projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: move btrfs_truncate_page to btrfs_cont_expand instead of btrfs_truncate
[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>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
53b381b3 43#include <linux/blkdev.h>
4b4e25f2 44#include "compat.h"
39279cc3
CM		 45#include "ctree.h"
46#include "disk-io.h"
47#include "transaction.h"
48#include "btrfs_inode.h"
39279cc3 49#include "print-tree.h"
e6dcd2dc 50#include "ordered-data.h"
95819c05 51#include "xattr.h"
e02119d5 52#include "tree-log.h"
4a54c8c1 53#include "volumes.h"
c8b97818 54#include "compression.h"
b4ce94de 55#include "locking.h"
dc89e982 56#include "free-space-cache.h"
581bb050 57#include "inode-map.h"
38c227d8 58#include "backref.h"
39279cc3
CM		 59
60struct btrfs_iget_args {
61 u64 ino;
62 struct btrfs_root *root;
63};
64
6e1d5dcc
AD		 65static const struct inode_operations btrfs_dir_inode_operations;
66static const struct inode_operations btrfs_symlink_inode_operations;
67static const struct inode_operations btrfs_dir_ro_inode_operations;
68static const struct inode_operations btrfs_special_inode_operations;
69static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 70static const struct address_space_operations btrfs_aops;
71static const struct address_space_operations btrfs_symlink_aops;
828c0950 72static const struct file_operations btrfs_dir_file_operations;
d1310b2e 73static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 74
75static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 76static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 77struct kmem_cache *btrfs_trans_handle_cachep;
78struct kmem_cache *btrfs_transaction_cachep;
39279cc3 79struct kmem_cache *btrfs_path_cachep;
dc89e982 80struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 81
82#define S_SHIFT 12
83static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
84 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
85 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
86 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
87 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
88 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
89 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
90 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91};
92
3972f260 93static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 94static int btrfs_truncate(struct inode *inode);
5fd02043 95static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 96static noinline int cow_file_range(struct inode *inode,
97 struct page *locked_page,
98 u64 start, u64 end, int *page_started,
99 unsigned long *nr_written, int unlock);
70c8a91c
JB		 100static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
101 u64 len, u64 orig_start,
102 u64 block_start, u64 block_len,
cc95bef6
JB		 103 u64 orig_block_len, u64 ram_bytes,
104 int type);
7b128766 105
48a3b636
ES		 106static int btrfs_dirty_inode(struct inode *inode);
107
f34f57a3 108static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 109 struct inode *inode, struct inode *dir,
110 const struct qstr *qstr)
0279b4cd
JO		 111{
112 int err;
113
f34f57a3 114 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 115 if (!err)
2a7dba39 116 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 117 return err;
118}
119
c8b97818
CM		 120/*
121 * this does all the hard work for inserting an inline extent into
122 * the btree. The caller should have done a btrfs_drop_extents so that
123 * no overlapping inline items exist in the btree
124 */
d397712b 125static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 126 struct btrfs_root *root, struct inode *inode,
127 u64 start, size_t size, size_t compressed_size,
fe3f566c 128 int compress_type,
c8b97818
CM		 129 struct page **compressed_pages)
130{
131 struct btrfs_key key;
132 struct btrfs_path *path;
133 struct extent_buffer *leaf;
134 struct page *page = NULL;
135 char *kaddr;
136 unsigned long ptr;
137 struct btrfs_file_extent_item *ei;
138 int err = 0;
139 int ret;
140 size_t cur_size = size;
141 size_t datasize;
142 unsigned long offset;
c8b97818 143
fe3f566c 144 if (compressed_size && compressed_pages)
c8b97818 145 cur_size = compressed_size;
c8b97818 146
d397712b
CM		 147 path = btrfs_alloc_path();
148 if (!path)
c8b97818
CM		 149 return -ENOMEM;
150
b9473439 151 path->leave_spinning = 1;
c8b97818 152
33345d01 153 key.objectid = btrfs_ino(inode);
c8b97818
CM		 154 key.offset = start;
155 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 156 datasize = btrfs_file_extent_calc_inline_size(cur_size);
157
158 inode_add_bytes(inode, size);
159 ret = btrfs_insert_empty_item(trans, root, path, &key,
160 datasize);
c8b97818
CM		 161 if (ret) {
162 err = ret;
c8b97818
CM		 163 goto fail;
164 }
165 leaf = path->nodes[0];
166 ei = btrfs_item_ptr(leaf, path->slots[0],
167 struct btrfs_file_extent_item);
168 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
169 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
170 btrfs_set_file_extent_encryption(leaf, ei, 0);
171 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
172 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
173 ptr = btrfs_file_extent_inline_start(ei);
174
261507a0 175 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 176 struct page *cpage;
177 int i = 0;
d397712b 178 while (compressed_size > 0) {
c8b97818 179 cpage = compressed_pages[i];
5b050f04 180 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 181 PAGE_CACHE_SIZE);
182
7ac687d9 183 kaddr = kmap_atomic(cpage);
c8b97818 184 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 185 kunmap_atomic(kaddr);
c8b97818
CM		 186
187 i++;
188 ptr += cur_size;
189 compressed_size -= cur_size;
190 }
191 btrfs_set_file_extent_compression(leaf, ei,
261507a0 192 compress_type);
c8b97818
CM		 193 } else {
194 page = find_get_page(inode->i_mapping,
195 start >> PAGE_CACHE_SHIFT);
196 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 197 kaddr = kmap_atomic(page);
c8b97818
CM		 198 offset = start & (PAGE_CACHE_SIZE - 1);
199 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 200 kunmap_atomic(kaddr);
c8b97818
CM		 201 page_cache_release(page);
202 }
203 btrfs_mark_buffer_dirty(leaf);
204 btrfs_free_path(path);
205
c2167754
YZ		 206 /*
207 * we're an inline extent, so nobody can
208 * extend the file past i_size without locking
209 * a page we already have locked.
210 *
211 * We must do any isize and inode updates
212 * before we unlock the pages. Otherwise we
213 * could end up racing with unlink.
214 */
c8b97818 215 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 216 ret = btrfs_update_inode(trans, root, inode);
c2167754 217
79787eaa 218 return ret;
c8b97818
CM		 219fail:
220 btrfs_free_path(path);
221 return err;
222}
223
224
225/*
226 * conditionally insert an inline extent into the file. This
227 * does the checks required to make sure the data is small enough
228 * to fit as an inline extent.
229 */
7f366cfe 230static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 231 struct btrfs_root *root,
232 struct inode *inode, u64 start, u64 end,
fe3f566c 233 size_t compressed_size, int compress_type,
c8b97818
CM		 234 struct page **compressed_pages)
235{
236 u64 isize = i_size_read(inode);
237 u64 actual_end = min(end + 1, isize);
238 u64 inline_len = actual_end - start;
fda2832f 239 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 240 u64 data_len = inline_len;
241 int ret;
242
243 if (compressed_size)
244 data_len = compressed_size;
245
246 if (start > 0 ||
70b99e69 247 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 248 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (!compressed_size &&
250 (actual_end & (root->sectorsize - 1)) == 0) ||
251 end + 1 < isize ||
252 data_len > root->fs_info->max_inline) {
253 return 1;
254 }
255
2671485d 256 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM		 257 if (ret)
258 return ret;
c8b97818
CM		 259
260 if (isize > actual_end)
261 inline_len = min_t(u64, isize, actual_end);
262 ret = insert_inline_extent(trans, root, inode, start,
263 inline_len, compressed_size,
fe3f566c 264 compress_type, compressed_pages);
2adcac1a 265 if (ret && ret != -ENOSPC) {
79787eaa
JM		 266 btrfs_abort_transaction(trans, root, ret);
267 return ret;
2adcac1a
JB		 268 } else if (ret == -ENOSPC) {
269 return 1;
79787eaa 270 }
2adcac1a 271
bdc20e67 272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 273 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 274 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 275 return 0;
276}
277
771ed689
CM		 278struct async_extent {
279 u64 start;
280 u64 ram_size;
281 u64 compressed_size;
282 struct page **pages;
283 unsigned long nr_pages;
261507a0 284 int compress_type;
771ed689
CM		 285 struct list_head list;
286};
287
288struct async_cow {
289 struct inode *inode;
290 struct btrfs_root *root;
291 struct page *locked_page;
292 u64 start;
293 u64 end;
294 struct list_head extents;
295 struct btrfs_work work;
296};
297
298static noinline int add_async_extent(struct async_cow *cow,
299 u64 start, u64 ram_size,
300 u64 compressed_size,
301 struct page **pages,
261507a0
LZ		 302 unsigned long nr_pages,
303 int compress_type)
771ed689
CM		 304{
305 struct async_extent *async_extent;
306
307 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 308 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 309 async_extent->start = start;
310 async_extent->ram_size = ram_size;
311 async_extent->compressed_size = compressed_size;
312 async_extent->pages = pages;
313 async_extent->nr_pages = nr_pages;
261507a0 314 async_extent->compress_type = compress_type;
771ed689
CM		 315 list_add_tail(&async_extent->list, &cow->extents);
316 return 0;
317}
318
d352ac68 319/*
771ed689
CM		 320 * we create compressed extents in two phases. The first
321 * phase compresses a range of pages that have already been
322 * locked (both pages and state bits are locked).
c8b97818 323 *
771ed689
CM		 324 * This is done inside an ordered work queue, and the compression
325 * is spread across many cpus. The actual IO submission is step
326 * two, and the ordered work queue takes care of making sure that
327 * happens in the same order things were put onto the queue by
328 * writepages and friends.
c8b97818 329 *
771ed689
CM		 330 * If this code finds it can't get good compression, it puts an
331 * entry onto the work queue to write the uncompressed bytes. This
332 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 333 * are written in the same order that the flusher thread sent them
334 * down.
d352ac68 335 */
771ed689
CM		 336static noinline int compress_file_range(struct inode *inode,
337 struct page *locked_page,
338 u64 start, u64 end,
339 struct async_cow *async_cow,
340 int *num_added)
b888db2b
CM		 341{
342 struct btrfs_root *root = BTRFS_I(inode)->root;
343 struct btrfs_trans_handle *trans;
db94535d 344 u64 num_bytes;
db94535d 345 u64 blocksize = root->sectorsize;
c8b97818 346 u64 actual_end;
42dc7bab 347 u64 isize = i_size_read(inode);
e6dcd2dc 348 int ret = 0;
c8b97818
CM		 349 struct page **pages = NULL;
350 unsigned long nr_pages;
351 unsigned long nr_pages_ret = 0;
352 unsigned long total_compressed = 0;
353 unsigned long total_in = 0;
354 unsigned long max_compressed = 128 * 1024;
771ed689 355 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 356 int i;
357 int will_compress;
261507a0 358 int compress_type = root->fs_info->compress_type;
4adaa611 359 int redirty = 0;
b888db2b 360
4cb13e5d
LB		 361 /* if this is a small write inside eof, kick off a defrag */
362 if ((end - start + 1) < 16 * 1024 &&
363 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 364 btrfs_add_inode_defrag(NULL, inode);
365
42dc7bab 366 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 367again:
368 will_compress = 0;
369 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
370 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 371
f03d9301
CM		 372 /*
373 * we don't want to send crud past the end of i_size through
374 * compression, that's just a waste of CPU time. So, if the
375 * end of the file is before the start of our current
376 * requested range of bytes, we bail out to the uncompressed
377 * cleanup code that can deal with all of this.
378 *
379 * It isn't really the fastest way to fix things, but this is a
380 * very uncommon corner.
381 */
382 if (actual_end <= start)
383 goto cleanup_and_bail_uncompressed;
384
c8b97818
CM		 385 total_compressed = actual_end - start;
386
387 /* we want to make sure that amount of ram required to uncompress
388 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 389 * of a compressed extent to 128k. This is a crucial number
390 * because it also controls how easily we can spread reads across
391 * cpus for decompression.
392 *
393 * We also want to make sure the amount of IO required to do
394 * a random read is reasonably small, so we limit the size of
395 * a compressed extent to 128k.
c8b97818
CM		 396 */
397 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 398 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 399 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 400 total_in = 0;
401 ret = 0;
db94535d 402
771ed689
CM		 403 /*
404 * we do compression for mount -o compress and when the
405 * inode has not been flagged as nocompress. This flag can
406 * change at any time if we discover bad compression ratios.
c8b97818 407 */
6cbff00f 408 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 409 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 410 (BTRFS_I(inode)->force_compress) ||
411 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 412 WARN_ON(pages);
cfbc246e 413 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 414 if (!pages) {
415 /* just bail out to the uncompressed code */
416 goto cont;
417 }
c8b97818 418
261507a0
LZ		 419 if (BTRFS_I(inode)->force_compress)
420 compress_type = BTRFS_I(inode)->force_compress;
421
4adaa611
CM		 422 /*
423 * we need to call clear_page_dirty_for_io on each
424 * page in the range. Otherwise applications with the file
425 * mmap'd can wander in and change the page contents while
426 * we are compressing them.
427 *
428 * If the compression fails for any reason, we set the pages
429 * dirty again later on.
430 */
431 extent_range_clear_dirty_for_io(inode, start, end);
432 redirty = 1;
261507a0
LZ		 433 ret = btrfs_compress_pages(compress_type,
434 inode->i_mapping, start,
435 total_compressed, pages,
436 nr_pages, &nr_pages_ret,
437 &total_in,
438 &total_compressed,
439 max_compressed);
c8b97818
CM		 440
441 if (!ret) {
442 unsigned long offset = total_compressed &
443 (PAGE_CACHE_SIZE - 1);
444 struct page *page = pages[nr_pages_ret - 1];
445 char *kaddr;
446
447 /* zero the tail end of the last page, we might be
448 * sending it down to disk
449 */
450 if (offset) {
7ac687d9 451 kaddr = kmap_atomic(page);
c8b97818
CM		 452 memset(kaddr + offset, 0,
453 PAGE_CACHE_SIZE - offset);
7ac687d9 454 kunmap_atomic(kaddr);
c8b97818
CM		 455 }
456 will_compress = 1;
457 }
458 }
560f7d75 459cont:
c8b97818 460 if (start == 0) {
7a7eaa40 461 trans = btrfs_join_transaction(root);
79787eaa
JM		 462 if (IS_ERR(trans)) {
463 ret = PTR_ERR(trans);
464 trans = NULL;
465 goto cleanup_and_out;
466 }
0ca1f7ce 467 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 468
c8b97818 469 /* lets try to make an inline extent */
771ed689 470 if (ret || total_in < (actual_end - start)) {
c8b97818 471 /* we didn't compress the entire range, try
771ed689 472 * to make an uncompressed inline extent.
c8b97818
CM		 473 */
474 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 475 start, end, 0, 0, NULL);
c8b97818 476 } else {
771ed689 477 /* try making a compressed inline extent */
c8b97818
CM		 478 ret = cow_file_range_inline(trans, root, inode,
479 start, end,
fe3f566c
LZ		 480 total_compressed,
481 compress_type, pages);
c8b97818 482 }
79787eaa 483 if (ret <= 0) {
771ed689 484 /*
79787eaa
JM		 485 * inline extent creation worked or returned error,
486 * we don't need to create any more async work items.
487 * Unlock and free up our temp pages.
771ed689 488 */
c8b97818 489 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 490 &BTRFS_I(inode)->io_tree,
491 start, end, NULL,
492 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 493 EXTENT_CLEAR_DELALLOC |
a791e35e 494 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 495
496 btrfs_end_transaction(trans, root);
c8b97818
CM		 497 goto free_pages_out;
498 }
c2167754 499 btrfs_end_transaction(trans, root);
c8b97818
CM		 500 }
501
502 if (will_compress) {
503 /*
504 * we aren't doing an inline extent round the compressed size
505 * up to a block size boundary so the allocator does sane
506 * things
507 */
fda2832f 508 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 509
510 /*
511 * one last check to make sure the compression is really a
512 * win, compare the page count read with the blocks on disk
513 */
fda2832f 514 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 515 if (total_compressed >= total_in) {
516 will_compress = 0;
517 } else {
c8b97818
CM		 518 num_bytes = total_in;
519 }
520 }
521 if (!will_compress && pages) {
522 /*
523 * the compression code ran but failed to make things smaller,
524 * free any pages it allocated and our page pointer array
525 */
526 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 527 WARN_ON(pages[i]->mapping);
c8b97818
CM		 528 page_cache_release(pages[i]);
529 }
530 kfree(pages);
531 pages = NULL;
532 total_compressed = 0;
533 nr_pages_ret = 0;
534
535 /* flag the file so we don't compress in the future */
1e701a32
CM		 536 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
537 !(BTRFS_I(inode)->force_compress)) {
a555f810 538 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 539 }
c8b97818 540 }
771ed689
CM		 541 if (will_compress) {
542 *num_added += 1;
c8b97818 543
771ed689
CM		 544 /* the async work queues will take care of doing actual
545 * allocation on disk for these compressed pages,
546 * and will submit them to the elevator.
547 */
548 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 549 total_compressed, pages, nr_pages_ret,
550 compress_type);
179e29e4 551
24ae6365 552 if (start + num_bytes < end) {
771ed689
CM		 553 start += num_bytes;
554 pages = NULL;
555 cond_resched();
556 goto again;
557 }
558 } else {
f03d9301 559cleanup_and_bail_uncompressed:
771ed689
CM		 560 /*
561 * No compression, but we still need to write the pages in
562 * the file we've been given so far. redirty the locked
563 * page if it corresponds to our extent and set things up
564 * for the async work queue to run cow_file_range to do
565 * the normal delalloc dance
566 */
567 if (page_offset(locked_page) >= start &&
568 page_offset(locked_page) <= end) {
569 __set_page_dirty_nobuffers(locked_page);
570 /* unlocked later on in the async handlers */
571 }
4adaa611
CM		 572 if (redirty)
573 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 574 add_async_extent(async_cow, start, end - start + 1,
575 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 576 *num_added += 1;
577 }
3b951516 578
771ed689 579out:
79787eaa 580 return ret;
771ed689
CM		 581
582free_pages_out:
583 for (i = 0; i < nr_pages_ret; i++) {
584 WARN_ON(pages[i]->mapping);
585 page_cache_release(pages[i]);
586 }
d397712b 587 kfree(pages);
771ed689
CM		 588
589 goto out;
79787eaa
JM		 590
591cleanup_and_out:
592 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
593 start, end, NULL,
594 EXTENT_CLEAR_UNLOCK_PAGE |
595 EXTENT_CLEAR_DIRTY |
596 EXTENT_CLEAR_DELALLOC |
597 EXTENT_SET_WRITEBACK |
598 EXTENT_END_WRITEBACK);
599 if (!trans || IS_ERR(trans))
600 btrfs_error(root->fs_info, ret, "Failed to join transaction");
601 else
602 btrfs_abort_transaction(trans, root, ret);
603 goto free_pages_out;
771ed689
CM		 604}
605
606/*
607 * phase two of compressed writeback. This is the ordered portion
608 * of the code, which only gets called in the order the work was
609 * queued. We walk all the async extents created by compress_file_range
610 * and send them down to the disk.
611 */
612static noinline int submit_compressed_extents(struct inode *inode,
613 struct async_cow *async_cow)
614{
615 struct async_extent *async_extent;
616 u64 alloc_hint = 0;
617 struct btrfs_trans_handle *trans;
618 struct btrfs_key ins;
619 struct extent_map *em;
620 struct btrfs_root *root = BTRFS_I(inode)->root;
621 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
622 struct extent_io_tree *io_tree;
f5a84ee3 623 int ret = 0;
771ed689
CM		 624
625 if (list_empty(&async_cow->extents))
626 return 0;
627
3e04e7f1 628again:
d397712b 629 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 630 async_extent = list_entry(async_cow->extents.next,
631 struct async_extent, list);
632 list_del(&async_extent->list);
c8b97818 633
771ed689
CM		 634 io_tree = &BTRFS_I(inode)->io_tree;
635
f5a84ee3 636retry:
771ed689
CM		 637 /* did the compression code fall back to uncompressed IO? */
638 if (!async_extent->pages) {
639 int page_started = 0;
640 unsigned long nr_written = 0;
641
642 lock_extent(io_tree, async_extent->start,
2ac55d41 643 async_extent->start +
d0082371 644 async_extent->ram_size - 1);
771ed689
CM		 645
646 /* allocate blocks */
f5a84ee3
JB		 647 ret = cow_file_range(inode, async_cow->locked_page,
648 async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 &page_started, &nr_written, 0);
771ed689 652
79787eaa
JM		 653 /* JDM XXX */
654
771ed689
CM		 655 /*
656 * if page_started, cow_file_range inserted an
657 * inline extent and took care of all the unlocking
658 * and IO for us. Otherwise, we need to submit
659 * all those pages down to the drive.
660 */
f5a84ee3 661 if (!page_started && !ret)
771ed689
CM		 662 extent_write_locked_range(io_tree,
663 inode, async_extent->start,
d397712b 664 async_extent->start +
771ed689
CM		 665 async_extent->ram_size - 1,
666 btrfs_get_extent,
667 WB_SYNC_ALL);
3e04e7f1
JB		 668 else if (ret)
669 unlock_page(async_cow->locked_page);
771ed689
CM		 670 kfree(async_extent);
671 cond_resched();
672 continue;
673 }
674
675 lock_extent(io_tree, async_extent->start,
d0082371 676 async_extent->start + async_extent->ram_size - 1);
771ed689 677
7a7eaa40 678 trans = btrfs_join_transaction(root);
79787eaa
JM		 679 if (IS_ERR(trans)) {
680 ret = PTR_ERR(trans);
681 } else {
682 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
683 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 684 async_extent->compressed_size,
685 async_extent->compressed_size,
81c9ad23 686 0, alloc_hint, &ins, 1);
962197ba 687 if (ret && ret != -ENOSPC)
79787eaa
JM		 688 btrfs_abort_transaction(trans, root, ret);
689 btrfs_end_transaction(trans, root);
690 }
c2167754 691
f5a84ee3
JB		 692 if (ret) {
693 int i;
3e04e7f1 694
f5a84ee3
JB		 695 for (i = 0; i < async_extent->nr_pages; i++) {
696 WARN_ON(async_extent->pages[i]->mapping);
697 page_cache_release(async_extent->pages[i]);
698 }
699 kfree(async_extent->pages);
700 async_extent->nr_pages = 0;
701 async_extent->pages = NULL;
3e04e7f1 702
fdf8e2ea
JB		 703 if (ret == -ENOSPC) {
704 unlock_extent(io_tree, async_extent->start,
705 async_extent->start +
706 async_extent->ram_size - 1);
79787eaa 707 goto retry;
fdf8e2ea 708 }
3e04e7f1 709 goto out_free;
f5a84ee3
JB		 710 }
711
c2167754
YZ		 712 /*
713 * here we're doing allocation and writeback of the
714 * compressed pages
715 */
716 btrfs_drop_extent_cache(inode, async_extent->start,
717 async_extent->start +
718 async_extent->ram_size - 1, 0);
719
172ddd60 720 em = alloc_extent_map();
b9aa55be
LB		 721 if (!em) {
722 ret = -ENOMEM;
3e04e7f1 723 goto out_free_reserve;
b9aa55be 724 }
771ed689
CM		 725 em->start = async_extent->start;
726 em->len = async_extent->ram_size;
445a6944 727 em->orig_start = em->start;
2ab28f32
JB		 728 em->mod_start = em->start;
729 em->mod_len = em->len;
c8b97818 730
771ed689
CM		 731 em->block_start = ins.objectid;
732 em->block_len = ins.offset;
b4939680 733 em->orig_block_len = ins.offset;
cc95bef6 734 em->ram_bytes = async_extent->ram_size;
771ed689 735 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 736 em->compress_type = async_extent->compress_type;
771ed689
CM		 737 set_bit(EXTENT_FLAG_PINNED, &em->flags);
738 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 739 em->generation = -1;
771ed689 740
d397712b 741 while (1) {
890871be 742 write_lock(&em_tree->lock);
09a2a8f9 743 ret = add_extent_mapping(em_tree, em, 1);
890871be 744 write_unlock(&em_tree->lock);
771ed689
CM		 745 if (ret != -EEXIST) {
746 free_extent_map(em);
747 break;
748 }
749 btrfs_drop_extent_cache(inode, async_extent->start,
750 async_extent->start +
751 async_extent->ram_size - 1, 0);
752 }
753
3e04e7f1
JB		 754 if (ret)
755 goto out_free_reserve;
756
261507a0
LZ		 757 ret = btrfs_add_ordered_extent_compress(inode,
758 async_extent->start,
759 ins.objectid,
760 async_extent->ram_size,
761 ins.offset,
762 BTRFS_ORDERED_COMPRESSED,
763 async_extent->compress_type);
3e04e7f1
JB		 764 if (ret)
765 goto out_free_reserve;
771ed689 766
771ed689
CM		 767 /*
768 * clear dirty, set writeback and unlock the pages.
769 */
770 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 771 &BTRFS_I(inode)->io_tree,
772 async_extent->start,
773 async_extent->start +
774 async_extent->ram_size - 1,
775 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
776 EXTENT_CLEAR_UNLOCK |
a3429ab7 777 EXTENT_CLEAR_DELALLOC |
a791e35e 778 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 779
780 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 781 async_extent->start,
782 async_extent->ram_size,
783 ins.objectid,
784 ins.offset, async_extent->pages,
785 async_extent->nr_pages);
771ed689
CM		 786 alloc_hint = ins.objectid + ins.offset;
787 kfree(async_extent);
3e04e7f1
JB		 788 if (ret)
789 goto out;
771ed689
CM		 790 cond_resched();
791 }
79787eaa
JM		 792 ret = 0;
793out:
794 return ret;
3e04e7f1
JB		 795out_free_reserve:
796 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 797out_free:
3e04e7f1
JB		 798 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
799 async_extent->start,
800 async_extent->start +
801 async_extent->ram_size - 1,
802 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
803 EXTENT_CLEAR_UNLOCK |
804 EXTENT_CLEAR_DELALLOC |
805 EXTENT_CLEAR_DIRTY |
806 EXTENT_SET_WRITEBACK |
807 EXTENT_END_WRITEBACK);
79787eaa 808 kfree(async_extent);
3e04e7f1 809 goto again;
771ed689
CM		 810}
811
4b46fce2
JB		 812static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
813 u64 num_bytes)
814{
815 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
816 struct extent_map *em;
817 u64 alloc_hint = 0;
818
819 read_lock(&em_tree->lock);
820 em = search_extent_mapping(em_tree, start, num_bytes);
821 if (em) {
822 /*
823 * if block start isn't an actual block number then find the
824 * first block in this inode and use that as a hint. If that
825 * block is also bogus then just don't worry about it.
826 */
827 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
828 free_extent_map(em);
829 em = search_extent_mapping(em_tree, 0, 0);
830 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
831 alloc_hint = em->block_start;
832 if (em)
833 free_extent_map(em);
834 } else {
835 alloc_hint = em->block_start;
836 free_extent_map(em);
837 }
838 }
839 read_unlock(&em_tree->lock);
840
841 return alloc_hint;
842}
843
771ed689
CM		 844/*
845 * when extent_io.c finds a delayed allocation range in the file,
846 * the call backs end up in this code. The basic idea is to
847 * allocate extents on disk for the range, and create ordered data structs
848 * in ram to track those extents.
849 *
850 * locked_page is the page that writepage had locked already. We use
851 * it to make sure we don't do extra locks or unlocks.
852 *
853 * *page_started is set to one if we unlock locked_page and do everything
854 * required to start IO on it. It may be clean and already done with
855 * IO when we return.
856 */
b7d5b0a8
MX		 857static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
858 struct inode *inode,
859 struct btrfs_root *root,
860 struct page *locked_page,
861 u64 start, u64 end, int *page_started,
862 unsigned long *nr_written,
863 int unlock)
771ed689 864{
771ed689
CM		 865 u64 alloc_hint = 0;
866 u64 num_bytes;
867 unsigned long ram_size;
868 u64 disk_num_bytes;
869 u64 cur_alloc_size;
870 u64 blocksize = root->sectorsize;
771ed689
CM		 871 struct btrfs_key ins;
872 struct extent_map *em;
873 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
874 int ret = 0;
875
83eea1f1 876 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 877
fda2832f 878 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 879 num_bytes = max(blocksize, num_bytes);
880 disk_num_bytes = num_bytes;
771ed689 881
4cb5300b 882 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 883 if (num_bytes < 64 * 1024 &&
884 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 885 btrfs_add_inode_defrag(trans, inode);
886
771ed689
CM		 887 if (start == 0) {
888 /* lets try to make an inline extent */
889 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 890 start, end, 0, 0, NULL);
771ed689
CM		 891 if (ret == 0) {
892 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 893 &BTRFS_I(inode)->io_tree,
894 start, end, NULL,
895 EXTENT_CLEAR_UNLOCK_PAGE |
896 EXTENT_CLEAR_UNLOCK |
897 EXTENT_CLEAR_DELALLOC |
898 EXTENT_CLEAR_DIRTY |
899 EXTENT_SET_WRITEBACK |
900 EXTENT_END_WRITEBACK);
c2167754 901
771ed689
CM		 902 *nr_written = *nr_written +
903 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
904 *page_started = 1;
771ed689 905 goto out;
79787eaa
JM		 906 } else if (ret < 0) {
907 btrfs_abort_transaction(trans, root, ret);
908 goto out_unlock;
771ed689
CM		 909 }
910 }
911
912 BUG_ON(disk_num_bytes >
6c41761f 913 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 914
4b46fce2 915 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 916 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
917
d397712b 918 while (disk_num_bytes > 0) {
a791e35e
CM		 919 unsigned long op;
920
287a0ab9 921 cur_alloc_size = disk_num_bytes;
e6dcd2dc 922 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 923 root->sectorsize, 0, alloc_hint,
81c9ad23 924 &ins, 1);
79787eaa
JM		 925 if (ret < 0) {
926 btrfs_abort_transaction(trans, root, ret);
927 goto out_unlock;
928 }
d397712b 929
172ddd60 930 em = alloc_extent_map();
b9aa55be
LB		 931 if (!em) {
932 ret = -ENOMEM;
ace68bac 933 goto out_reserve;
b9aa55be 934 }
e6dcd2dc 935 em->start = start;
445a6944 936 em->orig_start = em->start;
771ed689
CM		 937 ram_size = ins.offset;
938 em->len = ins.offset;
2ab28f32
JB		 939 em->mod_start = em->start;
940 em->mod_len = em->len;
c8b97818 941
e6dcd2dc 942 em->block_start = ins.objectid;
c8b97818 943 em->block_len = ins.offset;
b4939680 944 em->orig_block_len = ins.offset;
cc95bef6 945 em->ram_bytes = ram_size;
e6dcd2dc 946 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 947 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 948 em->generation = -1;
c8b97818 949
d397712b 950 while (1) {
890871be 951 write_lock(&em_tree->lock);
09a2a8f9 952 ret = add_extent_mapping(em_tree, em, 1);
890871be 953 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 954 if (ret != -EEXIST) {
955 free_extent_map(em);
956 break;
957 }
958 btrfs_drop_extent_cache(inode, start,
c8b97818 959 start + ram_size - 1, 0);
e6dcd2dc 960 }
ace68bac
LB		 961 if (ret)
962 goto out_reserve;
e6dcd2dc 963
98d20f67 964 cur_alloc_size = ins.offset;
e6dcd2dc 965 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 966 ram_size, cur_alloc_size, 0);
ace68bac
LB		 967 if (ret)
968 goto out_reserve;
c8b97818 969
17d217fe
YZ		 970 if (root->root_key.objectid ==
971 BTRFS_DATA_RELOC_TREE_OBJECTID) {
972 ret = btrfs_reloc_clone_csums(inode, start,
973 cur_alloc_size);
79787eaa
JM		 974 if (ret) {
975 btrfs_abort_transaction(trans, root, ret);
ace68bac 976 goto out_reserve;
79787eaa 977 }
17d217fe
YZ		 978 }
979
d397712b 980 if (disk_num_bytes < cur_alloc_size)
3b951516 981 break;
d397712b 982
c8b97818
CM		 983 /* we're not doing compressed IO, don't unlock the first
984 * page (which the caller expects to stay locked), don't
985 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 986 *
987 * Do set the Private2 bit so we know this page was properly
988 * setup for writepage
c8b97818 989 */
a791e35e
CM		 990 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
991 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
992 EXTENT_SET_PRIVATE2;
993
c8b97818
CM		 994 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
995 start, start + ram_size - 1,
a791e35e 996 locked_page, op);
c8b97818 997 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 998 num_bytes -= cur_alloc_size;
999 alloc_hint = ins.objectid + ins.offset;
1000 start += cur_alloc_size;
b888db2b 1001 }
79787eaa 1002out:
be20aa9d 1003 return ret;
b7d5b0a8 1004
ace68bac
LB		 1005out_reserve:
1006 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa
JM		 1007out_unlock:
1008 extent_clear_unlock_delalloc(inode,
1009 &BTRFS_I(inode)->io_tree,
beb42dd7 1010 start, end, locked_page,
79787eaa
JM		 1011 EXTENT_CLEAR_UNLOCK_PAGE |
1012 EXTENT_CLEAR_UNLOCK |
1013 EXTENT_CLEAR_DELALLOC |
1014 EXTENT_CLEAR_DIRTY |
1015 EXTENT_SET_WRITEBACK |
1016 EXTENT_END_WRITEBACK);
1017
1018 goto out;
771ed689 1019}
c8b97818 1020
b7d5b0a8
MX		 1021static noinline int cow_file_range(struct inode *inode,
1022 struct page *locked_page,
1023 u64 start, u64 end, int *page_started,
1024 unsigned long *nr_written,
1025 int unlock)
1026{
1027 struct btrfs_trans_handle *trans;
1028 struct btrfs_root *root = BTRFS_I(inode)->root;
1029 int ret;
1030
1031 trans = btrfs_join_transaction(root);
1032 if (IS_ERR(trans)) {
1033 extent_clear_unlock_delalloc(inode,
1034 &BTRFS_I(inode)->io_tree,
1035 start, end, locked_page,
1036 EXTENT_CLEAR_UNLOCK_PAGE |
1037 EXTENT_CLEAR_UNLOCK |
1038 EXTENT_CLEAR_DELALLOC |
1039 EXTENT_CLEAR_DIRTY |
1040 EXTENT_SET_WRITEBACK |
1041 EXTENT_END_WRITEBACK);
1042 return PTR_ERR(trans);
1043 }
1044 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1045
1046 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1047 page_started, nr_written, unlock);
1048
1049 btrfs_end_transaction(trans, root);
1050
1051 return ret;
1052}
1053
771ed689
CM		 1054/*
1055 * work queue call back to started compression on a file and pages
1056 */
1057static noinline void async_cow_start(struct btrfs_work *work)
1058{
1059 struct async_cow *async_cow;
1060 int num_added = 0;
1061 async_cow = container_of(work, struct async_cow, work);
1062
1063 compress_file_range(async_cow->inode, async_cow->locked_page,
1064 async_cow->start, async_cow->end, async_cow,
1065 &num_added);
8180ef88 1066 if (num_added == 0) {
cb77fcd8 1067 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1068 async_cow->inode = NULL;
8180ef88 1069 }
771ed689
CM		 1070}
1071
1072/*
1073 * work queue call back to submit previously compressed pages
1074 */
1075static noinline void async_cow_submit(struct btrfs_work *work)
1076{
1077 struct async_cow *async_cow;
1078 struct btrfs_root *root;
1079 unsigned long nr_pages;
1080
1081 async_cow = container_of(work, struct async_cow, work);
1082
1083 root = async_cow->root;
1084 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1085 PAGE_CACHE_SHIFT;
1086
66657b31 1087 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1088 5 * 1024 * 1024 &&
771ed689
CM		 1089 waitqueue_active(&root->fs_info->async_submit_wait))
1090 wake_up(&root->fs_info->async_submit_wait);
1091
d397712b 1092 if (async_cow->inode)
771ed689 1093 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1094}
c8b97818 1095
771ed689
CM		 1096static noinline void async_cow_free(struct btrfs_work *work)
1097{
1098 struct async_cow *async_cow;
1099 async_cow = container_of(work, struct async_cow, work);
8180ef88 1100 if (async_cow->inode)
cb77fcd8 1101 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1102 kfree(async_cow);
1103}
1104
1105static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1106 u64 start, u64 end, int *page_started,
1107 unsigned long *nr_written)
1108{
1109 struct async_cow *async_cow;
1110 struct btrfs_root *root = BTRFS_I(inode)->root;
1111 unsigned long nr_pages;
1112 u64 cur_end;
287082b0 1113 int limit = 10 * 1024 * 1024;
771ed689 1114
a3429ab7
CM		 1115 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1116 1, 0, NULL, GFP_NOFS);
d397712b 1117 while (start < end) {
771ed689 1118 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1119 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1120 async_cow->inode = igrab(inode);
771ed689
CM		 1121 async_cow->root = root;
1122 async_cow->locked_page = locked_page;
1123 async_cow->start = start;
1124
6cbff00f 1125 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1126 cur_end = end;
1127 else
1128 cur_end = min(end, start + 512 * 1024 - 1);
1129
1130 async_cow->end = cur_end;
1131 INIT_LIST_HEAD(&async_cow->extents);
1132
1133 async_cow->work.func = async_cow_start;
1134 async_cow->work.ordered_func = async_cow_submit;
1135 async_cow->work.ordered_free = async_cow_free;
1136 async_cow->work.flags = 0;
1137
771ed689
CM		 1138 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1139 PAGE_CACHE_SHIFT;
1140 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1141
1142 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1143 &async_cow->work);
1144
1145 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1146 wait_event(root->fs_info->async_submit_wait,
1147 (atomic_read(&root->fs_info->async_delalloc_pages) <
1148 limit));
1149 }
1150
d397712b 1151 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1152 atomic_read(&root->fs_info->async_delalloc_pages)) {
1153 wait_event(root->fs_info->async_submit_wait,
1154 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1155 0));
1156 }
1157
1158 *nr_written += nr_pages;
1159 start = cur_end + 1;
1160 }
1161 *page_started = 1;
1162 return 0;
be20aa9d
CM		 1163}
1164
d397712b 1165static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1166 u64 bytenr, u64 num_bytes)
1167{
1168 int ret;
1169 struct btrfs_ordered_sum *sums;
1170 LIST_HEAD(list);
1171
07d400a6 1172 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1173 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1174 if (ret == 0 && list_empty(&list))
1175 return 0;
1176
1177 while (!list_empty(&list)) {
1178 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1179 list_del(&sums->list);
1180 kfree(sums);
1181 }
1182 return 1;
1183}
1184
d352ac68
CM		 1185/*
1186 * when nowcow writeback call back. This checks for snapshots or COW copies
1187 * of the extents that exist in the file, and COWs the file as required.
1188 *
1189 * If no cow copies or snapshots exist, we write directly to the existing
1190 * blocks on disk
1191 */
7f366cfe
CM		 1192static noinline int run_delalloc_nocow(struct inode *inode,
1193 struct page *locked_page,
771ed689
CM		 1194 u64 start, u64 end, int *page_started, int force,
1195 unsigned long *nr_written)
be20aa9d 1196{
be20aa9d 1197 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1198 struct btrfs_trans_handle *trans;
be20aa9d 1199 struct extent_buffer *leaf;
be20aa9d 1200 struct btrfs_path *path;
80ff3856 1201 struct btrfs_file_extent_item *fi;
be20aa9d 1202 struct btrfs_key found_key;
80ff3856
YZ		 1203 u64 cow_start;
1204 u64 cur_offset;
1205 u64 extent_end;
5d4f98a2 1206 u64 extent_offset;
80ff3856
YZ		 1207 u64 disk_bytenr;
1208 u64 num_bytes;
b4939680 1209 u64 disk_num_bytes;
cc95bef6 1210 u64 ram_bytes;
80ff3856 1211 int extent_type;
79787eaa 1212 int ret, err;
d899e052 1213 int type;
80ff3856
YZ		 1214 int nocow;
1215 int check_prev = 1;
82d5902d 1216 bool nolock;
33345d01 1217 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1218
1219 path = btrfs_alloc_path();
17ca04af
JB		 1220 if (!path) {
1221 extent_clear_unlock_delalloc(inode,
1222 &BTRFS_I(inode)->io_tree,
1223 start, end, locked_page,
1224 EXTENT_CLEAR_UNLOCK_PAGE |
1225 EXTENT_CLEAR_UNLOCK |
1226 EXTENT_CLEAR_DELALLOC |
1227 EXTENT_CLEAR_DIRTY |
1228 EXTENT_SET_WRITEBACK |
1229 EXTENT_END_WRITEBACK);
d8926bb3 1230 return -ENOMEM;
17ca04af 1231 }
82d5902d 1232
83eea1f1 1233 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1234
1235 if (nolock)
7a7eaa40 1236 trans = btrfs_join_transaction_nolock(root);
82d5902d 1237 else
7a7eaa40 1238 trans = btrfs_join_transaction(root);
ff5714cc 1239
79787eaa 1240 if (IS_ERR(trans)) {
17ca04af
JB		 1241 extent_clear_unlock_delalloc(inode,
1242 &BTRFS_I(inode)->io_tree,
1243 start, end, locked_page,
1244 EXTENT_CLEAR_UNLOCK_PAGE |
1245 EXTENT_CLEAR_UNLOCK |
1246 EXTENT_CLEAR_DELALLOC |
1247 EXTENT_CLEAR_DIRTY |
1248 EXTENT_SET_WRITEBACK |
1249 EXTENT_END_WRITEBACK);
79787eaa
JM		 1250 btrfs_free_path(path);
1251 return PTR_ERR(trans);
1252 }
1253
74b21075 1254 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1255
80ff3856
YZ		 1256 cow_start = (u64)-1;
1257 cur_offset = start;
1258 while (1) {
33345d01 1259 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1260 cur_offset, 0);
79787eaa
JM		 1261 if (ret < 0) {
1262 btrfs_abort_transaction(trans, root, ret);
1263 goto error;
1264 }
80ff3856
YZ		 1265 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266 leaf = path->nodes[0];
1267 btrfs_item_key_to_cpu(leaf, &found_key,
1268 path->slots[0] - 1);
33345d01 1269 if (found_key.objectid == ino &&
80ff3856
YZ		 1270 found_key.type == BTRFS_EXTENT_DATA_KEY)
1271 path->slots[0]--;
1272 }
1273 check_prev = 0;
1274next_slot:
1275 leaf = path->nodes[0];
1276 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1278 if (ret < 0) {
1279 btrfs_abort_transaction(trans, root, ret);
1280 goto error;
1281 }
80ff3856
YZ		 1282 if (ret > 0)
1283 break;
1284 leaf = path->nodes[0];
1285 }
be20aa9d 1286
80ff3856
YZ		 1287 nocow = 0;
1288 disk_bytenr = 0;
17d217fe 1289 num_bytes = 0;
80ff3856
YZ		 1290 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1291
33345d01 1292 if (found_key.objectid > ino ||
80ff3856
YZ		 1293 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1294 found_key.offset > end)
1295 break;
1296
1297 if (found_key.offset > cur_offset) {
1298 extent_end = found_key.offset;
e9061e21 1299 extent_type = 0;
80ff3856
YZ		 1300 goto out_check;
1301 }
1302
1303 fi = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_file_extent_item);
1305 extent_type = btrfs_file_extent_type(leaf, fi);
1306
cc95bef6 1307 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1308 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1309 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1310 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1311 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1312 extent_end = found_key.offset +
1313 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1314 disk_num_bytes =
1315 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1316 if (extent_end <= start) {
1317 path->slots[0]++;
1318 goto next_slot;
1319 }
17d217fe
YZ		 1320 if (disk_bytenr == 0)
1321 goto out_check;
80ff3856
YZ		 1322 if (btrfs_file_extent_compression(leaf, fi) ||
1323 btrfs_file_extent_encryption(leaf, fi) ||
1324 btrfs_file_extent_other_encoding(leaf, fi))
1325 goto out_check;
d899e052
YZ		 1326 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1327 goto out_check;
d2fb3437 1328 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1329 goto out_check;
33345d01 1330 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1331 found_key.offset -
1332 extent_offset, disk_bytenr))
17d217fe 1333 goto out_check;
5d4f98a2 1334 disk_bytenr += extent_offset;
17d217fe
YZ		 1335 disk_bytenr += cur_offset - found_key.offset;
1336 num_bytes = min(end + 1, extent_end) - cur_offset;
1337 /*
1338 * force cow if csum exists in the range.
1339 * this ensure that csum for a given extent are
1340 * either valid or do not exist.
1341 */
1342 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1343 goto out_check;
80ff3856
YZ		 1344 nocow = 1;
1345 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1346 extent_end = found_key.offset +
1347 btrfs_file_extent_inline_len(leaf, fi);
1348 extent_end = ALIGN(extent_end, root->sectorsize);
1349 } else {
1350 BUG_ON(1);
1351 }
1352out_check:
1353 if (extent_end <= start) {
1354 path->slots[0]++;
1355 goto next_slot;
1356 }
1357 if (!nocow) {
1358 if (cow_start == (u64)-1)
1359 cow_start = cur_offset;
1360 cur_offset = extent_end;
1361 if (cur_offset > end)
1362 break;
1363 path->slots[0]++;
1364 goto next_slot;
7ea394f1
YZ		 1365 }
1366
b3b4aa74 1367 btrfs_release_path(path);
80ff3856 1368 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1369 ret = __cow_file_range(trans, inode, root, locked_page,
1370 cow_start, found_key.offset - 1,
1371 page_started, nr_written, 1);
79787eaa
JM		 1372 if (ret) {
1373 btrfs_abort_transaction(trans, root, ret);
1374 goto error;
1375 }
80ff3856 1376 cow_start = (u64)-1;
7ea394f1 1377 }
80ff3856 1378
d899e052
YZ		 1379 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1380 struct extent_map *em;
1381 struct extent_map_tree *em_tree;
1382 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1383 em = alloc_extent_map();
79787eaa 1384 BUG_ON(!em); /* -ENOMEM */
d899e052 1385 em->start = cur_offset;
70c8a91c 1386 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1387 em->len = num_bytes;
1388 em->block_len = num_bytes;
1389 em->block_start = disk_bytenr;
b4939680 1390 em->orig_block_len = disk_num_bytes;
cc95bef6 1391 em->ram_bytes = ram_bytes;
d899e052 1392 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1393 em->mod_start = em->start;
1394 em->mod_len = em->len;
d899e052 1395 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1396 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1397 em->generation = -1;
d899e052 1398 while (1) {
890871be 1399 write_lock(&em_tree->lock);
09a2a8f9 1400 ret = add_extent_mapping(em_tree, em, 1);
890871be 1401 write_unlock(&em_tree->lock);
d899e052
YZ		 1402 if (ret != -EEXIST) {
1403 free_extent_map(em);
1404 break;
1405 }
1406 btrfs_drop_extent_cache(inode, em->start,
1407 em->start + em->len - 1, 0);
1408 }
1409 type = BTRFS_ORDERED_PREALLOC;
1410 } else {
1411 type = BTRFS_ORDERED_NOCOW;
1412 }
80ff3856
YZ		 1413
1414 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1415 num_bytes, num_bytes, type);
79787eaa 1416 BUG_ON(ret); /* -ENOMEM */
771ed689 1417
efa56464
YZ		 1418 if (root->root_key.objectid ==
1419 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1420 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1421 num_bytes);
79787eaa
JM		 1422 if (ret) {
1423 btrfs_abort_transaction(trans, root, ret);
1424 goto error;
1425 }
efa56464
YZ		 1426 }
1427
d899e052 1428 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1429 cur_offset, cur_offset + num_bytes - 1,
1430 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1431 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1432 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1433 cur_offset = extent_end;
1434 if (cur_offset > end)
1435 break;
be20aa9d 1436 }
b3b4aa74 1437 btrfs_release_path(path);
80ff3856 1438
17ca04af 1439 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1440 cow_start = cur_offset;
17ca04af
JB		 1441 cur_offset = end;
1442 }
1443
80ff3856 1444 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1445 ret = __cow_file_range(trans, inode, root, locked_page,
1446 cow_start, end,
1447 page_started, nr_written, 1);
79787eaa
JM		 1448 if (ret) {
1449 btrfs_abort_transaction(trans, root, ret);
1450 goto error;
1451 }
80ff3856
YZ		 1452 }
1453
79787eaa 1454error:
a698d075 1455 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1456 if (!ret)
1457 ret = err;
1458
17ca04af
JB		 1459 if (ret && cur_offset < end)
1460 extent_clear_unlock_delalloc(inode,
1461 &BTRFS_I(inode)->io_tree,
1462 cur_offset, end, locked_page,
1463 EXTENT_CLEAR_UNLOCK_PAGE |
1464 EXTENT_CLEAR_UNLOCK |
1465 EXTENT_CLEAR_DELALLOC |
1466 EXTENT_CLEAR_DIRTY |
1467 EXTENT_SET_WRITEBACK |
1468 EXTENT_END_WRITEBACK);
1469
7ea394f1 1470 btrfs_free_path(path);
79787eaa 1471 return ret;
be20aa9d
CM		 1472}
1473
d352ac68
CM		 1474/*
1475 * extent_io.c call back to do delayed allocation processing
1476 */
c8b97818 1477static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1478 u64 start, u64 end, int *page_started,
1479 unsigned long *nr_written)
be20aa9d 1480{
be20aa9d 1481 int ret;
7f366cfe 1482 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1483
7ddf5a42 1484 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1485 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1486 page_started, 1, nr_written);
7ddf5a42 1487 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1488 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1489 page_started, 0, nr_written);
7ddf5a42
JB		 1490 } else if (!btrfs_test_opt(root, COMPRESS) &&
1491 !(BTRFS_I(inode)->force_compress) &&
1492 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1493 ret = cow_file_range(inode, locked_page, start, end,
1494 page_started, nr_written, 1);
7ddf5a42
JB		 1495 } else {
1496 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1497 &BTRFS_I(inode)->runtime_flags);
771ed689 1498 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1499 page_started, nr_written);
7ddf5a42 1500 }
b888db2b
CM		 1501 return ret;
1502}
1503
1bf85046
JM		 1504static void btrfs_split_extent_hook(struct inode *inode,
1505 struct extent_state *orig, u64 split)
9ed74f2d 1506{
0ca1f7ce 1507 /* not delalloc, ignore it */
9ed74f2d 1508 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1509 return;
9ed74f2d 1510
9e0baf60
JB		 1511 spin_lock(&BTRFS_I(inode)->lock);
1512 BTRFS_I(inode)->outstanding_extents++;
1513 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1514}
1515
1516/*
1517 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1518 * extents so we can keep track of new extents that are just merged onto old
1519 * extents, such as when we are doing sequential writes, so we can properly
1520 * account for the metadata space we'll need.
1521 */
1bf85046
JM		 1522static void btrfs_merge_extent_hook(struct inode *inode,
1523 struct extent_state *new,
1524 struct extent_state *other)
9ed74f2d 1525{
9ed74f2d
JB		 1526 /* not delalloc, ignore it */
1527 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1528 return;
9ed74f2d 1529
9e0baf60
JB		 1530 spin_lock(&BTRFS_I(inode)->lock);
1531 BTRFS_I(inode)->outstanding_extents--;
1532 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1533}
1534
eb73c1b7
MX		 1535static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1536 struct inode *inode)
1537{
1538 spin_lock(&root->delalloc_lock);
1539 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1540 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1541 &root->delalloc_inodes);
1542 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1543 &BTRFS_I(inode)->runtime_flags);
1544 root->nr_delalloc_inodes++;
1545 if (root->nr_delalloc_inodes == 1) {
1546 spin_lock(&root->fs_info->delalloc_root_lock);
1547 BUG_ON(!list_empty(&root->delalloc_root));
1548 list_add_tail(&root->delalloc_root,
1549 &root->fs_info->delalloc_roots);
1550 spin_unlock(&root->fs_info->delalloc_root_lock);
1551 }
1552 }
1553 spin_unlock(&root->delalloc_lock);
1554}
1555
1556static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1557 struct inode *inode)
1558{
1559 spin_lock(&root->delalloc_lock);
1560 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1561 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1562 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1563 &BTRFS_I(inode)->runtime_flags);
1564 root->nr_delalloc_inodes--;
1565 if (!root->nr_delalloc_inodes) {
1566 spin_lock(&root->fs_info->delalloc_root_lock);
1567 BUG_ON(list_empty(&root->delalloc_root));
1568 list_del_init(&root->delalloc_root);
1569 spin_unlock(&root->fs_info->delalloc_root_lock);
1570 }
1571 }
1572 spin_unlock(&root->delalloc_lock);
1573}
1574
d352ac68
CM		 1575/*
1576 * extent_io.c set_bit_hook, used to track delayed allocation
1577 * bytes in this file, and to maintain the list of inodes that
1578 * have pending delalloc work to be done.
1579 */
1bf85046 1580static void btrfs_set_bit_hook(struct inode *inode,
41074888 1581 struct extent_state *state, unsigned long *bits)
291d673e 1582{
9ed74f2d 1583
75eff68e
CM		 1584 /*
1585 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1586 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1587 * bit, which is only set or cleared with irqs on
1588 */
0ca1f7ce 1589 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1590 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1591 u64 len = state->end + 1 - state->start;
83eea1f1 1592 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1593
9e0baf60 1594 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1595 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1596 } else {
1597 spin_lock(&BTRFS_I(inode)->lock);
1598 BTRFS_I(inode)->outstanding_extents++;
1599 spin_unlock(&BTRFS_I(inode)->lock);
1600 }
287a0ab9 1601
963d678b
MX		 1602 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1603 root->fs_info->delalloc_batch);
df0af1a5 1604 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1605 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5 1606 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1607 &BTRFS_I(inode)->runtime_flags))
1608 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1609 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1610 }
291d673e
CM		 1611}
1612
d352ac68
CM		 1613/*
1614 * extent_io.c clear_bit_hook, see set_bit_hook for why
1615 */
1bf85046 1616static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1617 struct extent_state *state,
1618 unsigned long *bits)
291d673e 1619{
75eff68e
CM		 1620 /*
1621 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1622 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1623 * bit, which is only set or cleared with irqs on
1624 */
0ca1f7ce 1625 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1626 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1627 u64 len = state->end + 1 - state->start;
83eea1f1 1628 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1629
9e0baf60 1630 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1631 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1632 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1633 spin_lock(&BTRFS_I(inode)->lock);
1634 BTRFS_I(inode)->outstanding_extents--;
1635 spin_unlock(&BTRFS_I(inode)->lock);
1636 }
0ca1f7ce
YZ		 1637
1638 if (*bits & EXTENT_DO_ACCOUNTING)
1639 btrfs_delalloc_release_metadata(inode, len);
1640
0cb59c99
JB		 1641 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1642 && do_list)
0ca1f7ce 1643 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1644
963d678b
MX		 1645 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1646 root->fs_info->delalloc_batch);
df0af1a5 1647 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1648 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1649 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1650 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1651 &BTRFS_I(inode)->runtime_flags))
1652 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1653 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1654 }
291d673e
CM		 1655}
1656
d352ac68
CM		 1657/*
1658 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1659 * we don't create bios that span stripes or chunks
1660 */
64a16701 1661int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1662 size_t size, struct bio *bio,
1663 unsigned long bio_flags)
239b14b3
CM		 1664{
1665 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1666 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1667 u64 length = 0;
1668 u64 map_length;
239b14b3
CM		 1669 int ret;
1670
771ed689
CM		 1671 if (bio_flags & EXTENT_BIO_COMPRESSED)
1672 return 0;
1673
f2d8d74d 1674 length = bio->bi_size;
239b14b3 1675 map_length = length;
64a16701 1676 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1677 &map_length, NULL, 0);
3ec706c8 1678 /* Will always return 0 with map_multi == NULL */
3444a972 1679 BUG_ON(ret < 0);
d397712b 1680 if (map_length < length + size)
239b14b3 1681 return 1;
3444a972 1682 return 0;
239b14b3
CM		 1683}
1684
d352ac68
CM		 1685/*
1686 * in order to insert checksums into the metadata in large chunks,
1687 * we wait until bio submission time. All the pages in the bio are
1688 * checksummed and sums are attached onto the ordered extent record.
1689 *
1690 * At IO completion time the cums attached on the ordered extent record
1691 * are inserted into the btree
1692 */
d397712b
CM		 1693static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1694 struct bio *bio, int mirror_num,
eaf25d93
CM		 1695 unsigned long bio_flags,
1696 u64 bio_offset)
065631f6 1697{
065631f6 1698 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1699 int ret = 0;
e015640f 1700
d20f7043 1701 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1702 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1703 return 0;
1704}
e015640f 1705
4a69a410
CM		 1706/*
1707 * in order to insert checksums into the metadata in large chunks,
1708 * we wait until bio submission time. All the pages in the bio are
1709 * checksummed and sums are attached onto the ordered extent record.
1710 *
1711 * At IO completion time the cums attached on the ordered extent record
1712 * are inserted into the btree
1713 */
b2950863 1714static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1715 int mirror_num, unsigned long bio_flags,
1716 u64 bio_offset)
4a69a410
CM		 1717{
1718 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1719 int ret;
1720
1721 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1722 if (ret)
1723 bio_endio(bio, ret);
1724 return ret;
44b8bd7e
CM		 1725}
1726
d352ac68 1727/*
cad321ad
CM		 1728 * extent_io.c submission hook. This does the right thing for csum calculation
1729 * on write, or reading the csums from the tree before a read
d352ac68 1730 */
b2950863 1731static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1732 int mirror_num, unsigned long bio_flags,
1733 u64 bio_offset)
44b8bd7e
CM		 1734{
1735 struct btrfs_root *root = BTRFS_I(inode)->root;
1736 int ret = 0;
19b9bdb0 1737 int skip_sum;
0417341e 1738 int metadata = 0;
b812ce28 1739 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1740
6cbff00f 1741 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1742
83eea1f1 1743 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1744 metadata = 2;
1745
7b6d91da 1746 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1747 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1748 if (ret)
61891923 1749 goto out;
5fd02043 1750
d20f7043 1751 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1752 ret = btrfs_submit_compressed_read(inode, bio,
1753 mirror_num,
1754 bio_flags);
1755 goto out;
c2db1073
TI		 1756 } else if (!skip_sum) {
1757 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1758 if (ret)
61891923 1759 goto out;
c2db1073 1760 }
4d1b5fb4 1761 goto mapit;
b812ce28 1762 } else if (async && !skip_sum) {
17d217fe
YZ		 1763 /* csum items have already been cloned */
1764 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1765 goto mapit;
19b9bdb0 1766 /* we're doing a write, do the async checksumming */
61891923 1767 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1768 inode, rw, bio, mirror_num,
eaf25d93
CM		 1769 bio_flags, bio_offset,
1770 __btrfs_submit_bio_start,
4a69a410 1771 __btrfs_submit_bio_done);
61891923 1772 goto out;
b812ce28
JB		 1773 } else if (!skip_sum) {
1774 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1775 if (ret)
1776 goto out;
19b9bdb0
CM		 1777 }
1778
0b86a832 1779mapit:
61891923
SB		 1780 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1781
1782out:
1783 if (ret < 0)
1784 bio_endio(bio, ret);
1785 return ret;
065631f6 1786}
6885f308 1787
d352ac68
CM		 1788/*
1789 * given a list of ordered sums record them in the inode. This happens
1790 * at IO completion time based on sums calculated at bio submission time.
1791 */
ba1da2f4 1792static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1793 struct inode *inode, u64 file_offset,
1794 struct list_head *list)
1795{
e6dcd2dc
CM		 1796 struct btrfs_ordered_sum *sum;
1797
c6e30871 1798 list_for_each_entry(sum, list, list) {
39847c4d 1799 trans->adding_csums = 1;
d20f7043
CM		 1800 btrfs_csum_file_blocks(trans,
1801 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1802 trans->adding_csums = 0;
e6dcd2dc
CM		 1803 }
1804 return 0;
1805}
1806
2ac55d41
JB		 1807int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1808 struct extent_state **cached_state)
ea8c2819 1809{
6c1500f2 1810 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1811 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1812 cached_state, GFP_NOFS);
ea8c2819
CM		 1813}
1814
d352ac68 1815/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1816struct btrfs_writepage_fixup {
1817 struct page *page;
1818 struct btrfs_work work;
1819};
1820
b2950863 1821static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1822{
1823 struct btrfs_writepage_fixup *fixup;
1824 struct btrfs_ordered_extent *ordered;
2ac55d41 1825 struct extent_state *cached_state = NULL;
247e743c
CM		 1826 struct page *page;
1827 struct inode *inode;
1828 u64 page_start;
1829 u64 page_end;
87826df0 1830 int ret;
247e743c
CM		 1831
1832 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1833 page = fixup->page;
4a096752 1834again:
247e743c
CM		 1835 lock_page(page);
1836 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1837 ClearPageChecked(page);
1838 goto out_page;
1839 }
1840
1841 inode = page->mapping->host;
1842 page_start = page_offset(page);
1843 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1844
2ac55d41 1845 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1846 &cached_state);
4a096752
CM		 1847
1848 /* already ordered? We're done */
8b62b72b 1849 if (PagePrivate2(page))
247e743c 1850 goto out;
4a096752
CM		 1851
1852 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1853 if (ordered) {
2ac55d41
JB		 1854 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1855 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1856 unlock_page(page);
1857 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1858 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1859 goto again;
1860 }
247e743c 1861
87826df0
JM		 1862 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1863 if (ret) {
1864 mapping_set_error(page->mapping, ret);
1865 end_extent_writepage(page, ret, page_start, page_end);
1866 ClearPageChecked(page);
1867 goto out;
1868 }
1869
2ac55d41 1870 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1871 ClearPageChecked(page);
87826df0 1872 set_page_dirty(page);
247e743c 1873out:
2ac55d41
JB		 1874 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1875 &cached_state, GFP_NOFS);
247e743c
CM		 1876out_page:
1877 unlock_page(page);
1878 page_cache_release(page);
b897abec 1879 kfree(fixup);
247e743c
CM		 1880}
1881
1882/*
1883 * There are a few paths in the higher layers of the kernel that directly
1884 * set the page dirty bit without asking the filesystem if it is a
1885 * good idea. This causes problems because we want to make sure COW
1886 * properly happens and the data=ordered rules are followed.
1887 *
c8b97818 1888 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1889 * hasn't been properly setup for IO. We kick off an async process
1890 * to fix it up. The async helper will wait for ordered extents, set
1891 * the delalloc bit and make it safe to write the page.
1892 */
b2950863 1893static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1894{
1895 struct inode *inode = page->mapping->host;
1896 struct btrfs_writepage_fixup *fixup;
1897 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1898
8b62b72b
CM		 1899 /* this page is properly in the ordered list */
1900 if (TestClearPagePrivate2(page))
247e743c
CM		 1901 return 0;
1902
1903 if (PageChecked(page))
1904 return -EAGAIN;
1905
1906 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1907 if (!fixup)
1908 return -EAGAIN;
f421950f 1909
247e743c
CM		 1910 SetPageChecked(page);
1911 page_cache_get(page);
1912 fixup->work.func = btrfs_writepage_fixup_worker;
1913 fixup->page = page;
1914 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1915 return -EBUSY;
247e743c
CM		 1916}
1917
d899e052
YZ		 1918static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1919 struct inode *inode, u64 file_pos,
1920 u64 disk_bytenr, u64 disk_num_bytes,
1921 u64 num_bytes, u64 ram_bytes,
1922 u8 compression, u8 encryption,
1923 u16 other_encoding, int extent_type)
1924{
1925 struct btrfs_root *root = BTRFS_I(inode)->root;
1926 struct btrfs_file_extent_item *fi;
1927 struct btrfs_path *path;
1928 struct extent_buffer *leaf;
1929 struct btrfs_key ins;
d899e052
YZ		 1930 int ret;
1931
1932 path = btrfs_alloc_path();
d8926bb3
MF		 1933 if (!path)
1934 return -ENOMEM;
d899e052 1935
b9473439 1936 path->leave_spinning = 1;
a1ed835e
CM		 1937
1938 /*
1939 * we may be replacing one extent in the tree with another.
1940 * The new extent is pinned in the extent map, and we don't want
1941 * to drop it from the cache until it is completely in the btree.
1942 *
1943 * So, tell btrfs_drop_extents to leave this extent in the cache.
1944 * the caller is expected to unpin it and allow it to be merged
1945 * with the others.
1946 */
5dc562c5 1947 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1948 file_pos + num_bytes, 0);
79787eaa
JM		 1949 if (ret)
1950 goto out;
d899e052 1951
33345d01 1952 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1953 ins.offset = file_pos;
1954 ins.type = BTRFS_EXTENT_DATA_KEY;
1955 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1956 if (ret)
1957 goto out;
d899e052
YZ		 1958 leaf = path->nodes[0];
1959 fi = btrfs_item_ptr(leaf, path->slots[0],
1960 struct btrfs_file_extent_item);
1961 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1962 btrfs_set_file_extent_type(leaf, fi, extent_type);
1963 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1964 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1965 btrfs_set_file_extent_offset(leaf, fi, 0);
1966 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1967 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1968 btrfs_set_file_extent_compression(leaf, fi, compression);
1969 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1970 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1971
d899e052 1972 btrfs_mark_buffer_dirty(leaf);
ce195332 1973 btrfs_release_path(path);
d899e052
YZ		 1974
1975 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1976
1977 ins.objectid = disk_bytenr;
1978 ins.offset = disk_num_bytes;
1979 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1980 ret = btrfs_alloc_reserved_file_extent(trans, root,
1981 root->root_key.objectid,
33345d01 1982 btrfs_ino(inode), file_pos, &ins);
79787eaa 1983out:
d899e052 1984 btrfs_free_path(path);
b9473439 1985
79787eaa 1986 return ret;
d899e052
YZ		 1987}
1988
38c227d8
LB		 1989/* snapshot-aware defrag */
1990struct sa_defrag_extent_backref {
1991 struct rb_node node;
1992 struct old_sa_defrag_extent *old;
1993 u64 root_id;
1994 u64 inum;
1995 u64 file_pos;
1996 u64 extent_offset;
1997 u64 num_bytes;
1998 u64 generation;
1999};
2000
2001struct old_sa_defrag_extent {
2002 struct list_head list;
2003 struct new_sa_defrag_extent *new;
2004
2005 u64 extent_offset;
2006 u64 bytenr;
2007 u64 offset;
2008 u64 len;
2009 int count;
2010};
2011
2012struct new_sa_defrag_extent {
2013 struct rb_root root;
2014 struct list_head head;
2015 struct btrfs_path *path;
2016 struct inode *inode;
2017 u64 file_pos;
2018 u64 len;
2019 u64 bytenr;
2020 u64 disk_len;
2021 u8 compress_type;
2022};
2023
2024static int backref_comp(struct sa_defrag_extent_backref *b1,
2025 struct sa_defrag_extent_backref *b2)
2026{
2027 if (b1->root_id < b2->root_id)
2028 return -1;
2029 else if (b1->root_id > b2->root_id)
2030 return 1;
2031
2032 if (b1->inum < b2->inum)
2033 return -1;
2034 else if (b1->inum > b2->inum)
2035 return 1;
2036
2037 if (b1->file_pos < b2->file_pos)
2038 return -1;
2039 else if (b1->file_pos > b2->file_pos)
2040 return 1;
2041
2042 /*
2043 * [------------------------------] ===> (a range of space)
2044 * |<--->| |<---->| =============> (fs/file tree A)
2045 * |<---------------------------->| ===> (fs/file tree B)
2046 *
2047 * A range of space can refer to two file extents in one tree while
2048 * refer to only one file extent in another tree.
2049 *
2050 * So we may process a disk offset more than one time(two extents in A)
2051 * and locate at the same extent(one extent in B), then insert two same
2052 * backrefs(both refer to the extent in B).
2053 */
2054 return 0;
2055}
2056
2057static void backref_insert(struct rb_root *root,
2058 struct sa_defrag_extent_backref *backref)
2059{
2060 struct rb_node **p = &root->rb_node;
2061 struct rb_node *parent = NULL;
2062 struct sa_defrag_extent_backref *entry;
2063 int ret;
2064
2065 while (*p) {
2066 parent = *p;
2067 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2068
2069 ret = backref_comp(backref, entry);
2070 if (ret < 0)
2071 p = &(*p)->rb_left;
2072 else
2073 p = &(*p)->rb_right;
2074 }
2075
2076 rb_link_node(&backref->node, parent, p);
2077 rb_insert_color(&backref->node, root);
2078}
2079
2080/*
2081 * Note the backref might has changed, and in this case we just return 0.
2082 */
2083static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2084 void *ctx)
2085{
2086 struct btrfs_file_extent_item *extent;
2087 struct btrfs_fs_info *fs_info;
2088 struct old_sa_defrag_extent *old = ctx;
2089 struct new_sa_defrag_extent *new = old->new;
2090 struct btrfs_path *path = new->path;
2091 struct btrfs_key key;
2092 struct btrfs_root *root;
2093 struct sa_defrag_extent_backref *backref;
2094 struct extent_buffer *leaf;
2095 struct inode *inode = new->inode;
2096 int slot;
2097 int ret;
2098 u64 extent_offset;
2099 u64 num_bytes;
2100
2101 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2102 inum == btrfs_ino(inode))
2103 return 0;
2104
2105 key.objectid = root_id;
2106 key.type = BTRFS_ROOT_ITEM_KEY;
2107 key.offset = (u64)-1;
2108
2109 fs_info = BTRFS_I(inode)->root->fs_info;
2110 root = btrfs_read_fs_root_no_name(fs_info, &key);
2111 if (IS_ERR(root)) {
2112 if (PTR_ERR(root) == -ENOENT)
2113 return 0;
2114 WARN_ON(1);
2115 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2116 inum, offset, root_id);
2117 return PTR_ERR(root);
2118 }
2119
2120 key.objectid = inum;
2121 key.type = BTRFS_EXTENT_DATA_KEY;
2122 if (offset > (u64)-1 << 32)
2123 key.offset = 0;
2124 else
2125 key.offset = offset;
2126
2127 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2128 if (ret < 0) {
2129 WARN_ON(1);
2130 return ret;
2131 }
2132
2133 while (1) {
2134 cond_resched();
2135
2136 leaf = path->nodes[0];
2137 slot = path->slots[0];
2138
2139 if (slot >= btrfs_header_nritems(leaf)) {
2140 ret = btrfs_next_leaf(root, path);
2141 if (ret < 0) {
2142 goto out;
2143 } else if (ret > 0) {
2144 ret = 0;
2145 goto out;
2146 }
2147 continue;
2148 }
2149
2150 path->slots[0]++;
2151
2152 btrfs_item_key_to_cpu(leaf, &key, slot);
2153
2154 if (key.objectid > inum)
2155 goto out;
2156
2157 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2158 continue;
2159
2160 extent = btrfs_item_ptr(leaf, slot,
2161 struct btrfs_file_extent_item);
2162
2163 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2164 continue;
2165
2166 extent_offset = btrfs_file_extent_offset(leaf, extent);
2167 if (key.offset - extent_offset != offset)
2168 continue;
2169
2170 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2171 if (extent_offset >= old->extent_offset + old->offset +
2172 old->len || extent_offset + num_bytes <=
2173 old->extent_offset + old->offset)
2174 continue;
2175
2176 break;
2177 }
2178
2179 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2180 if (!backref) {
2181 ret = -ENOENT;
2182 goto out;
2183 }
2184
2185 backref->root_id = root_id;
2186 backref->inum = inum;
2187 backref->file_pos = offset + extent_offset;
2188 backref->num_bytes = num_bytes;
2189 backref->extent_offset = extent_offset;
2190 backref->generation = btrfs_file_extent_generation(leaf, extent);
2191 backref->old = old;
2192 backref_insert(&new->root, backref);
2193 old->count++;
2194out:
2195 btrfs_release_path(path);
2196 WARN_ON(ret);
2197 return ret;
2198}
2199
2200static noinline bool record_extent_backrefs(struct btrfs_path *path,
2201 struct new_sa_defrag_extent *new)
2202{
2203 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2204 struct old_sa_defrag_extent *old, *tmp;
2205 int ret;
2206
2207 new->path = path;
2208
2209 list_for_each_entry_safe(old, tmp, &new->head, list) {
2210 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2211 path, record_one_backref,
2212 old);
2213 BUG_ON(ret < 0 && ret != -ENOENT);
2214
2215 /* no backref to be processed for this extent */
2216 if (!old->count) {
2217 list_del(&old->list);
2218 kfree(old);
2219 }
2220 }
2221
2222 if (list_empty(&new->head))
2223 return false;
2224
2225 return true;
2226}
2227
2228static int relink_is_mergable(struct extent_buffer *leaf,
2229 struct btrfs_file_extent_item *fi,
2230 u64 disk_bytenr)
2231{
2232 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2233 return 0;
2234
2235 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2236 return 0;
2237
2238 if (btrfs_file_extent_compression(leaf, fi) ||
2239 btrfs_file_extent_encryption(leaf, fi) ||
2240 btrfs_file_extent_other_encoding(leaf, fi))
2241 return 0;
2242
2243 return 1;
2244}
2245
2246/*
2247 * Note the backref might has changed, and in this case we just return 0.
2248 */
2249static noinline int relink_extent_backref(struct btrfs_path *path,
2250 struct sa_defrag_extent_backref *prev,
2251 struct sa_defrag_extent_backref *backref)
2252{
2253 struct btrfs_file_extent_item *extent;
2254 struct btrfs_file_extent_item *item;
2255 struct btrfs_ordered_extent *ordered;
2256 struct btrfs_trans_handle *trans;
2257 struct btrfs_fs_info *fs_info;
2258 struct btrfs_root *root;
2259 struct btrfs_key key;
2260 struct extent_buffer *leaf;
2261 struct old_sa_defrag_extent *old = backref->old;
2262 struct new_sa_defrag_extent *new = old->new;
2263 struct inode *src_inode = new->inode;
2264 struct inode *inode;
2265 struct extent_state *cached = NULL;
2266 int ret = 0;
2267 u64 start;
2268 u64 len;
2269 u64 lock_start;
2270 u64 lock_end;
2271 bool merge = false;
2272 int index;
2273
2274 if (prev && prev->root_id == backref->root_id &&
2275 prev->inum == backref->inum &&
2276 prev->file_pos + prev->num_bytes == backref->file_pos)
2277 merge = true;
2278
2279 /* step 1: get root */
2280 key.objectid = backref->root_id;
2281 key.type = BTRFS_ROOT_ITEM_KEY;
2282 key.offset = (u64)-1;
2283
2284 fs_info = BTRFS_I(src_inode)->root->fs_info;
2285 index = srcu_read_lock(&fs_info->subvol_srcu);
2286
2287 root = btrfs_read_fs_root_no_name(fs_info, &key);
2288 if (IS_ERR(root)) {
2289 srcu_read_unlock(&fs_info->subvol_srcu, index);
2290 if (PTR_ERR(root) == -ENOENT)
2291 return 0;
2292 return PTR_ERR(root);
2293 }
38c227d8
LB		 2294
2295 /* step 2: get inode */
2296 key.objectid = backref->inum;
2297 key.type = BTRFS_INODE_ITEM_KEY;
2298 key.offset = 0;
2299
2300 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2301 if (IS_ERR(inode)) {
2302 srcu_read_unlock(&fs_info->subvol_srcu, index);
2303 return 0;
2304 }
2305
2306 srcu_read_unlock(&fs_info->subvol_srcu, index);
2307
2308 /* step 3: relink backref */
2309 lock_start = backref->file_pos;
2310 lock_end = backref->file_pos + backref->num_bytes - 1;
2311 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2312 0, &cached);
2313
2314 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2315 if (ordered) {
2316 btrfs_put_ordered_extent(ordered);
2317 goto out_unlock;
2318 }
2319
2320 trans = btrfs_join_transaction(root);
2321 if (IS_ERR(trans)) {
2322 ret = PTR_ERR(trans);
2323 goto out_unlock;
2324 }
2325
2326 key.objectid = backref->inum;
2327 key.type = BTRFS_EXTENT_DATA_KEY;
2328 key.offset = backref->file_pos;
2329
2330 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2331 if (ret < 0) {
2332 goto out_free_path;
2333 } else if (ret > 0) {
2334 ret = 0;
2335 goto out_free_path;
2336 }
2337
2338 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2339 struct btrfs_file_extent_item);
2340
2341 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2342 backref->generation)
2343 goto out_free_path;
2344
2345 btrfs_release_path(path);
2346
2347 start = backref->file_pos;
2348 if (backref->extent_offset < old->extent_offset + old->offset)
2349 start += old->extent_offset + old->offset -
2350 backref->extent_offset;
2351
2352 len = min(backref->extent_offset + backref->num_bytes,
2353 old->extent_offset + old->offset + old->len);
2354 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2355
2356 ret = btrfs_drop_extents(trans, root, inode, start,
2357 start + len, 1);
2358 if (ret)
2359 goto out_free_path;
2360again:
2361 key.objectid = btrfs_ino(inode);
2362 key.type = BTRFS_EXTENT_DATA_KEY;
2363 key.offset = start;
2364
a09a0a70 2365 path->leave_spinning = 1;
38c227d8
LB		 2366 if (merge) {
2367 struct btrfs_file_extent_item *fi;
2368 u64 extent_len;
2369 struct btrfs_key found_key;
2370
2371 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2372 if (ret < 0)
2373 goto out_free_path;
2374
2375 path->slots[0]--;
2376 leaf = path->nodes[0];
2377 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2378
2379 fi = btrfs_item_ptr(leaf, path->slots[0],
2380 struct btrfs_file_extent_item);
2381 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2382
2383 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2384 extent_len + found_key.offset == start) {
2385 btrfs_set_file_extent_num_bytes(leaf, fi,
2386 extent_len + len);
2387 btrfs_mark_buffer_dirty(leaf);
2388 inode_add_bytes(inode, len);
2389
2390 ret = 1;
2391 goto out_free_path;
2392 } else {
2393 merge = false;
2394 btrfs_release_path(path);
2395 goto again;
2396 }
2397 }
2398
2399 ret = btrfs_insert_empty_item(trans, root, path, &key,
2400 sizeof(*extent));
2401 if (ret) {
2402 btrfs_abort_transaction(trans, root, ret);
2403 goto out_free_path;
2404 }
2405
2406 leaf = path->nodes[0];
2407 item = btrfs_item_ptr(leaf, path->slots[0],
2408 struct btrfs_file_extent_item);
2409 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2410 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2411 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2412 btrfs_set_file_extent_num_bytes(leaf, item, len);
2413 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2414 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2415 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2416 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2417 btrfs_set_file_extent_encryption(leaf, item, 0);
2418 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2419
2420 btrfs_mark_buffer_dirty(leaf);
2421 inode_add_bytes(inode, len);
a09a0a70 2422 btrfs_release_path(path);
38c227d8
LB		 2423
2424 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2425 new->disk_len, 0,
2426 backref->root_id, backref->inum,
2427 new->file_pos, 0); /* start - extent_offset */
2428 if (ret) {
2429 btrfs_abort_transaction(trans, root, ret);
2430 goto out_free_path;
2431 }
2432
2433 ret = 1;
2434out_free_path:
2435 btrfs_release_path(path);
a09a0a70 2436 path->leave_spinning = 0;
38c227d8
LB		 2437 btrfs_end_transaction(trans, root);
2438out_unlock:
2439 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2440 &cached, GFP_NOFS);
2441 iput(inode);
2442 return ret;
2443}
2444
2445static void relink_file_extents(struct new_sa_defrag_extent *new)
2446{
2447 struct btrfs_path *path;
2448 struct old_sa_defrag_extent *old, *tmp;
2449 struct sa_defrag_extent_backref *backref;
2450 struct sa_defrag_extent_backref *prev = NULL;
2451 struct inode *inode;
2452 struct btrfs_root *root;
2453 struct rb_node *node;
2454 int ret;
2455
2456 inode = new->inode;
2457 root = BTRFS_I(inode)->root;
2458
2459 path = btrfs_alloc_path();
2460 if (!path)
2461 return;
2462
2463 if (!record_extent_backrefs(path, new)) {
2464 btrfs_free_path(path);
2465 goto out;
2466 }
2467 btrfs_release_path(path);
2468
2469 while (1) {
2470 node = rb_first(&new->root);
2471 if (!node)
2472 break;
2473 rb_erase(node, &new->root);
2474
2475 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2476
2477 ret = relink_extent_backref(path, prev, backref);
2478 WARN_ON(ret < 0);
2479
2480 kfree(prev);
2481
2482 if (ret == 1)
2483 prev = backref;
2484 else
2485 prev = NULL;
2486 cond_resched();
2487 }
2488 kfree(prev);
2489
2490 btrfs_free_path(path);
2491
2492 list_for_each_entry_safe(old, tmp, &new->head, list) {
2493 list_del(&old->list);
2494 kfree(old);
2495 }
2496out:
2497 atomic_dec(&root->fs_info->defrag_running);
2498 wake_up(&root->fs_info->transaction_wait);
2499
2500 kfree(new);
2501}
2502
2503static struct new_sa_defrag_extent *
2504record_old_file_extents(struct inode *inode,
2505 struct btrfs_ordered_extent *ordered)
2506{
2507 struct btrfs_root *root = BTRFS_I(inode)->root;
2508 struct btrfs_path *path;
2509 struct btrfs_key key;
2510 struct old_sa_defrag_extent *old, *tmp;
2511 struct new_sa_defrag_extent *new;
2512 int ret;
2513
2514 new = kmalloc(sizeof(*new), GFP_NOFS);
2515 if (!new)
2516 return NULL;
2517
2518 new->inode = inode;
2519 new->file_pos = ordered->file_offset;
2520 new->len = ordered->len;
2521 new->bytenr = ordered->start;
2522 new->disk_len = ordered->disk_len;
2523 new->compress_type = ordered->compress_type;
2524 new->root = RB_ROOT;
2525 INIT_LIST_HEAD(&new->head);
2526
2527 path = btrfs_alloc_path();
2528 if (!path)
2529 goto out_kfree;
2530
2531 key.objectid = btrfs_ino(inode);
2532 key.type = BTRFS_EXTENT_DATA_KEY;
2533 key.offset = new->file_pos;
2534
2535 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2536 if (ret < 0)
2537 goto out_free_path;
2538 if (ret > 0 && path->slots[0] > 0)
2539 path->slots[0]--;
2540
2541 /* find out all the old extents for the file range */
2542 while (1) {
2543 struct btrfs_file_extent_item *extent;
2544 struct extent_buffer *l;
2545 int slot;
2546 u64 num_bytes;
2547 u64 offset;
2548 u64 end;
2549 u64 disk_bytenr;
2550 u64 extent_offset;
2551
2552 l = path->nodes[0];
2553 slot = path->slots[0];
2554
2555 if (slot >= btrfs_header_nritems(l)) {
2556 ret = btrfs_next_leaf(root, path);
2557 if (ret < 0)
2558 goto out_free_list;
2559 else if (ret > 0)
2560 break;
2561 continue;
2562 }
2563
2564 btrfs_item_key_to_cpu(l, &key, slot);
2565
2566 if (key.objectid != btrfs_ino(inode))
2567 break;
2568 if (key.type != BTRFS_EXTENT_DATA_KEY)
2569 break;
2570 if (key.offset >= new->file_pos + new->len)
2571 break;
2572
2573 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2574
2575 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2576 if (key.offset + num_bytes < new->file_pos)
2577 goto next;
2578
2579 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2580 if (!disk_bytenr)
2581 goto next;
2582
2583 extent_offset = btrfs_file_extent_offset(l, extent);
2584
2585 old = kmalloc(sizeof(*old), GFP_NOFS);
2586 if (!old)
2587 goto out_free_list;
2588
2589 offset = max(new->file_pos, key.offset);
2590 end = min(new->file_pos + new->len, key.offset + num_bytes);
2591
2592 old->bytenr = disk_bytenr;
2593 old->extent_offset = extent_offset;
2594 old->offset = offset - key.offset;
2595 old->len = end - offset;
2596 old->new = new;
2597 old->count = 0;
2598 list_add_tail(&old->list, &new->head);
2599next:
2600 path->slots[0]++;
2601 cond_resched();
2602 }
2603
2604 btrfs_free_path(path);
2605 atomic_inc(&root->fs_info->defrag_running);
2606
2607 return new;
2608
2609out_free_list:
2610 list_for_each_entry_safe(old, tmp, &new->head, list) {
2611 list_del(&old->list);
2612 kfree(old);
2613 }
2614out_free_path:
2615 btrfs_free_path(path);
2616out_kfree:
2617 kfree(new);
2618 return NULL;
2619}
2620
5d13a98f
CM		 2621/*
2622 * helper function for btrfs_finish_ordered_io, this
2623 * just reads in some of the csum leaves to prime them into ram
2624 * before we start the transaction. It limits the amount of btree
2625 * reads required while inside the transaction.
2626 */
d352ac68
CM		 2627/* as ordered data IO finishes, this gets called so we can finish
2628 * an ordered extent if the range of bytes in the file it covers are
2629 * fully written.
2630 */
5fd02043 2631static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2632{
5fd02043 2633 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2634 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2635 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2636 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2637 struct extent_state *cached_state = NULL;
38c227d8 2638 struct new_sa_defrag_extent *new = NULL;
261507a0 2639 int compress_type = 0;
e6dcd2dc 2640 int ret;
82d5902d 2641 bool nolock;
e6dcd2dc 2642
83eea1f1 2643 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2644
5fd02043
JB		 2645 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2646 ret = -EIO;
2647 goto out;
2648 }
2649
c2167754 2650 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2651 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2652 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2653 if (nolock)
2654 trans = btrfs_join_transaction_nolock(root);
2655 else
2656 trans = btrfs_join_transaction(root);
2657 if (IS_ERR(trans)) {
2658 ret = PTR_ERR(trans);
2659 trans = NULL;
2660 goto out;
c2167754 2661 }
6c760c07
JB		 2662 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2663 ret = btrfs_update_inode_fallback(trans, root, inode);
2664 if (ret) /* -ENOMEM or corruption */
2665 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2666 goto out;
2667 }
e6dcd2dc 2668
2ac55d41
JB		 2669 lock_extent_bits(io_tree, ordered_extent->file_offset,
2670 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2671 0, &cached_state);
e6dcd2dc 2672
38c227d8
LB		 2673 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2674 ordered_extent->file_offset + ordered_extent->len - 1,
2675 EXTENT_DEFRAG, 1, cached_state);
2676 if (ret) {
2677 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2678 if (last_snapshot >= BTRFS_I(inode)->generation)
2679 /* the inode is shared */
2680 new = record_old_file_extents(inode, ordered_extent);
2681
2682 clear_extent_bit(io_tree, ordered_extent->file_offset,
2683 ordered_extent->file_offset + ordered_extent->len - 1,
2684 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2685 }
2686
0cb59c99 2687 if (nolock)
7a7eaa40 2688 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2689 else
7a7eaa40 2690 trans = btrfs_join_transaction(root);
79787eaa
JM		 2691 if (IS_ERR(trans)) {
2692 ret = PTR_ERR(trans);
2693 trans = NULL;
2694 goto out_unlock;
2695 }
0ca1f7ce 2696 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2697
c8b97818 2698 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2699 compress_type = ordered_extent->compress_type;
d899e052 2700 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2701 BUG_ON(compress_type);
920bbbfb 2702 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2703 ordered_extent->file_offset,
2704 ordered_extent->file_offset +
2705 ordered_extent->len);
d899e052 2706 } else {
0af3d00b 2707 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2708 ret = insert_reserved_file_extent(trans, inode,
2709 ordered_extent->file_offset,
2710 ordered_extent->start,
2711 ordered_extent->disk_len,
2712 ordered_extent->len,
2713 ordered_extent->len,
261507a0 2714 compress_type, 0, 0,
d899e052 2715 BTRFS_FILE_EXTENT_REG);
d899e052 2716 }
5dc562c5
JB		 2717 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2718 ordered_extent->file_offset, ordered_extent->len,
2719 trans->transid);
79787eaa
JM		 2720 if (ret < 0) {
2721 btrfs_abort_transaction(trans, root, ret);
5fd02043 2722 goto out_unlock;
79787eaa 2723 }
2ac55d41 2724
e6dcd2dc
CM		 2725 add_pending_csums(trans, inode, ordered_extent->file_offset,
2726 &ordered_extent->list);
2727
6c760c07
JB		 2728 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2729 ret = btrfs_update_inode_fallback(trans, root, inode);
2730 if (ret) { /* -ENOMEM or corruption */
2731 btrfs_abort_transaction(trans, root, ret);
2732 goto out_unlock;
1ef30be1
JB		 2733 }
2734 ret = 0;
5fd02043
JB		 2735out_unlock:
2736 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2737 ordered_extent->file_offset +
2738 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2739out:
5b0e95bf 2740 if (root != root->fs_info->tree_root)
0cb59c99 2741 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2742 if (trans)
2743 btrfs_end_transaction(trans, root);
0cb59c99 2744
0bec9ef5 2745 if (ret) {
5fd02043
JB		 2746 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2747 ordered_extent->file_offset +
2748 ordered_extent->len - 1, NULL, GFP_NOFS);
2749
0bec9ef5
JB		 2750 /*
2751 * If the ordered extent had an IOERR or something else went
2752 * wrong we need to return the space for this ordered extent
2753 * back to the allocator.
2754 */
2755 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2756 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2757 btrfs_free_reserved_extent(root, ordered_extent->start,
2758 ordered_extent->disk_len);
2759 }
2760
2761
5fd02043 2762 /*
8bad3c02
LB		 2763 * This needs to be done to make sure anybody waiting knows we are done
2764 * updating everything for this ordered extent.
5fd02043
JB		 2765 */
2766 btrfs_remove_ordered_extent(inode, ordered_extent);
2767
38c227d8
LB		 2768 /* for snapshot-aware defrag */
2769 if (new)
2770 relink_file_extents(new);
2771
e6dcd2dc
CM		 2772 /* once for us */
2773 btrfs_put_ordered_extent(ordered_extent);
2774 /* once for the tree */
2775 btrfs_put_ordered_extent(ordered_extent);
2776
5fd02043
JB		 2777 return ret;
2778}
2779
2780static void finish_ordered_fn(struct btrfs_work *work)
2781{
2782 struct btrfs_ordered_extent *ordered_extent;
2783 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2784 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2785}
2786
b2950863 2787static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2788 struct extent_state *state, int uptodate)
2789{
5fd02043
JB		 2790 struct inode *inode = page->mapping->host;
2791 struct btrfs_root *root = BTRFS_I(inode)->root;
2792 struct btrfs_ordered_extent *ordered_extent = NULL;
2793 struct btrfs_workers *workers;
2794
1abe9b8a 2795 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2796
8b62b72b 2797 ClearPagePrivate2(page);
5fd02043
JB		 2798 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2799 end - start + 1, uptodate))
2800 return 0;
2801
2802 ordered_extent->work.func = finish_ordered_fn;
2803 ordered_extent->work.flags = 0;
2804
83eea1f1 2805 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2806 workers = &root->fs_info->endio_freespace_worker;
2807 else
2808 workers = &root->fs_info->endio_write_workers;
2809 btrfs_queue_worker(workers, &ordered_extent->work);
2810
2811 return 0;
211f90e6
CM		 2812}
2813
d352ac68
CM		 2814/*
2815 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2816 * if there's a match, we allow the bio to finish. If not, the code in
2817 * extent_io.c will try to find good copies for us.
d352ac68 2818 */
b2950863 2819static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2820 struct extent_state *state, int mirror)
07157aac 2821{
4eee4fa4 2822 size_t offset = start - page_offset(page);
07157aac 2823 struct inode *inode = page->mapping->host;
d1310b2e 2824 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2825 char *kaddr;
aadfeb6e 2826 u64 private = ~(u32)0;
07157aac 2827 int ret;
ff79f819
CM		 2828 struct btrfs_root *root = BTRFS_I(inode)->root;
2829 u32 csum = ~(u32)0;
c2cf52eb
SK		 2830 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2831 DEFAULT_RATELIMIT_BURST);
d1310b2e 2832
d20f7043
CM		 2833 if (PageChecked(page)) {
2834 ClearPageChecked(page);
2835 goto good;
2836 }
6cbff00f
CH		 2837
2838 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2839 goto good;
17d217fe
YZ		 2840
2841 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2842 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2843 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2844 GFP_NOFS);
b6cda9bc 2845 return 0;
17d217fe 2846 }
d20f7043 2847
c2e639f0 2848 if (state && state->start == start) {
70dec807
CM		 2849 private = state->private;
2850 ret = 0;
2851 } else {
2852 ret = get_state_private(io_tree, start, &private);
2853 }
7ac687d9 2854 kaddr = kmap_atomic(page);
d397712b 2855 if (ret)
07157aac 2856 goto zeroit;
d397712b 2857
b0496686 2858 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2859 btrfs_csum_final(csum, (char *)&csum);
d397712b 2860 if (csum != private)
07157aac 2861 goto zeroit;
d397712b 2862
7ac687d9 2863 kunmap_atomic(kaddr);
d20f7043 2864good:
07157aac
CM		 2865 return 0;
2866
2867zeroit:
c2cf52eb
SK		 2868 if (__ratelimit(&_rs))
2869 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2870 (unsigned long long)btrfs_ino(page->mapping->host),
2871 (unsigned long long)start, csum,
2872 (unsigned long long)private);
db94535d
CM		 2873 memset(kaddr + offset, 1, end - start + 1);
2874 flush_dcache_page(page);
7ac687d9 2875 kunmap_atomic(kaddr);
3b951516
CM		 2876 if (private == 0)
2877 return 0;
7e38326f 2878 return -EIO;
07157aac 2879}
b888db2b 2880
24bbcf04
YZ		 2881struct delayed_iput {
2882 struct list_head list;
2883 struct inode *inode;
2884};
2885
79787eaa
JM		 2886/* JDM: If this is fs-wide, why can't we add a pointer to
2887 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2888void btrfs_add_delayed_iput(struct inode *inode)
2889{
2890 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2891 struct delayed_iput *delayed;
2892
2893 if (atomic_add_unless(&inode->i_count, -1, 1))
2894 return;
2895
2896 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2897 delayed->inode = inode;
2898
2899 spin_lock(&fs_info->delayed_iput_lock);
2900 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2901 spin_unlock(&fs_info->delayed_iput_lock);
2902}
2903
2904void btrfs_run_delayed_iputs(struct btrfs_root *root)
2905{
2906 LIST_HEAD(list);
2907 struct btrfs_fs_info *fs_info = root->fs_info;
2908 struct delayed_iput *delayed;
2909 int empty;
2910
2911 spin_lock(&fs_info->delayed_iput_lock);
2912 empty = list_empty(&fs_info->delayed_iputs);
2913 spin_unlock(&fs_info->delayed_iput_lock);
2914 if (empty)
2915 return;
2916
24bbcf04
YZ		 2917 spin_lock(&fs_info->delayed_iput_lock);
2918 list_splice_init(&fs_info->delayed_iputs, &list);
2919 spin_unlock(&fs_info->delayed_iput_lock);
2920
2921 while (!list_empty(&list)) {
2922 delayed = list_entry(list.next, struct delayed_iput, list);
2923 list_del(&delayed->list);
2924 iput(delayed->inode);
2925 kfree(delayed);
2926 }
24bbcf04
YZ		 2927}
2928
d68fc57b 2929/*
42b2aa86 2930 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2931 * files in the subvolume, it removes orphan item and frees block_rsv
2932 * structure.
2933 */
2934void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2935 struct btrfs_root *root)
2936{
90290e19 2937 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2938 int ret;
2939
8a35d95f 2940 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2941 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2942 return;
2943
90290e19 2944 spin_lock(&root->orphan_lock);
8a35d95f 2945 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2946 spin_unlock(&root->orphan_lock);
2947 return;
2948 }
2949
2950 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2951 spin_unlock(&root->orphan_lock);
2952 return;
2953 }
2954
2955 block_rsv = root->orphan_block_rsv;
2956 root->orphan_block_rsv = NULL;
2957 spin_unlock(&root->orphan_lock);
2958
d68fc57b
YZ		 2959 if (root->orphan_item_inserted &&
2960 btrfs_root_refs(&root->root_item) > 0) {
2961 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2962 root->root_key.objectid);
2963 BUG_ON(ret);
2964 root->orphan_item_inserted = 0;
2965 }
2966
90290e19
JB		 2967 if (block_rsv) {
2968 WARN_ON(block_rsv->size > 0);
2969 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2970 }
2971}
2972
7b128766
JB		 2973/*
2974 * This creates an orphan entry for the given inode in case something goes
2975 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2976 *
2977 * NOTE: caller of this function should reserve 5 units of metadata for
2978 * this function.
7b128766
JB		 2979 */
2980int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2981{
2982 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2983 struct btrfs_block_rsv *block_rsv = NULL;
2984 int reserve = 0;
2985 int insert = 0;
2986 int ret;
7b128766 2987
d68fc57b 2988 if (!root->orphan_block_rsv) {
66d8f3dd 2989 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2990 if (!block_rsv)
2991 return -ENOMEM;
d68fc57b 2992 }
7b128766 2993
d68fc57b
YZ		 2994 spin_lock(&root->orphan_lock);
2995 if (!root->orphan_block_rsv) {
2996 root->orphan_block_rsv = block_rsv;
2997 } else if (block_rsv) {
2998 btrfs_free_block_rsv(root, block_rsv);
2999 block_rsv = NULL;
7b128766 3000 }
7b128766 3001
8a35d95f
JB		 3002 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3003 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3004#if 0
3005 /*
3006 * For proper ENOSPC handling, we should do orphan
3007 * cleanup when mounting. But this introduces backward
3008 * compatibility issue.
3009 */
3010 if (!xchg(&root->orphan_item_inserted, 1))
3011 insert = 2;
3012 else
3013 insert = 1;
3014#endif
3015 insert = 1;
321f0e70 3016 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3017 }
3018
72ac3c0d
JB		 3019 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3020 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3021 reserve = 1;
d68fc57b 3022 spin_unlock(&root->orphan_lock);
7b128766 3023
d68fc57b
YZ		 3024 /* grab metadata reservation from transaction handle */
3025 if (reserve) {
3026 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3027 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3028 }
7b128766 3029
d68fc57b
YZ		 3030 /* insert an orphan item to track this unlinked/truncated file */
3031 if (insert >= 1) {
33345d01 3032 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3033 if (ret && ret != -EEXIST) {
8a35d95f
JB		 3034 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3035 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 3036 btrfs_abort_transaction(trans, root, ret);
3037 return ret;
3038 }
3039 ret = 0;
d68fc57b
YZ		 3040 }
3041
3042 /* insert an orphan item to track subvolume contains orphan files */
3043 if (insert >= 2) {
3044 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3045 root->root_key.objectid);
79787eaa
JM		 3046 if (ret && ret != -EEXIST) {
3047 btrfs_abort_transaction(trans, root, ret);
3048 return ret;
3049 }
d68fc57b
YZ		 3050 }
3051 return 0;
7b128766
JB		 3052}
3053
3054/*
3055 * We have done the truncate/delete so we can go ahead and remove the orphan
3056 * item for this particular inode.
3057 */
48a3b636
ES		 3058static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3059 struct inode *inode)
7b128766
JB		 3060{
3061 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3062 int delete_item = 0;
3063 int release_rsv = 0;
7b128766
JB		 3064 int ret = 0;
3065
d68fc57b 3066 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3067 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3068 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3069 delete_item = 1;
7b128766 3070
72ac3c0d
JB		 3071 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3072 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3073 release_rsv = 1;
d68fc57b 3074 spin_unlock(&root->orphan_lock);
7b128766 3075
d68fc57b 3076 if (trans && delete_item) {
33345d01 3077 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3078 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 3079 }
7b128766 3080
8a35d95f 3081 if (release_rsv) {
d68fc57b 3082 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 3083 atomic_dec(&root->orphan_inodes);
3084 }
7b128766 3085
d68fc57b 3086 return 0;
7b128766
JB		 3087}
3088
3089/*
3090 * this cleans up any orphans that may be left on the list from the last use
3091 * of this root.
3092 */
66b4ffd1 3093int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3094{
3095 struct btrfs_path *path;
3096 struct extent_buffer *leaf;
7b128766
JB		 3097 struct btrfs_key key, found_key;
3098 struct btrfs_trans_handle *trans;
3099 struct inode *inode;
8f6d7f4f 3100 u64 last_objectid = 0;
7b128766
JB		 3101 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3102
d68fc57b 3103 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3104 return 0;
c71bf099
YZ		 3105
3106 path = btrfs_alloc_path();
66b4ffd1
JB		 3107 if (!path) {
3108 ret = -ENOMEM;
3109 goto out;
3110 }
7b128766
JB		 3111 path->reada = -1;
3112
3113 key.objectid = BTRFS_ORPHAN_OBJECTID;
3114 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3115 key.offset = (u64)-1;
3116
7b128766
JB		 3117 while (1) {
3118 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3119 if (ret < 0)
3120 goto out;
7b128766
JB		 3121
3122 /*
3123 * if ret == 0 means we found what we were searching for, which
25985edc 3124 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3125 * find the key and see if we have stuff that matches
3126 */
3127 if (ret > 0) {
66b4ffd1 3128 ret = 0;
7b128766
JB		 3129 if (path->slots[0] == 0)
3130 break;
3131 path->slots[0]--;
3132 }
3133
3134 /* pull out the item */
3135 leaf = path->nodes[0];
7b128766
JB		 3136 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3137
3138 /* make sure the item matches what we want */
3139 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3140 break;
3141 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3142 break;
3143
3144 /* release the path since we're done with it */
b3b4aa74 3145 btrfs_release_path(path);
7b128766
JB		 3146
3147 /*
3148 * this is where we are basically btrfs_lookup, without the
3149 * crossing root thing. we store the inode number in the
3150 * offset of the orphan item.
3151 */
8f6d7f4f
JB		 3152
3153 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3154 btrfs_err(root->fs_info,
3155 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3156 ret = -EINVAL;
3157 goto out;
3158 }
3159
3160 last_objectid = found_key.offset;
3161
5d4f98a2
YZ		 3162 found_key.objectid = found_key.offset;
3163 found_key.type = BTRFS_INODE_ITEM_KEY;
3164 found_key.offset = 0;
73f73415 3165 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 3166 ret = PTR_RET(inode);
3167 if (ret && ret != -ESTALE)
66b4ffd1 3168 goto out;
7b128766 3169
f8e9e0b0
AJ		 3170 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3171 struct btrfs_root *dead_root;
3172 struct btrfs_fs_info *fs_info = root->fs_info;
3173 int is_dead_root = 0;
3174
3175 /*
3176 * this is an orphan in the tree root. Currently these
3177 * could come from 2 sources:
3178 * a) a snapshot deletion in progress
3179 * b) a free space cache inode
3180 * We need to distinguish those two, as the snapshot
3181 * orphan must not get deleted.
3182 * find_dead_roots already ran before us, so if this
3183 * is a snapshot deletion, we should find the root
3184 * in the dead_roots list
3185 */
3186 spin_lock(&fs_info->trans_lock);
3187 list_for_each_entry(dead_root, &fs_info->dead_roots,
3188 root_list) {
3189 if (dead_root->root_key.objectid ==
3190 found_key.objectid) {
3191 is_dead_root = 1;
3192 break;
3193 }
3194 }
3195 spin_unlock(&fs_info->trans_lock);
3196 if (is_dead_root) {
3197 /* prevent this orphan from being found again */
3198 key.offset = found_key.objectid - 1;
3199 continue;
3200 }
3201 }
7b128766 3202 /*
a8c9e576
JB		 3203 * Inode is already gone but the orphan item is still there,
3204 * kill the orphan item.
7b128766 3205 */
a8c9e576
JB		 3206 if (ret == -ESTALE) {
3207 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3208 if (IS_ERR(trans)) {
3209 ret = PTR_ERR(trans);
3210 goto out;
3211 }
c2cf52eb
SK		 3212 btrfs_debug(root->fs_info, "auto deleting %Lu",
3213 found_key.objectid);
a8c9e576
JB		 3214 ret = btrfs_del_orphan_item(trans, root,
3215 found_key.objectid);
79787eaa 3216 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 3217 btrfs_end_transaction(trans, root);
7b128766
JB		 3218 continue;
3219 }
3220
a8c9e576
JB		 3221 /*
3222 * add this inode to the orphan list so btrfs_orphan_del does
3223 * the proper thing when we hit it
3224 */
8a35d95f
JB		 3225 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3226 &BTRFS_I(inode)->runtime_flags);
925396ec 3227 atomic_inc(&root->orphan_inodes);
a8c9e576 3228
7b128766
JB		 3229 /* if we have links, this was a truncate, lets do that */
3230 if (inode->i_nlink) {
a41ad394
JB		 3231 if (!S_ISREG(inode->i_mode)) {
3232 WARN_ON(1);
3233 iput(inode);
3234 continue;
3235 }
7b128766 3236 nr_truncate++;
f3fe820c
JB		 3237
3238 /* 1 for the orphan item deletion. */
3239 trans = btrfs_start_transaction(root, 1);
3240 if (IS_ERR(trans)) {
c69b26b0 3241 iput(inode);
f3fe820c
JB		 3242 ret = PTR_ERR(trans);
3243 goto out;
3244 }
3245 ret = btrfs_orphan_add(trans, inode);
3246 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3247 if (ret) {
3248 iput(inode);
f3fe820c 3249 goto out;
c69b26b0 3250 }
f3fe820c 3251
66b4ffd1 3252 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3253 if (ret)
3254 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3255 } else {
3256 nr_unlink++;
3257 }
3258
3259 /* this will do delete_inode and everything for us */
3260 iput(inode);
66b4ffd1
JB		 3261 if (ret)
3262 goto out;
7b128766 3263 }
3254c876
MX		 3264 /* release the path since we're done with it */
3265 btrfs_release_path(path);
3266
d68fc57b
YZ		 3267 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3268
3269 if (root->orphan_block_rsv)
3270 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3271 (u64)-1);
3272
3273 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3274 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3275 if (!IS_ERR(trans))
3276 btrfs_end_transaction(trans, root);
d68fc57b 3277 }
7b128766
JB		 3278
3279 if (nr_unlink)
4884b476 3280 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3281 if (nr_truncate)
4884b476 3282 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3283
3284out:
3285 if (ret)
c2cf52eb
SK		 3286 btrfs_crit(root->fs_info,
3287 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3288 btrfs_free_path(path);
3289 return ret;
7b128766
JB		 3290}
3291
46a53cca
CM		 3292/*
3293 * very simple check to peek ahead in the leaf looking for xattrs. If we
3294 * don't find any xattrs, we know there can't be any acls.
3295 *
3296 * slot is the slot the inode is in, objectid is the objectid of the inode
3297 */
3298static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3299 int slot, u64 objectid)
3300{
3301 u32 nritems = btrfs_header_nritems(leaf);
3302 struct btrfs_key found_key;
3303 int scanned = 0;
3304
3305 slot++;
3306 while (slot < nritems) {
3307 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3308
3309 /* we found a different objectid, there must not be acls */
3310 if (found_key.objectid != objectid)
3311 return 0;
3312
3313 /* we found an xattr, assume we've got an acl */
3314 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3315 return 1;
3316
3317 /*
3318 * we found a key greater than an xattr key, there can't
3319 * be any acls later on
3320 */
3321 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3322 return 0;
3323
3324 slot++;
3325 scanned++;
3326
3327 /*
3328 * it goes inode, inode backrefs, xattrs, extents,
3329 * so if there are a ton of hard links to an inode there can
3330 * be a lot of backrefs. Don't waste time searching too hard,
3331 * this is just an optimization
3332 */
3333 if (scanned >= 8)
3334 break;
3335 }
3336 /* we hit the end of the leaf before we found an xattr or
3337 * something larger than an xattr. We have to assume the inode
3338 * has acls
3339 */
3340 return 1;
3341}
3342
d352ac68
CM		 3343/*
3344 * read an inode from the btree into the in-memory inode
3345 */
5d4f98a2 3346static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3347{
3348 struct btrfs_path *path;
5f39d397 3349 struct extent_buffer *leaf;
39279cc3 3350 struct btrfs_inode_item *inode_item;
0b86a832 3351 struct btrfs_timespec *tspec;
39279cc3
CM		 3352 struct btrfs_root *root = BTRFS_I(inode)->root;
3353 struct btrfs_key location;
46a53cca 3354 int maybe_acls;
618e21d5 3355 u32 rdev;
39279cc3 3356 int ret;
2f7e33d4
MX		 3357 bool filled = false;
3358
3359 ret = btrfs_fill_inode(inode, &rdev);
3360 if (!ret)
3361 filled = true;
39279cc3
CM		 3362
3363 path = btrfs_alloc_path();
1748f843
MF		 3364 if (!path)
3365 goto make_bad;
3366
d90c7321 3367 path->leave_spinning = 1;
39279cc3 3368 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3369
39279cc3 3370 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3371 if (ret)
39279cc3 3372 goto make_bad;
39279cc3 3373
5f39d397 3374 leaf = path->nodes[0];
2f7e33d4
MX		 3375
3376 if (filled)
3377 goto cache_acl;
3378
5f39d397
CM		 3379 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3380 struct btrfs_inode_item);
5f39d397 3381 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3382 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3383 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3384 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3385 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3386
3387 tspec = btrfs_inode_atime(inode_item);
3388 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3389 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3390
3391 tspec = btrfs_inode_mtime(inode_item);
3392 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3393 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3394
3395 tspec = btrfs_inode_ctime(inode_item);
3396 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3397 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3398
a76a3cd4 3399 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3400 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3401 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3402
3403 /*
3404 * If we were modified in the current generation and evicted from memory
3405 * and then re-read we need to do a full sync since we don't have any
3406 * idea about which extents were modified before we were evicted from
3407 * cache.
3408 */
3409 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3410 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3411 &BTRFS_I(inode)->runtime_flags);
3412
0c4d2d95 3413 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3414 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3415 inode->i_rdev = 0;
5f39d397
CM		 3416 rdev = btrfs_inode_rdev(leaf, inode_item);
3417
aec7477b 3418 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3419 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3420cache_acl:
46a53cca
CM		 3421 /*
3422 * try to precache a NULL acl entry for files that don't have
3423 * any xattrs or acls
3424 */
33345d01
LZ		 3425 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3426 btrfs_ino(inode));
72c04902
AV		 3427 if (!maybe_acls)
3428 cache_no_acl(inode);
46a53cca 3429
39279cc3 3430 btrfs_free_path(path);
39279cc3 3431
39279cc3 3432 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3433 case S_IFREG:
3434 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3435 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3436 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3437 inode->i_fop = &btrfs_file_operations;
3438 inode->i_op = &btrfs_file_inode_operations;
3439 break;
3440 case S_IFDIR:
3441 inode->i_fop = &btrfs_dir_file_operations;
3442 if (root == root->fs_info->tree_root)
3443 inode->i_op = &btrfs_dir_ro_inode_operations;
3444 else
3445 inode->i_op = &btrfs_dir_inode_operations;
3446 break;
3447 case S_IFLNK:
3448 inode->i_op = &btrfs_symlink_inode_operations;
3449 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3450 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3451 break;
618e21d5 3452 default:
0279b4cd 3453 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3454 init_special_inode(inode, inode->i_mode, rdev);
3455 break;
39279cc3 3456 }
6cbff00f
CH		 3457
3458 btrfs_update_iflags(inode);
39279cc3
CM		 3459 return;
3460
3461make_bad:
39279cc3 3462 btrfs_free_path(path);
39279cc3
CM		 3463 make_bad_inode(inode);
3464}
3465
d352ac68
CM		 3466/*
3467 * given a leaf and an inode, copy the inode fields into the leaf
3468 */
e02119d5
CM		 3469static void fill_inode_item(struct btrfs_trans_handle *trans,
3470 struct extent_buffer *leaf,
5f39d397 3471 struct btrfs_inode_item *item,
39279cc3
CM		 3472 struct inode *inode)
3473{
51fab693
LB		 3474 struct btrfs_map_token token;
3475
3476 btrfs_init_map_token(&token);
5f39d397 3477
51fab693
LB		 3478 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3479 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3480 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3481 &token);
3482 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3483 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3484
51fab693
LB		 3485 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3486 inode->i_atime.tv_sec, &token);
3487 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3488 inode->i_atime.tv_nsec, &token);
5f39d397 3489
51fab693
LB		 3490 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3491 inode->i_mtime.tv_sec, &token);
3492 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3493 inode->i_mtime.tv_nsec, &token);
5f39d397 3494
51fab693
LB		 3495 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3496 inode->i_ctime.tv_sec, &token);
3497 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3498 inode->i_ctime.tv_nsec, &token);
5f39d397 3499
51fab693
LB		 3500 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3501 &token);
3502 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3503 &token);
3504 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3505 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3506 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3507 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3508 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3509}
3510
d352ac68
CM		 3511/*
3512 * copy everything in the in-memory inode into the btree.
3513 */
2115133f 3514static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3515 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3516{
3517 struct btrfs_inode_item *inode_item;
3518 struct btrfs_path *path;
5f39d397 3519 struct extent_buffer *leaf;
39279cc3
CM		 3520 int ret;
3521
3522 path = btrfs_alloc_path();
16cdcec7
MX		 3523 if (!path)
3524 return -ENOMEM;
3525
b9473439 3526 path->leave_spinning = 1;
16cdcec7
MX		 3527 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3528 1);
39279cc3
CM		 3529 if (ret) {
3530 if (ret > 0)
3531 ret = -ENOENT;
3532 goto failed;
3533 }
3534
b4ce94de 3535 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3536 leaf = path->nodes[0];
3537 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3538 struct btrfs_inode_item);
39279cc3 3539
e02119d5 3540 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3541 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3542 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3543 ret = 0;
3544failed:
39279cc3
CM		 3545 btrfs_free_path(path);
3546 return ret;
3547}
3548
2115133f
CM		 3549/*
3550 * copy everything in the in-memory inode into the btree.
3551 */
3552noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3553 struct btrfs_root *root, struct inode *inode)
3554{
3555 int ret;
3556
3557 /*
3558 * If the inode is a free space inode, we can deadlock during commit
3559 * if we put it into the delayed code.
3560 *
3561 * The data relocation inode should also be directly updated
3562 * without delay
3563 */
83eea1f1 3564 if (!btrfs_is_free_space_inode(inode)
2115133f 3565 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3566 btrfs_update_root_times(trans, root);
3567
2115133f
CM		 3568 ret = btrfs_delayed_update_inode(trans, root, inode);
3569 if (!ret)
3570 btrfs_set_inode_last_trans(trans, inode);
3571 return ret;
3572 }
3573
3574 return btrfs_update_inode_item(trans, root, inode);
3575}
3576
be6aef60
JB		 3577noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3578 struct btrfs_root *root,
3579 struct inode *inode)
2115133f
CM		 3580{
3581 int ret;
3582
3583 ret = btrfs_update_inode(trans, root, inode);
3584 if (ret == -ENOSPC)
3585 return btrfs_update_inode_item(trans, root, inode);
3586 return ret;
3587}
3588
d352ac68
CM		 3589/*
3590 * unlink helper that gets used here in inode.c and in the tree logging
3591 * recovery code. It remove a link in a directory with a given name, and
3592 * also drops the back refs in the inode to the directory
3593 */
92986796
AV		 3594static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3595 struct btrfs_root *root,
3596 struct inode *dir, struct inode *inode,
3597 const char *name, int name_len)
39279cc3
CM		 3598{
3599 struct btrfs_path *path;
39279cc3 3600 int ret = 0;
5f39d397 3601 struct extent_buffer *leaf;
39279cc3 3602 struct btrfs_dir_item *di;
5f39d397 3603 struct btrfs_key key;
aec7477b 3604 u64 index;
33345d01
LZ		 3605 u64 ino = btrfs_ino(inode);
3606 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3607
3608 path = btrfs_alloc_path();
54aa1f4d
CM		 3609 if (!path) {
3610 ret = -ENOMEM;
554233a6 3611 goto out;
54aa1f4d
CM		 3612 }
3613
b9473439 3614 path->leave_spinning = 1;
33345d01 3615 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3616 name, name_len, -1);
3617 if (IS_ERR(di)) {
3618 ret = PTR_ERR(di);
3619 goto err;
3620 }
3621 if (!di) {
3622 ret = -ENOENT;
3623 goto err;
3624 }
5f39d397
CM		 3625 leaf = path->nodes[0];
3626 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3627 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3628 if (ret)
3629 goto err;
b3b4aa74 3630 btrfs_release_path(path);
39279cc3 3631
33345d01
LZ		 3632 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3633 dir_ino, &index);
aec7477b 3634 if (ret) {
c2cf52eb
SK		 3635 btrfs_info(root->fs_info,
3636 "failed to delete reference to %.*s, inode %llu parent %llu",
3637 name_len, name,
3638 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 3639 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3640 goto err;
3641 }
3642
16cdcec7 3643 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3644 if (ret) {
3645 btrfs_abort_transaction(trans, root, ret);
39279cc3 3646 goto err;
79787eaa 3647 }
39279cc3 3648
e02119d5 3649 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3650 inode, dir_ino);
79787eaa
JM		 3651 if (ret != 0 && ret != -ENOENT) {
3652 btrfs_abort_transaction(trans, root, ret);
3653 goto err;
3654 }
e02119d5
CM		 3655
3656 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3657 dir, index);
6418c961
CM		 3658 if (ret == -ENOENT)
3659 ret = 0;
d4e3991b
ZB		 3660 else if (ret)
3661 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3662err:
3663 btrfs_free_path(path);
e02119d5
CM		 3664 if (ret)
3665 goto out;
3666
3667 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3668 inode_inc_iversion(inode);
3669 inode_inc_iversion(dir);
e02119d5 3670 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3671 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3672out:
39279cc3
CM		 3673 return ret;
3674}
3675
92986796
AV		 3676int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3677 struct btrfs_root *root,
3678 struct inode *dir, struct inode *inode,
3679 const char *name, int name_len)
3680{
3681 int ret;
3682 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3683 if (!ret) {
3684 btrfs_drop_nlink(inode);
3685 ret = btrfs_update_inode(trans, root, inode);
3686 }
3687 return ret;
3688}
39279cc3 3689
a22285a6
YZ		 3690/*
3691 * helper to start transaction for unlink and rmdir.
3692 *
d52be818
JB		 3693 * unlink and rmdir are special in btrfs, they do not always free space, so
3694 * if we cannot make our reservations the normal way try and see if there is
3695 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3696 * allow the unlink to occur.
a22285a6 3697 */
d52be818 3698static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3699{
39279cc3 3700 struct btrfs_trans_handle *trans;
a22285a6 3701 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3702 int ret;
3703
e70bea5f
JB		 3704 /*
3705 * 1 for the possible orphan item
3706 * 1 for the dir item
3707 * 1 for the dir index
3708 * 1 for the inode ref
e70bea5f
JB		 3709 * 1 for the inode
3710 */
6e137ed3 3711 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3712 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3713 return trans;
4df27c4d 3714
d52be818
JB		 3715 if (PTR_ERR(trans) == -ENOSPC) {
3716 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
a22285a6 3717
d52be818
JB		 3718 trans = btrfs_start_transaction(root, 0);
3719 if (IS_ERR(trans))
3720 return trans;
3721 ret = btrfs_cond_migrate_bytes(root->fs_info,
3722 &root->fs_info->trans_block_rsv,
3723 num_bytes, 5);
3724 if (ret) {
3725 btrfs_end_transaction(trans, root);
3726 return ERR_PTR(ret);
a22285a6 3727 }
5a77d76c 3728 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3729 trans->bytes_reserved = num_bytes;
a22285a6 3730 }
d52be818 3731 return trans;
a22285a6
YZ		 3732}
3733
3734static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3735{
3736 struct btrfs_root *root = BTRFS_I(dir)->root;
3737 struct btrfs_trans_handle *trans;
3738 struct inode *inode = dentry->d_inode;
3739 int ret;
a22285a6 3740
d52be818 3741 trans = __unlink_start_trans(dir);
a22285a6
YZ		 3742 if (IS_ERR(trans))
3743 return PTR_ERR(trans);
5f39d397 3744
12fcfd22
CM		 3745 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3746
e02119d5
CM		 3747 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3748 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3749 if (ret)
3750 goto out;
7b128766 3751
a22285a6 3752 if (inode->i_nlink == 0) {
7b128766 3753 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3754 if (ret)
3755 goto out;
a22285a6 3756 }
7b128766 3757
b532402e 3758out:
d52be818 3759 btrfs_end_transaction(trans, root);
b53d3f5d 3760 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3761 return ret;
3762}
3763
4df27c4d
YZ		 3764int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3765 struct btrfs_root *root,
3766 struct inode *dir, u64 objectid,
3767 const char *name, int name_len)
3768{
3769 struct btrfs_path *path;
3770 struct extent_buffer *leaf;
3771 struct btrfs_dir_item *di;
3772 struct btrfs_key key;
3773 u64 index;
3774 int ret;
33345d01 3775 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3776
3777 path = btrfs_alloc_path();
3778 if (!path)
3779 return -ENOMEM;
3780
33345d01 3781 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3782 name, name_len, -1);
79787eaa
JM		 3783 if (IS_ERR_OR_NULL(di)) {
3784 if (!di)
3785 ret = -ENOENT;
3786 else
3787 ret = PTR_ERR(di);
3788 goto out;
3789 }
4df27c4d
YZ		 3790
3791 leaf = path->nodes[0];
3792 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3793 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3794 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3795 if (ret) {
3796 btrfs_abort_transaction(trans, root, ret);
3797 goto out;
3798 }
b3b4aa74 3799 btrfs_release_path(path);
4df27c4d
YZ		 3800
3801 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3802 objectid, root->root_key.objectid,
33345d01 3803 dir_ino, &index, name, name_len);
4df27c4d 3804 if (ret < 0) {
79787eaa
JM		 3805 if (ret != -ENOENT) {
3806 btrfs_abort_transaction(trans, root, ret);
3807 goto out;
3808 }
33345d01 3809 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3810 name, name_len);
79787eaa
JM		 3811 if (IS_ERR_OR_NULL(di)) {
3812 if (!di)
3813 ret = -ENOENT;
3814 else
3815 ret = PTR_ERR(di);
3816 btrfs_abort_transaction(trans, root, ret);
3817 goto out;
3818 }
4df27c4d
YZ		 3819
3820 leaf = path->nodes[0];
3821 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3822 btrfs_release_path(path);
4df27c4d
YZ		 3823 index = key.offset;
3824 }
945d8962 3825 btrfs_release_path(path);
4df27c4d 3826
16cdcec7 3827 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3828 if (ret) {
3829 btrfs_abort_transaction(trans, root, ret);
3830 goto out;
3831 }
4df27c4d
YZ		 3832
3833 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3834 inode_inc_iversion(dir);
4df27c4d 3835 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3836 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3837 if (ret)
3838 btrfs_abort_transaction(trans, root, ret);
3839out:
71d7aed0 3840 btrfs_free_path(path);
79787eaa 3841 return ret;
4df27c4d
YZ		 3842}
3843
39279cc3
CM		 3844static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3845{
3846 struct inode *inode = dentry->d_inode;
1832a6d5 3847 int err = 0;
39279cc3 3848 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3849 struct btrfs_trans_handle *trans;
39279cc3 3850
b3ae244e 3851 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3852 return -ENOTEMPTY;
b3ae244e
DS		 3853 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3854 return -EPERM;
134d4512 3855
d52be818 3856 trans = __unlink_start_trans(dir);
a22285a6 3857 if (IS_ERR(trans))
5df6a9f6 3858 return PTR_ERR(trans);
5df6a9f6 3859
33345d01 3860 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3861 err = btrfs_unlink_subvol(trans, root, dir,
3862 BTRFS_I(inode)->location.objectid,
3863 dentry->d_name.name,
3864 dentry->d_name.len);
3865 goto out;
3866 }
3867
7b128766
JB		 3868 err = btrfs_orphan_add(trans, inode);
3869 if (err)
4df27c4d 3870 goto out;
7b128766 3871
39279cc3 3872 /* now the directory is empty */
e02119d5
CM		 3873 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3874 dentry->d_name.name, dentry->d_name.len);
d397712b 3875 if (!err)
dbe674a9 3876 btrfs_i_size_write(inode, 0);
4df27c4d 3877out:
d52be818 3878 btrfs_end_transaction(trans, root);
b53d3f5d 3879 btrfs_btree_balance_dirty(root);
3954401f 3880
39279cc3
CM		 3881 return err;
3882}
3883
39279cc3
CM		 3884/*
3885 * this can truncate away extent items, csum items and directory items.
3886 * It starts at a high offset and removes keys until it can't find
d352ac68 3887 * any higher than new_size
39279cc3
CM		 3888 *
3889 * csum items that cross the new i_size are truncated to the new size
3890 * as well.
7b128766
JB		 3891 *
3892 * min_type is the minimum key type to truncate down to. If set to 0, this
3893 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3894 */
8082510e
YZ		 3895int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3896 struct btrfs_root *root,
3897 struct inode *inode,
3898 u64 new_size, u32 min_type)
39279cc3 3899{
39279cc3 3900 struct btrfs_path *path;
5f39d397 3901 struct extent_buffer *leaf;
39279cc3 3902 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3903 struct btrfs_key key;
3904 struct btrfs_key found_key;
39279cc3 3905 u64 extent_start = 0;
db94535d 3906 u64 extent_num_bytes = 0;
5d4f98a2 3907 u64 extent_offset = 0;
39279cc3 3908 u64 item_end = 0;
8082510e 3909 u32 found_type = (u8)-1;
39279cc3
CM		 3910 int found_extent;
3911 int del_item;
85e21bac
CM		 3912 int pending_del_nr = 0;
3913 int pending_del_slot = 0;
179e29e4 3914 int extent_type = -1;
8082510e
YZ		 3915 int ret;
3916 int err = 0;
33345d01 3917 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3918
3919 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3920
0eb0e19c
MF		 3921 path = btrfs_alloc_path();
3922 if (!path)
3923 return -ENOMEM;
3924 path->reada = -1;
3925
5dc562c5
JB		 3926 /*
3927 * We want to drop from the next block forward in case this new size is
3928 * not block aligned since we will be keeping the last block of the
3929 * extent just the way it is.
3930 */
0af3d00b 3931 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 3932 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3933 root->sectorsize), (u64)-1, 0);
8082510e 3934
16cdcec7
MX		 3935 /*
3936 * This function is also used to drop the items in the log tree before
3937 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3938 * it is used to drop the loged items. So we shouldn't kill the delayed
3939 * items.
3940 */
3941 if (min_type == 0 && root == BTRFS_I(inode)->root)
3942 btrfs_kill_delayed_inode_items(inode);
3943
33345d01 3944 key.objectid = ino;
39279cc3 3945 key.offset = (u64)-1;
5f39d397
CM		 3946 key.type = (u8)-1;
3947
85e21bac 3948search_again:
b9473439 3949 path->leave_spinning = 1;
85e21bac 3950 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3951 if (ret < 0) {
3952 err = ret;
3953 goto out;
3954 }
d397712b 3955
85e21bac 3956 if (ret > 0) {
e02119d5
CM		 3957 /* there are no items in the tree for us to truncate, we're
3958 * done
3959 */
8082510e
YZ		 3960 if (path->slots[0] == 0)
3961 goto out;
85e21bac
CM		 3962 path->slots[0]--;
3963 }
3964
d397712b 3965 while (1) {
39279cc3 3966 fi = NULL;
5f39d397
CM		 3967 leaf = path->nodes[0];
3968 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3969 found_type = btrfs_key_type(&found_key);
39279cc3 3970
33345d01 3971 if (found_key.objectid != ino)
39279cc3 3972 break;
5f39d397 3973
85e21bac 3974 if (found_type < min_type)
39279cc3
CM		 3975 break;
3976
5f39d397 3977 item_end = found_key.offset;
39279cc3 3978 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3979 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3980 struct btrfs_file_extent_item);
179e29e4
CM		 3981 extent_type = btrfs_file_extent_type(leaf, fi);
3982 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3983 item_end +=
db94535d 3984 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3985 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3986 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3987 fi);
39279cc3 3988 }
008630c1 3989 item_end--;
39279cc3 3990 }
8082510e
YZ		 3991 if (found_type > min_type) {
3992 del_item = 1;
3993 } else {
3994 if (item_end < new_size)
b888db2b 3995 break;
8082510e
YZ		 3996 if (found_key.offset >= new_size)
3997 del_item = 1;
3998 else
3999 del_item = 0;
39279cc3 4000 }
39279cc3 4001 found_extent = 0;
39279cc3 4002 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4003 if (found_type != BTRFS_EXTENT_DATA_KEY)
4004 goto delete;
4005
4006 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4007 u64 num_dec;
db94535d 4008 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4009 if (!del_item) {
db94535d
CM		 4010 u64 orig_num_bytes =
4011 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4012 extent_num_bytes = ALIGN(new_size -
4013 found_key.offset,
4014 root->sectorsize);
db94535d
CM		 4015 btrfs_set_file_extent_num_bytes(leaf, fi,
4016 extent_num_bytes);
4017 num_dec = (orig_num_bytes -
9069218d 4018 extent_num_bytes);
e02119d5 4019 if (root->ref_cows && extent_start != 0)
a76a3cd4 4020 inode_sub_bytes(inode, num_dec);
5f39d397 4021 btrfs_mark_buffer_dirty(leaf);
39279cc3 4022 } else {
db94535d
CM		 4023 extent_num_bytes =
4024 btrfs_file_extent_disk_num_bytes(leaf,
4025 fi);
5d4f98a2
YZ		 4026 extent_offset = found_key.offset -
4027 btrfs_file_extent_offset(leaf, fi);
4028
39279cc3 4029 /* FIXME blocksize != 4096 */
9069218d 4030 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4031 if (extent_start != 0) {
4032 found_extent = 1;
e02119d5 4033 if (root->ref_cows)
a76a3cd4 4034 inode_sub_bytes(inode, num_dec);
e02119d5 4035 }
39279cc3 4036 }
9069218d 4037 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4038 /*
4039 * we can't truncate inline items that have had
4040 * special encodings
4041 */
4042 if (!del_item &&
4043 btrfs_file_extent_compression(leaf, fi) == 0 &&
4044 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4045 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4046 u32 size = new_size - found_key.offset;
4047
4048 if (root->ref_cows) {
a76a3cd4
YZ		 4049 inode_sub_bytes(inode, item_end + 1 -
4050 new_size);
e02119d5
CM		 4051 }
4052 size =
4053 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4054 btrfs_truncate_item(root, path, size, 1);
e02119d5 4055 } else if (root->ref_cows) {
a76a3cd4
YZ		 4056 inode_sub_bytes(inode, item_end + 1 -
4057 found_key.offset);
9069218d 4058 }
39279cc3 4059 }
179e29e4 4060delete:
39279cc3 4061 if (del_item) {
85e21bac
CM		 4062 if (!pending_del_nr) {
4063 /* no pending yet, add ourselves */
4064 pending_del_slot = path->slots[0];
4065 pending_del_nr = 1;
4066 } else if (pending_del_nr &&
4067 path->slots[0] + 1 == pending_del_slot) {
4068 /* hop on the pending chunk */
4069 pending_del_nr++;
4070 pending_del_slot = path->slots[0];
4071 } else {
d397712b 4072 BUG();
85e21bac 4073 }
39279cc3
CM		 4074 } else {
4075 break;
4076 }
0af3d00b
JB		 4077 if (found_extent && (root->ref_cows ||
4078 root == root->fs_info->tree_root)) {
b9473439 4079 btrfs_set_path_blocking(path);
39279cc3 4080 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4081 extent_num_bytes, 0,
4082 btrfs_header_owner(leaf),
66d7e7f0 4083 ino, extent_offset, 0);
39279cc3
CM		 4084 BUG_ON(ret);
4085 }
85e21bac 4086
8082510e
YZ		 4087 if (found_type == BTRFS_INODE_ITEM_KEY)
4088 break;
4089
4090 if (path->slots[0] == 0 ||
4091 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4092 if (pending_del_nr) {
4093 ret = btrfs_del_items(trans, root, path,
4094 pending_del_slot,
4095 pending_del_nr);
79787eaa
JM		 4096 if (ret) {
4097 btrfs_abort_transaction(trans,
4098 root, ret);
4099 goto error;
4100 }
8082510e
YZ		 4101 pending_del_nr = 0;
4102 }
b3b4aa74 4103 btrfs_release_path(path);
85e21bac 4104 goto search_again;
8082510e
YZ		 4105 } else {
4106 path->slots[0]--;
85e21bac 4107 }
39279cc3 4108 }
8082510e 4109out:
85e21bac
CM		 4110 if (pending_del_nr) {
4111 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4112 pending_del_nr);
79787eaa
JM		 4113 if (ret)
4114 btrfs_abort_transaction(trans, root, ret);
85e21bac 4115 }
79787eaa 4116error:
39279cc3 4117 btrfs_free_path(path);
8082510e 4118 return err;
39279cc3
CM		 4119}
4120
4121/*
2aaa6655
JB		 4122 * btrfs_truncate_page - read, zero a chunk and write a page
4123 * @inode - inode that we're zeroing
4124 * @from - the offset to start zeroing
4125 * @len - the length to zero, 0 to zero the entire range respective to the
4126 * offset
4127 * @front - zero up to the offset instead of from the offset on
4128 *
4129 * This will find the page for the "from" offset and cow the page and zero the
4130 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4131 */
2aaa6655
JB		 4132int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4133 int front)
39279cc3 4134{
2aaa6655 4135 struct address_space *mapping = inode->i_mapping;
db94535d 4136 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4137 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4138 struct btrfs_ordered_extent *ordered;
2ac55d41 4139 struct extent_state *cached_state = NULL;
e6dcd2dc 4140 char *kaddr;
db94535d 4141 u32 blocksize = root->sectorsize;
39279cc3
CM		 4142 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4143 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4144 struct page *page;
3b16a4e3 4145 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4146 int ret = 0;
a52d9a80 4147 u64 page_start;
e6dcd2dc 4148 u64 page_end;
39279cc3 4149
2aaa6655
JB		 4150 if ((offset & (blocksize - 1)) == 0 &&
4151 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4152 goto out;
0ca1f7ce 4153 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4154 if (ret)
4155 goto out;
39279cc3 4156
211c17f5 4157again:
3b16a4e3 4158 page = find_or_create_page(mapping, index, mask);
5d5e103a 4159 if (!page) {
0ca1f7ce 4160 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4161 ret = -ENOMEM;
39279cc3 4162 goto out;
5d5e103a 4163 }
e6dcd2dc
CM		 4164
4165 page_start = page_offset(page);
4166 page_end = page_start + PAGE_CACHE_SIZE - 1;
4167
39279cc3 4168 if (!PageUptodate(page)) {
9ebefb18 4169 ret = btrfs_readpage(NULL, page);
39279cc3 4170 lock_page(page);
211c17f5
CM		 4171 if (page->mapping != mapping) {
4172 unlock_page(page);
4173 page_cache_release(page);
4174 goto again;
4175 }
39279cc3
CM		 4176 if (!PageUptodate(page)) {
4177 ret = -EIO;
89642229 4178 goto out_unlock;
39279cc3
CM		 4179 }
4180 }
211c17f5 4181 wait_on_page_writeback(page);
e6dcd2dc 4182
d0082371 4183 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4184 set_page_extent_mapped(page);
4185
4186 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4187 if (ordered) {
2ac55d41
JB		 4188 unlock_extent_cached(io_tree, page_start, page_end,
4189 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4190 unlock_page(page);
4191 page_cache_release(page);
eb84ae03 4192 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4193 btrfs_put_ordered_extent(ordered);
4194 goto again;
4195 }
4196
2ac55d41 4197 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4198 EXTENT_DIRTY | EXTENT_DELALLOC |
4199 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4200 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4201
2ac55d41
JB		 4202 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4203 &cached_state);
9ed74f2d 4204 if (ret) {
2ac55d41
JB		 4205 unlock_extent_cached(io_tree, page_start, page_end,
4206 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4207 goto out_unlock;
4208 }
4209
e6dcd2dc 4210 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4211 if (!len)
4212 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4213 kaddr = kmap(page);
2aaa6655
JB		 4214 if (front)
4215 memset(kaddr, 0, offset);
4216 else
4217 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4218 flush_dcache_page(page);
4219 kunmap(page);
4220 }
247e743c 4221 ClearPageChecked(page);
e6dcd2dc 4222 set_page_dirty(page);
2ac55d41
JB		 4223 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4224 GFP_NOFS);
39279cc3 4225
89642229 4226out_unlock:
5d5e103a 4227 if (ret)
0ca1f7ce 4228 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4229 unlock_page(page);
4230 page_cache_release(page);
4231out:
4232 return ret;
4233}
4234
695a0d0d
JB		 4235/*
4236 * This function puts in dummy file extents for the area we're creating a hole
4237 * for. So if we are truncating this file to a larger size we need to insert
4238 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4239 * the range between oldsize and size
4240 */
a41ad394 4241int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4242{
9036c102
YZ		 4243 struct btrfs_trans_handle *trans;
4244 struct btrfs_root *root = BTRFS_I(inode)->root;
4245 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4246 struct extent_map *em = NULL;
2ac55d41 4247 struct extent_state *cached_state = NULL;
5dc562c5 4248 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4249 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4250 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4251 u64 last_byte;
4252 u64 cur_offset;
4253 u64 hole_size;
9ed74f2d 4254 int err = 0;
39279cc3 4255
a71754fc
JB		 4256 /*
4257 * If our size started in the middle of a page we need to zero out the
4258 * rest of the page before we expand the i_size, otherwise we could
4259 * expose stale data.
4260 */
4261 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4262 if (err)
4263 return err;
4264
9036c102
YZ		 4265 if (size <= hole_start)
4266 return 0;
4267
9036c102
YZ		 4268 while (1) {
4269 struct btrfs_ordered_extent *ordered;
4270 btrfs_wait_ordered_range(inode, hole_start,
4271 block_end - hole_start);
2ac55d41 4272 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4273 &cached_state);
9036c102
YZ		 4274 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4275 if (!ordered)
4276 break;
2ac55d41
JB		 4277 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4278 &cached_state, GFP_NOFS);
9036c102
YZ		 4279 btrfs_put_ordered_extent(ordered);
4280 }
39279cc3 4281
9036c102
YZ		 4282 cur_offset = hole_start;
4283 while (1) {
4284 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4285 block_end - cur_offset, 0);
79787eaa
JM		 4286 if (IS_ERR(em)) {
4287 err = PTR_ERR(em);
f2767956 4288 em = NULL;
79787eaa
JM		 4289 break;
4290 }
9036c102 4291 last_byte = min(extent_map_end(em), block_end);
fda2832f 4292 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4293 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4294 struct extent_map *hole_em;
9036c102 4295 hole_size = last_byte - cur_offset;
9ed74f2d 4296
3642320e 4297 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4298 if (IS_ERR(trans)) {
4299 err = PTR_ERR(trans);
9ed74f2d 4300 break;
a22285a6 4301 }
8082510e 4302
5dc562c5
JB		 4303 err = btrfs_drop_extents(trans, root, inode,
4304 cur_offset,
2671485d 4305 cur_offset + hole_size, 1);
5b397377 4306 if (err) {
79787eaa 4307 btrfs_abort_transaction(trans, root, err);
5b397377 4308 btrfs_end_transaction(trans, root);
3893e33b 4309 break;
5b397377 4310 }
8082510e 4311
9036c102 4312 err = btrfs_insert_file_extent(trans, root,
33345d01 4313 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4314 0, hole_size, 0, hole_size,
4315 0, 0, 0);
5b397377 4316 if (err) {
79787eaa 4317 btrfs_abort_transaction(trans, root, err);
5b397377 4318 btrfs_end_transaction(trans, root);
3893e33b 4319 break;
5b397377 4320 }
8082510e 4321
5dc562c5
JB		 4322 btrfs_drop_extent_cache(inode, cur_offset,
4323 cur_offset + hole_size - 1, 0);
4324 hole_em = alloc_extent_map();
4325 if (!hole_em) {
4326 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4327 &BTRFS_I(inode)->runtime_flags);
4328 goto next;
4329 }
4330 hole_em->start = cur_offset;
4331 hole_em->len = hole_size;
4332 hole_em->orig_start = cur_offset;
8082510e 4333
5dc562c5
JB		 4334 hole_em->block_start = EXTENT_MAP_HOLE;
4335 hole_em->block_len = 0;
b4939680 4336 hole_em->orig_block_len = 0;
cc95bef6 4337 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4338 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4339 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4340 hole_em->generation = trans->transid;
8082510e 4341
5dc562c5
JB		 4342 while (1) {
4343 write_lock(&em_tree->lock);
09a2a8f9 4344 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4345 write_unlock(&em_tree->lock);
4346 if (err != -EEXIST)
4347 break;
4348 btrfs_drop_extent_cache(inode, cur_offset,
4349 cur_offset +
4350 hole_size - 1, 0);
4351 }
4352 free_extent_map(hole_em);
4353next:
3642320e 4354 btrfs_update_inode(trans, root, inode);
8082510e 4355 btrfs_end_transaction(trans, root);
9036c102
YZ		 4356 }
4357 free_extent_map(em);
a22285a6 4358 em = NULL;
9036c102 4359 cur_offset = last_byte;
8082510e 4360 if (cur_offset >= block_end)
9036c102
YZ		 4361 break;
4362 }
1832a6d5 4363
a22285a6 4364 free_extent_map(em);
2ac55d41
JB		 4365 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4366 GFP_NOFS);
9036c102
YZ		 4367 return err;
4368}
39279cc3 4369
3972f260 4370static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4371{
f4a2f4c5
MX		 4372 struct btrfs_root *root = BTRFS_I(inode)->root;
4373 struct btrfs_trans_handle *trans;
a41ad394 4374 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4375 loff_t newsize = attr->ia_size;
4376 int mask = attr->ia_valid;
8082510e
YZ		 4377 int ret;
4378
a41ad394 4379 if (newsize == oldsize)
8082510e
YZ		 4380 return 0;
4381
3972f260
ES		 4382 /*
4383 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4384 * special case where we need to update the times despite not having
4385 * these flags set. For all other operations the VFS set these flags
4386 * explicitly if it wants a timestamp update.
4387 */
4388 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4389 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4390
a41ad394 4391 if (newsize > oldsize) {
a41ad394
JB		 4392 truncate_pagecache(inode, oldsize, newsize);
4393 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4394 if (ret)
8082510e 4395 return ret;
8082510e 4396
f4a2f4c5
MX		 4397 trans = btrfs_start_transaction(root, 1);
4398 if (IS_ERR(trans))
4399 return PTR_ERR(trans);
4400
4401 i_size_write(inode, newsize);
4402 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4403 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4404 btrfs_end_transaction(trans, root);
a41ad394 4405 } else {
8082510e 4406
a41ad394
JB		 4407 /*
4408 * We're truncating a file that used to have good data down to
4409 * zero. Make sure it gets into the ordered flush list so that
4410 * any new writes get down to disk quickly.
4411 */
4412 if (newsize == 0)
72ac3c0d
JB		 4413 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4414 &BTRFS_I(inode)->runtime_flags);
8082510e 4415
f3fe820c
JB		 4416 /*
4417 * 1 for the orphan item we're going to add
4418 * 1 for the orphan item deletion.
4419 */
4420 trans = btrfs_start_transaction(root, 2);
4421 if (IS_ERR(trans))
4422 return PTR_ERR(trans);
4423
4424 /*
4425 * We need to do this in case we fail at _any_ point during the
4426 * actual truncate. Once we do the truncate_setsize we could
4427 * invalidate pages which forces any outstanding ordered io to
4428 * be instantly completed which will give us extents that need
4429 * to be truncated. If we fail to get an orphan inode down we
4430 * could have left over extents that were never meant to live,
4431 * so we need to garuntee from this point on that everything
4432 * will be consistent.
4433 */
4434 ret = btrfs_orphan_add(trans, inode);
4435 btrfs_end_transaction(trans, root);
4436 if (ret)
4437 return ret;
4438
a41ad394
JB		 4439 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4440 truncate_setsize(inode, newsize);
2e60a51e
MX		 4441
4442 /* Disable nonlocked read DIO to avoid the end less truncate */
4443 btrfs_inode_block_unlocked_dio(inode);
4444 inode_dio_wait(inode);
4445 btrfs_inode_resume_unlocked_dio(inode);
4446
a41ad394 4447 ret = btrfs_truncate(inode);
f3fe820c
JB		 4448 if (ret && inode->i_nlink)
4449 btrfs_orphan_del(NULL, inode);
8082510e
YZ		 4450 }
4451
a41ad394 4452 return ret;
8082510e
YZ		 4453}
4454
9036c102
YZ		 4455static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4456{
4457 struct inode *inode = dentry->d_inode;
b83cc969 4458 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4459 int err;
39279cc3 4460
b83cc969
LZ		 4461 if (btrfs_root_readonly(root))
4462 return -EROFS;
4463
9036c102
YZ		 4464 err = inode_change_ok(inode, attr);
4465 if (err)
4466 return err;
2bf5a725 4467
5a3f23d5 4468 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4469 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4470 if (err)
4471 return err;
39279cc3 4472 }
9036c102 4473
1025774c
CH		 4474 if (attr->ia_valid) {
4475 setattr_copy(inode, attr);
0c4d2d95 4476 inode_inc_iversion(inode);
22c44fe6 4477 err = btrfs_dirty_inode(inode);
1025774c 4478
22c44fe6 4479 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4480 err = btrfs_acl_chmod(inode);
4481 }
33268eaf 4482
39279cc3
CM		 4483 return err;
4484}
61295eb8 4485
bd555975 4486void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4487{
4488 struct btrfs_trans_handle *trans;
4489 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4490 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4491 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4492 int ret;
4493
1abe9b8a 4494 trace_btrfs_inode_evict(inode);
4495
39279cc3 4496 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 4497 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 4498 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4499 goto no_delete;
4500
39279cc3 4501 if (is_bad_inode(inode)) {
7b128766 4502 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4503 goto no_delete;
4504 }
bd555975 4505 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4506 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4507
c71bf099 4508 if (root->fs_info->log_root_recovering) {
6bf02314 4509 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4510 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4511 goto no_delete;
4512 }
4513
76dda93c
YZ		 4514 if (inode->i_nlink > 0) {
4515 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4516 goto no_delete;
4517 }
4518
0e8c36a9
MX		 4519 ret = btrfs_commit_inode_delayed_inode(inode);
4520 if (ret) {
4521 btrfs_orphan_del(NULL, inode);
4522 goto no_delete;
4523 }
4524
66d8f3dd 4525 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4526 if (!rsv) {
4527 btrfs_orphan_del(NULL, inode);
4528 goto no_delete;
4529 }
4a338542 4530 rsv->size = min_size;
ca7e70f5 4531 rsv->failfast = 1;
726c35fa 4532 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4533
dbe674a9 4534 btrfs_i_size_write(inode, 0);
5f39d397 4535
4289a667 4536 /*
8407aa46
MX		 4537 * This is a bit simpler than btrfs_truncate since we've already
4538 * reserved our space for our orphan item in the unlink, so we just
4539 * need to reserve some slack space in case we add bytes and update
4540 * inode item when doing the truncate.
4289a667 4541 */
8082510e 4542 while (1) {
08e007d2
MX		 4543 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4544 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4545
4546 /*
4547 * Try and steal from the global reserve since we will
4548 * likely not use this space anyway, we want to try as
4549 * hard as possible to get this to work.
4550 */
4551 if (ret)
4552 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4553
d68fc57b 4554 if (ret) {
c2cf52eb
SK		 4555 btrfs_warn(root->fs_info,
4556 "Could not get space for a delete, will truncate on mount %d",
4557 ret);
4289a667
JB		 4558 btrfs_orphan_del(NULL, inode);
4559 btrfs_free_block_rsv(root, rsv);
4560 goto no_delete;
d68fc57b 4561 }
7b128766 4562
0e8c36a9 4563 trans = btrfs_join_transaction(root);
4289a667
JB		 4564 if (IS_ERR(trans)) {
4565 btrfs_orphan_del(NULL, inode);
4566 btrfs_free_block_rsv(root, rsv);
4567 goto no_delete;
d68fc57b 4568 }
7b128766 4569
4289a667
JB		 4570 trans->block_rsv = rsv;
4571
d68fc57b 4572 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4573 if (ret != -ENOSPC)
8082510e 4574 break;
85e21bac 4575
8407aa46 4576 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4577 btrfs_end_transaction(trans, root);
4578 trans = NULL;
b53d3f5d 4579 btrfs_btree_balance_dirty(root);
8082510e 4580 }
5f39d397 4581
4289a667
JB		 4582 btrfs_free_block_rsv(root, rsv);
4583
8082510e 4584 if (ret == 0) {
4289a667 4585 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 4586 ret = btrfs_orphan_del(trans, inode);
4587 BUG_ON(ret);
4588 }
54aa1f4d 4589
4289a667 4590 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4591 if (!(root == root->fs_info->tree_root ||
4592 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4593 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4594
54aa1f4d 4595 btrfs_end_transaction(trans, root);
b53d3f5d 4596 btrfs_btree_balance_dirty(root);
39279cc3 4597no_delete:
89042e5a 4598 btrfs_remove_delayed_node(inode);
dbd5768f 4599 clear_inode(inode);
8082510e 4600 return;
39279cc3
CM		 4601}
4602
4603/*
4604 * this returns the key found in the dir entry in the location pointer.
4605 * If no dir entries were found, location->objectid is 0.
4606 */
4607static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4608 struct btrfs_key *location)
4609{
4610 const char *name = dentry->d_name.name;
4611 int namelen = dentry->d_name.len;
4612 struct btrfs_dir_item *di;
4613 struct btrfs_path *path;
4614 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4615 int ret = 0;
39279cc3
CM		 4616
4617 path = btrfs_alloc_path();
d8926bb3
MF		 4618 if (!path)
4619 return -ENOMEM;
3954401f 4620
33345d01 4621 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4622 namelen, 0);
0d9f7f3e
Y		 4623 if (IS_ERR(di))
4624 ret = PTR_ERR(di);
d397712b 4625
c704005d 4626 if (IS_ERR_OR_NULL(di))
3954401f 4627 goto out_err;
d397712b 4628
5f39d397 4629 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4630out:
39279cc3
CM		 4631 btrfs_free_path(path);
4632 return ret;
3954401f
CM		 4633out_err:
4634 location->objectid = 0;
4635 goto out;
39279cc3
CM		 4636}
4637
4638/*
4639 * when we hit a tree root in a directory, the btrfs part of the inode
4640 * needs to be changed to reflect the root directory of the tree root. This
4641 * is kind of like crossing a mount point.
4642 */
4643static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4644 struct inode *dir,
4645 struct dentry *dentry,
4646 struct btrfs_key *location,
4647 struct btrfs_root **sub_root)
39279cc3 4648{
4df27c4d
YZ		 4649 struct btrfs_path *path;
4650 struct btrfs_root *new_root;
4651 struct btrfs_root_ref *ref;
4652 struct extent_buffer *leaf;
4653 int ret;
4654 int err = 0;
39279cc3 4655
4df27c4d
YZ		 4656 path = btrfs_alloc_path();
4657 if (!path) {
4658 err = -ENOMEM;
4659 goto out;
4660 }
39279cc3 4661
4df27c4d
YZ		 4662 err = -ENOENT;
4663 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4664 BTRFS_I(dir)->root->root_key.objectid,
4665 location->objectid);
4666 if (ret) {
4667 if (ret < 0)
4668 err = ret;
4669 goto out;
4670 }
39279cc3 4671
4df27c4d
YZ		 4672 leaf = path->nodes[0];
4673 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4674 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4675 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4676 goto out;
39279cc3 4677
4df27c4d
YZ		 4678 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4679 (unsigned long)(ref + 1),
4680 dentry->d_name.len);
4681 if (ret)
4682 goto out;
4683
b3b4aa74 4684 btrfs_release_path(path);
4df27c4d
YZ		 4685
4686 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4687 if (IS_ERR(new_root)) {
4688 err = PTR_ERR(new_root);
4689 goto out;
4690 }
4691
4df27c4d
YZ		 4692 *sub_root = new_root;
4693 location->objectid = btrfs_root_dirid(&new_root->root_item);
4694 location->type = BTRFS_INODE_ITEM_KEY;
4695 location->offset = 0;
4696 err = 0;
4697out:
4698 btrfs_free_path(path);
4699 return err;
39279cc3
CM		 4700}
4701
5d4f98a2
YZ		 4702static void inode_tree_add(struct inode *inode)
4703{
4704 struct btrfs_root *root = BTRFS_I(inode)->root;
4705 struct btrfs_inode *entry;
03e860bd
FNP		 4706 struct rb_node **p;
4707 struct rb_node *parent;
33345d01 4708 u64 ino = btrfs_ino(inode);
5d4f98a2 4709
1d3382cb 4710 if (inode_unhashed(inode))
76dda93c 4711 return;
e1409cef
MX		 4712again:
4713 parent = NULL;
5d4f98a2 4714 spin_lock(&root->inode_lock);
e1409cef 4715 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 4716 while (*p) {
4717 parent = *p;
4718 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4719
33345d01 4720 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4721 p = &parent->rb_left;
33345d01 4722 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4723 p = &parent->rb_right;
5d4f98a2
YZ		 4724 else {
4725 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4726 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 4727 rb_erase(parent, &root->inode_tree);
4728 RB_CLEAR_NODE(parent);
4729 spin_unlock(&root->inode_lock);
4730 goto again;
5d4f98a2
YZ		 4731 }
4732 }
4733 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4734 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4735 spin_unlock(&root->inode_lock);
4736}
4737
4738static void inode_tree_del(struct inode *inode)
4739{
4740 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4741 int empty = 0;
5d4f98a2 4742
03e860bd 4743 spin_lock(&root->inode_lock);
5d4f98a2 4744 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4745 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4746 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4747 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4748 }
03e860bd 4749 spin_unlock(&root->inode_lock);
76dda93c 4750
0af3d00b
JB		 4751 /*
4752 * Free space cache has inodes in the tree root, but the tree root has a
4753 * root_refs of 0, so this could end up dropping the tree root as a
4754 * snapshot, so we need the extra !root->fs_info->tree_root check to
4755 * make sure we don't drop it.
4756 */
4757 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4758 root != root->fs_info->tree_root) {
76dda93c
YZ		 4759 synchronize_srcu(&root->fs_info->subvol_srcu);
4760 spin_lock(&root->inode_lock);
4761 empty = RB_EMPTY_ROOT(&root->inode_tree);
4762 spin_unlock(&root->inode_lock);
4763 if (empty)
4764 btrfs_add_dead_root(root);
4765 }
4766}
4767
143bede5 4768void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4769{
4770 struct rb_node *node;
4771 struct rb_node *prev;
4772 struct btrfs_inode *entry;
4773 struct inode *inode;
4774 u64 objectid = 0;
4775
4776 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4777
4778 spin_lock(&root->inode_lock);
4779again:
4780 node = root->inode_tree.rb_node;
4781 prev = NULL;
4782 while (node) {
4783 prev = node;
4784 entry = rb_entry(node, struct btrfs_inode, rb_node);
4785
33345d01 4786 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4787 node = node->rb_left;
33345d01 4788 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4789 node = node->rb_right;
4790 else
4791 break;
4792 }
4793 if (!node) {
4794 while (prev) {
4795 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4796 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4797 node = prev;
4798 break;
4799 }
4800 prev = rb_next(prev);
4801 }
4802 }
4803 while (node) {
4804 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4805 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4806 inode = igrab(&entry->vfs_inode);
4807 if (inode) {
4808 spin_unlock(&root->inode_lock);
4809 if (atomic_read(&inode->i_count) > 1)
4810 d_prune_aliases(inode);
4811 /*
45321ac5 4812 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4813 * the inode cache when its usage count
4814 * hits zero.
4815 */
4816 iput(inode);
4817 cond_resched();
4818 spin_lock(&root->inode_lock);
4819 goto again;
4820 }
4821
4822 if (cond_resched_lock(&root->inode_lock))
4823 goto again;
4824
4825 node = rb_next(node);
4826 }
4827 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4828}
4829
e02119d5
CM		 4830static int btrfs_init_locked_inode(struct inode *inode, void *p)
4831{
4832 struct btrfs_iget_args *args = p;
4833 inode->i_ino = args->ino;
e02119d5 4834 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4835 return 0;
4836}
4837
4838static int btrfs_find_actor(struct inode *inode, void *opaque)
4839{
4840 struct btrfs_iget_args *args = opaque;
33345d01 4841 return args->ino == btrfs_ino(inode) &&
d397712b 4842 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4843}
4844
5d4f98a2
YZ		 4845static struct inode *btrfs_iget_locked(struct super_block *s,
4846 u64 objectid,
4847 struct btrfs_root *root)
39279cc3
CM		 4848{
4849 struct inode *inode;
4850 struct btrfs_iget_args args;
4851 args.ino = objectid;
4852 args.root = root;
4853
4854 inode = iget5_locked(s, objectid, btrfs_find_actor,
4855 btrfs_init_locked_inode,
4856 (void *)&args);
4857 return inode;
4858}
4859
1a54ef8c
BR		 4860/* Get an inode object given its location and corresponding root.
4861 * Returns in *is_new if the inode was read from disk
4862 */
4863struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4864 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4865{
4866 struct inode *inode;
4867
4868 inode = btrfs_iget_locked(s, location->objectid, root);
4869 if (!inode)
5d4f98a2 4870 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4871
4872 if (inode->i_state & I_NEW) {
4873 BTRFS_I(inode)->root = root;
4874 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4875 btrfs_read_locked_inode(inode);
1748f843
MF		 4876 if (!is_bad_inode(inode)) {
4877 inode_tree_add(inode);
4878 unlock_new_inode(inode);
4879 if (new)
4880 *new = 1;
4881 } else {
e0b6d65b
ST		 4882 unlock_new_inode(inode);
4883 iput(inode);
4884 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4885 }
4886 }
4887
1a54ef8c
BR		 4888 return inode;
4889}
4890
4df27c4d
YZ		 4891static struct inode *new_simple_dir(struct super_block *s,
4892 struct btrfs_key *key,
4893 struct btrfs_root *root)
4894{
4895 struct inode *inode = new_inode(s);
4896
4897 if (!inode)
4898 return ERR_PTR(-ENOMEM);
4899
4df27c4d
YZ		 4900 BTRFS_I(inode)->root = root;
4901 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4902 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4903
4904 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4905 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4906 inode->i_fop = &simple_dir_operations;
4907 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4908 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4909
4910 return inode;
4911}
4912
3de4586c 4913struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4914{
d397712b 4915 struct inode *inode;
4df27c4d 4916 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4917 struct btrfs_root *sub_root = root;
4918 struct btrfs_key location;
76dda93c 4919 int index;
b4aff1f8 4920 int ret = 0;
39279cc3
CM		 4921
4922 if (dentry->d_name.len > BTRFS_NAME_LEN)
4923 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4924
39e3c955 4925 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 4926 if (ret < 0)
4927 return ERR_PTR(ret);
5f39d397 4928
4df27c4d
YZ		 4929 if (location.objectid == 0)
4930 return NULL;
4931
4932 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4933 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4934 return inode;
4935 }
4936
4937 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4938
76dda93c 4939 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4940 ret = fixup_tree_root_location(root, dir, dentry,
4941 &location, &sub_root);
4942 if (ret < 0) {
4943 if (ret != -ENOENT)
4944 inode = ERR_PTR(ret);
4945 else
4946 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4947 } else {
73f73415 4948 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4949 }
76dda93c
YZ		 4950 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4951
34d19bad 4952 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4953 down_read(&root->fs_info->cleanup_work_sem);
4954 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4955 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4956 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 4957 if (ret) {
4958 iput(inode);
66b4ffd1 4959 inode = ERR_PTR(ret);
01cd3367 4960 }
c71bf099
YZ		 4961 }
4962
3de4586c
CM		 4963 return inode;
4964}
4965
fe15ce44 4966static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4967{
4968 struct btrfs_root *root;
848cce0d 4969 struct inode *inode = dentry->d_inode;
76dda93c 4970
848cce0d
LZ		 4971 if (!inode && !IS_ROOT(dentry))
4972 inode = dentry->d_parent->d_inode;
76dda93c 4973
848cce0d
LZ		 4974 if (inode) {
4975 root = BTRFS_I(inode)->root;
efefb143
YZ		 4976 if (btrfs_root_refs(&root->root_item) == 0)
4977 return 1;
848cce0d
LZ		 4978
4979 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4980 return 1;
efefb143 4981 }
76dda93c
YZ		 4982 return 0;
4983}
4984
b4aff1f8
JB		 4985static void btrfs_dentry_release(struct dentry *dentry)
4986{
4987 if (dentry->d_fsdata)
4988 kfree(dentry->d_fsdata);
4989}
4990
3de4586c 4991static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 4992 unsigned int flags)
3de4586c 4993{
a66e7cc6
JB		 4994 struct dentry *ret;
4995
4996 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 4997 return ret;
39279cc3
CM		 4998}
4999
16cdcec7 5000unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5001 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5002};
5003
cbdf5a24
DW		 5004static int btrfs_real_readdir(struct file *filp, void *dirent,
5005 filldir_t filldir)
39279cc3 5006{
496ad9aa 5007 struct inode *inode = file_inode(filp);
39279cc3
CM		 5008 struct btrfs_root *root = BTRFS_I(inode)->root;
5009 struct btrfs_item *item;
5010 struct btrfs_dir_item *di;
5011 struct btrfs_key key;
5f39d397 5012 struct btrfs_key found_key;
39279cc3 5013 struct btrfs_path *path;
16cdcec7
MX		 5014 struct list_head ins_list;
5015 struct list_head del_list;
39279cc3 5016 int ret;
5f39d397 5017 struct extent_buffer *leaf;
39279cc3 5018 int slot;
39279cc3
CM		 5019 unsigned char d_type;
5020 int over = 0;
5021 u32 di_cur;
5022 u32 di_total;
5023 u32 di_len;
5024 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5025 char tmp_name[32];
5026 char *name_ptr;
5027 int name_len;
16cdcec7 5028 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 5029
5030 /* FIXME, use a real flag for deciding about the key type */
5031 if (root->fs_info->tree_root == root)
5032 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5033
3954401f
CM		 5034 /* special case for "." */
5035 if (filp->f_pos == 0) {
3765fefa
HS		 5036 over = filldir(dirent, ".", 1,
5037 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 5038 if (over)
5039 return 0;
5040 filp->f_pos = 1;
5041 }
3954401f
CM		 5042 /* special case for .., just use the back ref */
5043 if (filp->f_pos == 1) {
5ecc7e5d 5044 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 5045 over = filldir(dirent, "..", 2,
3765fefa 5046 filp->f_pos, pino, DT_DIR);
3954401f 5047 if (over)
49593bfa 5048 return 0;
3954401f
CM		 5049 filp->f_pos = 2;
5050 }
49593bfa 5051 path = btrfs_alloc_path();
16cdcec7
MX		 5052 if (!path)
5053 return -ENOMEM;
ff5714cc 5054
026fd317 5055 path->reada = 1;
49593bfa 5056
16cdcec7
MX		 5057 if (key_type == BTRFS_DIR_INDEX_KEY) {
5058 INIT_LIST_HEAD(&ins_list);
5059 INIT_LIST_HEAD(&del_list);
5060 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5061 }
5062
39279cc3
CM		 5063 btrfs_set_key_type(&key, key_type);
5064 key.offset = filp->f_pos;
33345d01 5065 key.objectid = btrfs_ino(inode);
5f39d397 5066
39279cc3
CM		 5067 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5068 if (ret < 0)
5069 goto err;
49593bfa
DW		 5070
5071 while (1) {
5f39d397 5072 leaf = path->nodes[0];
39279cc3 5073 slot = path->slots[0];
b9e03af0
LZ		 5074 if (slot >= btrfs_header_nritems(leaf)) {
5075 ret = btrfs_next_leaf(root, path);
5076 if (ret < 0)
5077 goto err;
5078 else if (ret > 0)
5079 break;
5080 continue;
39279cc3 5081 }
3de4586c 5082
5f39d397
CM		 5083 item = btrfs_item_nr(leaf, slot);
5084 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5085
5086 if (found_key.objectid != key.objectid)
39279cc3 5087 break;
5f39d397 5088 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5089 break;
5f39d397 5090 if (found_key.offset < filp->f_pos)
b9e03af0 5091 goto next;
16cdcec7
MX		 5092 if (key_type == BTRFS_DIR_INDEX_KEY &&
5093 btrfs_should_delete_dir_index(&del_list,
5094 found_key.offset))
5095 goto next;
5f39d397
CM		 5096
5097 filp->f_pos = found_key.offset;
16cdcec7 5098 is_curr = 1;
49593bfa 5099
39279cc3
CM		 5100 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5101 di_cur = 0;
5f39d397 5102 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5103
5104 while (di_cur < di_total) {
5f39d397
CM		 5105 struct btrfs_key location;
5106
22a94d44
JB		 5107 if (verify_dir_item(root, leaf, di))
5108 break;
5109
5f39d397 5110 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5111 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5112 name_ptr = tmp_name;
5113 } else {
5114 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5115 if (!name_ptr) {
5116 ret = -ENOMEM;
5117 goto err;
5118 }
5f39d397
CM		 5119 }
5120 read_extent_buffer(leaf, name_ptr,
5121 (unsigned long)(di + 1), name_len);
5122
5123 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5124 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5125
fede766f 5126
3de4586c 5127 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5128 * skip it.
5129 *
5130 * In contrast to old kernels, we insert the snapshot's
5131 * dir item and dir index after it has been created, so
5132 * we won't find a reference to our own snapshot. We
5133 * still keep the following code for backward
5134 * compatibility.
3de4586c
CM		 5135 */
5136 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5137 location.objectid == root->root_key.objectid) {
5138 over = 0;
5139 goto skip;
5140 }
5f39d397 5141 over = filldir(dirent, name_ptr, name_len,
49593bfa 5142 found_key.offset, location.objectid,
39279cc3 5143 d_type);
5f39d397 5144
3de4586c 5145skip:
5f39d397
CM		 5146 if (name_ptr != tmp_name)
5147 kfree(name_ptr);
5148
39279cc3
CM		 5149 if (over)
5150 goto nopos;
5103e947 5151 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5152 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5153 di_cur += di_len;
5154 di = (struct btrfs_dir_item *)((char *)di + di_len);
5155 }
b9e03af0
LZ		 5156next:
5157 path->slots[0]++;
39279cc3 5158 }
49593bfa 5159
16cdcec7
MX		 5160 if (key_type == BTRFS_DIR_INDEX_KEY) {
5161 if (is_curr)
5162 filp->f_pos++;
5163 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5164 &ins_list);
5165 if (ret)
5166 goto nopos;
5167 }
5168
49593bfa 5169 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 5170 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 5171 /*
5172 * 32-bit glibc will use getdents64, but then strtol -
5173 * so the last number we can serve is this.
5174 */
5175 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 5176 else
5177 filp->f_pos++;
39279cc3
CM		 5178nopos:
5179 ret = 0;
5180err:
16cdcec7
MX		 5181 if (key_type == BTRFS_DIR_INDEX_KEY)
5182 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5183 btrfs_free_path(path);
39279cc3
CM		 5184 return ret;
5185}
5186
a9185b41 5187int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5188{
5189 struct btrfs_root *root = BTRFS_I(inode)->root;
5190 struct btrfs_trans_handle *trans;
5191 int ret = 0;
0af3d00b 5192 bool nolock = false;
39279cc3 5193
72ac3c0d 5194 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5195 return 0;
5196
83eea1f1 5197 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5198 nolock = true;
0af3d00b 5199
a9185b41 5200 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5201 if (nolock)
7a7eaa40 5202 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5203 else
7a7eaa40 5204 trans = btrfs_join_transaction(root);
3612b495
TI		 5205 if (IS_ERR(trans))
5206 return PTR_ERR(trans);
a698d075 5207 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5208 }
5209 return ret;
5210}
5211
5212/*
54aa1f4d 5213 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5214 * inode changes. But, it is most likely to find the inode in cache.
5215 * FIXME, needs more benchmarking...there are no reasons other than performance
5216 * to keep or drop this code.
5217 */
48a3b636 5218static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5219{
5220 struct btrfs_root *root = BTRFS_I(inode)->root;
5221 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5222 int ret;
5223
72ac3c0d 5224 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5225 return 0;
39279cc3 5226
7a7eaa40 5227 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5228 if (IS_ERR(trans))
5229 return PTR_ERR(trans);
8929ecfa
YZ		 5230
5231 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5232 if (ret && ret == -ENOSPC) {
5233 /* whoops, lets try again with the full transaction */
5234 btrfs_end_transaction(trans, root);
5235 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5236 if (IS_ERR(trans))
5237 return PTR_ERR(trans);
8929ecfa 5238
94b60442 5239 ret = btrfs_update_inode(trans, root, inode);
94b60442 5240 }
39279cc3 5241 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5242 if (BTRFS_I(inode)->delayed_node)
5243 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5244
5245 return ret;
5246}
5247
5248/*
5249 * This is a copy of file_update_time. We need this so we can return error on
5250 * ENOSPC for updating the inode in the case of file write and mmap writes.
5251 */
e41f941a
JB		 5252static int btrfs_update_time(struct inode *inode, struct timespec *now,
5253 int flags)
22c44fe6 5254{
2bc55652
AB		 5255 struct btrfs_root *root = BTRFS_I(inode)->root;
5256
5257 if (btrfs_root_readonly(root))
5258 return -EROFS;
5259
e41f941a 5260 if (flags & S_VERSION)
22c44fe6 5261 inode_inc_iversion(inode);
e41f941a
JB		 5262 if (flags & S_CTIME)
5263 inode->i_ctime = *now;
5264 if (flags & S_MTIME)
5265 inode->i_mtime = *now;
5266 if (flags & S_ATIME)
5267 inode->i_atime = *now;
5268 return btrfs_dirty_inode(inode);
39279cc3
CM		 5269}
5270
d352ac68
CM		 5271/*
5272 * find the highest existing sequence number in a directory
5273 * and then set the in-memory index_cnt variable to reflect
5274 * free sequence numbers
5275 */
aec7477b
JB		 5276static int btrfs_set_inode_index_count(struct inode *inode)
5277{
5278 struct btrfs_root *root = BTRFS_I(inode)->root;
5279 struct btrfs_key key, found_key;
5280 struct btrfs_path *path;
5281 struct extent_buffer *leaf;
5282 int ret;
5283
33345d01 5284 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5285 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5286 key.offset = (u64)-1;
5287
5288 path = btrfs_alloc_path();
5289 if (!path)
5290 return -ENOMEM;
5291
5292 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5293 if (ret < 0)
5294 goto out;
5295 /* FIXME: we should be able to handle this */
5296 if (ret == 0)
5297 goto out;
5298 ret = 0;
5299
5300 /*
5301 * MAGIC NUMBER EXPLANATION:
5302 * since we search a directory based on f_pos we have to start at 2
5303 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5304 * else has to start at 2
5305 */
5306 if (path->slots[0] == 0) {
5307 BTRFS_I(inode)->index_cnt = 2;
5308 goto out;
5309 }
5310
5311 path->slots[0]--;
5312
5313 leaf = path->nodes[0];
5314 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5315
33345d01 5316 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5317 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5318 BTRFS_I(inode)->index_cnt = 2;
5319 goto out;
5320 }
5321
5322 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5323out:
5324 btrfs_free_path(path);
5325 return ret;
5326}
5327
d352ac68
CM		 5328/*
5329 * helper to find a free sequence number in a given directory. This current
5330 * code is very simple, later versions will do smarter things in the btree
5331 */
3de4586c 5332int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5333{
5334 int ret = 0;
5335
5336 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5337 ret = btrfs_inode_delayed_dir_index_count(dir);
5338 if (ret) {
5339 ret = btrfs_set_inode_index_count(dir);
5340 if (ret)
5341 return ret;
5342 }
aec7477b
JB		 5343 }
5344
00e4e6b3 5345 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5346 BTRFS_I(dir)->index_cnt++;
5347
5348 return ret;
5349}
5350
39279cc3
CM		 5351static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5352 struct btrfs_root *root,
aec7477b 5353 struct inode *dir,
9c58309d 5354 const char *name, int name_len,
175a4eb7
AV		 5355 u64 ref_objectid, u64 objectid,
5356 umode_t mode, u64 *index)
39279cc3
CM		 5357{
5358 struct inode *inode;
5f39d397 5359 struct btrfs_inode_item *inode_item;
39279cc3 5360 struct btrfs_key *location;
5f39d397 5361 struct btrfs_path *path;
9c58309d
CM		 5362 struct btrfs_inode_ref *ref;
5363 struct btrfs_key key[2];
5364 u32 sizes[2];
5365 unsigned long ptr;
39279cc3
CM		 5366 int ret;
5367 int owner;
5368
5f39d397 5369 path = btrfs_alloc_path();
d8926bb3
MF		 5370 if (!path)
5371 return ERR_PTR(-ENOMEM);
5f39d397 5372
39279cc3 5373 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5374 if (!inode) {
5375 btrfs_free_path(path);
39279cc3 5376 return ERR_PTR(-ENOMEM);
8fb27640 5377 }
39279cc3 5378
581bb050
LZ		 5379 /*
5380 * we have to initialize this early, so we can reclaim the inode
5381 * number if we fail afterwards in this function.
5382 */
5383 inode->i_ino = objectid;
5384
aec7477b 5385 if (dir) {
1abe9b8a 5386 trace_btrfs_inode_request(dir);
5387
3de4586c 5388 ret = btrfs_set_inode_index(dir, index);
09771430 5389 if (ret) {
8fb27640 5390 btrfs_free_path(path);
09771430 5391 iput(inode);
aec7477b 5392 return ERR_PTR(ret);
09771430 5393 }
aec7477b
JB		 5394 }
5395 /*
5396 * index_cnt is ignored for everything but a dir,
5397 * btrfs_get_inode_index_count has an explanation for the magic
5398 * number
5399 */
5400 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5401 BTRFS_I(inode)->root = root;
e02119d5 5402 BTRFS_I(inode)->generation = trans->transid;
76195853 5403 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5404
5dc562c5
JB		 5405 /*
5406 * We could have gotten an inode number from somebody who was fsynced
5407 * and then removed in this same transaction, so let's just set full
5408 * sync since it will be a full sync anyway and this will blow away the
5409 * old info in the log.
5410 */
5411 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5412
569254b0 5413 if (S_ISDIR(mode))
39279cc3
CM		 5414 owner = 0;
5415 else
5416 owner = 1;
9c58309d
CM		 5417
5418 key[0].objectid = objectid;
5419 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5420 key[0].offset = 0;
5421
f186373f
MF		 5422 /*
5423 * Start new inodes with an inode_ref. This is slightly more
5424 * efficient for small numbers of hard links since they will
5425 * be packed into one item. Extended refs will kick in if we
5426 * add more hard links than can fit in the ref item.
5427 */
9c58309d
CM		 5428 key[1].objectid = objectid;
5429 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5430 key[1].offset = ref_objectid;
5431
5432 sizes[0] = sizeof(struct btrfs_inode_item);
5433 sizes[1] = name_len + sizeof(*ref);
5434
b9473439 5435 path->leave_spinning = 1;
9c58309d
CM		 5436 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5437 if (ret != 0)
5f39d397
CM		 5438 goto fail;
5439
ecc11fab 5440 inode_init_owner(inode, dir, mode);
a76a3cd4 5441 inode_set_bytes(inode, 0);
39279cc3 5442 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5443 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5444 struct btrfs_inode_item);
293f7e07
LZ		 5445 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5446 sizeof(*inode_item));
e02119d5 5447 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5448
5449 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5450 struct btrfs_inode_ref);
5451 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5452 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5453 ptr = (unsigned long)(ref + 1);
5454 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5455
5f39d397
CM		 5456 btrfs_mark_buffer_dirty(path->nodes[0]);
5457 btrfs_free_path(path);
5458
39279cc3
CM		 5459 location = &BTRFS_I(inode)->location;
5460 location->objectid = objectid;
39279cc3
CM		 5461 location->offset = 0;
5462 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5463
6cbff00f
CH		 5464 btrfs_inherit_iflags(inode, dir);
5465
569254b0 5466 if (S_ISREG(mode)) {
94272164
CM		 5467 if (btrfs_test_opt(root, NODATASUM))
5468 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5469 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5470 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5471 BTRFS_INODE_NODATASUM;
94272164
CM		 5472 }
5473
39279cc3 5474 insert_inode_hash(inode);
5d4f98a2 5475 inode_tree_add(inode);
1abe9b8a 5476
5477 trace_btrfs_inode_new(inode);
1973f0fa 5478 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5479
8ea05e3a
AB		 5480 btrfs_update_root_times(trans, root);
5481
39279cc3 5482 return inode;
5f39d397 5483fail:
aec7477b
JB		 5484 if (dir)
5485 BTRFS_I(dir)->index_cnt--;
5f39d397 5486 btrfs_free_path(path);
09771430 5487 iput(inode);
5f39d397 5488 return ERR_PTR(ret);
39279cc3
CM		 5489}
5490
5491static inline u8 btrfs_inode_type(struct inode *inode)
5492{
5493 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5494}
5495
d352ac68
CM		 5496/*
5497 * utility function to add 'inode' into 'parent_inode' with
5498 * a give name and a given sequence number.
5499 * if 'add_backref' is true, also insert a backref from the
5500 * inode to the parent directory.
5501 */
e02119d5
CM		 5502int btrfs_add_link(struct btrfs_trans_handle *trans,
5503 struct inode *parent_inode, struct inode *inode,
5504 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5505{
4df27c4d 5506 int ret = 0;
39279cc3 5507 struct btrfs_key key;
e02119d5 5508 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5509 u64 ino = btrfs_ino(inode);
5510 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5511
33345d01 5512 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5513 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5514 } else {
33345d01 5515 key.objectid = ino;
4df27c4d
YZ		 5516 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5517 key.offset = 0;
5518 }
5519
33345d01 5520 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5521 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5522 key.objectid, root->root_key.objectid,
33345d01 5523 parent_ino, index, name, name_len);
4df27c4d 5524 } else if (add_backref) {
33345d01
LZ		 5525 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5526 parent_ino, index);
4df27c4d 5527 }
39279cc3 5528
79787eaa
JM		 5529 /* Nothing to clean up yet */
5530 if (ret)
5531 return ret;
4df27c4d 5532
79787eaa
JM		 5533 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5534 parent_inode, &key,
5535 btrfs_inode_type(inode), index);
9c52057c 5536 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5537 goto fail_dir_item;
5538 else if (ret) {
5539 btrfs_abort_transaction(trans, root, ret);
5540 return ret;
39279cc3 5541 }
79787eaa
JM		 5542
5543 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5544 name_len * 2);
0c4d2d95 5545 inode_inc_iversion(parent_inode);
79787eaa
JM		 5546 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5547 ret = btrfs_update_inode(trans, root, parent_inode);
5548 if (ret)
5549 btrfs_abort_transaction(trans, root, ret);
39279cc3 5550 return ret;
fe66a05a
CM		 5551
5552fail_dir_item:
5553 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5554 u64 local_index;
5555 int err;
5556 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5557 key.objectid, root->root_key.objectid,
5558 parent_ino, &local_index, name, name_len);
5559
5560 } else if (add_backref) {
5561 u64 local_index;
5562 int err;
5563
5564 err = btrfs_del_inode_ref(trans, root, name, name_len,
5565 ino, parent_ino, &local_index);
5566 }
5567 return ret;
39279cc3
CM		 5568}
5569
5570static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5571 struct inode *dir, struct dentry *dentry,
5572 struct inode *inode, int backref, u64 index)
39279cc3 5573{
a1b075d2
JB		 5574 int err = btrfs_add_link(trans, dir, inode,
5575 dentry->d_name.name, dentry->d_name.len,
5576 backref, index);
39279cc3
CM		 5577 if (err > 0)
5578 err = -EEXIST;
5579 return err;
5580}
5581
618e21d5 5582static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5583 umode_t mode, dev_t rdev)
618e21d5
JB		 5584{
5585 struct btrfs_trans_handle *trans;
5586 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5587 struct inode *inode = NULL;
618e21d5
JB		 5588 int err;
5589 int drop_inode = 0;
5590 u64 objectid;
00e4e6b3 5591 u64 index = 0;
618e21d5
JB		 5592
5593 if (!new_valid_dev(rdev))
5594 return -EINVAL;
5595
9ed74f2d
JB		 5596 /*
5597 * 2 for inode item and ref
5598 * 2 for dir items
5599 * 1 for xattr if selinux is on
5600 */
a22285a6
YZ		 5601 trans = btrfs_start_transaction(root, 5);
5602 if (IS_ERR(trans))
5603 return PTR_ERR(trans);
1832a6d5 5604
581bb050
LZ		 5605 err = btrfs_find_free_ino(root, &objectid);
5606 if (err)
5607 goto out_unlock;
5608
aec7477b 5609 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5610 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5611 mode, &index);
7cf96da3
TI		 5612 if (IS_ERR(inode)) {
5613 err = PTR_ERR(inode);
618e21d5 5614 goto out_unlock;
7cf96da3 5615 }
618e21d5 5616
2a7dba39 5617 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5618 if (err) {
5619 drop_inode = 1;
5620 goto out_unlock;
5621 }
5622
ad19db71
CS		 5623 /*
5624 * If the active LSM wants to access the inode during
5625 * d_instantiate it needs these. Smack checks to see
5626 * if the filesystem supports xattrs by looking at the
5627 * ops vector.
5628 */
5629
5630 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5631 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5632 if (err)
5633 drop_inode = 1;
5634 else {
618e21d5 5635 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5636 btrfs_update_inode(trans, root, inode);
08c422c2 5637 d_instantiate(dentry, inode);
618e21d5 5638 }
618e21d5 5639out_unlock:
7ad85bb7 5640 btrfs_end_transaction(trans, root);
b53d3f5d 5641 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5642 if (drop_inode) {
5643 inode_dec_link_count(inode);
5644 iput(inode);
5645 }
618e21d5
JB		 5646 return err;
5647}
5648
39279cc3 5649static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5650 umode_t mode, bool excl)
39279cc3
CM		 5651{
5652 struct btrfs_trans_handle *trans;
5653 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5654 struct inode *inode = NULL;
43baa579 5655 int drop_inode_on_err = 0;
a22285a6 5656 int err;
39279cc3 5657 u64 objectid;
00e4e6b3 5658 u64 index = 0;
39279cc3 5659
9ed74f2d
JB		 5660 /*
5661 * 2 for inode item and ref
5662 * 2 for dir items
5663 * 1 for xattr if selinux is on
5664 */
a22285a6
YZ		 5665 trans = btrfs_start_transaction(root, 5);
5666 if (IS_ERR(trans))
5667 return PTR_ERR(trans);
9ed74f2d 5668
581bb050
LZ		 5669 err = btrfs_find_free_ino(root, &objectid);
5670 if (err)
5671 goto out_unlock;
5672
aec7477b 5673 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5674 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5675 mode, &index);
7cf96da3
TI		 5676 if (IS_ERR(inode)) {
5677 err = PTR_ERR(inode);
39279cc3 5678 goto out_unlock;
7cf96da3 5679 }
43baa579 5680 drop_inode_on_err = 1;
39279cc3 5681
2a7dba39 5682 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5683 if (err)
33268eaf 5684 goto out_unlock;
33268eaf 5685
9185aa58
FB		 5686 err = btrfs_update_inode(trans, root, inode);
5687 if (err)
5688 goto out_unlock;
5689
ad19db71
CS		 5690 /*
5691 * If the active LSM wants to access the inode during
5692 * d_instantiate it needs these. Smack checks to see
5693 * if the filesystem supports xattrs by looking at the
5694 * ops vector.
5695 */
5696 inode->i_fop = &btrfs_file_operations;
5697 inode->i_op = &btrfs_file_inode_operations;
5698
a1b075d2 5699 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5700 if (err)
43baa579
FB		 5701 goto out_unlock;
5702
5703 inode->i_mapping->a_ops = &btrfs_aops;
5704 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5705 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5706 d_instantiate(dentry, inode);
5707
39279cc3 5708out_unlock:
7ad85bb7 5709 btrfs_end_transaction(trans, root);
43baa579 5710 if (err && drop_inode_on_err) {
39279cc3
CM		 5711 inode_dec_link_count(inode);
5712 iput(inode);
5713 }
b53d3f5d 5714 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5715 return err;
5716}
5717
5718static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5719 struct dentry *dentry)
5720{
5721 struct btrfs_trans_handle *trans;
5722 struct btrfs_root *root = BTRFS_I(dir)->root;
5723 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5724 u64 index;
39279cc3
CM		 5725 int err;
5726 int drop_inode = 0;
5727
4a8be425
TH		 5728 /* do not allow sys_link's with other subvols of the same device */
5729 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5730 return -EXDEV;
4a8be425 5731
f186373f 5732 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5733 return -EMLINK;
4a8be425 5734
3de4586c 5735 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5736 if (err)
5737 goto fail;
5738
a22285a6 5739 /*
7e6b6465 5740 * 2 items for inode and inode ref
a22285a6 5741 * 2 items for dir items
7e6b6465 5742 * 1 item for parent inode
a22285a6 5743 */
7e6b6465 5744 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5745 if (IS_ERR(trans)) {
5746 err = PTR_ERR(trans);
5747 goto fail;
5748 }
5f39d397 5749
3153495d 5750 btrfs_inc_nlink(inode);
0c4d2d95 5751 inode_inc_iversion(inode);
3153495d 5752 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5753 ihold(inode);
e9976151 5754 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5755
a1b075d2 5756 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5757
a5719521 5758 if (err) {
54aa1f4d 5759 drop_inode = 1;
a5719521 5760 } else {
10d9f309 5761 struct dentry *parent = dentry->d_parent;
a5719521 5762 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5763 if (err)
5764 goto fail;
08c422c2 5765 d_instantiate(dentry, inode);
6a912213 5766 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5767 }
39279cc3 5768
7ad85bb7 5769 btrfs_end_transaction(trans, root);
1832a6d5 5770fail:
39279cc3
CM		 5771 if (drop_inode) {
5772 inode_dec_link_count(inode);
5773 iput(inode);
5774 }
b53d3f5d 5775 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5776 return err;
5777}
5778
18bb1db3 5779static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5780{
b9d86667 5781 struct inode *inode = NULL;
39279cc3
CM		 5782 struct btrfs_trans_handle *trans;
5783 struct btrfs_root *root = BTRFS_I(dir)->root;
5784 int err = 0;
5785 int drop_on_err = 0;
b9d86667 5786 u64 objectid = 0;
00e4e6b3 5787 u64 index = 0;
39279cc3 5788
9ed74f2d
JB		 5789 /*
5790 * 2 items for inode and ref
5791 * 2 items for dir items
5792 * 1 for xattr if selinux is on
5793 */
a22285a6
YZ		 5794 trans = btrfs_start_transaction(root, 5);
5795 if (IS_ERR(trans))
5796 return PTR_ERR(trans);
39279cc3 5797
581bb050
LZ		 5798 err = btrfs_find_free_ino(root, &objectid);
5799 if (err)
5800 goto out_fail;
5801
aec7477b 5802 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5803 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5804 S_IFDIR | mode, &index);
39279cc3
CM		 5805 if (IS_ERR(inode)) {
5806 err = PTR_ERR(inode);
5807 goto out_fail;
5808 }
5f39d397 5809
39279cc3 5810 drop_on_err = 1;
33268eaf 5811
2a7dba39 5812 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5813 if (err)
5814 goto out_fail;
5815
39279cc3
CM		 5816 inode->i_op = &btrfs_dir_inode_operations;
5817 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5818
dbe674a9 5819 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5820 err = btrfs_update_inode(trans, root, inode);
5821 if (err)
5822 goto out_fail;
5f39d397 5823
a1b075d2
JB		 5824 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5825 dentry->d_name.len, 0, index);
39279cc3
CM		 5826 if (err)
5827 goto out_fail;
5f39d397 5828
39279cc3
CM		 5829 d_instantiate(dentry, inode);
5830 drop_on_err = 0;
39279cc3
CM		 5831
5832out_fail:
7ad85bb7 5833 btrfs_end_transaction(trans, root);
39279cc3
CM		 5834 if (drop_on_err)
5835 iput(inode);
b53d3f5d 5836 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5837 return err;
5838}
5839
d352ac68
CM		 5840/* helper for btfs_get_extent. Given an existing extent in the tree,
5841 * and an extent that you want to insert, deal with overlap and insert
5842 * the new extent into the tree.
5843 */
3b951516
CM		 5844static int merge_extent_mapping(struct extent_map_tree *em_tree,
5845 struct extent_map *existing,
e6dcd2dc
CM		 5846 struct extent_map *em,
5847 u64 map_start, u64 map_len)
3b951516
CM		 5848{
5849 u64 start_diff;
3b951516 5850
e6dcd2dc
CM		 5851 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5852 start_diff = map_start - em->start;
5853 em->start = map_start;
5854 em->len = map_len;
c8b97818
CM		 5855 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5856 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5857 em->block_start += start_diff;
c8b97818
CM		 5858 em->block_len -= start_diff;
5859 }
09a2a8f9 5860 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 5861}
5862
c8b97818
CM		 5863static noinline int uncompress_inline(struct btrfs_path *path,
5864 struct inode *inode, struct page *page,
5865 size_t pg_offset, u64 extent_offset,
5866 struct btrfs_file_extent_item *item)
5867{
5868 int ret;
5869 struct extent_buffer *leaf = path->nodes[0];
5870 char *tmp;
5871 size_t max_size;
5872 unsigned long inline_size;
5873 unsigned long ptr;
261507a0 5874 int compress_type;
c8b97818
CM		 5875
5876 WARN_ON(pg_offset != 0);
261507a0 5877 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5878 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5879 inline_size = btrfs_file_extent_inline_item_len(leaf,
5880 btrfs_item_nr(leaf, path->slots[0]));
5881 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5882 if (!tmp)
5883 return -ENOMEM;
c8b97818
CM		 5884 ptr = btrfs_file_extent_inline_start(item);
5885
5886 read_extent_buffer(leaf, tmp, ptr, inline_size);
5887
5b050f04 5888 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5889 ret = btrfs_decompress(compress_type, tmp, page,
5890 extent_offset, inline_size, max_size);
c8b97818 5891 if (ret) {
7ac687d9 5892 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5893 unsigned long copy_size = min_t(u64,
5894 PAGE_CACHE_SIZE - pg_offset,
5895 max_size - extent_offset);
5896 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5897 kunmap_atomic(kaddr);
c8b97818
CM		 5898 }
5899 kfree(tmp);
5900 return 0;
5901}
5902
d352ac68
CM		 5903/*
5904 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5905 * the ugly parts come from merging extents from the disk with the in-ram
5906 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5907 * where the in-ram extents might be locked pending data=ordered completion.
5908 *
5909 * This also copies inline extents directly into the page.
5910 */
d397712b 5911
a52d9a80 5912struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5913 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5914 int create)
5915{
5916 int ret;
5917 int err = 0;
db94535d 5918 u64 bytenr;
a52d9a80
CM		 5919 u64 extent_start = 0;
5920 u64 extent_end = 0;
33345d01 5921 u64 objectid = btrfs_ino(inode);
a52d9a80 5922 u32 found_type;
f421950f 5923 struct btrfs_path *path = NULL;
a52d9a80
CM		 5924 struct btrfs_root *root = BTRFS_I(inode)->root;
5925 struct btrfs_file_extent_item *item;
5f39d397
CM		 5926 struct extent_buffer *leaf;
5927 struct btrfs_key found_key;
a52d9a80
CM		 5928 struct extent_map *em = NULL;
5929 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5930 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5931 struct btrfs_trans_handle *trans = NULL;
261507a0 5932 int compress_type;
a52d9a80 5933
a52d9a80 5934again:
890871be 5935 read_lock(&em_tree->lock);
d1310b2e 5936 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5937 if (em)
5938 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5939 read_unlock(&em_tree->lock);
d1310b2e 5940
a52d9a80 5941 if (em) {
e1c4b745
CM		 5942 if (em->start > start || em->start + em->len <= start)
5943 free_extent_map(em);
5944 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5945 free_extent_map(em);
5946 else
5947 goto out;
a52d9a80 5948 }
172ddd60 5949 em = alloc_extent_map();
a52d9a80 5950 if (!em) {
d1310b2e
CM		 5951 err = -ENOMEM;
5952 goto out;
a52d9a80 5953 }
e6dcd2dc 5954 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5955 em->start = EXTENT_MAP_HOLE;
445a6944 5956 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5957 em->len = (u64)-1;
c8b97818 5958 em->block_len = (u64)-1;
f421950f
CM		 5959
5960 if (!path) {
5961 path = btrfs_alloc_path();
026fd317
JB		 5962 if (!path) {
5963 err = -ENOMEM;
5964 goto out;
5965 }
5966 /*
5967 * Chances are we'll be called again, so go ahead and do
5968 * readahead
5969 */
5970 path->reada = 1;
f421950f
CM		 5971 }
5972
179e29e4
CM		 5973 ret = btrfs_lookup_file_extent(trans, root, path,
5974 objectid, start, trans != NULL);
a52d9a80
CM		 5975 if (ret < 0) {
5976 err = ret;
5977 goto out;
5978 }
5979
5980 if (ret != 0) {
5981 if (path->slots[0] == 0)
5982 goto not_found;
5983 path->slots[0]--;
5984 }
5985
5f39d397
CM		 5986 leaf = path->nodes[0];
5987 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5988 struct btrfs_file_extent_item);
a52d9a80 5989 /* are we inside the extent that was found? */
5f39d397
CM		 5990 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5991 found_type = btrfs_key_type(&found_key);
5992 if (found_key.objectid != objectid ||
a52d9a80
CM		 5993 found_type != BTRFS_EXTENT_DATA_KEY) {
5994 goto not_found;
5995 }
5996
5f39d397
CM		 5997 found_type = btrfs_file_extent_type(leaf, item);
5998 extent_start = found_key.offset;
261507a0 5999 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 6000 if (found_type == BTRFS_FILE_EXTENT_REG ||
6001 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6002 extent_end = extent_start +
db94535d 6003 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6004 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6005 size_t size;
6006 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 6007 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102
YZ		 6008 }
6009
6010 if (start >= extent_end) {
6011 path->slots[0]++;
6012 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6013 ret = btrfs_next_leaf(root, path);
6014 if (ret < 0) {
6015 err = ret;
6016 goto out;
a52d9a80 6017 }
9036c102
YZ		 6018 if (ret > 0)
6019 goto not_found;
6020 leaf = path->nodes[0];
a52d9a80 6021 }
9036c102
YZ		 6022 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6023 if (found_key.objectid != objectid ||
6024 found_key.type != BTRFS_EXTENT_DATA_KEY)
6025 goto not_found;
6026 if (start + len <= found_key.offset)
6027 goto not_found;
6028 em->start = start;
70c8a91c 6029 em->orig_start = start;
9036c102
YZ		 6030 em->len = found_key.offset - start;
6031 goto not_found_em;
6032 }
6033
cc95bef6 6034 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6035 if (found_type == BTRFS_FILE_EXTENT_REG ||
6036 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6037 em->start = extent_start;
6038 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6039 em->orig_start = extent_start -
6040 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6041 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6042 item);
db94535d
CM		 6043 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6044 if (bytenr == 0) {
5f39d397 6045 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6046 goto insert;
6047 }
261507a0 6048 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6049 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6050 em->compress_type = compress_type;
c8b97818 6051 em->block_start = bytenr;
b4939680 6052 em->block_len = em->orig_block_len;
c8b97818
CM		 6053 } else {
6054 bytenr += btrfs_file_extent_offset(leaf, item);
6055 em->block_start = bytenr;
6056 em->block_len = em->len;
d899e052
YZ		 6057 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6058 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6059 }
a52d9a80
CM		 6060 goto insert;
6061 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6062 unsigned long ptr;
a52d9a80 6063 char *map;
3326d1b0
CM		 6064 size_t size;
6065 size_t extent_offset;
6066 size_t copy_size;
a52d9a80 6067
689f9346 6068 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6069 if (!page || create) {
689f9346 6070 em->start = extent_start;
9036c102 6071 em->len = extent_end - extent_start;
689f9346
Y		 6072 goto out;
6073 }
5f39d397 6074
9036c102
YZ		 6075 size = btrfs_file_extent_inline_len(leaf, item);
6076 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6077 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6078 size - extent_offset);
3326d1b0 6079 em->start = extent_start + extent_offset;
fda2832f 6080 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6081 em->orig_block_len = em->len;
70c8a91c 6082 em->orig_start = em->start;
261507a0 6083 if (compress_type) {
c8b97818 6084 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6085 em->compress_type = compress_type;
6086 }
689f9346 6087 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6088 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6089 if (btrfs_file_extent_compression(leaf, item) !=
6090 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6091 ret = uncompress_inline(path, inode, page,
6092 pg_offset,
6093 extent_offset, item);
79787eaa 6094 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6095 } else {
6096 map = kmap(page);
6097 read_extent_buffer(leaf, map + pg_offset, ptr,
6098 copy_size);
93c82d57
CM		 6099 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6100 memset(map + pg_offset + copy_size, 0,
6101 PAGE_CACHE_SIZE - pg_offset -
6102 copy_size);
6103 }
c8b97818
CM		 6104 kunmap(page);
6105 }
179e29e4
CM		 6106 flush_dcache_page(page);
6107 } else if (create && PageUptodate(page)) {
6bf7e080 6108 BUG();
179e29e4
CM		 6109 if (!trans) {
6110 kunmap(page);
6111 free_extent_map(em);
6112 em = NULL;
ff5714cc 6113
b3b4aa74 6114 btrfs_release_path(path);
7a7eaa40 6115 trans = btrfs_join_transaction(root);
ff5714cc 6116
3612b495
TI		 6117 if (IS_ERR(trans))
6118 return ERR_CAST(trans);
179e29e4
CM		 6119 goto again;
6120 }
c8b97818 6121 map = kmap(page);
70dec807 6122 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6123 copy_size);
c8b97818 6124 kunmap(page);
179e29e4 6125 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6126 }
d1310b2e 6127 set_extent_uptodate(io_tree, em->start,
507903b8 6128 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6129 goto insert;
6130 } else {
31b1a2bd 6131 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6132 }
6133not_found:
6134 em->start = start;
70c8a91c 6135 em->orig_start = start;
d1310b2e 6136 em->len = len;
a52d9a80 6137not_found_em:
5f39d397 6138 em->block_start = EXTENT_MAP_HOLE;
9036c102 6139 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6140insert:
b3b4aa74 6141 btrfs_release_path(path);
d1310b2e 6142 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb
SK		 6143 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6144 (unsigned long long)em->start,
6145 (unsigned long long)em->len,
6146 (unsigned long long)start,
6147 (unsigned long long)len);
a52d9a80
CM		 6148 err = -EIO;
6149 goto out;
6150 }
d1310b2e
CM		 6151
6152 err = 0;
890871be 6153 write_lock(&em_tree->lock);
09a2a8f9 6154 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6155 /* it is possible that someone inserted the extent into the tree
6156 * while we had the lock dropped. It is also possible that
6157 * an overlapping map exists in the tree
6158 */
a52d9a80 6159 if (ret == -EEXIST) {
3b951516 6160 struct extent_map *existing;
e6dcd2dc
CM		 6161
6162 ret = 0;
6163
3b951516 6164 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6165 if (existing && (existing->start > start ||
6166 existing->start + existing->len <= start)) {
6167 free_extent_map(existing);
6168 existing = NULL;
6169 }
3b951516
CM		 6170 if (!existing) {
6171 existing = lookup_extent_mapping(em_tree, em->start,
6172 em->len);
6173 if (existing) {
6174 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6175 em, start,
6176 root->sectorsize);
3b951516
CM		 6177 free_extent_map(existing);
6178 if (err) {
6179 free_extent_map(em);
6180 em = NULL;
6181 }
6182 } else {
6183 err = -EIO;
3b951516
CM		 6184 free_extent_map(em);
6185 em = NULL;
6186 }
6187 } else {
6188 free_extent_map(em);
6189 em = existing;
e6dcd2dc 6190 err = 0;
a52d9a80 6191 }
a52d9a80 6192 }
890871be 6193 write_unlock(&em_tree->lock);
a52d9a80 6194out:
1abe9b8a 6195
f0bd95ea
TI		 6196 if (em)
6197 trace_btrfs_get_extent(root, em);
1abe9b8a 6198
f421950f
CM		 6199 if (path)
6200 btrfs_free_path(path);
a52d9a80
CM		 6201 if (trans) {
6202 ret = btrfs_end_transaction(trans, root);
d397712b 6203 if (!err)
a52d9a80
CM		 6204 err = ret;
6205 }
a52d9a80
CM		 6206 if (err) {
6207 free_extent_map(em);
a52d9a80
CM		 6208 return ERR_PTR(err);
6209 }
79787eaa 6210 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6211 return em;
6212}
6213
ec29ed5b
CM		 6214struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6215 size_t pg_offset, u64 start, u64 len,
6216 int create)
6217{
6218 struct extent_map *em;
6219 struct extent_map *hole_em = NULL;
6220 u64 range_start = start;
6221 u64 end;
6222 u64 found;
6223 u64 found_end;
6224 int err = 0;
6225
6226 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6227 if (IS_ERR(em))
6228 return em;
6229 if (em) {
6230 /*
f9e4fb53
LB		 6231 * if our em maps to
6232 * - a hole or
6233 * - a pre-alloc extent,
6234 * there might actually be delalloc bytes behind it.
ec29ed5b 6235 */
f9e4fb53
LB		 6236 if (em->block_start != EXTENT_MAP_HOLE &&
6237 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6238 return em;
6239 else
6240 hole_em = em;
6241 }
6242
6243 /* check to see if we've wrapped (len == -1 or similar) */
6244 end = start + len;
6245 if (end < start)
6246 end = (u64)-1;
6247 else
6248 end -= 1;
6249
6250 em = NULL;
6251
6252 /* ok, we didn't find anything, lets look for delalloc */
6253 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6254 end, len, EXTENT_DELALLOC, 1);
6255 found_end = range_start + found;
6256 if (found_end < range_start)
6257 found_end = (u64)-1;
6258
6259 /*
6260 * we didn't find anything useful, return
6261 * the original results from get_extent()
6262 */
6263 if (range_start > end || found_end <= start) {
6264 em = hole_em;
6265 hole_em = NULL;
6266 goto out;
6267 }
6268
6269 /* adjust the range_start to make sure it doesn't
6270 * go backwards from the start they passed in
6271 */
6272 range_start = max(start,range_start);
6273 found = found_end - range_start;
6274
6275 if (found > 0) {
6276 u64 hole_start = start;
6277 u64 hole_len = len;
6278
172ddd60 6279 em = alloc_extent_map();
ec29ed5b
CM		 6280 if (!em) {
6281 err = -ENOMEM;
6282 goto out;
6283 }
6284 /*
6285 * when btrfs_get_extent can't find anything it
6286 * returns one huge hole
6287 *
6288 * make sure what it found really fits our range, and
6289 * adjust to make sure it is based on the start from
6290 * the caller
6291 */
6292 if (hole_em) {
6293 u64 calc_end = extent_map_end(hole_em);
6294
6295 if (calc_end <= start || (hole_em->start > end)) {
6296 free_extent_map(hole_em);
6297 hole_em = NULL;
6298 } else {
6299 hole_start = max(hole_em->start, start);
6300 hole_len = calc_end - hole_start;
6301 }
6302 }
6303 em->bdev = NULL;
6304 if (hole_em && range_start > hole_start) {
6305 /* our hole starts before our delalloc, so we
6306 * have to return just the parts of the hole
6307 * that go until the delalloc starts
6308 */
6309 em->len = min(hole_len,
6310 range_start - hole_start);
6311 em->start = hole_start;
6312 em->orig_start = hole_start;
6313 /*
6314 * don't adjust block start at all,
6315 * it is fixed at EXTENT_MAP_HOLE
6316 */
6317 em->block_start = hole_em->block_start;
6318 em->block_len = hole_len;
f9e4fb53
LB		 6319 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6320 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6321 } else {
6322 em->start = range_start;
6323 em->len = found;
6324 em->orig_start = range_start;
6325 em->block_start = EXTENT_MAP_DELALLOC;
6326 em->block_len = found;
6327 }
6328 } else if (hole_em) {
6329 return hole_em;
6330 }
6331out:
6332
6333 free_extent_map(hole_em);
6334 if (err) {
6335 free_extent_map(em);
6336 return ERR_PTR(err);
6337 }
6338 return em;
6339}
6340
4b46fce2
JB		 6341static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6342 u64 start, u64 len)
6343{
6344 struct btrfs_root *root = BTRFS_I(inode)->root;
6345 struct btrfs_trans_handle *trans;
70c8a91c 6346 struct extent_map *em;
4b46fce2
JB		 6347 struct btrfs_key ins;
6348 u64 alloc_hint;
6349 int ret;
4b46fce2 6350
7a7eaa40 6351 trans = btrfs_join_transaction(root);
3612b495
TI		 6352 if (IS_ERR(trans))
6353 return ERR_CAST(trans);
4b46fce2
JB		 6354
6355 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6356
6357 alloc_hint = get_extent_allocation_hint(inode, start, len);
6358 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 6359 alloc_hint, &ins, 1);
4b46fce2
JB		 6360 if (ret) {
6361 em = ERR_PTR(ret);
6362 goto out;
6363 }
6364
70c8a91c 6365 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6366 ins.offset, ins.offset, ins.offset, 0);
70c8a91c
JB		 6367 if (IS_ERR(em))
6368 goto out;
4b46fce2
JB		 6369
6370 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6371 ins.offset, ins.offset, 0);
6372 if (ret) {
6373 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6374 em = ERR_PTR(ret);
6375 }
6376out:
6377 btrfs_end_transaction(trans, root);
6378 return em;
6379}
6380
46bfbb5c
CM		 6381/*
6382 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6383 * block must be cow'd
6384 */
6385static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
eb384b55
JB		 6386 struct inode *inode, u64 offset, u64 *len,
6387 u64 *orig_start, u64 *orig_block_len,
6388 u64 *ram_bytes)
46bfbb5c
CM		 6389{
6390 struct btrfs_path *path;
6391 int ret;
6392 struct extent_buffer *leaf;
6393 struct btrfs_root *root = BTRFS_I(inode)->root;
6394 struct btrfs_file_extent_item *fi;
6395 struct btrfs_key key;
6396 u64 disk_bytenr;
6397 u64 backref_offset;
6398 u64 extent_end;
6399 u64 num_bytes;
6400 int slot;
6401 int found_type;
6402
6403 path = btrfs_alloc_path();
6404 if (!path)
6405 return -ENOMEM;
6406
33345d01 6407 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6408 offset, 0);
6409 if (ret < 0)
6410 goto out;
6411
6412 slot = path->slots[0];
6413 if (ret == 1) {
6414 if (slot == 0) {
6415 /* can't find the item, must cow */
6416 ret = 0;
6417 goto out;
6418 }
6419 slot--;
6420 }
6421 ret = 0;
6422 leaf = path->nodes[0];
6423 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6424 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6425 key.type != BTRFS_EXTENT_DATA_KEY) {
6426 /* not our file or wrong item type, must cow */
6427 goto out;
6428 }
6429
6430 if (key.offset > offset) {
6431 /* Wrong offset, must cow */
6432 goto out;
6433 }
6434
6435 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6436 found_type = btrfs_file_extent_type(leaf, fi);
6437 if (found_type != BTRFS_FILE_EXTENT_REG &&
6438 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6439 /* not a regular extent, must cow */
6440 goto out;
6441 }
6442 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6443 backref_offset = btrfs_file_extent_offset(leaf, fi);
6444
eb384b55
JB		 6445 *orig_start = key.offset - backref_offset;
6446 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6447 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6448
46bfbb5c 6449 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
eb384b55 6450 if (extent_end < offset + *len) {
46bfbb5c
CM		 6451 /* extent doesn't include our full range, must cow */
6452 goto out;
6453 }
6454
6455 if (btrfs_extent_readonly(root, disk_bytenr))
6456 goto out;
6457
6458 /*
6459 * look for other files referencing this extent, if we
6460 * find any we must cow
6461 */
33345d01 6462 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 6463 key.offset - backref_offset, disk_bytenr))
6464 goto out;
6465
6466 /*
6467 * adjust disk_bytenr and num_bytes to cover just the bytes
6468 * in this extent we are about to write. If there
6469 * are any csums in that range we have to cow in order
6470 * to keep the csums correct
6471 */
6472 disk_bytenr += backref_offset;
6473 disk_bytenr += offset - key.offset;
eb384b55 6474 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6475 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6476 goto out;
6477 /*
6478 * all of the above have passed, it is safe to overwrite this extent
6479 * without cow
6480 */
eb384b55 6481 *len = num_bytes;
46bfbb5c
CM		 6482 ret = 1;
6483out:
6484 btrfs_free_path(path);
6485 return ret;
6486}
6487
eb838e73
JB		 6488static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6489 struct extent_state **cached_state, int writing)
6490{
6491 struct btrfs_ordered_extent *ordered;
6492 int ret = 0;
6493
6494 while (1) {
6495 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6496 0, cached_state);
6497 /*
6498 * We're concerned with the entire range that we're going to be
6499 * doing DIO to, so we need to make sure theres no ordered
6500 * extents in this range.
6501 */
6502 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6503 lockend - lockstart + 1);
6504
6505 /*
6506 * We need to make sure there are no buffered pages in this
6507 * range either, we could have raced between the invalidate in
6508 * generic_file_direct_write and locking the extent. The
6509 * invalidate needs to happen so that reads after a write do not
6510 * get stale data.
6511 */
6512 if (!ordered && (!writing ||
6513 !test_range_bit(&BTRFS_I(inode)->io_tree,
6514 lockstart, lockend, EXTENT_UPTODATE, 0,
6515 *cached_state)))
6516 break;
6517
6518 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6519 cached_state, GFP_NOFS);
6520
6521 if (ordered) {
6522 btrfs_start_ordered_extent(inode, ordered, 1);
6523 btrfs_put_ordered_extent(ordered);
6524 } else {
6525 /* Screw you mmap */
6526 ret = filemap_write_and_wait_range(inode->i_mapping,
6527 lockstart,
6528 lockend);
6529 if (ret)
6530 break;
6531
6532 /*
6533 * If we found a page that couldn't be invalidated just
6534 * fall back to buffered.
6535 */
6536 ret = invalidate_inode_pages2_range(inode->i_mapping,
6537 lockstart >> PAGE_CACHE_SHIFT,
6538 lockend >> PAGE_CACHE_SHIFT);
6539 if (ret)
6540 break;
6541 }
6542
6543 cond_resched();
6544 }
6545
6546 return ret;
6547}
6548
69ffb543
JB		 6549static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6550 u64 len, u64 orig_start,
6551 u64 block_start, u64 block_len,
cc95bef6
JB		 6552 u64 orig_block_len, u64 ram_bytes,
6553 int type)
69ffb543
JB		 6554{
6555 struct extent_map_tree *em_tree;
6556 struct extent_map *em;
6557 struct btrfs_root *root = BTRFS_I(inode)->root;
6558 int ret;
6559
6560 em_tree = &BTRFS_I(inode)->extent_tree;
6561 em = alloc_extent_map();
6562 if (!em)
6563 return ERR_PTR(-ENOMEM);
6564
6565 em->start = start;
6566 em->orig_start = orig_start;
2ab28f32
JB		 6567 em->mod_start = start;
6568 em->mod_len = len;
69ffb543
JB		 6569 em->len = len;
6570 em->block_len = block_len;
6571 em->block_start = block_start;
6572 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6573 em->orig_block_len = orig_block_len;
cc95bef6 6574 em->ram_bytes = ram_bytes;
70c8a91c 6575 em->generation = -1;
69ffb543
JB		 6576 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6577 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6578 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6579
6580 do {
6581 btrfs_drop_extent_cache(inode, em->start,
6582 em->start + em->len - 1, 0);
6583 write_lock(&em_tree->lock);
09a2a8f9 6584 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6585 write_unlock(&em_tree->lock);
6586 } while (ret == -EEXIST);
6587
6588 if (ret) {
6589 free_extent_map(em);
6590 return ERR_PTR(ret);
6591 }
6592
6593 return em;
6594}
6595
6596
4b46fce2
JB		 6597static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6598 struct buffer_head *bh_result, int create)
6599{
6600 struct extent_map *em;
6601 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6602 struct extent_state *cached_state = NULL;
4b46fce2 6603 u64 start = iblock << inode->i_blkbits;
eb838e73 6604 u64 lockstart, lockend;
4b46fce2 6605 u64 len = bh_result->b_size;
46bfbb5c 6606 struct btrfs_trans_handle *trans;
eb838e73 6607 int unlock_bits = EXTENT_LOCKED;
0934856d 6608 int ret = 0;
eb838e73 6609
172a5049 6610 if (create)
eb838e73 6611 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6612 else
c329861d 6613 len = min_t(u64, len, root->sectorsize);
eb838e73 6614
c329861d
JB		 6615 lockstart = start;
6616 lockend = start + len - 1;
6617
eb838e73
JB		 6618 /*
6619 * If this errors out it's because we couldn't invalidate pagecache for
6620 * this range and we need to fallback to buffered.
6621 */
6622 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6623 return -ENOTBLK;
6624
4b46fce2 6625 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6626 if (IS_ERR(em)) {
6627 ret = PTR_ERR(em);
6628 goto unlock_err;
6629 }
4b46fce2
JB		 6630
6631 /*
6632 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6633 * io. INLINE is special, and we could probably kludge it in here, but
6634 * it's still buffered so for safety lets just fall back to the generic
6635 * buffered path.
6636 *
6637 * For COMPRESSED we _have_ to read the entire extent in so we can
6638 * decompress it, so there will be buffering required no matter what we
6639 * do, so go ahead and fallback to buffered.
6640 *
6641 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6642 * to buffered IO. Don't blame me, this is the price we pay for using
6643 * the generic code.
6644 */
6645 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6646 em->block_start == EXTENT_MAP_INLINE) {
6647 free_extent_map(em);
eb838e73
JB		 6648 ret = -ENOTBLK;
6649 goto unlock_err;
4b46fce2
JB		 6650 }
6651
6652 /* Just a good old fashioned hole, return */
6653 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6654 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6655 free_extent_map(em);
eb838e73 6656 goto unlock_err;
4b46fce2
JB		 6657 }
6658
6659 /*
6660 * We don't allocate a new extent in the following cases
6661 *
6662 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6663 * existing extent.
6664 * 2) The extent is marked as PREALLOC. We're good to go here and can
6665 * just use the extent.
6666 *
6667 */
46bfbb5c 6668 if (!create) {
eb838e73
JB		 6669 len = min(len, em->len - (start - em->start));
6670 lockstart = start + len;
6671 goto unlock;
46bfbb5c 6672 }
4b46fce2
JB		 6673
6674 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6675 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6676 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6677 int type;
6678 int ret;
eb384b55 6679 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6680
6681 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6682 type = BTRFS_ORDERED_PREALLOC;
6683 else
6684 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6685 len = min(len, em->len - (start - em->start));
4b46fce2 6686 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 6687
6688 /*
6689 * we're not going to log anything, but we do need
6690 * to make sure the current transaction stays open
6691 * while we look for nocow cross refs
6692 */
7a7eaa40 6693 trans = btrfs_join_transaction(root);
3612b495 6694 if (IS_ERR(trans))
46bfbb5c
CM		 6695 goto must_cow;
6696
eb384b55
JB		 6697 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6698 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6699 if (type == BTRFS_ORDERED_PREALLOC) {
6700 free_extent_map(em);
6701 em = create_pinned_em(inode, start, len,
6702 orig_start,
b4939680 6703 block_start, len,
cc95bef6
JB		 6704 orig_block_len,
6705 ram_bytes, type);
69ffb543
JB		 6706 if (IS_ERR(em)) {
6707 btrfs_end_transaction(trans, root);
6708 goto unlock_err;
6709 }
6710 }
6711
46bfbb5c
CM		 6712 ret = btrfs_add_ordered_extent_dio(inode, start,
6713 block_start, len, len, type);
6714 btrfs_end_transaction(trans, root);
6715 if (ret) {
6716 free_extent_map(em);
eb838e73 6717 goto unlock_err;
46bfbb5c
CM		 6718 }
6719 goto unlock;
4b46fce2 6720 }
46bfbb5c 6721 btrfs_end_transaction(trans, root);
4b46fce2 6722 }
46bfbb5c
CM		 6723must_cow:
6724 /*
6725 * this will cow the extent, reset the len in case we changed
6726 * it above
6727 */
6728 len = bh_result->b_size;
70c8a91c
JB		 6729 free_extent_map(em);
6730 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6731 if (IS_ERR(em)) {
6732 ret = PTR_ERR(em);
6733 goto unlock_err;
6734 }
46bfbb5c
CM		 6735 len = min(len, em->len - (start - em->start));
6736unlock:
4b46fce2
JB		 6737 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6738 inode->i_blkbits;
46bfbb5c 6739 bh_result->b_size = len;
4b46fce2
JB		 6740 bh_result->b_bdev = em->bdev;
6741 set_buffer_mapped(bh_result);
c3473e83
JB		 6742 if (create) {
6743 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6744 set_buffer_new(bh_result);
6745
6746 /*
6747 * Need to update the i_size under the extent lock so buffered
6748 * readers will get the updated i_size when we unlock.
6749 */
6750 if (start + len > i_size_read(inode))
6751 i_size_write(inode, start + len);
0934856d 6752
172a5049
MX		 6753 spin_lock(&BTRFS_I(inode)->lock);
6754 BTRFS_I(inode)->outstanding_extents++;
6755 spin_unlock(&BTRFS_I(inode)->lock);
6756
0934856d
MX		 6757 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6758 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6759 &cached_state, GFP_NOFS);
6760 BUG_ON(ret);
c3473e83 6761 }
4b46fce2 6762
eb838e73
JB		 6763 /*
6764 * In the case of write we need to clear and unlock the entire range,
6765 * in the case of read we need to unlock only the end area that we
6766 * aren't using if there is any left over space.
6767 */
24c03fa5 6768 if (lockstart < lockend) {
0934856d
MX		 6769 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6770 lockend, unlock_bits, 1, 0,
6771 &cached_state, GFP_NOFS);
24c03fa5 6772 } else {
eb838e73 6773 free_extent_state(cached_state);
24c03fa5 6774 }
eb838e73 6775
4b46fce2
JB		 6776 free_extent_map(em);
6777
6778 return 0;
eb838e73
JB		 6779
6780unlock_err:
eb838e73
JB		 6781 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6782 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6783 return ret;
4b46fce2
JB		 6784}
6785
6786struct btrfs_dio_private {
6787 struct inode *inode;
6788 u64 logical_offset;
6789 u64 disk_bytenr;
6790 u64 bytes;
4b46fce2 6791 void *private;
e65e1535
MX		 6792
6793 /* number of bios pending for this dio */
6794 atomic_t pending_bios;
6795
6796 /* IO errors */
6797 int errors;
6798
9be3395b 6799 /* orig_bio is our btrfs_io_bio */
e65e1535 6800 struct bio *orig_bio;
9be3395b
CM		 6801
6802 /* dio_bio came from fs/direct-io.c */
6803 struct bio *dio_bio;
4b46fce2
JB		 6804};
6805
6806static void btrfs_endio_direct_read(struct bio *bio, int err)
6807{
e65e1535 6808 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6809 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6810 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6811 struct inode *inode = dip->inode;
c2cf52eb 6812 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 6813 struct bio *dio_bio;
4b46fce2 6814 u64 start;
4b46fce2
JB		 6815
6816 start = dip->logical_offset;
6817 do {
6818 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6819 struct page *page = bvec->bv_page;
6820 char *kaddr;
6821 u32 csum = ~(u32)0;
c329861d 6822 u64 private = ~(u32)0;
4b46fce2
JB		 6823 unsigned long flags;
6824
c329861d
JB		 6825 if (get_state_private(&BTRFS_I(inode)->io_tree,
6826 start, &private))
6827 goto failed;
4b46fce2 6828 local_irq_save(flags);
7ac687d9 6829 kaddr = kmap_atomic(page);
b0496686 6830 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6831 csum, bvec->bv_len);
6832 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6833 kunmap_atomic(kaddr);
4b46fce2
JB		 6834 local_irq_restore(flags);
6835
6836 flush_dcache_page(bvec->bv_page);
c329861d
JB		 6837 if (csum != private) {
6838failed:
c2cf52eb
SK		 6839 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6840 (unsigned long long)btrfs_ino(inode),
6841 (unsigned long long)start,
6842 csum, (unsigned)private);
4b46fce2
JB		 6843 err = -EIO;
6844 }
6845 }
6846
6847 start += bvec->bv_len;
4b46fce2
JB		 6848 bvec++;
6849 } while (bvec <= bvec_end);
6850
6851 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6852 dip->logical_offset + dip->bytes - 1);
9be3395b 6853 dio_bio = dip->dio_bio;
4b46fce2 6854
4b46fce2 6855 kfree(dip);
c0da7aa1
JB		 6856
6857 /* If we had a csum failure make sure to clear the uptodate flag */
6858 if (err)
9be3395b
CM		 6859 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6860 dio_end_io(dio_bio, err);
6861 bio_put(bio);
4b46fce2
JB		 6862}
6863
6864static void btrfs_endio_direct_write(struct bio *bio, int err)
6865{
6866 struct btrfs_dio_private *dip = bio->bi_private;
6867 struct inode *inode = dip->inode;
6868 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6869 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 6870 u64 ordered_offset = dip->logical_offset;
6871 u64 ordered_bytes = dip->bytes;
9be3395b 6872 struct bio *dio_bio;
4b46fce2
JB		 6873 int ret;
6874
6875 if (err)
6876 goto out_done;
163cf09c
CM		 6877again:
6878 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6879 &ordered_offset,
5fd02043 6880 ordered_bytes, !err);
4b46fce2 6881 if (!ret)
163cf09c 6882 goto out_test;
4b46fce2 6883
5fd02043
JB		 6884 ordered->work.func = finish_ordered_fn;
6885 ordered->work.flags = 0;
6886 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6887 &ordered->work);
163cf09c
CM		 6888out_test:
6889 /*
6890 * our bio might span multiple ordered extents. If we haven't
6891 * completed the accounting for the whole dio, go back and try again
6892 */
6893 if (ordered_offset < dip->logical_offset + dip->bytes) {
6894 ordered_bytes = dip->logical_offset + dip->bytes -
6895 ordered_offset;
5fd02043 6896 ordered = NULL;
163cf09c
CM		 6897 goto again;
6898 }
4b46fce2 6899out_done:
9be3395b 6900 dio_bio = dip->dio_bio;
4b46fce2 6901
4b46fce2 6902 kfree(dip);
c0da7aa1
JB		 6903
6904 /* If we had an error make sure to clear the uptodate flag */
6905 if (err)
9be3395b
CM		 6906 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6907 dio_end_io(dio_bio, err);
6908 bio_put(bio);
4b46fce2
JB		 6909}
6910
eaf25d93
CM		 6911static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6912 struct bio *bio, int mirror_num,
6913 unsigned long bio_flags, u64 offset)
6914{
6915 int ret;
6916 struct btrfs_root *root = BTRFS_I(inode)->root;
6917 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6918 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6919 return 0;
6920}
6921
e65e1535
MX		 6922static void btrfs_end_dio_bio(struct bio *bio, int err)
6923{
6924 struct btrfs_dio_private *dip = bio->bi_private;
6925
6926 if (err) {
33345d01 6927 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6928 "sector %#Lx len %u err no %d\n",
33345d01 6929 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6930 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6931 dip->errors = 1;
6932
6933 /*
6934 * before atomic variable goto zero, we must make sure
6935 * dip->errors is perceived to be set.
6936 */
6937 smp_mb__before_atomic_dec();
6938 }
6939
6940 /* if there are more bios still pending for this dio, just exit */
6941 if (!atomic_dec_and_test(&dip->pending_bios))
6942 goto out;
6943
9be3395b 6944 if (dip->errors) {
e65e1535 6945 bio_io_error(dip->orig_bio);
9be3395b
CM		 6946 } else {
6947 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 6948 bio_endio(dip->orig_bio, 0);
6949 }
6950out:
6951 bio_put(bio);
6952}
6953
6954static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6955 u64 first_sector, gfp_t gfp_flags)
6956{
6957 int nr_vecs = bio_get_nr_vecs(bdev);
6958 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6959}
6960
6961static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6962 int rw, u64 file_offset, int skip_sum,
c329861d 6963 int async_submit)
e65e1535
MX		 6964{
6965 int write = rw & REQ_WRITE;
6966 struct btrfs_root *root = BTRFS_I(inode)->root;
6967 int ret;
6968
b812ce28
JB		 6969 if (async_submit)
6970 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6971
e65e1535 6972 bio_get(bio);
5fd02043
JB		 6973
6974 if (!write) {
6975 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6976 if (ret)
6977 goto err;
6978 }
e65e1535 6979
1ae39938
JB		 6980 if (skip_sum)
6981 goto map;
6982
6983 if (write && async_submit) {
e65e1535
MX		 6984 ret = btrfs_wq_submit_bio(root->fs_info,
6985 inode, rw, bio, 0, 0,
6986 file_offset,
6987 __btrfs_submit_bio_start_direct_io,
6988 __btrfs_submit_bio_done);
6989 goto err;
1ae39938
JB		 6990 } else if (write) {
6991 /*
6992 * If we aren't doing async submit, calculate the csum of the
6993 * bio now.
6994 */
6995 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6996 if (ret)
6997 goto err;
c2db1073 6998 } else if (!skip_sum) {
c329861d 6999 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI		 7000 if (ret)
7001 goto err;
7002 }
e65e1535 7003
1ae39938
JB		 7004map:
7005 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7006err:
7007 bio_put(bio);
7008 return ret;
7009}
7010
7011static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7012 int skip_sum)
7013{
7014 struct inode *inode = dip->inode;
7015 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7016 struct bio *bio;
7017 struct bio *orig_bio = dip->orig_bio;
7018 struct bio_vec *bvec = orig_bio->bi_io_vec;
7019 u64 start_sector = orig_bio->bi_sector;
7020 u64 file_offset = dip->logical_offset;
7021 u64 submit_len = 0;
7022 u64 map_length;
7023 int nr_pages = 0;
e65e1535 7024 int ret = 0;
1ae39938 7025 int async_submit = 0;
e65e1535 7026
e65e1535 7027 map_length = orig_bio->bi_size;
53b381b3 7028 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 7029 &map_length, NULL, 0);
7030 if (ret) {
64728bbb 7031 bio_put(orig_bio);
e65e1535
MX		 7032 return -EIO;
7033 }
02f57c7a
JB		 7034 if (map_length >= orig_bio->bi_size) {
7035 bio = orig_bio;
7036 goto submit;
7037 }
7038
53b381b3
DW		 7039 /* async crcs make it difficult to collect full stripe writes. */
7040 if (btrfs_get_alloc_profile(root, 1) &
7041 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7042 async_submit = 0;
7043 else
7044 async_submit = 1;
7045
02f57c7a
JB		 7046 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7047 if (!bio)
7048 return -ENOMEM;
7049 bio->bi_private = dip;
7050 bio->bi_end_io = btrfs_end_dio_bio;
7051 atomic_inc(&dip->pending_bios);
7052
e65e1535
MX		 7053 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7054 if (unlikely(map_length < submit_len + bvec->bv_len ||
7055 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7056 bvec->bv_offset) < bvec->bv_len)) {
7057 /*
7058 * inc the count before we submit the bio so
7059 * we know the end IO handler won't happen before
7060 * we inc the count. Otherwise, the dip might get freed
7061 * before we're done setting it up
7062 */
7063 atomic_inc(&dip->pending_bios);
7064 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7065 file_offset, skip_sum,
c329861d 7066 async_submit);
e65e1535
MX		 7067 if (ret) {
7068 bio_put(bio);
7069 atomic_dec(&dip->pending_bios);
7070 goto out_err;
7071 }
7072
e65e1535
MX		 7073 start_sector += submit_len >> 9;
7074 file_offset += submit_len;
7075
7076 submit_len = 0;
7077 nr_pages = 0;
7078
7079 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7080 start_sector, GFP_NOFS);
7081 if (!bio)
7082 goto out_err;
7083 bio->bi_private = dip;
7084 bio->bi_end_io = btrfs_end_dio_bio;
7085
7086 map_length = orig_bio->bi_size;
53b381b3 7087 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7088 start_sector << 9,
e65e1535
MX		 7089 &map_length, NULL, 0);
7090 if (ret) {
7091 bio_put(bio);
7092 goto out_err;
7093 }
7094 } else {
7095 submit_len += bvec->bv_len;
7096 nr_pages ++;
7097 bvec++;
7098 }
7099 }
7100
02f57c7a 7101submit:
e65e1535 7102 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7103 async_submit);
e65e1535
MX		 7104 if (!ret)
7105 return 0;
7106
7107 bio_put(bio);
7108out_err:
7109 dip->errors = 1;
7110 /*
7111 * before atomic variable goto zero, we must
7112 * make sure dip->errors is perceived to be set.
7113 */
7114 smp_mb__before_atomic_dec();
7115 if (atomic_dec_and_test(&dip->pending_bios))
7116 bio_io_error(dip->orig_bio);
7117
7118 /* bio_end_io() will handle error, so we needn't return it */
7119 return 0;
7120}
7121
9be3395b
CM		 7122static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7123 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7124{
7125 struct btrfs_root *root = BTRFS_I(inode)->root;
7126 struct btrfs_dio_private *dip;
9be3395b
CM		 7127 struct bio_vec *bvec = dio_bio->bi_io_vec;
7128 struct bio *io_bio;
4b46fce2 7129 int skip_sum;
7b6d91da 7130 int write = rw & REQ_WRITE;
4b46fce2
JB		 7131 int ret = 0;
7132
7133 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7134
9be3395b
CM		 7135 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7136
7137 if (!io_bio) {
7138 ret = -ENOMEM;
7139 goto free_ordered;
7140 }
7141
4b46fce2
JB		 7142 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7143 if (!dip) {
7144 ret = -ENOMEM;
9be3395b 7145 goto free_io_bio;
4b46fce2 7146 }
4b46fce2 7147
9be3395b
CM		 7148 dip->private = dio_bio->bi_private;
7149 io_bio->bi_private = dio_bio->bi_private;
4b46fce2
JB		 7150 dip->inode = inode;
7151 dip->logical_offset = file_offset;
7152
4b46fce2
JB		 7153 dip->bytes = 0;
7154 do {
7155 dip->bytes += bvec->bv_len;
7156 bvec++;
9be3395b 7157 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
4b46fce2 7158
9be3395b
CM		 7159 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7160 io_bio->bi_private = dip;
e65e1535 7161 dip->errors = 0;
9be3395b
CM		 7162 dip->orig_bio = io_bio;
7163 dip->dio_bio = dio_bio;
e65e1535 7164 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7165
7166 if (write)
9be3395b 7167 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7168 else
9be3395b 7169 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7170
e65e1535
MX		 7171 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7172 if (!ret)
eaf25d93 7173 return;
9be3395b
CM		 7174
7175free_io_bio:
7176 bio_put(io_bio);
7177
4b46fce2
JB		 7178free_ordered:
7179 /*
7180 * If this is a write, we need to clean up the reserved space and kill
7181 * the ordered extent.
7182 */
7183 if (write) {
7184 struct btrfs_ordered_extent *ordered;
955256f2 7185 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7186 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7187 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7188 btrfs_free_reserved_extent(root, ordered->start,
7189 ordered->disk_len);
7190 btrfs_put_ordered_extent(ordered);
7191 btrfs_put_ordered_extent(ordered);
7192 }
9be3395b 7193 bio_endio(dio_bio, ret);
4b46fce2
JB		 7194}
7195
5a5f79b5
CM		 7196static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7197 const struct iovec *iov, loff_t offset,
7198 unsigned long nr_segs)
7199{
7200 int seg;
a1b75f7d 7201 int i;
5a5f79b5
CM		 7202 size_t size;
7203 unsigned long addr;
7204 unsigned blocksize_mask = root->sectorsize - 1;
7205 ssize_t retval = -EINVAL;
7206 loff_t end = offset;
7207
7208 if (offset & blocksize_mask)
7209 goto out;
7210
7211 /* Check the memory alignment. Blocks cannot straddle pages */
7212 for (seg = 0; seg < nr_segs; seg++) {
7213 addr = (unsigned long)iov[seg].iov_base;
7214 size = iov[seg].iov_len;
7215 end += size;
a1b75f7d 7216 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7217 goto out;
a1b75f7d
JB		 7218
7219 /* If this is a write we don't need to check anymore */
7220 if (rw & WRITE)
7221 continue;
7222
7223 /*
7224 * Check to make sure we don't have duplicate iov_base's in this
7225 * iovec, if so return EINVAL, otherwise we'll get csum errors
7226 * when reading back.
7227 */
7228 for (i = seg + 1; i < nr_segs; i++) {
7229 if (iov[seg].iov_base == iov[i].iov_base)
7230 goto out;
7231 }
5a5f79b5
CM		 7232 }
7233 retval = 0;
7234out:
7235 return retval;
7236}
eb838e73 7237
16432985
CM		 7238static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7239 const struct iovec *iov, loff_t offset,
7240 unsigned long nr_segs)
7241{
4b46fce2
JB		 7242 struct file *file = iocb->ki_filp;
7243 struct inode *inode = file->f_mapping->host;
0934856d 7244 size_t count = 0;
2e60a51e 7245 int flags = 0;
38851cc1
MX		 7246 bool wakeup = true;
7247 bool relock = false;
0934856d 7248 ssize_t ret;
4b46fce2 7249
5a5f79b5 7250 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7251 offset, nr_segs))
5a5f79b5 7252 return 0;
3f7c579c 7253
38851cc1
MX		 7254 atomic_inc(&inode->i_dio_count);
7255 smp_mb__after_atomic_inc();
7256
0934856d
MX		 7257 if (rw & WRITE) {
7258 count = iov_length(iov, nr_segs);
38851cc1
MX		 7259 /*
7260 * If the write DIO is beyond the EOF, we need update
7261 * the isize, but it is protected by i_mutex. So we can
7262 * not unlock the i_mutex at this case.
7263 */
7264 if (offset + count <= inode->i_size) {
7265 mutex_unlock(&inode->i_mutex);
7266 relock = true;
7267 }
0934856d
MX		 7268 ret = btrfs_delalloc_reserve_space(inode, count);
7269 if (ret)
38851cc1
MX		 7270 goto out;
7271 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7272 &BTRFS_I(inode)->runtime_flags))) {
7273 inode_dio_done(inode);
7274 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7275 wakeup = false;
0934856d
MX		 7276 }
7277
7278 ret = __blockdev_direct_IO(rw, iocb, inode,
7279 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7280 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7281 btrfs_submit_direct, flags);
0934856d
MX		 7282 if (rw & WRITE) {
7283 if (ret < 0 && ret != -EIOCBQUEUED)
7284 btrfs_delalloc_release_space(inode, count);
172a5049 7285 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7286 btrfs_delalloc_release_space(inode,
7287 count - (size_t)ret);
172a5049
MX		 7288 else
7289 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7290 }
38851cc1 7291out:
2e60a51e
MX		 7292 if (wakeup)
7293 inode_dio_done(inode);
38851cc1
MX		 7294 if (relock)
7295 mutex_lock(&inode->i_mutex);
0934856d
MX		 7296
7297 return ret;
16432985
CM		 7298}
7299
05dadc09
TI		 7300#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7301
1506fcc8
YS		 7302static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7303 __u64 start, __u64 len)
7304{
05dadc09
TI		 7305 int ret;
7306
7307 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7308 if (ret)
7309 return ret;
7310
ec29ed5b 7311 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7312}
7313
a52d9a80 7314int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7315{
d1310b2e
CM		 7316 struct extent_io_tree *tree;
7317 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7318 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7319}
1832a6d5 7320
a52d9a80 7321static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7322{
d1310b2e 7323 struct extent_io_tree *tree;
b888db2b
CM		 7324
7325
7326 if (current->flags & PF_MEMALLOC) {
7327 redirty_page_for_writepage(wbc, page);
7328 unlock_page(page);
7329 return 0;
7330 }
d1310b2e 7331 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7332 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7333}
7334
48a3b636
ES		 7335static int btrfs_writepages(struct address_space *mapping,
7336 struct writeback_control *wbc)
b293f02e 7337{
d1310b2e 7338 struct extent_io_tree *tree;
771ed689 7339
d1310b2e 7340 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7341 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7342}
7343
3ab2fb5a
CM		 7344static int
7345btrfs_readpages(struct file *file, struct address_space *mapping,
7346 struct list_head *pages, unsigned nr_pages)
7347{
d1310b2e
CM		 7348 struct extent_io_tree *tree;
7349 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7350 return extent_readpages(tree, mapping, pages, nr_pages,
7351 btrfs_get_extent);
7352}
e6dcd2dc 7353static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7354{
d1310b2e
CM		 7355 struct extent_io_tree *tree;
7356 struct extent_map_tree *map;
a52d9a80 7357 int ret;
8c2383c3 7358
d1310b2e
CM		 7359 tree = &BTRFS_I(page->mapping->host)->io_tree;
7360 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7361 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7362 if (ret == 1) {
7363 ClearPagePrivate(page);
7364 set_page_private(page, 0);
7365 page_cache_release(page);
39279cc3 7366 }
a52d9a80 7367 return ret;
39279cc3
CM		 7368}
7369
e6dcd2dc
CM		 7370static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7371{
98509cfc
CM		 7372 if (PageWriteback(page) || PageDirty(page))
7373 return 0;
b335b003 7374 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7375}
7376
a52d9a80 7377static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 7378{
5fd02043 7379 struct inode *inode = page->mapping->host;
d1310b2e 7380 struct extent_io_tree *tree;
e6dcd2dc 7381 struct btrfs_ordered_extent *ordered;
2ac55d41 7382 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7383 u64 page_start = page_offset(page);
7384 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7385
8b62b72b
CM		 7386 /*
7387 * we have the page locked, so new writeback can't start,
7388 * and the dirty bit won't be cleared while we are here.
7389 *
7390 * Wait for IO on this page so that we can safely clear
7391 * the PagePrivate2 bit and do ordered accounting
7392 */
e6dcd2dc 7393 wait_on_page_writeback(page);
8b62b72b 7394
5fd02043 7395 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7396 if (offset) {
7397 btrfs_releasepage(page, GFP_NOFS);
7398 return;
7399 }
d0082371 7400 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7401 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7402 if (ordered) {
eb84ae03
CM		 7403 /*
7404 * IO on this page will never be started, so we need
7405 * to account for any ordered extents now
7406 */
e6dcd2dc
CM		 7407 clear_extent_bit(tree, page_start, page_end,
7408 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7409 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7410 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7411 /*
7412 * whoever cleared the private bit is responsible
7413 * for the finish_ordered_io
7414 */
5fd02043
JB		 7415 if (TestClearPagePrivate2(page) &&
7416 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7417 PAGE_CACHE_SIZE, 1)) {
7418 btrfs_finish_ordered_io(ordered);
8b62b72b 7419 }
e6dcd2dc 7420 btrfs_put_ordered_extent(ordered);
2ac55d41 7421 cached_state = NULL;
d0082371 7422 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7423 }
7424 clear_extent_bit(tree, page_start, page_end,
32c00aff 7425 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7426 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7427 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7428 __btrfs_releasepage(page, GFP_NOFS);
7429
4a096752 7430 ClearPageChecked(page);
9ad6b7bc 7431 if (PagePrivate(page)) {
9ad6b7bc
CM		 7432 ClearPagePrivate(page);
7433 set_page_private(page, 0);
7434 page_cache_release(page);
7435 }
39279cc3
CM		 7436}
7437
9ebefb18
CM		 7438/*
7439 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7440 * called from a page fault handler when a page is first dirtied. Hence we must
7441 * be careful to check for EOF conditions here. We set the page up correctly
7442 * for a written page which means we get ENOSPC checking when writing into
7443 * holes and correct delalloc and unwritten extent mapping on filesystems that
7444 * support these features.
7445 *
7446 * We are not allowed to take the i_mutex here so we have to play games to
7447 * protect against truncate races as the page could now be beyond EOF. Because
7448 * vmtruncate() writes the inode size before removing pages, once we have the
7449 * page lock we can determine safely if the page is beyond EOF. If it is not
7450 * beyond EOF, then the page is guaranteed safe against truncation until we
7451 * unlock the page.
7452 */
c2ec175c 7453int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7454{
c2ec175c 7455 struct page *page = vmf->page;
496ad9aa 7456 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7457 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7458 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7459 struct btrfs_ordered_extent *ordered;
2ac55d41 7460 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7461 char *kaddr;
7462 unsigned long zero_start;
9ebefb18 7463 loff_t size;
1832a6d5 7464 int ret;
9998eb70 7465 int reserved = 0;
a52d9a80 7466 u64 page_start;
e6dcd2dc 7467 u64 page_end;
9ebefb18 7468
b2b5ef5c 7469 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7470 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7471 if (!ret) {
e41f941a 7472 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7473 reserved = 1;
7474 }
56a76f82
NP		 7475 if (ret) {
7476 if (ret == -ENOMEM)
7477 ret = VM_FAULT_OOM;
7478 else /* -ENOSPC, -EIO, etc */
7479 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7480 if (reserved)
7481 goto out;
7482 goto out_noreserve;
56a76f82 7483 }
1832a6d5 7484
56a76f82 7485 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7486again:
9ebefb18 7487 lock_page(page);
9ebefb18 7488 size = i_size_read(inode);
e6dcd2dc
CM		 7489 page_start = page_offset(page);
7490 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7491
9ebefb18 7492 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7493 (page_start >= size)) {
9ebefb18
CM		 7494 /* page got truncated out from underneath us */
7495 goto out_unlock;
7496 }
e6dcd2dc
CM		 7497 wait_on_page_writeback(page);
7498
d0082371 7499 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7500 set_page_extent_mapped(page);
7501
eb84ae03
CM		 7502 /*
7503 * we can't set the delalloc bits if there are pending ordered
7504 * extents. Drop our locks and wait for them to finish
7505 */
e6dcd2dc
CM		 7506 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7507 if (ordered) {
2ac55d41
JB		 7508 unlock_extent_cached(io_tree, page_start, page_end,
7509 &cached_state, GFP_NOFS);
e6dcd2dc 7510 unlock_page(page);
eb84ae03 7511 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7512 btrfs_put_ordered_extent(ordered);
7513 goto again;
7514 }
7515
fbf19087
JB		 7516 /*
7517 * XXX - page_mkwrite gets called every time the page is dirtied, even
7518 * if it was already dirty, so for space accounting reasons we need to
7519 * clear any delalloc bits for the range we are fixing to save. There
7520 * is probably a better way to do this, but for now keep consistent with
7521 * prepare_pages in the normal write path.
7522 */
2ac55d41 7523 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7524 EXTENT_DIRTY | EXTENT_DELALLOC |
7525 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7526 0, 0, &cached_state, GFP_NOFS);
fbf19087 7527
2ac55d41
JB		 7528 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7529 &cached_state);
9ed74f2d 7530 if (ret) {
2ac55d41
JB		 7531 unlock_extent_cached(io_tree, page_start, page_end,
7532 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7533 ret = VM_FAULT_SIGBUS;
7534 goto out_unlock;
7535 }
e6dcd2dc 7536 ret = 0;
9ebefb18
CM		 7537
7538 /* page is wholly or partially inside EOF */
a52d9a80 7539 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7540 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7541 else
e6dcd2dc 7542 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7543
e6dcd2dc
CM		 7544 if (zero_start != PAGE_CACHE_SIZE) {
7545 kaddr = kmap(page);
7546 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7547 flush_dcache_page(page);
7548 kunmap(page);
7549 }
247e743c 7550 ClearPageChecked(page);
e6dcd2dc 7551 set_page_dirty(page);
50a9b214 7552 SetPageUptodate(page);
5a3f23d5 7553
257c62e1
CM		 7554 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7555 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7556 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7557
2ac55d41 7558 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7559
7560out_unlock:
b2b5ef5c
JK		 7561 if (!ret) {
7562 sb_end_pagefault(inode->i_sb);
50a9b214 7563 return VM_FAULT_LOCKED;
b2b5ef5c 7564 }
9ebefb18 7565 unlock_page(page);
1832a6d5 7566out:
ec39e180 7567 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7568out_noreserve:
b2b5ef5c 7569 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7570 return ret;
7571}
7572
a41ad394 7573static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7574{
7575 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7576 struct btrfs_block_rsv *rsv;
a71754fc 7577 int ret = 0;
3893e33b 7578 int err = 0;
39279cc3 7579 struct btrfs_trans_handle *trans;
dbe674a9 7580 u64 mask = root->sectorsize - 1;
07127184 7581 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7582
4a096752 7583 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 7584 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
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
fa0d7e3d
NP		 7808static void btrfs_i_callback(struct rcu_head *head)
7809{
7810 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7811 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7812}
7813
39279cc3
CM		 7814void btrfs_destroy_inode(struct inode *inode)
7815{
e6dcd2dc 7816 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7817 struct btrfs_root *root = BTRFS_I(inode)->root;
7818
b3d9b7a3 7819 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7820 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7821 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7822 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7823 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7824 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7825
a6dbd429
JB		 7826 /*
7827 * This can happen where we create an inode, but somebody else also
7828 * created the same inode and we need to destroy the one we already
7829 * created.
7830 */
7831 if (!root)
7832 goto free;
7833
5a3f23d5
CM		 7834 /*
7835 * Make sure we're properly removed from the ordered operation
7836 * lists.
7837 */
7838 smp_mb();
7839 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
199c2a9c 7840 spin_lock(&root->fs_info->ordered_root_lock);
5a3f23d5 7841 list_del_init(&BTRFS_I(inode)->ordered_operations);
199c2a9c 7842 spin_unlock(&root->fs_info->ordered_root_lock);
5a3f23d5
CM		 7843 }
7844
8a35d95f
JB		 7845 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7846 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb
SK		 7847 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7848 (unsigned long long)btrfs_ino(inode));
8a35d95f 7849 atomic_dec(&root->orphan_inodes);
7b128766 7850 }
7b128766 7851
d397712b 7852 while (1) {
e6dcd2dc
CM		 7853 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7854 if (!ordered)
7855 break;
7856 else {
c2cf52eb
SK		 7857 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7858 (unsigned long long)ordered->file_offset,
7859 (unsigned long long)ordered->len);
e6dcd2dc
CM		 7860 btrfs_remove_ordered_extent(inode, ordered);
7861 btrfs_put_ordered_extent(ordered);
7862 btrfs_put_ordered_extent(ordered);
7863 }
7864 }
5d4f98a2 7865 inode_tree_del(inode);
5b21f2ed 7866 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7867free:
fa0d7e3d 7868 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7869}
7870
45321ac5 7871int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7872{
7873 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7874
6379ef9f
NA		 7875 if (root == NULL)
7876 return 1;
7877
fa6ac876 7878 /* the snap/subvol tree is on deleting */
0af3d00b 7879 if (btrfs_root_refs(&root->root_item) == 0 &&
fa6ac876 7880 root != root->fs_info->tree_root)
45321ac5 7881 return 1;
76dda93c 7882 else
45321ac5 7883 return generic_drop_inode(inode);
76dda93c
YZ		 7884}
7885
0ee0fda0 7886static void init_once(void *foo)
39279cc3
CM		 7887{
7888 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7889
7890 inode_init_once(&ei->vfs_inode);
7891}
7892
7893void btrfs_destroy_cachep(void)
7894{
8c0a8537
KS		 7895 /*
7896 * Make sure all delayed rcu free inodes are flushed before we
7897 * destroy cache.
7898 */
7899 rcu_barrier();
39279cc3
CM		 7900 if (btrfs_inode_cachep)
7901 kmem_cache_destroy(btrfs_inode_cachep);
7902 if (btrfs_trans_handle_cachep)
7903 kmem_cache_destroy(btrfs_trans_handle_cachep);
7904 if (btrfs_transaction_cachep)
7905 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7906 if (btrfs_path_cachep)
7907 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7908 if (btrfs_free_space_cachep)
7909 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 7910 if (btrfs_delalloc_work_cachep)
7911 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 7912}
7913
7914int btrfs_init_cachep(void)
7915{
837e1972 7916 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 7917 sizeof(struct btrfs_inode), 0,
7918 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7919 if (!btrfs_inode_cachep)
7920 goto fail;
9601e3f6 7921
837e1972 7922 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 7923 sizeof(struct btrfs_trans_handle), 0,
7924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7925 if (!btrfs_trans_handle_cachep)
7926 goto fail;
9601e3f6 7927
837e1972 7928 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 7929 sizeof(struct btrfs_transaction), 0,
7930 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7931 if (!btrfs_transaction_cachep)
7932 goto fail;
9601e3f6 7933
837e1972 7934 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 7935 sizeof(struct btrfs_path), 0,
7936 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7937 if (!btrfs_path_cachep)
7938 goto fail;
9601e3f6 7939
837e1972 7940 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 7941 sizeof(struct btrfs_free_space), 0,
7942 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7943 if (!btrfs_free_space_cachep)
7944 goto fail;
7945
8ccf6f19
MX		 7946 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7947 sizeof(struct btrfs_delalloc_work), 0,
7948 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7949 NULL);
7950 if (!btrfs_delalloc_work_cachep)
7951 goto fail;
7952
39279cc3
CM		 7953 return 0;
7954fail:
7955 btrfs_destroy_cachep();
7956 return -ENOMEM;
7957}
7958
7959static int btrfs_getattr(struct vfsmount *mnt,
7960 struct dentry *dentry, struct kstat *stat)
7961{
df0af1a5 7962 u64 delalloc_bytes;
39279cc3 7963 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7964 u32 blocksize = inode->i_sb->s_blocksize;
7965
39279cc3 7966 generic_fillattr(inode, stat);
0ee5dc67 7967 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7968 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 7969
7970 spin_lock(&BTRFS_I(inode)->lock);
7971 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7972 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 7973 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 7974 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7975 return 0;
7976}
7977
d397712b
CM		 7978static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7979 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7980{
7981 struct btrfs_trans_handle *trans;
7982 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7983 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7984 struct inode *new_inode = new_dentry->d_inode;
7985 struct inode *old_inode = old_dentry->d_inode;
7986 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7987 u64 index = 0;
4df27c4d 7988 u64 root_objectid;
39279cc3 7989 int ret;
33345d01 7990 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7991
33345d01 7992 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 7993 return -EPERM;
7994
4df27c4d 7995 /* we only allow rename subvolume link between subvolumes */
33345d01 7996 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 7997 return -EXDEV;
7998
33345d01
LZ		 7999 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8000 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8001 return -ENOTEMPTY;
5f39d397 8002
4df27c4d
YZ		 8003 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8004 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8005 return -ENOTEMPTY;
9c52057c
CM		 8006
8007
8008 /* check for collisions, even if the name isn't there */
8009 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8010 new_dentry->d_name.name,
8011 new_dentry->d_name.len);
8012
8013 if (ret) {
8014 if (ret == -EEXIST) {
8015 /* we shouldn't get
8016 * eexist without a new_inode */
8017 if (!new_inode) {
8018 WARN_ON(1);
8019 return ret;
8020 }
8021 } else {
8022 /* maybe -EOVERFLOW */
8023 return ret;
8024 }
8025 }
8026 ret = 0;
8027
5a3f23d5
CM		 8028 /*
8029 * we're using rename to replace one file with another.
8030 * and the replacement file is large. Start IO on it now so
8031 * we don't add too much work to the end of the transaction
8032 */
4baf8c92 8033 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8034 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8035 filemap_flush(old_inode->i_mapping);
8036
76dda93c 8037 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8038 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8039 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8040 /*
8041 * We want to reserve the absolute worst case amount of items. So if
8042 * both inodes are subvols and we need to unlink them then that would
8043 * require 4 item modifications, but if they are both normal inodes it
8044 * would require 5 item modifications, so we'll assume their normal
8045 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8046 * should cover the worst case number of items we'll modify.
8047 */
6e137ed3 8048 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8049 if (IS_ERR(trans)) {
8050 ret = PTR_ERR(trans);
8051 goto out_notrans;
8052 }
76dda93c 8053
4df27c4d
YZ		 8054 if (dest != root)
8055 btrfs_record_root_in_trans(trans, dest);
5f39d397 8056
a5719521
YZ		 8057 ret = btrfs_set_inode_index(new_dir, &index);
8058 if (ret)
8059 goto out_fail;
5a3f23d5 8060
33345d01 8061 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8062 /* force full log commit if subvolume involved. */
8063 root->fs_info->last_trans_log_full_commit = trans->transid;
8064 } else {
a5719521
YZ		 8065 ret = btrfs_insert_inode_ref(trans, dest,
8066 new_dentry->d_name.name,
8067 new_dentry->d_name.len,
33345d01
LZ		 8068 old_ino,
8069 btrfs_ino(new_dir), index);
a5719521
YZ		 8070 if (ret)
8071 goto out_fail;
4df27c4d
YZ		 8072 /*
8073 * this is an ugly little race, but the rename is required
8074 * to make sure that if we crash, the inode is either at the
8075 * old name or the new one. pinning the log transaction lets
8076 * us make sure we don't allow a log commit to come in after
8077 * we unlink the name but before we add the new name back in.
8078 */
8079 btrfs_pin_log_trans(root);
8080 }
5a3f23d5
CM		 8081 /*
8082 * make sure the inode gets flushed if it is replacing
8083 * something.
8084 */
33345d01 8085 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8086 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8087
0c4d2d95
JB		 8088 inode_inc_iversion(old_dir);
8089 inode_inc_iversion(new_dir);
8090 inode_inc_iversion(old_inode);
39279cc3
CM		 8091 old_dir->i_ctime = old_dir->i_mtime = ctime;
8092 new_dir->i_ctime = new_dir->i_mtime = ctime;
8093 old_inode->i_ctime = ctime;
5f39d397 8094
12fcfd22
CM		 8095 if (old_dentry->d_parent != new_dentry->d_parent)
8096 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8097
33345d01 8098 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8099 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8100 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8101 old_dentry->d_name.name,
8102 old_dentry->d_name.len);
8103 } else {
92986796
AV		 8104 ret = __btrfs_unlink_inode(trans, root, old_dir,
8105 old_dentry->d_inode,
8106 old_dentry->d_name.name,
8107 old_dentry->d_name.len);
8108 if (!ret)
8109 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8110 }
79787eaa
JM		 8111 if (ret) {
8112 btrfs_abort_transaction(trans, root, ret);
8113 goto out_fail;
8114 }
39279cc3
CM		 8115
8116 if (new_inode) {
0c4d2d95 8117 inode_inc_iversion(new_inode);
39279cc3 8118 new_inode->i_ctime = CURRENT_TIME;
33345d01 8119 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8120 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8121 root_objectid = BTRFS_I(new_inode)->location.objectid;
8122 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8123 root_objectid,
8124 new_dentry->d_name.name,
8125 new_dentry->d_name.len);
8126 BUG_ON(new_inode->i_nlink == 0);
8127 } else {
8128 ret = btrfs_unlink_inode(trans, dest, new_dir,
8129 new_dentry->d_inode,
8130 new_dentry->d_name.name,
8131 new_dentry->d_name.len);
8132 }
79787eaa 8133 if (!ret && new_inode->i_nlink == 0) {
e02119d5 8134 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 8135 BUG_ON(ret);
7b128766 8136 }
79787eaa
JM		 8137 if (ret) {
8138 btrfs_abort_transaction(trans, root, ret);
8139 goto out_fail;
8140 }
39279cc3 8141 }
aec7477b 8142
4df27c4d
YZ		 8143 ret = btrfs_add_link(trans, new_dir, old_inode,
8144 new_dentry->d_name.name,
a5719521 8145 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8146 if (ret) {
8147 btrfs_abort_transaction(trans, root, ret);
8148 goto out_fail;
8149 }
39279cc3 8150
33345d01 8151 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8152 struct dentry *parent = new_dentry->d_parent;
6a912213 8153 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8154 btrfs_end_log_trans(root);
8155 }
39279cc3 8156out_fail:
7ad85bb7 8157 btrfs_end_transaction(trans, root);
b44c59a8 8158out_notrans:
33345d01 8159 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8160 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8161
39279cc3
CM		 8162 return ret;
8163}
8164
8ccf6f19
MX		 8165static void btrfs_run_delalloc_work(struct btrfs_work *work)
8166{
8167 struct btrfs_delalloc_work *delalloc_work;
8168
8169 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8170 work);
8171 if (delalloc_work->wait)
8172 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8173 else
8174 filemap_flush(delalloc_work->inode->i_mapping);
8175
8176 if (delalloc_work->delay_iput)
8177 btrfs_add_delayed_iput(delalloc_work->inode);
8178 else
8179 iput(delalloc_work->inode);
8180 complete(&delalloc_work->completion);
8181}
8182
8183struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8184 int wait, int delay_iput)
8185{
8186 struct btrfs_delalloc_work *work;
8187
8188 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8189 if (!work)
8190 return NULL;
8191
8192 init_completion(&work->completion);
8193 INIT_LIST_HEAD(&work->list);
8194 work->inode = inode;
8195 work->wait = wait;
8196 work->delay_iput = delay_iput;
8197 work->work.func = btrfs_run_delalloc_work;
8198
8199 return work;
8200}
8201
8202void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8203{
8204 wait_for_completion(&work->completion);
8205 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8206}
8207
d352ac68
CM		 8208/*
8209 * some fairly slow code that needs optimization. This walks the list
8210 * of all the inodes with pending delalloc and forces them to disk.
8211 */
eb73c1b7 8212static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8213{
ea8c2819 8214 struct btrfs_inode *binode;
5b21f2ed 8215 struct inode *inode;
8ccf6f19
MX		 8216 struct btrfs_delalloc_work *work, *next;
8217 struct list_head works;
1eafa6c7 8218 struct list_head splice;
8ccf6f19 8219 int ret = 0;
ea8c2819 8220
8ccf6f19 8221 INIT_LIST_HEAD(&works);
1eafa6c7 8222 INIT_LIST_HEAD(&splice);
63607cc8 8223
eb73c1b7
MX		 8224 spin_lock(&root->delalloc_lock);
8225 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8226 while (!list_empty(&splice)) {
8227 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8228 delalloc_inodes);
1eafa6c7 8229
eb73c1b7
MX		 8230 list_move_tail(&binode->delalloc_inodes,
8231 &root->delalloc_inodes);
5b21f2ed 8232 inode = igrab(&binode->vfs_inode);
df0af1a5 8233 if (!inode) {
eb73c1b7 8234 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8235 continue;
df0af1a5 8236 }
eb73c1b7 8237 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8238
8239 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8240 if (unlikely(!work)) {
8241 ret = -ENOMEM;
8242 goto out;
5b21f2ed 8243 }
1eafa6c7
MX		 8244 list_add_tail(&work->list, &works);
8245 btrfs_queue_worker(&root->fs_info->flush_workers,
8246 &work->work);
8247
5b21f2ed 8248 cond_resched();
eb73c1b7 8249 spin_lock(&root->delalloc_lock);
ea8c2819 8250 }
eb73c1b7 8251 spin_unlock(&root->delalloc_lock);
8c8bee1d 8252
1eafa6c7
MX		 8253 list_for_each_entry_safe(work, next, &works, list) {
8254 list_del_init(&work->list);
8255 btrfs_wait_and_free_delalloc_work(work);
8256 }
eb73c1b7
MX		 8257 return 0;
8258out:
8259 list_for_each_entry_safe(work, next, &works, list) {
8260 list_del_init(&work->list);
8261 btrfs_wait_and_free_delalloc_work(work);
8262 }
8263
8264 if (!list_empty_careful(&splice)) {
8265 spin_lock(&root->delalloc_lock);
8266 list_splice_tail(&splice, &root->delalloc_inodes);
8267 spin_unlock(&root->delalloc_lock);
8268 }
8269 return ret;
8270}
1eafa6c7 8271
eb73c1b7
MX		 8272int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8273{
8274 int ret;
8275
8276 if (root->fs_info->sb->s_flags & MS_RDONLY)
8277 return -EROFS;
8278
8279 ret = __start_delalloc_inodes(root, delay_iput);
8280 /*
8281 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8282 * we have to make sure the IO is actually started and that
8283 * ordered extents get created before we return
8284 */
8285 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8286 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8287 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8288 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8289 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8290 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8291 }
8292 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8293 return ret;
8294}
8295
8296int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8297 int delay_iput)
8298{
8299 struct btrfs_root *root;
8300 struct list_head splice;
8301 int ret;
8302
8303 if (fs_info->sb->s_flags & MS_RDONLY)
8304 return -EROFS;
8305
8306 INIT_LIST_HEAD(&splice);
8307
8308 spin_lock(&fs_info->delalloc_root_lock);
8309 list_splice_init(&fs_info->delalloc_roots, &splice);
8310 while (!list_empty(&splice)) {
8311 root = list_first_entry(&splice, struct btrfs_root,
8312 delalloc_root);
8313 root = btrfs_grab_fs_root(root);
8314 BUG_ON(!root);
8315 list_move_tail(&root->delalloc_root,
8316 &fs_info->delalloc_roots);
8317 spin_unlock(&fs_info->delalloc_root_lock);
8318
8319 ret = __start_delalloc_inodes(root, delay_iput);
8320 btrfs_put_fs_root(root);
8321 if (ret)
8322 goto out;
8323
8324 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8325 }
eb73c1b7 8326 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8327
eb73c1b7
MX		 8328 atomic_inc(&fs_info->async_submit_draining);
8329 while (atomic_read(&fs_info->nr_async_submits) ||
8330 atomic_read(&fs_info->async_delalloc_pages)) {
8331 wait_event(fs_info->async_submit_wait,
8332 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8333 atomic_read(&fs_info->async_delalloc_pages) == 0));
8334 }
8335 atomic_dec(&fs_info->async_submit_draining);
8336 return 0;
8337out:
1eafa6c7 8338 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8339 spin_lock(&fs_info->delalloc_root_lock);
8340 list_splice_tail(&splice, &fs_info->delalloc_roots);
8341 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8342 }
8ccf6f19 8343 return ret;
ea8c2819
CM		 8344}
8345
39279cc3
CM		 8346static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8347 const char *symname)
8348{
8349 struct btrfs_trans_handle *trans;
8350 struct btrfs_root *root = BTRFS_I(dir)->root;
8351 struct btrfs_path *path;
8352 struct btrfs_key key;
1832a6d5 8353 struct inode *inode = NULL;
39279cc3
CM		 8354 int err;
8355 int drop_inode = 0;
8356 u64 objectid;
00e4e6b3 8357 u64 index = 0 ;
39279cc3
CM		 8358 int name_len;
8359 int datasize;
5f39d397 8360 unsigned long ptr;
39279cc3 8361 struct btrfs_file_extent_item *ei;
5f39d397 8362 struct extent_buffer *leaf;
39279cc3
CM		 8363
8364 name_len = strlen(symname) + 1;
8365 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8366 return -ENAMETOOLONG;
1832a6d5 8367
9ed74f2d
JB		 8368 /*
8369 * 2 items for inode item and ref
8370 * 2 items for dir items
8371 * 1 item for xattr if selinux is on
8372 */
a22285a6
YZ		 8373 trans = btrfs_start_transaction(root, 5);
8374 if (IS_ERR(trans))
8375 return PTR_ERR(trans);
1832a6d5 8376
581bb050
LZ		 8377 err = btrfs_find_free_ino(root, &objectid);
8378 if (err)
8379 goto out_unlock;
8380
aec7477b 8381 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8382 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8383 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8384 if (IS_ERR(inode)) {
8385 err = PTR_ERR(inode);
39279cc3 8386 goto out_unlock;
7cf96da3 8387 }
39279cc3 8388
2a7dba39 8389 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8390 if (err) {
8391 drop_inode = 1;
8392 goto out_unlock;
8393 }
8394
ad19db71
CS		 8395 /*
8396 * If the active LSM wants to access the inode during
8397 * d_instantiate it needs these. Smack checks to see
8398 * if the filesystem supports xattrs by looking at the
8399 * ops vector.
8400 */
8401 inode->i_fop = &btrfs_file_operations;
8402 inode->i_op = &btrfs_file_inode_operations;
8403
a1b075d2 8404 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8405 if (err)
8406 drop_inode = 1;
8407 else {
8408 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8409 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8410 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8411 }
39279cc3
CM		 8412 if (drop_inode)
8413 goto out_unlock;
8414
8415 path = btrfs_alloc_path();
d8926bb3
MF		 8416 if (!path) {
8417 err = -ENOMEM;
8418 drop_inode = 1;
8419 goto out_unlock;
8420 }
33345d01 8421 key.objectid = btrfs_ino(inode);
39279cc3 8422 key.offset = 0;
39279cc3
CM		 8423 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8424 datasize = btrfs_file_extent_calc_inline_size(name_len);
8425 err = btrfs_insert_empty_item(trans, root, path, &key,
8426 datasize);
54aa1f4d
CM		 8427 if (err) {
8428 drop_inode = 1;
b0839166 8429 btrfs_free_path(path);
54aa1f4d
CM		 8430 goto out_unlock;
8431 }
5f39d397
CM		 8432 leaf = path->nodes[0];
8433 ei = btrfs_item_ptr(leaf, path->slots[0],
8434 struct btrfs_file_extent_item);
8435 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8436 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8437 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8438 btrfs_set_file_extent_encryption(leaf, ei, 0);
8439 btrfs_set_file_extent_compression(leaf, ei, 0);
8440 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8441 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8442
39279cc3 8443 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8444 write_extent_buffer(leaf, symname, ptr, name_len);
8445 btrfs_mark_buffer_dirty(leaf);
39279cc3 8446 btrfs_free_path(path);
5f39d397 8447
39279cc3
CM		 8448 inode->i_op = &btrfs_symlink_inode_operations;
8449 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8450 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8451 inode_set_bytes(inode, name_len);
dbe674a9 8452 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 8453 err = btrfs_update_inode(trans, root, inode);
8454 if (err)
8455 drop_inode = 1;
39279cc3
CM		 8456
8457out_unlock:
08c422c2
AV		 8458 if (!err)
8459 d_instantiate(dentry, inode);
7ad85bb7 8460 btrfs_end_transaction(trans, root);
39279cc3
CM		 8461 if (drop_inode) {
8462 inode_dec_link_count(inode);
8463 iput(inode);
8464 }
b53d3f5d 8465 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8466 return err;
8467}
16432985 8468
0af3d00b
JB		 8469static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8470 u64 start, u64 num_bytes, u64 min_size,
8471 loff_t actual_len, u64 *alloc_hint,
8472 struct btrfs_trans_handle *trans)
d899e052 8473{
5dc562c5
JB		 8474 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8475 struct extent_map *em;
d899e052
YZ		 8476 struct btrfs_root *root = BTRFS_I(inode)->root;
8477 struct btrfs_key ins;
d899e052 8478 u64 cur_offset = start;
55a61d1d 8479 u64 i_size;
154ea289 8480 u64 cur_bytes;
d899e052 8481 int ret = 0;
0af3d00b 8482 bool own_trans = true;
d899e052 8483
0af3d00b
JB		 8484 if (trans)
8485 own_trans = false;
d899e052 8486 while (num_bytes > 0) {
0af3d00b
JB		 8487 if (own_trans) {
8488 trans = btrfs_start_transaction(root, 3);
8489 if (IS_ERR(trans)) {
8490 ret = PTR_ERR(trans);
8491 break;
8492 }
5a303d5d
YZ		 8493 }
8494
154ea289
CM		 8495 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8496 cur_bytes = max(cur_bytes, min_size);
8497 ret = btrfs_reserve_extent(trans, root, cur_bytes,
24542bf7 8498 min_size, 0, *alloc_hint, &ins, 1);
5a303d5d 8499 if (ret) {
0af3d00b
JB		 8500 if (own_trans)
8501 btrfs_end_transaction(trans, root);
a22285a6 8502 break;
d899e052 8503 }
5a303d5d 8504
d899e052
YZ		 8505 ret = insert_reserved_file_extent(trans, inode,
8506 cur_offset, ins.objectid,
8507 ins.offset, ins.offset,
920bbbfb 8508 ins.offset, 0, 0, 0,
d899e052 8509 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 8510 if (ret) {
8511 btrfs_abort_transaction(trans, root, ret);
8512 if (own_trans)
8513 btrfs_end_transaction(trans, root);
8514 break;
8515 }
a1ed835e
CM		 8516 btrfs_drop_extent_cache(inode, cur_offset,
8517 cur_offset + ins.offset -1, 0);
5a303d5d 8518
5dc562c5
JB		 8519 em = alloc_extent_map();
8520 if (!em) {
8521 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8522 &BTRFS_I(inode)->runtime_flags);
8523 goto next;
8524 }
8525
8526 em->start = cur_offset;
8527 em->orig_start = cur_offset;
8528 em->len = ins.offset;
8529 em->block_start = ins.objectid;
8530 em->block_len = ins.offset;
b4939680 8531 em->orig_block_len = ins.offset;
cc95bef6 8532 em->ram_bytes = ins.offset;
5dc562c5
JB		 8533 em->bdev = root->fs_info->fs_devices->latest_bdev;
8534 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8535 em->generation = trans->transid;
8536
8537 while (1) {
8538 write_lock(&em_tree->lock);
09a2a8f9 8539 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8540 write_unlock(&em_tree->lock);
8541 if (ret != -EEXIST)
8542 break;
8543 btrfs_drop_extent_cache(inode, cur_offset,
8544 cur_offset + ins.offset - 1,
8545 0);
8546 }
8547 free_extent_map(em);
8548next:
d899e052
YZ		 8549 num_bytes -= ins.offset;
8550 cur_offset += ins.offset;
efa56464 8551 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8552
0c4d2d95 8553 inode_inc_iversion(inode);
d899e052 8554 inode->i_ctime = CURRENT_TIME;
6cbff00f 8555 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8556 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8557 (actual_len > inode->i_size) &&
8558 (cur_offset > inode->i_size)) {
d1ea6a61 8559 if (cur_offset > actual_len)
55a61d1d 8560 i_size = actual_len;
d1ea6a61 8561 else
55a61d1d
JB		 8562 i_size = cur_offset;
8563 i_size_write(inode, i_size);
8564 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8565 }
8566
d899e052 8567 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8568
8569 if (ret) {
8570 btrfs_abort_transaction(trans, root, ret);
8571 if (own_trans)
8572 btrfs_end_transaction(trans, root);
8573 break;
8574 }
d899e052 8575
0af3d00b
JB		 8576 if (own_trans)
8577 btrfs_end_transaction(trans, root);
5a303d5d 8578 }
d899e052
YZ		 8579 return ret;
8580}
8581
0af3d00b
JB		 8582int btrfs_prealloc_file_range(struct inode *inode, int mode,
8583 u64 start, u64 num_bytes, u64 min_size,
8584 loff_t actual_len, u64 *alloc_hint)
8585{
8586 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8587 min_size, actual_len, alloc_hint,
8588 NULL);
8589}
8590
8591int btrfs_prealloc_file_range_trans(struct inode *inode,
8592 struct btrfs_trans_handle *trans, int mode,
8593 u64 start, u64 num_bytes, u64 min_size,
8594 loff_t actual_len, u64 *alloc_hint)
8595{
8596 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8597 min_size, actual_len, alloc_hint, trans);
8598}
8599
e6dcd2dc
CM		 8600static int btrfs_set_page_dirty(struct page *page)
8601{
e6dcd2dc
CM		 8602 return __set_page_dirty_nobuffers(page);
8603}
8604
10556cb2 8605static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8606{
b83cc969 8607 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8608 umode_t mode = inode->i_mode;
b83cc969 8609
cb6db4e5
JM		 8610 if (mask & MAY_WRITE &&
8611 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8612 if (btrfs_root_readonly(root))
8613 return -EROFS;
8614 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8615 return -EACCES;
8616 }
2830ba7f 8617 return generic_permission(inode, mask);
fdebe2bd 8618}
39279cc3 8619
6e1d5dcc 8620static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8621 .getattr = btrfs_getattr,
39279cc3
CM		 8622 .lookup = btrfs_lookup,
8623 .create = btrfs_create,
8624 .unlink = btrfs_unlink,
8625 .link = btrfs_link,
8626 .mkdir = btrfs_mkdir,
8627 .rmdir = btrfs_rmdir,
8628 .rename = btrfs_rename,
8629 .symlink = btrfs_symlink,
8630 .setattr = btrfs_setattr,
618e21d5 8631 .mknod = btrfs_mknod,
95819c05
CH		 8632 .setxattr = btrfs_setxattr,
8633 .getxattr = btrfs_getxattr,
5103e947 8634 .listxattr = btrfs_listxattr,
95819c05 8635 .removexattr = btrfs_removexattr,
fdebe2bd 8636 .permission = btrfs_permission,
4e34e719 8637 .get_acl = btrfs_get_acl,
39279cc3 8638};
6e1d5dcc 8639static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8640 .lookup = btrfs_lookup,
fdebe2bd 8641 .permission = btrfs_permission,
4e34e719 8642 .get_acl = btrfs_get_acl,
39279cc3 8643};
76dda93c 8644
828c0950 8645static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8646 .llseek = generic_file_llseek,
8647 .read = generic_read_dir,
cbdf5a24 8648 .readdir = btrfs_real_readdir,
34287aa3 8649 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8650#ifdef CONFIG_COMPAT
34287aa3 8651 .compat_ioctl = btrfs_ioctl,
39279cc3 8652#endif
6bf13c0c 8653 .release = btrfs_release_file,
e02119d5 8654 .fsync = btrfs_sync_file,
39279cc3
CM		 8655};
8656
d1310b2e 8657static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8658 .fill_delalloc = run_delalloc_range,
065631f6 8659 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8660 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8661 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8662 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8663 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8664 .set_bit_hook = btrfs_set_bit_hook,
8665 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8666 .merge_extent_hook = btrfs_merge_extent_hook,
8667 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8668};
8669
35054394
CM		 8670/*
8671 * btrfs doesn't support the bmap operation because swapfiles
8672 * use bmap to make a mapping of extents in the file. They assume
8673 * these extents won't change over the life of the file and they
8674 * use the bmap result to do IO directly to the drive.
8675 *
8676 * the btrfs bmap call would return logical addresses that aren't
8677 * suitable for IO and they also will change frequently as COW
8678 * operations happen. So, swapfile + btrfs == corruption.
8679 *
8680 * For now we're avoiding this by dropping bmap.
8681 */
7f09410b 8682static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8683 .readpage = btrfs_readpage,
8684 .writepage = btrfs_writepage,
b293f02e 8685 .writepages = btrfs_writepages,
3ab2fb5a 8686 .readpages = btrfs_readpages,
16432985 8687 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8688 .invalidatepage = btrfs_invalidatepage,
8689 .releasepage = btrfs_releasepage,
e6dcd2dc 8690 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8691 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8692};
8693
7f09410b 8694static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8695 .readpage = btrfs_readpage,
8696 .writepage = btrfs_writepage,
2bf5a725
CM		 8697 .invalidatepage = btrfs_invalidatepage,
8698 .releasepage = btrfs_releasepage,
39279cc3
CM		 8699};
8700
6e1d5dcc 8701static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8702 .getattr = btrfs_getattr,
8703 .setattr = btrfs_setattr,
95819c05
CH		 8704 .setxattr = btrfs_setxattr,
8705 .getxattr = btrfs_getxattr,
5103e947 8706 .listxattr = btrfs_listxattr,
95819c05 8707 .removexattr = btrfs_removexattr,
fdebe2bd 8708 .permission = btrfs_permission,
1506fcc8 8709 .fiemap = btrfs_fiemap,
4e34e719 8710 .get_acl = btrfs_get_acl,
e41f941a 8711 .update_time = btrfs_update_time,
39279cc3 8712};
6e1d5dcc 8713static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8714 .getattr = btrfs_getattr,
8715 .setattr = btrfs_setattr,
fdebe2bd 8716 .permission = btrfs_permission,
95819c05
CH		 8717 .setxattr = btrfs_setxattr,
8718 .getxattr = btrfs_getxattr,
33268eaf 8719 .listxattr = btrfs_listxattr,
95819c05 8720 .removexattr = btrfs_removexattr,
4e34e719 8721 .get_acl = btrfs_get_acl,
e41f941a 8722 .update_time = btrfs_update_time,
618e21d5 8723};
6e1d5dcc 8724static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8725 .readlink = generic_readlink,
8726 .follow_link = page_follow_link_light,
8727 .put_link = page_put_link,
f209561a 8728 .getattr = btrfs_getattr,
22c44fe6 8729 .setattr = btrfs_setattr,
fdebe2bd 8730 .permission = btrfs_permission,
0279b4cd
JO		 8731 .setxattr = btrfs_setxattr,
8732 .getxattr = btrfs_getxattr,
8733 .listxattr = btrfs_listxattr,
8734 .removexattr = btrfs_removexattr,
4e34e719 8735 .get_acl = btrfs_get_acl,
e41f941a 8736 .update_time = btrfs_update_time,
39279cc3 8737};
76dda93c 8738
82d339d9 8739const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8740 .d_delete = btrfs_dentry_delete,
b4aff1f8 8741 .d_release = btrfs_dentry_release,
76dda93c 8742};
Linux 6.x block layer and io_uring tree(s)