052 8516 }
5a303d5d 8517
d899e052
YZ		 8518 ret = insert_reserved_file_extent(trans, inode,
8519 cur_offset, ins.objectid,
8520 ins.offset, ins.offset,
920bbbfb 8521 ins.offset, 0, 0, 0,
d899e052 8522 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 8523 if (ret) {
857cc2fc
JB		 8524 btrfs_free_reserved_extent(root, ins.objectid,
8525 ins.offset);
79787eaa
JM		 8526 btrfs_abort_transaction(trans, root, ret);
8527 if (own_trans)
8528 btrfs_end_transaction(trans, root);
8529 break;
8530 }
a1ed835e
CM		 8531 btrfs_drop_extent_cache(inode, cur_offset,
8532 cur_offset + ins.offset -1, 0);
5a303d5d 8533
5dc562c5
JB		 8534 em = alloc_extent_map();
8535 if (!em) {
8536 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8537 &BTRFS_I(inode)->runtime_flags);
8538 goto next;
8539 }
8540
8541 em->start = cur_offset;
8542 em->orig_start = cur_offset;
8543 em->len = ins.offset;
8544 em->block_start = ins.objectid;
8545 em->block_len = ins.offset;
b4939680 8546 em->orig_block_len = ins.offset;
cc95bef6 8547 em->ram_bytes = ins.offset;
5dc562c5
JB		 8548 em->bdev = root->fs_info->fs_devices->latest_bdev;
8549 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8550 em->generation = trans->transid;
8551
8552 while (1) {
8553 write_lock(&em_tree->lock);
09a2a8f9 8554 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8555 write_unlock(&em_tree->lock);
8556 if (ret != -EEXIST)
8557 break;
8558 btrfs_drop_extent_cache(inode, cur_offset,
8559 cur_offset + ins.offset - 1,
8560 0);
8561 }
8562 free_extent_map(em);
8563next:
d899e052
YZ		 8564 num_bytes -= ins.offset;
8565 cur_offset += ins.offset;
efa56464 8566 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8567
0c4d2d95 8568 inode_inc_iversion(inode);
d899e052 8569 inode->i_ctime = CURRENT_TIME;
6cbff00f 8570 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8571 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8572 (actual_len > inode->i_size) &&
8573 (cur_offset > inode->i_size)) {
d1ea6a61 8574 if (cur_offset > actual_len)
55a61d1d 8575 i_size = actual_len;
d1ea6a61 8576 else
55a61d1d
JB		 8577 i_size = cur_offset;
8578 i_size_write(inode, i_size);
8579 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8580 }
8581
d899e052 8582 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8583
8584 if (ret) {
8585 btrfs_abort_transaction(trans, root, ret);
8586 if (own_trans)
8587 btrfs_end_transaction(trans, root);
8588 break;
8589 }
d899e052 8590
0af3d00b
JB		 8591 if (own_trans)
8592 btrfs_end_transaction(trans, root);
5a303d5d 8593 }
d899e052
YZ		 8594 return ret;
8595}
8596
0af3d00b
JB		 8597int btrfs_prealloc_file_range(struct inode *inode, int mode,
8598 u64 start, u64 num_bytes, u64 min_size,
8599 loff_t actual_len, u64 *alloc_hint)
8600{
8601 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8602 min_size, actual_len, alloc_hint,
8603 NULL);
8604}
8605
8606int btrfs_prealloc_file_range_trans(struct inode *inode,
8607 struct btrfs_trans_handle *trans, int mode,
8608 u64 start, u64 num_bytes, u64 min_size,
8609 loff_t actual_len, u64 *alloc_hint)
8610{
8611 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8612 min_size, actual_len, alloc_hint, trans);
8613}
8614
e6dcd2dc
CM		 8615static int btrfs_set_page_dirty(struct page *page)
8616{
e6dcd2dc
CM		 8617 return __set_page_dirty_nobuffers(page);
8618}
8619
10556cb2 8620static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8621{
b83cc969 8622 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8623 umode_t mode = inode->i_mode;
b83cc969 8624
cb6db4e5
JM		 8625 if (mask & MAY_WRITE &&
8626 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8627 if (btrfs_root_readonly(root))
8628 return -EROFS;
8629 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8630 return -EACCES;
8631 }
2830ba7f 8632 return generic_permission(inode, mask);
fdebe2bd 8633}
39279cc3 8634
6e1d5dcc 8635static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8636 .getattr = btrfs_getattr,
39279cc3
CM		 8637 .lookup = btrfs_lookup,
8638 .create = btrfs_create,
8639 .unlink = btrfs_unlink,
8640 .link = btrfs_link,
8641 .mkdir = btrfs_mkdir,
8642 .rmdir = btrfs_rmdir,
8643 .rename = btrfs_rename,
8644 .symlink = btrfs_symlink,
8645 .setattr = btrfs_setattr,
618e21d5 8646 .mknod = btrfs_mknod,
95819c05
CH		 8647 .setxattr = btrfs_setxattr,
8648 .getxattr = btrfs_getxattr,
5103e947 8649 .listxattr = btrfs_listxattr,
95819c05 8650 .removexattr = btrfs_removexattr,
fdebe2bd 8651 .permission = btrfs_permission,
4e34e719 8652 .get_acl = btrfs_get_acl,
93fd63c2 8653 .update_time = btrfs_update_time,
39279cc3 8654};
6e1d5dcc 8655static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8656 .lookup = btrfs_lookup,
fdebe2bd 8657 .permission = btrfs_permission,
4e34e719 8658 .get_acl = btrfs_get_acl,
93fd63c2 8659 .update_time = btrfs_update_time,
39279cc3 8660};
76dda93c 8661
828c0950 8662static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8663 .llseek = generic_file_llseek,
8664 .read = generic_read_dir,
9cdda8d3 8665 .iterate = btrfs_real_readdir,
34287aa3 8666 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8667#ifdef CONFIG_COMPAT
34287aa3 8668 .compat_ioctl = btrfs_ioctl,
39279cc3 8669#endif
6bf13c0c 8670 .release = btrfs_release_file,
e02119d5 8671 .fsync = btrfs_sync_file,
39279cc3
CM		 8672};
8673
d1310b2e 8674static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8675 .fill_delalloc = run_delalloc_range,
065631f6 8676 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8677 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8678 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8679 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8680 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8681 .set_bit_hook = btrfs_set_bit_hook,
8682 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8683 .merge_extent_hook = btrfs_merge_extent_hook,
8684 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8685};
8686
35054394
CM		 8687/*
8688 * btrfs doesn't support the bmap operation because swapfiles
8689 * use bmap to make a mapping of extents in the file. They assume
8690 * these extents won't change over the life of the file and they
8691 * use the bmap result to do IO directly to the drive.
8692 *
8693 * the btrfs bmap call would return logical addresses that aren't
8694 * suitable for IO and they also will change frequently as COW
8695 * operations happen. So, swapfile + btrfs == corruption.
8696 *
8697 * For now we're avoiding this by dropping bmap.
8698 */
7f09410b 8699static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8700 .readpage = btrfs_readpage,
8701 .writepage = btrfs_writepage,
b293f02e 8702 .writepages = btrfs_writepages,
3ab2fb5a 8703 .readpages = btrfs_readpages,
16432985 8704 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8705 .invalidatepage = btrfs_invalidatepage,
8706 .releasepage = btrfs_releasepage,
e6dcd2dc 8707 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8708 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8709};
8710
7f09410b 8711static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8712 .readpage = btrfs_readpage,
8713 .writepage = btrfs_writepage,
2bf5a725
CM		 8714 .invalidatepage = btrfs_invalidatepage,
8715 .releasepage = btrfs_releasepage,
39279cc3
CM		 8716};
8717
6e1d5dcc 8718static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8719 .getattr = btrfs_getattr,
8720 .setattr = btrfs_setattr,
95819c05
CH		 8721 .setxattr = btrfs_setxattr,
8722 .getxattr = btrfs_getxattr,
5103e947 8723 .listxattr = btrfs_listxattr,
95819c05 8724 .removexattr = btrfs_removexattr,
fdebe2bd 8725 .permission = btrfs_permission,
1506fcc8 8726 .fiemap = btrfs_fiemap,
4e34e719 8727 .get_acl = btrfs_get_acl,
e41f941a 8728 .update_time = btrfs_update_time,
39279cc3 8729};
6e1d5dcc 8730static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8731 .getattr = btrfs_getattr,
8732 .setattr = btrfs_setattr,
fdebe2bd 8733 .permission = btrfs_permission,
95819c05
CH		 8734 .setxattr = btrfs_setxattr,
8735 .getxattr = btrfs_getxattr,
33268eaf 8736 .listxattr = btrfs_listxattr,
95819c05 8737 .removexattr = btrfs_removexattr,
4e34e719 8738 .get_acl = btrfs_get_acl,
e41f941a 8739 .update_time = btrfs_update_time,
618e21d5 8740};
6e1d5dcc 8741static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8742 .readlink = generic_readlink,
8743 .follow_link = page_follow_link_light,
8744 .put_link = page_put_link,
f209561a 8745 .getattr = btrfs_getattr,
22c44fe6 8746 .setattr = btrfs_setattr,
fdebe2bd 8747 .permission = btrfs_permission,
0279b4cd
JO		 8748 .setxattr = btrfs_setxattr,
8749 .getxattr = btrfs_getxattr,
8750 .listxattr = btrfs_listxattr,
8751 .removexattr = btrfs_removexattr,
4e34e719 8752 .get_acl = btrfs_get_acl,
e41f941a 8753 .update_time = btrfs_update_time,
39279cc3 8754};
76dda93c 8755
82d339d9 8756const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8757 .d_delete = btrfs_dentry_delete,
b4aff1f8 8758 .d_release = btrfs_dentry_release,
76dda93c 8759};
Linux 6.x block layer and io_uring tree(s)