projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: stop using try_to_writeback_inodes_sb_nr to flush delalloc
[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>
f23b5a59 44#include <linux/posix_acl_xattr.h>
4b4e25f2 45#include "compat.h"
39279cc3
CM		 46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
39279cc3 50#include "print-tree.h"
e6dcd2dc 51#include "ordered-data.h"
95819c05 52#include "xattr.h"
e02119d5 53#include "tree-log.h"
4a54c8c1 54#include "volumes.h"
c8b97818 55#include "compression.h"
b4ce94de 56#include "locking.h"
dc89e982 57#include "free-space-cache.h"
581bb050 58#include "inode-map.h"
38c227d8 59#include "backref.h"
f23b5a59 60#include "hash.h"
39279cc3
CM		 61
62struct btrfs_iget_args {
63 u64 ino;
64 struct btrfs_root *root;
65};
66
6e1d5dcc
AD		 67static const struct inode_operations btrfs_dir_inode_operations;
68static const struct inode_operations btrfs_symlink_inode_operations;
69static const struct inode_operations btrfs_dir_ro_inode_operations;
70static const struct inode_operations btrfs_special_inode_operations;
71static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 72static const struct address_space_operations btrfs_aops;
73static const struct address_space_operations btrfs_symlink_aops;
828c0950 74static const struct file_operations btrfs_dir_file_operations;
d1310b2e 75static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 76
77static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 78static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 79struct kmem_cache *btrfs_trans_handle_cachep;
80struct kmem_cache *btrfs_transaction_cachep;
39279cc3 81struct kmem_cache *btrfs_path_cachep;
dc89e982 82struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 83
84#define S_SHIFT 12
85static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
86 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
87 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
88 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
89 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
90 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
91 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
92 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
93};
94
3972f260 95static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 96static int btrfs_truncate(struct inode *inode);
5fd02043 97static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 98static noinline int cow_file_range(struct inode *inode,
99 struct page *locked_page,
100 u64 start, u64 end, int *page_started,
101 unsigned long *nr_written, int unlock);
70c8a91c
JB		 102static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
103 u64 len, u64 orig_start,
104 u64 block_start, u64 block_len,
cc95bef6
JB		 105 u64 orig_block_len, u64 ram_bytes,
106 int type);
7b128766 107
48a3b636
ES		 108static int btrfs_dirty_inode(struct inode *inode);
109
f34f57a3 110static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 111 struct inode *inode, struct inode *dir,
112 const struct qstr *qstr)
0279b4cd
JO		 113{
114 int err;
115
f34f57a3 116 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 117 if (!err)
2a7dba39 118 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 119 return err;
120}
121
c8b97818
CM		 122/*
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
126 */
d397712b 127static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
fe3f566c 130 int compress_type,
c8b97818
CM		 131 struct page **compressed_pages)
132{
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
c8b97818 145
fe3f566c 146 if (compressed_size && compressed_pages)
c8b97818 147 cur_size = compressed_size;
c8b97818 148
d397712b
CM		 149 path = btrfs_alloc_path();
150 if (!path)
c8b97818
CM		 151 return -ENOMEM;
152
b9473439 153 path->leave_spinning = 1;
c8b97818 154
33345d01 155 key.objectid = btrfs_ino(inode);
c8b97818
CM		 156 key.offset = start;
157 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159
160 inode_add_bytes(inode, size);
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 datasize);
c8b97818
CM		 163 if (ret) {
164 err = ret;
c8b97818
CM		 165 goto fail;
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
261507a0 177 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 178 struct page *cpage;
179 int i = 0;
d397712b 180 while (compressed_size > 0) {
c8b97818 181 cpage = compressed_pages[i];
5b050f04 182 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 183 PAGE_CACHE_SIZE);
184
7ac687d9 185 kaddr = kmap_atomic(cpage);
c8b97818 186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 187 kunmap_atomic(kaddr);
c8b97818
CM		 188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
261507a0 194 compress_type);
c8b97818
CM		 195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 199 kaddr = kmap_atomic(page);
c8b97818
CM		 200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 202 kunmap_atomic(kaddr);
c8b97818
CM		 203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
207
c2167754
YZ		 208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
c8b97818 217 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 218 ret = btrfs_update_inode(trans, root, inode);
c2167754 219
79787eaa 220 return ret;
c8b97818
CM		 221fail:
222 btrfs_free_path(path);
223 return err;
224}
225
226
227/*
228 * conditionally insert an inline extent into the file. This
229 * does the checks required to make sure the data is small enough
230 * to fit as an inline extent.
231 */
7f366cfe 232static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 233 struct btrfs_root *root,
234 struct inode *inode, u64 start, u64 end,
fe3f566c 235 size_t compressed_size, int compress_type,
c8b97818
CM		 236 struct page **compressed_pages)
237{
238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
fda2832f 241 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 242 u64 data_len = inline_len;
243 int ret;
244
245 if (compressed_size)
246 data_len = compressed_size;
247
248 if (start > 0 ||
70b99e69 249 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 250 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
251 (!compressed_size &&
252 (actual_end & (root->sectorsize - 1)) == 0) ||
253 end + 1 < isize ||
254 data_len > root->fs_info->max_inline) {
255 return 1;
256 }
257
2671485d 258 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM		 259 if (ret)
260 return ret;
c8b97818
CM		 261
262 if (isize > actual_end)
263 inline_len = min_t(u64, isize, actual_end);
264 ret = insert_inline_extent(trans, root, inode, start,
265 inline_len, compressed_size,
fe3f566c 266 compress_type, compressed_pages);
2adcac1a 267 if (ret && ret != -ENOSPC) {
79787eaa
JM		 268 btrfs_abort_transaction(trans, root, ret);
269 return ret;
2adcac1a
JB		 270 } else if (ret == -ENOSPC) {
271 return 1;
79787eaa 272 }
2adcac1a 273
bdc20e67 274 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 275 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 276 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 277 return 0;
278}
279
771ed689
CM		 280struct async_extent {
281 u64 start;
282 u64 ram_size;
283 u64 compressed_size;
284 struct page **pages;
285 unsigned long nr_pages;
261507a0 286 int compress_type;
771ed689
CM		 287 struct list_head list;
288};
289
290struct async_cow {
291 struct inode *inode;
292 struct btrfs_root *root;
293 struct page *locked_page;
294 u64 start;
295 u64 end;
296 struct list_head extents;
297 struct btrfs_work work;
298};
299
300static noinline int add_async_extent(struct async_cow *cow,
301 u64 start, u64 ram_size,
302 u64 compressed_size,
303 struct page **pages,
261507a0
LZ		 304 unsigned long nr_pages,
305 int compress_type)
771ed689
CM		 306{
307 struct async_extent *async_extent;
308
309 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 310 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 311 async_extent->start = start;
312 async_extent->ram_size = ram_size;
313 async_extent->compressed_size = compressed_size;
314 async_extent->pages = pages;
315 async_extent->nr_pages = nr_pages;
261507a0 316 async_extent->compress_type = compress_type;
771ed689
CM		 317 list_add_tail(&async_extent->list, &cow->extents);
318 return 0;
319}
320
d352ac68 321/*
771ed689
CM		 322 * we create compressed extents in two phases. The first
323 * phase compresses a range of pages that have already been
324 * locked (both pages and state bits are locked).
c8b97818 325 *
771ed689
CM		 326 * This is done inside an ordered work queue, and the compression
327 * is spread across many cpus. The actual IO submission is step
328 * two, and the ordered work queue takes care of making sure that
329 * happens in the same order things were put onto the queue by
330 * writepages and friends.
c8b97818 331 *
771ed689
CM		 332 * If this code finds it can't get good compression, it puts an
333 * entry onto the work queue to write the uncompressed bytes. This
334 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 335 * are written in the same order that the flusher thread sent them
336 * down.
d352ac68 337 */
771ed689
CM		 338static noinline int compress_file_range(struct inode *inode,
339 struct page *locked_page,
340 u64 start, u64 end,
341 struct async_cow *async_cow,
342 int *num_added)
b888db2b
CM		 343{
344 struct btrfs_root *root = BTRFS_I(inode)->root;
345 struct btrfs_trans_handle *trans;
db94535d 346 u64 num_bytes;
db94535d 347 u64 blocksize = root->sectorsize;
c8b97818 348 u64 actual_end;
42dc7bab 349 u64 isize = i_size_read(inode);
e6dcd2dc 350 int ret = 0;
c8b97818
CM		 351 struct page **pages = NULL;
352 unsigned long nr_pages;
353 unsigned long nr_pages_ret = 0;
354 unsigned long total_compressed = 0;
355 unsigned long total_in = 0;
356 unsigned long max_compressed = 128 * 1024;
771ed689 357 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 358 int i;
359 int will_compress;
261507a0 360 int compress_type = root->fs_info->compress_type;
4adaa611 361 int redirty = 0;
b888db2b 362
4cb13e5d
LB		 363 /* if this is a small write inside eof, kick off a defrag */
364 if ((end - start + 1) < 16 * 1024 &&
365 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 366 btrfs_add_inode_defrag(NULL, inode);
367
42dc7bab 368 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 369again:
370 will_compress = 0;
371 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
372 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 373
f03d9301
CM		 374 /*
375 * we don't want to send crud past the end of i_size through
376 * compression, that's just a waste of CPU time. So, if the
377 * end of the file is before the start of our current
378 * requested range of bytes, we bail out to the uncompressed
379 * cleanup code that can deal with all of this.
380 *
381 * It isn't really the fastest way to fix things, but this is a
382 * very uncommon corner.
383 */
384 if (actual_end <= start)
385 goto cleanup_and_bail_uncompressed;
386
c8b97818
CM		 387 total_compressed = actual_end - start;
388
389 /* we want to make sure that amount of ram required to uncompress
390 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 391 * of a compressed extent to 128k. This is a crucial number
392 * because it also controls how easily we can spread reads across
393 * cpus for decompression.
394 *
395 * We also want to make sure the amount of IO required to do
396 * a random read is reasonably small, so we limit the size of
397 * a compressed extent to 128k.
c8b97818
CM		 398 */
399 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 400 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 401 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 402 total_in = 0;
403 ret = 0;
db94535d 404
771ed689
CM		 405 /*
406 * we do compression for mount -o compress and when the
407 * inode has not been flagged as nocompress. This flag can
408 * change at any time if we discover bad compression ratios.
c8b97818 409 */
6cbff00f 410 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 411 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 412 (BTRFS_I(inode)->force_compress) ||
413 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 414 WARN_ON(pages);
cfbc246e 415 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 416 if (!pages) {
417 /* just bail out to the uncompressed code */
418 goto cont;
419 }
c8b97818 420
261507a0
LZ		 421 if (BTRFS_I(inode)->force_compress)
422 compress_type = BTRFS_I(inode)->force_compress;
423
4adaa611
CM		 424 /*
425 * we need to call clear_page_dirty_for_io on each
426 * page in the range. Otherwise applications with the file
427 * mmap'd can wander in and change the page contents while
428 * we are compressing them.
429 *
430 * If the compression fails for any reason, we set the pages
431 * dirty again later on.
432 */
433 extent_range_clear_dirty_for_io(inode, start, end);
434 redirty = 1;
261507a0
LZ		 435 ret = btrfs_compress_pages(compress_type,
436 inode->i_mapping, start,
437 total_compressed, pages,
438 nr_pages, &nr_pages_ret,
439 &total_in,
440 &total_compressed,
441 max_compressed);
c8b97818
CM		 442
443 if (!ret) {
444 unsigned long offset = total_compressed &
445 (PAGE_CACHE_SIZE - 1);
446 struct page *page = pages[nr_pages_ret - 1];
447 char *kaddr;
448
449 /* zero the tail end of the last page, we might be
450 * sending it down to disk
451 */
452 if (offset) {
7ac687d9 453 kaddr = kmap_atomic(page);
c8b97818
CM		 454 memset(kaddr + offset, 0,
455 PAGE_CACHE_SIZE - offset);
7ac687d9 456 kunmap_atomic(kaddr);
c8b97818
CM		 457 }
458 will_compress = 1;
459 }
460 }
560f7d75 461cont:
c8b97818 462 if (start == 0) {
7a7eaa40 463 trans = btrfs_join_transaction(root);
79787eaa
JM		 464 if (IS_ERR(trans)) {
465 ret = PTR_ERR(trans);
466 trans = NULL;
467 goto cleanup_and_out;
468 }
0ca1f7ce 469 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 470
c8b97818 471 /* lets try to make an inline extent */
771ed689 472 if (ret || total_in < (actual_end - start)) {
c8b97818 473 /* we didn't compress the entire range, try
771ed689 474 * to make an uncompressed inline extent.
c8b97818
CM		 475 */
476 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 477 start, end, 0, 0, NULL);
c8b97818 478 } else {
771ed689 479 /* try making a compressed inline extent */
c8b97818
CM		 480 ret = cow_file_range_inline(trans, root, inode,
481 start, end,
fe3f566c
LZ		 482 total_compressed,
483 compress_type, pages);
c8b97818 484 }
79787eaa 485 if (ret <= 0) {
771ed689 486 /*
79787eaa
JM		 487 * inline extent creation worked or returned error,
488 * we don't need to create any more async work items.
489 * Unlock and free up our temp pages.
771ed689 490 */
c8b97818 491 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 492 &BTRFS_I(inode)->io_tree,
493 start, end, NULL,
494 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 495 EXTENT_CLEAR_DELALLOC |
a791e35e 496 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 497
498 btrfs_end_transaction(trans, root);
c8b97818
CM		 499 goto free_pages_out;
500 }
c2167754 501 btrfs_end_transaction(trans, root);
c8b97818
CM		 502 }
503
504 if (will_compress) {
505 /*
506 * we aren't doing an inline extent round the compressed size
507 * up to a block size boundary so the allocator does sane
508 * things
509 */
fda2832f 510 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 511
512 /*
513 * one last check to make sure the compression is really a
514 * win, compare the page count read with the blocks on disk
515 */
fda2832f 516 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 517 if (total_compressed >= total_in) {
518 will_compress = 0;
519 } else {
c8b97818
CM		 520 num_bytes = total_in;
521 }
522 }
523 if (!will_compress && pages) {
524 /*
525 * the compression code ran but failed to make things smaller,
526 * free any pages it allocated and our page pointer array
527 */
528 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 529 WARN_ON(pages[i]->mapping);
c8b97818
CM		 530 page_cache_release(pages[i]);
531 }
532 kfree(pages);
533 pages = NULL;
534 total_compressed = 0;
535 nr_pages_ret = 0;
536
537 /* flag the file so we don't compress in the future */
1e701a32
CM		 538 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
539 !(BTRFS_I(inode)->force_compress)) {
a555f810 540 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 541 }
c8b97818 542 }
771ed689
CM		 543 if (will_compress) {
544 *num_added += 1;
c8b97818 545
771ed689
CM		 546 /* the async work queues will take care of doing actual
547 * allocation on disk for these compressed pages,
548 * and will submit them to the elevator.
549 */
550 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 551 total_compressed, pages, nr_pages_ret,
552 compress_type);
179e29e4 553
24ae6365 554 if (start + num_bytes < end) {
771ed689
CM		 555 start += num_bytes;
556 pages = NULL;
557 cond_resched();
558 goto again;
559 }
560 } else {
f03d9301 561cleanup_and_bail_uncompressed:
771ed689
CM		 562 /*
563 * No compression, but we still need to write the pages in
564 * the file we've been given so far. redirty the locked
565 * page if it corresponds to our extent and set things up
566 * for the async work queue to run cow_file_range to do
567 * the normal delalloc dance
568 */
569 if (page_offset(locked_page) >= start &&
570 page_offset(locked_page) <= end) {
571 __set_page_dirty_nobuffers(locked_page);
572 /* unlocked later on in the async handlers */
573 }
4adaa611
CM		 574 if (redirty)
575 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 576 add_async_extent(async_cow, start, end - start + 1,
577 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 578 *num_added += 1;
579 }
3b951516 580
771ed689 581out:
79787eaa 582 return ret;
771ed689
CM		 583
584free_pages_out:
585 for (i = 0; i < nr_pages_ret; i++) {
586 WARN_ON(pages[i]->mapping);
587 page_cache_release(pages[i]);
588 }
d397712b 589 kfree(pages);
771ed689
CM		 590
591 goto out;
79787eaa
JM		 592
593cleanup_and_out:
594 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
595 start, end, NULL,
596 EXTENT_CLEAR_UNLOCK_PAGE |
597 EXTENT_CLEAR_DIRTY |
598 EXTENT_CLEAR_DELALLOC |
599 EXTENT_SET_WRITEBACK |
600 EXTENT_END_WRITEBACK);
601 if (!trans || IS_ERR(trans))
602 btrfs_error(root->fs_info, ret, "Failed to join transaction");
603 else
604 btrfs_abort_transaction(trans, root, ret);
605 goto free_pages_out;
771ed689
CM		 606}
607
608/*
609 * phase two of compressed writeback. This is the ordered portion
610 * of the code, which only gets called in the order the work was
611 * queued. We walk all the async extents created by compress_file_range
612 * and send them down to the disk.
613 */
614static noinline int submit_compressed_extents(struct inode *inode,
615 struct async_cow *async_cow)
616{
617 struct async_extent *async_extent;
618 u64 alloc_hint = 0;
619 struct btrfs_trans_handle *trans;
620 struct btrfs_key ins;
621 struct extent_map *em;
622 struct btrfs_root *root = BTRFS_I(inode)->root;
623 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
624 struct extent_io_tree *io_tree;
f5a84ee3 625 int ret = 0;
771ed689
CM		 626
627 if (list_empty(&async_cow->extents))
628 return 0;
629
3e04e7f1 630again:
d397712b 631 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 632 async_extent = list_entry(async_cow->extents.next,
633 struct async_extent, list);
634 list_del(&async_extent->list);
c8b97818 635
771ed689
CM		 636 io_tree = &BTRFS_I(inode)->io_tree;
637
f5a84ee3 638retry:
771ed689
CM		 639 /* did the compression code fall back to uncompressed IO? */
640 if (!async_extent->pages) {
641 int page_started = 0;
642 unsigned long nr_written = 0;
643
644 lock_extent(io_tree, async_extent->start,
2ac55d41 645 async_extent->start +
d0082371 646 async_extent->ram_size - 1);
771ed689
CM		 647
648 /* allocate blocks */
f5a84ee3
JB		 649 ret = cow_file_range(inode, async_cow->locked_page,
650 async_extent->start,
651 async_extent->start +
652 async_extent->ram_size - 1,
653 &page_started, &nr_written, 0);
771ed689 654
79787eaa
JM		 655 /* JDM XXX */
656
771ed689
CM		 657 /*
658 * if page_started, cow_file_range inserted an
659 * inline extent and took care of all the unlocking
660 * and IO for us. Otherwise, we need to submit
661 * all those pages down to the drive.
662 */
f5a84ee3 663 if (!page_started && !ret)
771ed689
CM		 664 extent_write_locked_range(io_tree,
665 inode, async_extent->start,
d397712b 666 async_extent->start +
771ed689
CM		 667 async_extent->ram_size - 1,
668 btrfs_get_extent,
669 WB_SYNC_ALL);
3e04e7f1
JB		 670 else if (ret)
671 unlock_page(async_cow->locked_page);
771ed689
CM		 672 kfree(async_extent);
673 cond_resched();
674 continue;
675 }
676
677 lock_extent(io_tree, async_extent->start,
d0082371 678 async_extent->start + async_extent->ram_size - 1);
771ed689 679
7a7eaa40 680 trans = btrfs_join_transaction(root);
79787eaa
JM		 681 if (IS_ERR(trans)) {
682 ret = PTR_ERR(trans);
683 } else {
684 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
685 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 686 async_extent->compressed_size,
687 async_extent->compressed_size,
81c9ad23 688 0, alloc_hint, &ins, 1);
962197ba 689 if (ret && ret != -ENOSPC)
79787eaa
JM		 690 btrfs_abort_transaction(trans, root, ret);
691 btrfs_end_transaction(trans, root);
692 }
c2167754 693
f5a84ee3
JB		 694 if (ret) {
695 int i;
3e04e7f1 696
f5a84ee3
JB		 697 for (i = 0; i < async_extent->nr_pages; i++) {
698 WARN_ON(async_extent->pages[i]->mapping);
699 page_cache_release(async_extent->pages[i]);
700 }
701 kfree(async_extent->pages);
702 async_extent->nr_pages = 0;
703 async_extent->pages = NULL;
3e04e7f1 704
fdf8e2ea
JB		 705 if (ret == -ENOSPC) {
706 unlock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
79787eaa 709 goto retry;
fdf8e2ea 710 }
3e04e7f1 711 goto out_free;
f5a84ee3
JB		 712 }
713
c2167754
YZ		 714 /*
715 * here we're doing allocation and writeback of the
716 * compressed pages
717 */
718 btrfs_drop_extent_cache(inode, async_extent->start,
719 async_extent->start +
720 async_extent->ram_size - 1, 0);
721
172ddd60 722 em = alloc_extent_map();
b9aa55be
LB		 723 if (!em) {
724 ret = -ENOMEM;
3e04e7f1 725 goto out_free_reserve;
b9aa55be 726 }
771ed689
CM		 727 em->start = async_extent->start;
728 em->len = async_extent->ram_size;
445a6944 729 em->orig_start = em->start;
2ab28f32
JB		 730 em->mod_start = em->start;
731 em->mod_len = em->len;
c8b97818 732
771ed689
CM		 733 em->block_start = ins.objectid;
734 em->block_len = ins.offset;
b4939680 735 em->orig_block_len = ins.offset;
cc95bef6 736 em->ram_bytes = async_extent->ram_size;
771ed689 737 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 738 em->compress_type = async_extent->compress_type;
771ed689
CM		 739 set_bit(EXTENT_FLAG_PINNED, &em->flags);
740 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 741 em->generation = -1;
771ed689 742
d397712b 743 while (1) {
890871be 744 write_lock(&em_tree->lock);
09a2a8f9 745 ret = add_extent_mapping(em_tree, em, 1);
890871be 746 write_unlock(&em_tree->lock);
771ed689
CM		 747 if (ret != -EEXIST) {
748 free_extent_map(em);
749 break;
750 }
751 btrfs_drop_extent_cache(inode, async_extent->start,
752 async_extent->start +
753 async_extent->ram_size - 1, 0);
754 }
755
3e04e7f1
JB		 756 if (ret)
757 goto out_free_reserve;
758
261507a0
LZ		 759 ret = btrfs_add_ordered_extent_compress(inode,
760 async_extent->start,
761 ins.objectid,
762 async_extent->ram_size,
763 ins.offset,
764 BTRFS_ORDERED_COMPRESSED,
765 async_extent->compress_type);
3e04e7f1
JB		 766 if (ret)
767 goto out_free_reserve;
771ed689 768
771ed689
CM		 769 /*
770 * clear dirty, set writeback and unlock the pages.
771 */
772 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 773 &BTRFS_I(inode)->io_tree,
774 async_extent->start,
775 async_extent->start +
776 async_extent->ram_size - 1,
777 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
778 EXTENT_CLEAR_UNLOCK |
a3429ab7 779 EXTENT_CLEAR_DELALLOC |
a791e35e 780 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 781
782 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 783 async_extent->start,
784 async_extent->ram_size,
785 ins.objectid,
786 ins.offset, async_extent->pages,
787 async_extent->nr_pages);
771ed689
CM		 788 alloc_hint = ins.objectid + ins.offset;
789 kfree(async_extent);
3e04e7f1
JB		 790 if (ret)
791 goto out;
771ed689
CM		 792 cond_resched();
793 }
79787eaa
JM		 794 ret = 0;
795out:
796 return ret;
3e04e7f1
JB		 797out_free_reserve:
798 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 799out_free:
3e04e7f1
JB		 800 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
801 async_extent->start,
802 async_extent->start +
803 async_extent->ram_size - 1,
804 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
805 EXTENT_CLEAR_UNLOCK |
806 EXTENT_CLEAR_DELALLOC |
807 EXTENT_CLEAR_DIRTY |
808 EXTENT_SET_WRITEBACK |
809 EXTENT_END_WRITEBACK);
79787eaa 810 kfree(async_extent);
3e04e7f1 811 goto again;
771ed689
CM		 812}
813
4b46fce2
JB		 814static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
815 u64 num_bytes)
816{
817 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
818 struct extent_map *em;
819 u64 alloc_hint = 0;
820
821 read_lock(&em_tree->lock);
822 em = search_extent_mapping(em_tree, start, num_bytes);
823 if (em) {
824 /*
825 * if block start isn't an actual block number then find the
826 * first block in this inode and use that as a hint. If that
827 * block is also bogus then just don't worry about it.
828 */
829 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
830 free_extent_map(em);
831 em = search_extent_mapping(em_tree, 0, 0);
832 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
833 alloc_hint = em->block_start;
834 if (em)
835 free_extent_map(em);
836 } else {
837 alloc_hint = em->block_start;
838 free_extent_map(em);
839 }
840 }
841 read_unlock(&em_tree->lock);
842
843 return alloc_hint;
844}
845
771ed689
CM		 846/*
847 * when extent_io.c finds a delayed allocation range in the file,
848 * the call backs end up in this code. The basic idea is to
849 * allocate extents on disk for the range, and create ordered data structs
850 * in ram to track those extents.
851 *
852 * locked_page is the page that writepage had locked already. We use
853 * it to make sure we don't do extra locks or unlocks.
854 *
855 * *page_started is set to one if we unlock locked_page and do everything
856 * required to start IO on it. It may be clean and already done with
857 * IO when we return.
858 */
b7d5b0a8
MX		 859static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
860 struct inode *inode,
861 struct btrfs_root *root,
862 struct page *locked_page,
863 u64 start, u64 end, int *page_started,
864 unsigned long *nr_written,
865 int unlock)
771ed689 866{
771ed689
CM		 867 u64 alloc_hint = 0;
868 u64 num_bytes;
869 unsigned long ram_size;
870 u64 disk_num_bytes;
871 u64 cur_alloc_size;
872 u64 blocksize = root->sectorsize;
771ed689
CM		 873 struct btrfs_key ins;
874 struct extent_map *em;
875 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
876 int ret = 0;
877
83eea1f1 878 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 879
fda2832f 880 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 881 num_bytes = max(blocksize, num_bytes);
882 disk_num_bytes = num_bytes;
771ed689 883
4cb5300b 884 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 885 if (num_bytes < 64 * 1024 &&
886 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 887 btrfs_add_inode_defrag(trans, inode);
888
771ed689
CM		 889 if (start == 0) {
890 /* lets try to make an inline extent */
891 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 892 start, end, 0, 0, NULL);
771ed689
CM		 893 if (ret == 0) {
894 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 895 &BTRFS_I(inode)->io_tree,
896 start, end, NULL,
897 EXTENT_CLEAR_UNLOCK_PAGE |
898 EXTENT_CLEAR_UNLOCK |
899 EXTENT_CLEAR_DELALLOC |
900 EXTENT_CLEAR_DIRTY |
901 EXTENT_SET_WRITEBACK |
902 EXTENT_END_WRITEBACK);
c2167754 903
771ed689
CM		 904 *nr_written = *nr_written +
905 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
906 *page_started = 1;
771ed689 907 goto out;
79787eaa
JM		 908 } else if (ret < 0) {
909 btrfs_abort_transaction(trans, root, ret);
910 goto out_unlock;
771ed689
CM		 911 }
912 }
913
914 BUG_ON(disk_num_bytes >
6c41761f 915 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 916
4b46fce2 917 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 918 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
919
d397712b 920 while (disk_num_bytes > 0) {
a791e35e
CM		 921 unsigned long op;
922
287a0ab9 923 cur_alloc_size = disk_num_bytes;
e6dcd2dc 924 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 925 root->sectorsize, 0, alloc_hint,
81c9ad23 926 &ins, 1);
79787eaa
JM		 927 if (ret < 0) {
928 btrfs_abort_transaction(trans, root, ret);
929 goto out_unlock;
930 }
d397712b 931
172ddd60 932 em = alloc_extent_map();
b9aa55be
LB		 933 if (!em) {
934 ret = -ENOMEM;
ace68bac 935 goto out_reserve;
b9aa55be 936 }
e6dcd2dc 937 em->start = start;
445a6944 938 em->orig_start = em->start;
771ed689
CM		 939 ram_size = ins.offset;
940 em->len = ins.offset;
2ab28f32
JB		 941 em->mod_start = em->start;
942 em->mod_len = em->len;
c8b97818 943
e6dcd2dc 944 em->block_start = ins.objectid;
c8b97818 945 em->block_len = ins.offset;
b4939680 946 em->orig_block_len = ins.offset;
cc95bef6 947 em->ram_bytes = ram_size;
e6dcd2dc 948 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 949 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 950 em->generation = -1;
c8b97818 951
d397712b 952 while (1) {
890871be 953 write_lock(&em_tree->lock);
09a2a8f9 954 ret = add_extent_mapping(em_tree, em, 1);
890871be 955 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 956 if (ret != -EEXIST) {
957 free_extent_map(em);
958 break;
959 }
960 btrfs_drop_extent_cache(inode, start,
c8b97818 961 start + ram_size - 1, 0);
e6dcd2dc 962 }
ace68bac
LB		 963 if (ret)
964 goto out_reserve;
e6dcd2dc 965
98d20f67 966 cur_alloc_size = ins.offset;
e6dcd2dc 967 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 968 ram_size, cur_alloc_size, 0);
ace68bac
LB		 969 if (ret)
970 goto out_reserve;
c8b97818 971
17d217fe
YZ		 972 if (root->root_key.objectid ==
973 BTRFS_DATA_RELOC_TREE_OBJECTID) {
974 ret = btrfs_reloc_clone_csums(inode, start,
975 cur_alloc_size);
79787eaa
JM		 976 if (ret) {
977 btrfs_abort_transaction(trans, root, ret);
ace68bac 978 goto out_reserve;
79787eaa 979 }
17d217fe
YZ		 980 }
981
d397712b 982 if (disk_num_bytes < cur_alloc_size)
3b951516 983 break;
d397712b 984
c8b97818
CM		 985 /* we're not doing compressed IO, don't unlock the first
986 * page (which the caller expects to stay locked), don't
987 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 988 *
989 * Do set the Private2 bit so we know this page was properly
990 * setup for writepage
c8b97818 991 */
a791e35e
CM		 992 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
993 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
994 EXTENT_SET_PRIVATE2;
995
c8b97818
CM		 996 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
997 start, start + ram_size - 1,
a791e35e 998 locked_page, op);
c8b97818 999 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1000 num_bytes -= cur_alloc_size;
1001 alloc_hint = ins.objectid + ins.offset;
1002 start += cur_alloc_size;
b888db2b 1003 }
79787eaa 1004out:
be20aa9d 1005 return ret;
b7d5b0a8 1006
ace68bac
LB		 1007out_reserve:
1008 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa
JM		 1009out_unlock:
1010 extent_clear_unlock_delalloc(inode,
1011 &BTRFS_I(inode)->io_tree,
beb42dd7 1012 start, end, locked_page,
79787eaa
JM		 1013 EXTENT_CLEAR_UNLOCK_PAGE |
1014 EXTENT_CLEAR_UNLOCK |
1015 EXTENT_CLEAR_DELALLOC |
1016 EXTENT_CLEAR_DIRTY |
1017 EXTENT_SET_WRITEBACK |
1018 EXTENT_END_WRITEBACK);
1019
1020 goto out;
771ed689 1021}
c8b97818 1022
b7d5b0a8
MX		 1023static noinline int cow_file_range(struct inode *inode,
1024 struct page *locked_page,
1025 u64 start, u64 end, int *page_started,
1026 unsigned long *nr_written,
1027 int unlock)
1028{
1029 struct btrfs_trans_handle *trans;
1030 struct btrfs_root *root = BTRFS_I(inode)->root;
1031 int ret;
1032
1033 trans = btrfs_join_transaction(root);
1034 if (IS_ERR(trans)) {
1035 extent_clear_unlock_delalloc(inode,
1036 &BTRFS_I(inode)->io_tree,
1037 start, end, locked_page,
1038 EXTENT_CLEAR_UNLOCK_PAGE |
1039 EXTENT_CLEAR_UNLOCK |
1040 EXTENT_CLEAR_DELALLOC |
1041 EXTENT_CLEAR_DIRTY |
1042 EXTENT_SET_WRITEBACK |
1043 EXTENT_END_WRITEBACK);
1044 return PTR_ERR(trans);
1045 }
1046 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1047
1048 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1049 page_started, nr_written, unlock);
1050
1051 btrfs_end_transaction(trans, root);
1052
1053 return ret;
1054}
1055
771ed689
CM		 1056/*
1057 * work queue call back to started compression on a file and pages
1058 */
1059static noinline void async_cow_start(struct btrfs_work *work)
1060{
1061 struct async_cow *async_cow;
1062 int num_added = 0;
1063 async_cow = container_of(work, struct async_cow, work);
1064
1065 compress_file_range(async_cow->inode, async_cow->locked_page,
1066 async_cow->start, async_cow->end, async_cow,
1067 &num_added);
8180ef88 1068 if (num_added == 0) {
cb77fcd8 1069 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1070 async_cow->inode = NULL;
8180ef88 1071 }
771ed689
CM		 1072}
1073
1074/*
1075 * work queue call back to submit previously compressed pages
1076 */
1077static noinline void async_cow_submit(struct btrfs_work *work)
1078{
1079 struct async_cow *async_cow;
1080 struct btrfs_root *root;
1081 unsigned long nr_pages;
1082
1083 async_cow = container_of(work, struct async_cow, work);
1084
1085 root = async_cow->root;
1086 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1087 PAGE_CACHE_SHIFT;
1088
66657b31 1089 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1090 5 * 1024 * 1024 &&
771ed689
CM		 1091 waitqueue_active(&root->fs_info->async_submit_wait))
1092 wake_up(&root->fs_info->async_submit_wait);
1093
d397712b 1094 if (async_cow->inode)
771ed689 1095 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1096}
c8b97818 1097
771ed689
CM		 1098static noinline void async_cow_free(struct btrfs_work *work)
1099{
1100 struct async_cow *async_cow;
1101 async_cow = container_of(work, struct async_cow, work);
8180ef88 1102 if (async_cow->inode)
cb77fcd8 1103 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1104 kfree(async_cow);
1105}
1106
1107static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1108 u64 start, u64 end, int *page_started,
1109 unsigned long *nr_written)
1110{
1111 struct async_cow *async_cow;
1112 struct btrfs_root *root = BTRFS_I(inode)->root;
1113 unsigned long nr_pages;
1114 u64 cur_end;
287082b0 1115 int limit = 10 * 1024 * 1024;
771ed689 1116
a3429ab7
CM		 1117 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1118 1, 0, NULL, GFP_NOFS);
d397712b 1119 while (start < end) {
771ed689 1120 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1121 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1122 async_cow->inode = igrab(inode);
771ed689
CM		 1123 async_cow->root = root;
1124 async_cow->locked_page = locked_page;
1125 async_cow->start = start;
1126
6cbff00f 1127 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1128 cur_end = end;
1129 else
1130 cur_end = min(end, start + 512 * 1024 - 1);
1131
1132 async_cow->end = cur_end;
1133 INIT_LIST_HEAD(&async_cow->extents);
1134
1135 async_cow->work.func = async_cow_start;
1136 async_cow->work.ordered_func = async_cow_submit;
1137 async_cow->work.ordered_free = async_cow_free;
1138 async_cow->work.flags = 0;
1139
771ed689
CM		 1140 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1141 PAGE_CACHE_SHIFT;
1142 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1143
1144 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1145 &async_cow->work);
1146
1147 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1148 wait_event(root->fs_info->async_submit_wait,
1149 (atomic_read(&root->fs_info->async_delalloc_pages) <
1150 limit));
1151 }
1152
d397712b 1153 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1154 atomic_read(&root->fs_info->async_delalloc_pages)) {
1155 wait_event(root->fs_info->async_submit_wait,
1156 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1157 0));
1158 }
1159
1160 *nr_written += nr_pages;
1161 start = cur_end + 1;
1162 }
1163 *page_started = 1;
1164 return 0;
be20aa9d
CM		 1165}
1166
d397712b 1167static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1168 u64 bytenr, u64 num_bytes)
1169{
1170 int ret;
1171 struct btrfs_ordered_sum *sums;
1172 LIST_HEAD(list);
1173
07d400a6 1174 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1175 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1176 if (ret == 0 && list_empty(&list))
1177 return 0;
1178
1179 while (!list_empty(&list)) {
1180 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1181 list_del(&sums->list);
1182 kfree(sums);
1183 }
1184 return 1;
1185}
1186
d352ac68
CM		 1187/*
1188 * when nowcow writeback call back. This checks for snapshots or COW copies
1189 * of the extents that exist in the file, and COWs the file as required.
1190 *
1191 * If no cow copies or snapshots exist, we write directly to the existing
1192 * blocks on disk
1193 */
7f366cfe
CM		 1194static noinline int run_delalloc_nocow(struct inode *inode,
1195 struct page *locked_page,
771ed689
CM		 1196 u64 start, u64 end, int *page_started, int force,
1197 unsigned long *nr_written)
be20aa9d 1198{
be20aa9d 1199 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1200 struct btrfs_trans_handle *trans;
be20aa9d 1201 struct extent_buffer *leaf;
be20aa9d 1202 struct btrfs_path *path;
80ff3856 1203 struct btrfs_file_extent_item *fi;
be20aa9d 1204 struct btrfs_key found_key;
80ff3856
YZ		 1205 u64 cow_start;
1206 u64 cur_offset;
1207 u64 extent_end;
5d4f98a2 1208 u64 extent_offset;
80ff3856
YZ		 1209 u64 disk_bytenr;
1210 u64 num_bytes;
b4939680 1211 u64 disk_num_bytes;
cc95bef6 1212 u64 ram_bytes;
80ff3856 1213 int extent_type;
79787eaa 1214 int ret, err;
d899e052 1215 int type;
80ff3856
YZ		 1216 int nocow;
1217 int check_prev = 1;
82d5902d 1218 bool nolock;
33345d01 1219 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1220
1221 path = btrfs_alloc_path();
17ca04af
JB		 1222 if (!path) {
1223 extent_clear_unlock_delalloc(inode,
1224 &BTRFS_I(inode)->io_tree,
1225 start, end, locked_page,
1226 EXTENT_CLEAR_UNLOCK_PAGE |
1227 EXTENT_CLEAR_UNLOCK |
1228 EXTENT_CLEAR_DELALLOC |
1229 EXTENT_CLEAR_DIRTY |
1230 EXTENT_SET_WRITEBACK |
1231 EXTENT_END_WRITEBACK);
d8926bb3 1232 return -ENOMEM;
17ca04af 1233 }
82d5902d 1234
83eea1f1 1235 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1236
1237 if (nolock)
7a7eaa40 1238 trans = btrfs_join_transaction_nolock(root);
82d5902d 1239 else
7a7eaa40 1240 trans = btrfs_join_transaction(root);
ff5714cc 1241
79787eaa 1242 if (IS_ERR(trans)) {
17ca04af
JB		 1243 extent_clear_unlock_delalloc(inode,
1244 &BTRFS_I(inode)->io_tree,
1245 start, end, locked_page,
1246 EXTENT_CLEAR_UNLOCK_PAGE |
1247 EXTENT_CLEAR_UNLOCK |
1248 EXTENT_CLEAR_DELALLOC |
1249 EXTENT_CLEAR_DIRTY |
1250 EXTENT_SET_WRITEBACK |
1251 EXTENT_END_WRITEBACK);
79787eaa
JM		 1252 btrfs_free_path(path);
1253 return PTR_ERR(trans);
1254 }
1255
74b21075 1256 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1257
80ff3856
YZ		 1258 cow_start = (u64)-1;
1259 cur_offset = start;
1260 while (1) {
33345d01 1261 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1262 cur_offset, 0);
79787eaa
JM		 1263 if (ret < 0) {
1264 btrfs_abort_transaction(trans, root, ret);
1265 goto error;
1266 }
80ff3856
YZ		 1267 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1268 leaf = path->nodes[0];
1269 btrfs_item_key_to_cpu(leaf, &found_key,
1270 path->slots[0] - 1);
33345d01 1271 if (found_key.objectid == ino &&
80ff3856
YZ		 1272 found_key.type == BTRFS_EXTENT_DATA_KEY)
1273 path->slots[0]--;
1274 }
1275 check_prev = 0;
1276next_slot:
1277 leaf = path->nodes[0];
1278 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1279 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1280 if (ret < 0) {
1281 btrfs_abort_transaction(trans, root, ret);
1282 goto error;
1283 }
80ff3856
YZ		 1284 if (ret > 0)
1285 break;
1286 leaf = path->nodes[0];
1287 }
be20aa9d 1288
80ff3856
YZ		 1289 nocow = 0;
1290 disk_bytenr = 0;
17d217fe 1291 num_bytes = 0;
80ff3856
YZ		 1292 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1293
33345d01 1294 if (found_key.objectid > ino ||
80ff3856
YZ		 1295 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1296 found_key.offset > end)
1297 break;
1298
1299 if (found_key.offset > cur_offset) {
1300 extent_end = found_key.offset;
e9061e21 1301 extent_type = 0;
80ff3856
YZ		 1302 goto out_check;
1303 }
1304
1305 fi = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_file_extent_item);
1307 extent_type = btrfs_file_extent_type(leaf, fi);
1308
cc95bef6 1309 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1310 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1311 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1312 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1313 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1314 extent_end = found_key.offset +
1315 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1316 disk_num_bytes =
1317 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1318 if (extent_end <= start) {
1319 path->slots[0]++;
1320 goto next_slot;
1321 }
17d217fe
YZ		 1322 if (disk_bytenr == 0)
1323 goto out_check;
80ff3856
YZ		 1324 if (btrfs_file_extent_compression(leaf, fi) ||
1325 btrfs_file_extent_encryption(leaf, fi) ||
1326 btrfs_file_extent_other_encoding(leaf, fi))
1327 goto out_check;
d899e052
YZ		 1328 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1329 goto out_check;
d2fb3437 1330 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1331 goto out_check;
33345d01 1332 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1333 found_key.offset -
1334 extent_offset, disk_bytenr))
17d217fe 1335 goto out_check;
5d4f98a2 1336 disk_bytenr += extent_offset;
17d217fe
YZ		 1337 disk_bytenr += cur_offset - found_key.offset;
1338 num_bytes = min(end + 1, extent_end) - cur_offset;
1339 /*
1340 * force cow if csum exists in the range.
1341 * this ensure that csum for a given extent are
1342 * either valid or do not exist.
1343 */
1344 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1345 goto out_check;
80ff3856
YZ		 1346 nocow = 1;
1347 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1348 extent_end = found_key.offset +
1349 btrfs_file_extent_inline_len(leaf, fi);
1350 extent_end = ALIGN(extent_end, root->sectorsize);
1351 } else {
1352 BUG_ON(1);
1353 }
1354out_check:
1355 if (extent_end <= start) {
1356 path->slots[0]++;
1357 goto next_slot;
1358 }
1359 if (!nocow) {
1360 if (cow_start == (u64)-1)
1361 cow_start = cur_offset;
1362 cur_offset = extent_end;
1363 if (cur_offset > end)
1364 break;
1365 path->slots[0]++;
1366 goto next_slot;
7ea394f1
YZ		 1367 }
1368
b3b4aa74 1369 btrfs_release_path(path);
80ff3856 1370 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1371 ret = __cow_file_range(trans, inode, root, locked_page,
1372 cow_start, found_key.offset - 1,
1373 page_started, nr_written, 1);
79787eaa
JM		 1374 if (ret) {
1375 btrfs_abort_transaction(trans, root, ret);
1376 goto error;
1377 }
80ff3856 1378 cow_start = (u64)-1;
7ea394f1 1379 }
80ff3856 1380
d899e052
YZ		 1381 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1382 struct extent_map *em;
1383 struct extent_map_tree *em_tree;
1384 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1385 em = alloc_extent_map();
79787eaa 1386 BUG_ON(!em); /* -ENOMEM */
d899e052 1387 em->start = cur_offset;
70c8a91c 1388 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1389 em->len = num_bytes;
1390 em->block_len = num_bytes;
1391 em->block_start = disk_bytenr;
b4939680 1392 em->orig_block_len = disk_num_bytes;
cc95bef6 1393 em->ram_bytes = ram_bytes;
d899e052 1394 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1395 em->mod_start = em->start;
1396 em->mod_len = em->len;
d899e052 1397 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1398 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1399 em->generation = -1;
d899e052 1400 while (1) {
890871be 1401 write_lock(&em_tree->lock);
09a2a8f9 1402 ret = add_extent_mapping(em_tree, em, 1);
890871be 1403 write_unlock(&em_tree->lock);
d899e052
YZ		 1404 if (ret != -EEXIST) {
1405 free_extent_map(em);
1406 break;
1407 }
1408 btrfs_drop_extent_cache(inode, em->start,
1409 em->start + em->len - 1, 0);
1410 }
1411 type = BTRFS_ORDERED_PREALLOC;
1412 } else {
1413 type = BTRFS_ORDERED_NOCOW;
1414 }
80ff3856
YZ		 1415
1416 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1417 num_bytes, num_bytes, type);
79787eaa 1418 BUG_ON(ret); /* -ENOMEM */
771ed689 1419
efa56464
YZ		 1420 if (root->root_key.objectid ==
1421 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1422 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1423 num_bytes);
79787eaa
JM		 1424 if (ret) {
1425 btrfs_abort_transaction(trans, root, ret);
1426 goto error;
1427 }
efa56464
YZ		 1428 }
1429
d899e052 1430 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1431 cur_offset, cur_offset + num_bytes - 1,
1432 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1433 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1434 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1435 cur_offset = extent_end;
1436 if (cur_offset > end)
1437 break;
be20aa9d 1438 }
b3b4aa74 1439 btrfs_release_path(path);
80ff3856 1440
17ca04af 1441 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1442 cow_start = cur_offset;
17ca04af
JB		 1443 cur_offset = end;
1444 }
1445
80ff3856 1446 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1447 ret = __cow_file_range(trans, inode, root, locked_page,
1448 cow_start, end,
1449 page_started, nr_written, 1);
79787eaa
JM		 1450 if (ret) {
1451 btrfs_abort_transaction(trans, root, ret);
1452 goto error;
1453 }
80ff3856
YZ		 1454 }
1455
79787eaa 1456error:
a698d075 1457 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1458 if (!ret)
1459 ret = err;
1460
17ca04af
JB		 1461 if (ret && cur_offset < end)
1462 extent_clear_unlock_delalloc(inode,
1463 &BTRFS_I(inode)->io_tree,
1464 cur_offset, end, locked_page,
1465 EXTENT_CLEAR_UNLOCK_PAGE |
1466 EXTENT_CLEAR_UNLOCK |
1467 EXTENT_CLEAR_DELALLOC |
1468 EXTENT_CLEAR_DIRTY |
1469 EXTENT_SET_WRITEBACK |
1470 EXTENT_END_WRITEBACK);
1471
7ea394f1 1472 btrfs_free_path(path);
79787eaa 1473 return ret;
be20aa9d
CM		 1474}
1475
d352ac68
CM		 1476/*
1477 * extent_io.c call back to do delayed allocation processing
1478 */
c8b97818 1479static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1480 u64 start, u64 end, int *page_started,
1481 unsigned long *nr_written)
be20aa9d 1482{
be20aa9d 1483 int ret;
7f366cfe 1484 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1485
7ddf5a42 1486 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1487 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1488 page_started, 1, nr_written);
7ddf5a42 1489 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1490 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1491 page_started, 0, nr_written);
7ddf5a42
JB		 1492 } else if (!btrfs_test_opt(root, COMPRESS) &&
1493 !(BTRFS_I(inode)->force_compress) &&
1494 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1495 ret = cow_file_range(inode, locked_page, start, end,
1496 page_started, nr_written, 1);
7ddf5a42
JB		 1497 } else {
1498 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1499 &BTRFS_I(inode)->runtime_flags);
771ed689 1500 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1501 page_started, nr_written);
7ddf5a42 1502 }
b888db2b
CM		 1503 return ret;
1504}
1505
1bf85046
JM		 1506static void btrfs_split_extent_hook(struct inode *inode,
1507 struct extent_state *orig, u64 split)
9ed74f2d 1508{
0ca1f7ce 1509 /* not delalloc, ignore it */
9ed74f2d 1510 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1511 return;
9ed74f2d 1512
9e0baf60
JB		 1513 spin_lock(&BTRFS_I(inode)->lock);
1514 BTRFS_I(inode)->outstanding_extents++;
1515 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1516}
1517
1518/*
1519 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1520 * extents so we can keep track of new extents that are just merged onto old
1521 * extents, such as when we are doing sequential writes, so we can properly
1522 * account for the metadata space we'll need.
1523 */
1bf85046
JM		 1524static void btrfs_merge_extent_hook(struct inode *inode,
1525 struct extent_state *new,
1526 struct extent_state *other)
9ed74f2d 1527{
9ed74f2d
JB		 1528 /* not delalloc, ignore it */
1529 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1530 return;
9ed74f2d 1531
9e0baf60
JB		 1532 spin_lock(&BTRFS_I(inode)->lock);
1533 BTRFS_I(inode)->outstanding_extents--;
1534 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1535}
1536
eb73c1b7
MX		 1537static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1538 struct inode *inode)
1539{
1540 spin_lock(&root->delalloc_lock);
1541 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1542 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1543 &root->delalloc_inodes);
1544 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1545 &BTRFS_I(inode)->runtime_flags);
1546 root->nr_delalloc_inodes++;
1547 if (root->nr_delalloc_inodes == 1) {
1548 spin_lock(&root->fs_info->delalloc_root_lock);
1549 BUG_ON(!list_empty(&root->delalloc_root));
1550 list_add_tail(&root->delalloc_root,
1551 &root->fs_info->delalloc_roots);
1552 spin_unlock(&root->fs_info->delalloc_root_lock);
1553 }
1554 }
1555 spin_unlock(&root->delalloc_lock);
1556}
1557
1558static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1559 struct inode *inode)
1560{
1561 spin_lock(&root->delalloc_lock);
1562 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1563 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1564 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1565 &BTRFS_I(inode)->runtime_flags);
1566 root->nr_delalloc_inodes--;
1567 if (!root->nr_delalloc_inodes) {
1568 spin_lock(&root->fs_info->delalloc_root_lock);
1569 BUG_ON(list_empty(&root->delalloc_root));
1570 list_del_init(&root->delalloc_root);
1571 spin_unlock(&root->fs_info->delalloc_root_lock);
1572 }
1573 }
1574 spin_unlock(&root->delalloc_lock);
1575}
1576
d352ac68
CM		 1577/*
1578 * extent_io.c set_bit_hook, used to track delayed allocation
1579 * bytes in this file, and to maintain the list of inodes that
1580 * have pending delalloc work to be done.
1581 */
1bf85046 1582static void btrfs_set_bit_hook(struct inode *inode,
41074888 1583 struct extent_state *state, unsigned long *bits)
291d673e 1584{
9ed74f2d 1585
75eff68e
CM		 1586 /*
1587 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1588 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1589 * bit, which is only set or cleared with irqs on
1590 */
0ca1f7ce 1591 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1592 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1593 u64 len = state->end + 1 - state->start;
83eea1f1 1594 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1595
9e0baf60 1596 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1597 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1598 } else {
1599 spin_lock(&BTRFS_I(inode)->lock);
1600 BTRFS_I(inode)->outstanding_extents++;
1601 spin_unlock(&BTRFS_I(inode)->lock);
1602 }
287a0ab9 1603
963d678b
MX		 1604 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1605 root->fs_info->delalloc_batch);
df0af1a5 1606 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1607 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5 1608 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1609 &BTRFS_I(inode)->runtime_flags))
1610 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1611 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1612 }
291d673e
CM		 1613}
1614
d352ac68
CM		 1615/*
1616 * extent_io.c clear_bit_hook, see set_bit_hook for why
1617 */
1bf85046 1618static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1619 struct extent_state *state,
1620 unsigned long *bits)
291d673e 1621{
75eff68e
CM		 1622 /*
1623 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1624 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1625 * bit, which is only set or cleared with irqs on
1626 */
0ca1f7ce 1627 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1628 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1629 u64 len = state->end + 1 - state->start;
83eea1f1 1630 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1631
9e0baf60 1632 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1633 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1634 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1635 spin_lock(&BTRFS_I(inode)->lock);
1636 BTRFS_I(inode)->outstanding_extents--;
1637 spin_unlock(&BTRFS_I(inode)->lock);
1638 }
0ca1f7ce
YZ		 1639
1640 if (*bits & EXTENT_DO_ACCOUNTING)
1641 btrfs_delalloc_release_metadata(inode, len);
1642
0cb59c99
JB		 1643 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1644 && do_list)
0ca1f7ce 1645 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1646
963d678b
MX		 1647 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1648 root->fs_info->delalloc_batch);
df0af1a5 1649 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1650 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1651 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1652 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1653 &BTRFS_I(inode)->runtime_flags))
1654 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1655 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1656 }
291d673e
CM		 1657}
1658
d352ac68
CM		 1659/*
1660 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1661 * we don't create bios that span stripes or chunks
1662 */
64a16701 1663int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1664 size_t size, struct bio *bio,
1665 unsigned long bio_flags)
239b14b3
CM		 1666{
1667 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1668 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1669 u64 length = 0;
1670 u64 map_length;
239b14b3
CM		 1671 int ret;
1672
771ed689
CM		 1673 if (bio_flags & EXTENT_BIO_COMPRESSED)
1674 return 0;
1675
f2d8d74d 1676 length = bio->bi_size;
239b14b3 1677 map_length = length;
64a16701 1678 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1679 &map_length, NULL, 0);
3ec706c8 1680 /* Will always return 0 with map_multi == NULL */
3444a972 1681 BUG_ON(ret < 0);
d397712b 1682 if (map_length < length + size)
239b14b3 1683 return 1;
3444a972 1684 return 0;
239b14b3
CM		 1685}
1686
d352ac68
CM		 1687/*
1688 * in order to insert checksums into the metadata in large chunks,
1689 * we wait until bio submission time. All the pages in the bio are
1690 * checksummed and sums are attached onto the ordered extent record.
1691 *
1692 * At IO completion time the cums attached on the ordered extent record
1693 * are inserted into the btree
1694 */
d397712b
CM		 1695static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1696 struct bio *bio, int mirror_num,
eaf25d93
CM		 1697 unsigned long bio_flags,
1698 u64 bio_offset)
065631f6 1699{
065631f6 1700 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1701 int ret = 0;
e015640f 1702
d20f7043 1703 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1704 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1705 return 0;
1706}
e015640f 1707
4a69a410
CM		 1708/*
1709 * in order to insert checksums into the metadata in large chunks,
1710 * we wait until bio submission time. All the pages in the bio are
1711 * checksummed and sums are attached onto the ordered extent record.
1712 *
1713 * At IO completion time the cums attached on the ordered extent record
1714 * are inserted into the btree
1715 */
b2950863 1716static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1717 int mirror_num, unsigned long bio_flags,
1718 u64 bio_offset)
4a69a410
CM		 1719{
1720 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1721 int ret;
1722
1723 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1724 if (ret)
1725 bio_endio(bio, ret);
1726 return ret;
44b8bd7e
CM		 1727}
1728
d352ac68 1729/*
cad321ad
CM		 1730 * extent_io.c submission hook. This does the right thing for csum calculation
1731 * on write, or reading the csums from the tree before a read
d352ac68 1732 */
b2950863 1733static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1734 int mirror_num, unsigned long bio_flags,
1735 u64 bio_offset)
44b8bd7e
CM		 1736{
1737 struct btrfs_root *root = BTRFS_I(inode)->root;
1738 int ret = 0;
19b9bdb0 1739 int skip_sum;
0417341e 1740 int metadata = 0;
b812ce28 1741 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1742
6cbff00f 1743 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1744
83eea1f1 1745 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1746 metadata = 2;
1747
7b6d91da 1748 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1749 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1750 if (ret)
61891923 1751 goto out;
5fd02043 1752
d20f7043 1753 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1754 ret = btrfs_submit_compressed_read(inode, bio,
1755 mirror_num,
1756 bio_flags);
1757 goto out;
c2db1073
TI		 1758 } else if (!skip_sum) {
1759 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1760 if (ret)
61891923 1761 goto out;
c2db1073 1762 }
4d1b5fb4 1763 goto mapit;
b812ce28 1764 } else if (async && !skip_sum) {
17d217fe
YZ		 1765 /* csum items have already been cloned */
1766 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1767 goto mapit;
19b9bdb0 1768 /* we're doing a write, do the async checksumming */
61891923 1769 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1770 inode, rw, bio, mirror_num,
eaf25d93
CM		 1771 bio_flags, bio_offset,
1772 __btrfs_submit_bio_start,
4a69a410 1773 __btrfs_submit_bio_done);
61891923 1774 goto out;
b812ce28
JB		 1775 } else if (!skip_sum) {
1776 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1777 if (ret)
1778 goto out;
19b9bdb0
CM		 1779 }
1780
0b86a832 1781mapit:
61891923
SB		 1782 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1783
1784out:
1785 if (ret < 0)
1786 bio_endio(bio, ret);
1787 return ret;
065631f6 1788}
6885f308 1789
d352ac68
CM		 1790/*
1791 * given a list of ordered sums record them in the inode. This happens
1792 * at IO completion time based on sums calculated at bio submission time.
1793 */
ba1da2f4 1794static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1795 struct inode *inode, u64 file_offset,
1796 struct list_head *list)
1797{
e6dcd2dc
CM		 1798 struct btrfs_ordered_sum *sum;
1799
c6e30871 1800 list_for_each_entry(sum, list, list) {
39847c4d 1801 trans->adding_csums = 1;
d20f7043
CM		 1802 btrfs_csum_file_blocks(trans,
1803 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1804 trans->adding_csums = 0;
e6dcd2dc
CM		 1805 }
1806 return 0;
1807}
1808
2ac55d41
JB		 1809int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1810 struct extent_state **cached_state)
ea8c2819 1811{
6c1500f2 1812 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1813 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1814 cached_state, GFP_NOFS);
ea8c2819
CM		 1815}
1816
d352ac68 1817/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1818struct btrfs_writepage_fixup {
1819 struct page *page;
1820 struct btrfs_work work;
1821};
1822
b2950863 1823static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1824{
1825 struct btrfs_writepage_fixup *fixup;
1826 struct btrfs_ordered_extent *ordered;
2ac55d41 1827 struct extent_state *cached_state = NULL;
247e743c
CM		 1828 struct page *page;
1829 struct inode *inode;
1830 u64 page_start;
1831 u64 page_end;
87826df0 1832 int ret;
247e743c
CM		 1833
1834 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1835 page = fixup->page;
4a096752 1836again:
247e743c
CM		 1837 lock_page(page);
1838 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1839 ClearPageChecked(page);
1840 goto out_page;
1841 }
1842
1843 inode = page->mapping->host;
1844 page_start = page_offset(page);
1845 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1846
2ac55d41 1847 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1848 &cached_state);
4a096752
CM		 1849
1850 /* already ordered? We're done */
8b62b72b 1851 if (PagePrivate2(page))
247e743c 1852 goto out;
4a096752
CM		 1853
1854 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1855 if (ordered) {
2ac55d41
JB		 1856 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1857 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1858 unlock_page(page);
1859 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1860 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1861 goto again;
1862 }
247e743c 1863
87826df0
JM		 1864 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1865 if (ret) {
1866 mapping_set_error(page->mapping, ret);
1867 end_extent_writepage(page, ret, page_start, page_end);
1868 ClearPageChecked(page);
1869 goto out;
1870 }
1871
2ac55d41 1872 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1873 ClearPageChecked(page);
87826df0 1874 set_page_dirty(page);
247e743c 1875out:
2ac55d41
JB		 1876 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1877 &cached_state, GFP_NOFS);
247e743c
CM		 1878out_page:
1879 unlock_page(page);
1880 page_cache_release(page);
b897abec 1881 kfree(fixup);
247e743c
CM		 1882}
1883
1884/*
1885 * There are a few paths in the higher layers of the kernel that directly
1886 * set the page dirty bit without asking the filesystem if it is a
1887 * good idea. This causes problems because we want to make sure COW
1888 * properly happens and the data=ordered rules are followed.
1889 *
c8b97818 1890 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1891 * hasn't been properly setup for IO. We kick off an async process
1892 * to fix it up. The async helper will wait for ordered extents, set
1893 * the delalloc bit and make it safe to write the page.
1894 */
b2950863 1895static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1896{
1897 struct inode *inode = page->mapping->host;
1898 struct btrfs_writepage_fixup *fixup;
1899 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1900
8b62b72b
CM		 1901 /* this page is properly in the ordered list */
1902 if (TestClearPagePrivate2(page))
247e743c
CM		 1903 return 0;
1904
1905 if (PageChecked(page))
1906 return -EAGAIN;
1907
1908 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1909 if (!fixup)
1910 return -EAGAIN;
f421950f 1911
247e743c
CM		 1912 SetPageChecked(page);
1913 page_cache_get(page);
1914 fixup->work.func = btrfs_writepage_fixup_worker;
1915 fixup->page = page;
1916 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1917 return -EBUSY;
247e743c
CM		 1918}
1919
d899e052
YZ		 1920static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1921 struct inode *inode, u64 file_pos,
1922 u64 disk_bytenr, u64 disk_num_bytes,
1923 u64 num_bytes, u64 ram_bytes,
1924 u8 compression, u8 encryption,
1925 u16 other_encoding, int extent_type)
1926{
1927 struct btrfs_root *root = BTRFS_I(inode)->root;
1928 struct btrfs_file_extent_item *fi;
1929 struct btrfs_path *path;
1930 struct extent_buffer *leaf;
1931 struct btrfs_key ins;
d899e052
YZ		 1932 int ret;
1933
1934 path = btrfs_alloc_path();
d8926bb3
MF		 1935 if (!path)
1936 return -ENOMEM;
d899e052 1937
b9473439 1938 path->leave_spinning = 1;
a1ed835e
CM		 1939
1940 /*
1941 * we may be replacing one extent in the tree with another.
1942 * The new extent is pinned in the extent map, and we don't want
1943 * to drop it from the cache until it is completely in the btree.
1944 *
1945 * So, tell btrfs_drop_extents to leave this extent in the cache.
1946 * the caller is expected to unpin it and allow it to be merged
1947 * with the others.
1948 */
5dc562c5 1949 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1950 file_pos + num_bytes, 0);
79787eaa
JM		 1951 if (ret)
1952 goto out;
d899e052 1953
33345d01 1954 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1955 ins.offset = file_pos;
1956 ins.type = BTRFS_EXTENT_DATA_KEY;
1957 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1958 if (ret)
1959 goto out;
d899e052
YZ		 1960 leaf = path->nodes[0];
1961 fi = btrfs_item_ptr(leaf, path->slots[0],
1962 struct btrfs_file_extent_item);
1963 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1964 btrfs_set_file_extent_type(leaf, fi, extent_type);
1965 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1966 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1967 btrfs_set_file_extent_offset(leaf, fi, 0);
1968 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1969 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1970 btrfs_set_file_extent_compression(leaf, fi, compression);
1971 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1972 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1973
d899e052 1974 btrfs_mark_buffer_dirty(leaf);
ce195332 1975 btrfs_release_path(path);
d899e052
YZ		 1976
1977 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1978
1979 ins.objectid = disk_bytenr;
1980 ins.offset = disk_num_bytes;
1981 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1982 ret = btrfs_alloc_reserved_file_extent(trans, root,
1983 root->root_key.objectid,
33345d01 1984 btrfs_ino(inode), file_pos, &ins);
79787eaa 1985out:
d899e052 1986 btrfs_free_path(path);
b9473439 1987
79787eaa 1988 return ret;
d899e052
YZ		 1989}
1990
38c227d8
LB		 1991/* snapshot-aware defrag */
1992struct sa_defrag_extent_backref {
1993 struct rb_node node;
1994 struct old_sa_defrag_extent *old;
1995 u64 root_id;
1996 u64 inum;
1997 u64 file_pos;
1998 u64 extent_offset;
1999 u64 num_bytes;
2000 u64 generation;
2001};
2002
2003struct old_sa_defrag_extent {
2004 struct list_head list;
2005 struct new_sa_defrag_extent *new;
2006
2007 u64 extent_offset;
2008 u64 bytenr;
2009 u64 offset;
2010 u64 len;
2011 int count;
2012};
2013
2014struct new_sa_defrag_extent {
2015 struct rb_root root;
2016 struct list_head head;
2017 struct btrfs_path *path;
2018 struct inode *inode;
2019 u64 file_pos;
2020 u64 len;
2021 u64 bytenr;
2022 u64 disk_len;
2023 u8 compress_type;
2024};
2025
2026static int backref_comp(struct sa_defrag_extent_backref *b1,
2027 struct sa_defrag_extent_backref *b2)
2028{
2029 if (b1->root_id < b2->root_id)
2030 return -1;
2031 else if (b1->root_id > b2->root_id)
2032 return 1;
2033
2034 if (b1->inum < b2->inum)
2035 return -1;
2036 else if (b1->inum > b2->inum)
2037 return 1;
2038
2039 if (b1->file_pos < b2->file_pos)
2040 return -1;
2041 else if (b1->file_pos > b2->file_pos)
2042 return 1;
2043
2044 /*
2045 * [------------------------------] ===> (a range of space)
2046 * |<--->| |<---->| =============> (fs/file tree A)
2047 * |<---------------------------->| ===> (fs/file tree B)
2048 *
2049 * A range of space can refer to two file extents in one tree while
2050 * refer to only one file extent in another tree.
2051 *
2052 * So we may process a disk offset more than one time(two extents in A)
2053 * and locate at the same extent(one extent in B), then insert two same
2054 * backrefs(both refer to the extent in B).
2055 */
2056 return 0;
2057}
2058
2059static void backref_insert(struct rb_root *root,
2060 struct sa_defrag_extent_backref *backref)
2061{
2062 struct rb_node **p = &root->rb_node;
2063 struct rb_node *parent = NULL;
2064 struct sa_defrag_extent_backref *entry;
2065 int ret;
2066
2067 while (*p) {
2068 parent = *p;
2069 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2070
2071 ret = backref_comp(backref, entry);
2072 if (ret < 0)
2073 p = &(*p)->rb_left;
2074 else
2075 p = &(*p)->rb_right;
2076 }
2077
2078 rb_link_node(&backref->node, parent, p);
2079 rb_insert_color(&backref->node, root);
2080}
2081
2082/*
2083 * Note the backref might has changed, and in this case we just return 0.
2084 */
2085static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2086 void *ctx)
2087{
2088 struct btrfs_file_extent_item *extent;
2089 struct btrfs_fs_info *fs_info;
2090 struct old_sa_defrag_extent *old = ctx;
2091 struct new_sa_defrag_extent *new = old->new;
2092 struct btrfs_path *path = new->path;
2093 struct btrfs_key key;
2094 struct btrfs_root *root;
2095 struct sa_defrag_extent_backref *backref;
2096 struct extent_buffer *leaf;
2097 struct inode *inode = new->inode;
2098 int slot;
2099 int ret;
2100 u64 extent_offset;
2101 u64 num_bytes;
2102
2103 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2104 inum == btrfs_ino(inode))
2105 return 0;
2106
2107 key.objectid = root_id;
2108 key.type = BTRFS_ROOT_ITEM_KEY;
2109 key.offset = (u64)-1;
2110
2111 fs_info = BTRFS_I(inode)->root->fs_info;
2112 root = btrfs_read_fs_root_no_name(fs_info, &key);
2113 if (IS_ERR(root)) {
2114 if (PTR_ERR(root) == -ENOENT)
2115 return 0;
2116 WARN_ON(1);
2117 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2118 inum, offset, root_id);
2119 return PTR_ERR(root);
2120 }
2121
2122 key.objectid = inum;
2123 key.type = BTRFS_EXTENT_DATA_KEY;
2124 if (offset > (u64)-1 << 32)
2125 key.offset = 0;
2126 else
2127 key.offset = offset;
2128
2129 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2130 if (ret < 0) {
2131 WARN_ON(1);
2132 return ret;
2133 }
2134
2135 while (1) {
2136 cond_resched();
2137
2138 leaf = path->nodes[0];
2139 slot = path->slots[0];
2140
2141 if (slot >= btrfs_header_nritems(leaf)) {
2142 ret = btrfs_next_leaf(root, path);
2143 if (ret < 0) {
2144 goto out;
2145 } else if (ret > 0) {
2146 ret = 0;
2147 goto out;
2148 }
2149 continue;
2150 }
2151
2152 path->slots[0]++;
2153
2154 btrfs_item_key_to_cpu(leaf, &key, slot);
2155
2156 if (key.objectid > inum)
2157 goto out;
2158
2159 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2160 continue;
2161
2162 extent = btrfs_item_ptr(leaf, slot,
2163 struct btrfs_file_extent_item);
2164
2165 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2166 continue;
2167
2168 extent_offset = btrfs_file_extent_offset(leaf, extent);
2169 if (key.offset - extent_offset != offset)
2170 continue;
2171
2172 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2173 if (extent_offset >= old->extent_offset + old->offset +
2174 old->len || extent_offset + num_bytes <=
2175 old->extent_offset + old->offset)
2176 continue;
2177
2178 break;
2179 }
2180
2181 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2182 if (!backref) {
2183 ret = -ENOENT;
2184 goto out;
2185 }
2186
2187 backref->root_id = root_id;
2188 backref->inum = inum;
2189 backref->file_pos = offset + extent_offset;
2190 backref->num_bytes = num_bytes;
2191 backref->extent_offset = extent_offset;
2192 backref->generation = btrfs_file_extent_generation(leaf, extent);
2193 backref->old = old;
2194 backref_insert(&new->root, backref);
2195 old->count++;
2196out:
2197 btrfs_release_path(path);
2198 WARN_ON(ret);
2199 return ret;
2200}
2201
2202static noinline bool record_extent_backrefs(struct btrfs_path *path,
2203 struct new_sa_defrag_extent *new)
2204{
2205 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2206 struct old_sa_defrag_extent *old, *tmp;
2207 int ret;
2208
2209 new->path = path;
2210
2211 list_for_each_entry_safe(old, tmp, &new->head, list) {
2212 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2213 path, record_one_backref,
2214 old);
2215 BUG_ON(ret < 0 && ret != -ENOENT);
2216
2217 /* no backref to be processed for this extent */
2218 if (!old->count) {
2219 list_del(&old->list);
2220 kfree(old);
2221 }
2222 }
2223
2224 if (list_empty(&new->head))
2225 return false;
2226
2227 return true;
2228}
2229
2230static int relink_is_mergable(struct extent_buffer *leaf,
2231 struct btrfs_file_extent_item *fi,
2232 u64 disk_bytenr)
2233{
2234 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2235 return 0;
2236
2237 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2238 return 0;
2239
2240 if (btrfs_file_extent_compression(leaf, fi) ||
2241 btrfs_file_extent_encryption(leaf, fi) ||
2242 btrfs_file_extent_other_encoding(leaf, fi))
2243 return 0;
2244
2245 return 1;
2246}
2247
2248/*
2249 * Note the backref might has changed, and in this case we just return 0.
2250 */
2251static noinline int relink_extent_backref(struct btrfs_path *path,
2252 struct sa_defrag_extent_backref *prev,
2253 struct sa_defrag_extent_backref *backref)
2254{
2255 struct btrfs_file_extent_item *extent;
2256 struct btrfs_file_extent_item *item;
2257 struct btrfs_ordered_extent *ordered;
2258 struct btrfs_trans_handle *trans;
2259 struct btrfs_fs_info *fs_info;
2260 struct btrfs_root *root;
2261 struct btrfs_key key;
2262 struct extent_buffer *leaf;
2263 struct old_sa_defrag_extent *old = backref->old;
2264 struct new_sa_defrag_extent *new = old->new;
2265 struct inode *src_inode = new->inode;
2266 struct inode *inode;
2267 struct extent_state *cached = NULL;
2268 int ret = 0;
2269 u64 start;
2270 u64 len;
2271 u64 lock_start;
2272 u64 lock_end;
2273 bool merge = false;
2274 int index;
2275
2276 if (prev && prev->root_id == backref->root_id &&
2277 prev->inum == backref->inum &&
2278 prev->file_pos + prev->num_bytes == backref->file_pos)
2279 merge = true;
2280
2281 /* step 1: get root */
2282 key.objectid = backref->root_id;
2283 key.type = BTRFS_ROOT_ITEM_KEY;
2284 key.offset = (u64)-1;
2285
2286 fs_info = BTRFS_I(src_inode)->root->fs_info;
2287 index = srcu_read_lock(&fs_info->subvol_srcu);
2288
2289 root = btrfs_read_fs_root_no_name(fs_info, &key);
2290 if (IS_ERR(root)) {
2291 srcu_read_unlock(&fs_info->subvol_srcu, index);
2292 if (PTR_ERR(root) == -ENOENT)
2293 return 0;
2294 return PTR_ERR(root);
2295 }
38c227d8
LB		 2296
2297 /* step 2: get inode */
2298 key.objectid = backref->inum;
2299 key.type = BTRFS_INODE_ITEM_KEY;
2300 key.offset = 0;
2301
2302 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2303 if (IS_ERR(inode)) {
2304 srcu_read_unlock(&fs_info->subvol_srcu, index);
2305 return 0;
2306 }
2307
2308 srcu_read_unlock(&fs_info->subvol_srcu, index);
2309
2310 /* step 3: relink backref */
2311 lock_start = backref->file_pos;
2312 lock_end = backref->file_pos + backref->num_bytes - 1;
2313 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2314 0, &cached);
2315
2316 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2317 if (ordered) {
2318 btrfs_put_ordered_extent(ordered);
2319 goto out_unlock;
2320 }
2321
2322 trans = btrfs_join_transaction(root);
2323 if (IS_ERR(trans)) {
2324 ret = PTR_ERR(trans);
2325 goto out_unlock;
2326 }
2327
2328 key.objectid = backref->inum;
2329 key.type = BTRFS_EXTENT_DATA_KEY;
2330 key.offset = backref->file_pos;
2331
2332 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2333 if (ret < 0) {
2334 goto out_free_path;
2335 } else if (ret > 0) {
2336 ret = 0;
2337 goto out_free_path;
2338 }
2339
2340 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2341 struct btrfs_file_extent_item);
2342
2343 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2344 backref->generation)
2345 goto out_free_path;
2346
2347 btrfs_release_path(path);
2348
2349 start = backref->file_pos;
2350 if (backref->extent_offset < old->extent_offset + old->offset)
2351 start += old->extent_offset + old->offset -
2352 backref->extent_offset;
2353
2354 len = min(backref->extent_offset + backref->num_bytes,
2355 old->extent_offset + old->offset + old->len);
2356 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2357
2358 ret = btrfs_drop_extents(trans, root, inode, start,
2359 start + len, 1);
2360 if (ret)
2361 goto out_free_path;
2362again:
2363 key.objectid = btrfs_ino(inode);
2364 key.type = BTRFS_EXTENT_DATA_KEY;
2365 key.offset = start;
2366
a09a0a70 2367 path->leave_spinning = 1;
38c227d8
LB		 2368 if (merge) {
2369 struct btrfs_file_extent_item *fi;
2370 u64 extent_len;
2371 struct btrfs_key found_key;
2372
2373 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2374 if (ret < 0)
2375 goto out_free_path;
2376
2377 path->slots[0]--;
2378 leaf = path->nodes[0];
2379 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2380
2381 fi = btrfs_item_ptr(leaf, path->slots[0],
2382 struct btrfs_file_extent_item);
2383 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2384
2385 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2386 extent_len + found_key.offset == start) {
2387 btrfs_set_file_extent_num_bytes(leaf, fi,
2388 extent_len + len);
2389 btrfs_mark_buffer_dirty(leaf);
2390 inode_add_bytes(inode, len);
2391
2392 ret = 1;
2393 goto out_free_path;
2394 } else {
2395 merge = false;
2396 btrfs_release_path(path);
2397 goto again;
2398 }
2399 }
2400
2401 ret = btrfs_insert_empty_item(trans, root, path, &key,
2402 sizeof(*extent));
2403 if (ret) {
2404 btrfs_abort_transaction(trans, root, ret);
2405 goto out_free_path;
2406 }
2407
2408 leaf = path->nodes[0];
2409 item = btrfs_item_ptr(leaf, path->slots[0],
2410 struct btrfs_file_extent_item);
2411 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2412 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2413 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2414 btrfs_set_file_extent_num_bytes(leaf, item, len);
2415 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2416 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2417 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2418 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2419 btrfs_set_file_extent_encryption(leaf, item, 0);
2420 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2421
2422 btrfs_mark_buffer_dirty(leaf);
2423 inode_add_bytes(inode, len);
a09a0a70 2424 btrfs_release_path(path);
38c227d8
LB		 2425
2426 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2427 new->disk_len, 0,
2428 backref->root_id, backref->inum,
2429 new->file_pos, 0); /* start - extent_offset */
2430 if (ret) {
2431 btrfs_abort_transaction(trans, root, ret);
2432 goto out_free_path;
2433 }
2434
2435 ret = 1;
2436out_free_path:
2437 btrfs_release_path(path);
a09a0a70 2438 path->leave_spinning = 0;
38c227d8
LB		 2439 btrfs_end_transaction(trans, root);
2440out_unlock:
2441 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2442 &cached, GFP_NOFS);
2443 iput(inode);
2444 return ret;
2445}
2446
2447static void relink_file_extents(struct new_sa_defrag_extent *new)
2448{
2449 struct btrfs_path *path;
2450 struct old_sa_defrag_extent *old, *tmp;
2451 struct sa_defrag_extent_backref *backref;
2452 struct sa_defrag_extent_backref *prev = NULL;
2453 struct inode *inode;
2454 struct btrfs_root *root;
2455 struct rb_node *node;
2456 int ret;
2457
2458 inode = new->inode;
2459 root = BTRFS_I(inode)->root;
2460
2461 path = btrfs_alloc_path();
2462 if (!path)
2463 return;
2464
2465 if (!record_extent_backrefs(path, new)) {
2466 btrfs_free_path(path);
2467 goto out;
2468 }
2469 btrfs_release_path(path);
2470
2471 while (1) {
2472 node = rb_first(&new->root);
2473 if (!node)
2474 break;
2475 rb_erase(node, &new->root);
2476
2477 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2478
2479 ret = relink_extent_backref(path, prev, backref);
2480 WARN_ON(ret < 0);
2481
2482 kfree(prev);
2483
2484 if (ret == 1)
2485 prev = backref;
2486 else
2487 prev = NULL;
2488 cond_resched();
2489 }
2490 kfree(prev);
2491
2492 btrfs_free_path(path);
2493
2494 list_for_each_entry_safe(old, tmp, &new->head, list) {
2495 list_del(&old->list);
2496 kfree(old);
2497 }
2498out:
2499 atomic_dec(&root->fs_info->defrag_running);
2500 wake_up(&root->fs_info->transaction_wait);
2501
2502 kfree(new);
2503}
2504
2505static struct new_sa_defrag_extent *
2506record_old_file_extents(struct inode *inode,
2507 struct btrfs_ordered_extent *ordered)
2508{
2509 struct btrfs_root *root = BTRFS_I(inode)->root;
2510 struct btrfs_path *path;
2511 struct btrfs_key key;
2512 struct old_sa_defrag_extent *old, *tmp;
2513 struct new_sa_defrag_extent *new;
2514 int ret;
2515
2516 new = kmalloc(sizeof(*new), GFP_NOFS);
2517 if (!new)
2518 return NULL;
2519
2520 new->inode = inode;
2521 new->file_pos = ordered->file_offset;
2522 new->len = ordered->len;
2523 new->bytenr = ordered->start;
2524 new->disk_len = ordered->disk_len;
2525 new->compress_type = ordered->compress_type;
2526 new->root = RB_ROOT;
2527 INIT_LIST_HEAD(&new->head);
2528
2529 path = btrfs_alloc_path();
2530 if (!path)
2531 goto out_kfree;
2532
2533 key.objectid = btrfs_ino(inode);
2534 key.type = BTRFS_EXTENT_DATA_KEY;
2535 key.offset = new->file_pos;
2536
2537 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2538 if (ret < 0)
2539 goto out_free_path;
2540 if (ret > 0 && path->slots[0] > 0)
2541 path->slots[0]--;
2542
2543 /* find out all the old extents for the file range */
2544 while (1) {
2545 struct btrfs_file_extent_item *extent;
2546 struct extent_buffer *l;
2547 int slot;
2548 u64 num_bytes;
2549 u64 offset;
2550 u64 end;
2551 u64 disk_bytenr;
2552 u64 extent_offset;
2553
2554 l = path->nodes[0];
2555 slot = path->slots[0];
2556
2557 if (slot >= btrfs_header_nritems(l)) {
2558 ret = btrfs_next_leaf(root, path);
2559 if (ret < 0)
2560 goto out_free_list;
2561 else if (ret > 0)
2562 break;
2563 continue;
2564 }
2565
2566 btrfs_item_key_to_cpu(l, &key, slot);
2567
2568 if (key.objectid != btrfs_ino(inode))
2569 break;
2570 if (key.type != BTRFS_EXTENT_DATA_KEY)
2571 break;
2572 if (key.offset >= new->file_pos + new->len)
2573 break;
2574
2575 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2576
2577 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2578 if (key.offset + num_bytes < new->file_pos)
2579 goto next;
2580
2581 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2582 if (!disk_bytenr)
2583 goto next;
2584
2585 extent_offset = btrfs_file_extent_offset(l, extent);
2586
2587 old = kmalloc(sizeof(*old), GFP_NOFS);
2588 if (!old)
2589 goto out_free_list;
2590
2591 offset = max(new->file_pos, key.offset);
2592 end = min(new->file_pos + new->len, key.offset + num_bytes);
2593
2594 old->bytenr = disk_bytenr;
2595 old->extent_offset = extent_offset;
2596 old->offset = offset - key.offset;
2597 old->len = end - offset;
2598 old->new = new;
2599 old->count = 0;
2600 list_add_tail(&old->list, &new->head);
2601next:
2602 path->slots[0]++;
2603 cond_resched();
2604 }
2605
2606 btrfs_free_path(path);
2607 atomic_inc(&root->fs_info->defrag_running);
2608
2609 return new;
2610
2611out_free_list:
2612 list_for_each_entry_safe(old, tmp, &new->head, list) {
2613 list_del(&old->list);
2614 kfree(old);
2615 }
2616out_free_path:
2617 btrfs_free_path(path);
2618out_kfree:
2619 kfree(new);
2620 return NULL;
2621}
2622
5d13a98f
CM		 2623/*
2624 * helper function for btrfs_finish_ordered_io, this
2625 * just reads in some of the csum leaves to prime them into ram
2626 * before we start the transaction. It limits the amount of btree
2627 * reads required while inside the transaction.
2628 */
d352ac68
CM		 2629/* as ordered data IO finishes, this gets called so we can finish
2630 * an ordered extent if the range of bytes in the file it covers are
2631 * fully written.
2632 */
5fd02043 2633static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2634{
5fd02043 2635 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2636 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2637 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2638 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2639 struct extent_state *cached_state = NULL;
38c227d8 2640 struct new_sa_defrag_extent *new = NULL;
261507a0 2641 int compress_type = 0;
e6dcd2dc 2642 int ret;
82d5902d 2643 bool nolock;
e6dcd2dc 2644
83eea1f1 2645 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2646
5fd02043
JB		 2647 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2648 ret = -EIO;
2649 goto out;
2650 }
2651
c2167754 2652 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2653 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2654 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2655 if (nolock)
2656 trans = btrfs_join_transaction_nolock(root);
2657 else
2658 trans = btrfs_join_transaction(root);
2659 if (IS_ERR(trans)) {
2660 ret = PTR_ERR(trans);
2661 trans = NULL;
2662 goto out;
c2167754 2663 }
6c760c07
JB		 2664 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2665 ret = btrfs_update_inode_fallback(trans, root, inode);
2666 if (ret) /* -ENOMEM or corruption */
2667 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2668 goto out;
2669 }
e6dcd2dc 2670
2ac55d41
JB		 2671 lock_extent_bits(io_tree, ordered_extent->file_offset,
2672 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2673 0, &cached_state);
e6dcd2dc 2674
38c227d8
LB		 2675 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2676 ordered_extent->file_offset + ordered_extent->len - 1,
2677 EXTENT_DEFRAG, 1, cached_state);
2678 if (ret) {
2679 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2680 if (last_snapshot >= BTRFS_I(inode)->generation)
2681 /* the inode is shared */
2682 new = record_old_file_extents(inode, ordered_extent);
2683
2684 clear_extent_bit(io_tree, ordered_extent->file_offset,
2685 ordered_extent->file_offset + ordered_extent->len - 1,
2686 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2687 }
2688
0cb59c99 2689 if (nolock)
7a7eaa40 2690 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2691 else
7a7eaa40 2692 trans = btrfs_join_transaction(root);
79787eaa
JM		 2693 if (IS_ERR(trans)) {
2694 ret = PTR_ERR(trans);
2695 trans = NULL;
2696 goto out_unlock;
2697 }
0ca1f7ce 2698 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2699
c8b97818 2700 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2701 compress_type = ordered_extent->compress_type;
d899e052 2702 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2703 BUG_ON(compress_type);
920bbbfb 2704 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2705 ordered_extent->file_offset,
2706 ordered_extent->file_offset +
2707 ordered_extent->len);
d899e052 2708 } else {
0af3d00b 2709 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2710 ret = insert_reserved_file_extent(trans, inode,
2711 ordered_extent->file_offset,
2712 ordered_extent->start,
2713 ordered_extent->disk_len,
2714 ordered_extent->len,
2715 ordered_extent->len,
261507a0 2716 compress_type, 0, 0,
d899e052 2717 BTRFS_FILE_EXTENT_REG);
d899e052 2718 }
5dc562c5
JB		 2719 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2720 ordered_extent->file_offset, ordered_extent->len,
2721 trans->transid);
79787eaa
JM		 2722 if (ret < 0) {
2723 btrfs_abort_transaction(trans, root, ret);
5fd02043 2724 goto out_unlock;
79787eaa 2725 }
2ac55d41 2726
e6dcd2dc
CM		 2727 add_pending_csums(trans, inode, ordered_extent->file_offset,
2728 &ordered_extent->list);
2729
6c760c07
JB		 2730 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2731 ret = btrfs_update_inode_fallback(trans, root, inode);
2732 if (ret) { /* -ENOMEM or corruption */
2733 btrfs_abort_transaction(trans, root, ret);
2734 goto out_unlock;
1ef30be1
JB		 2735 }
2736 ret = 0;
5fd02043
JB		 2737out_unlock:
2738 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2739 ordered_extent->file_offset +
2740 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2741out:
5b0e95bf 2742 if (root != root->fs_info->tree_root)
0cb59c99 2743 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2744 if (trans)
2745 btrfs_end_transaction(trans, root);
0cb59c99 2746
0bec9ef5 2747 if (ret) {
5fd02043
JB		 2748 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2749 ordered_extent->file_offset +
2750 ordered_extent->len - 1, NULL, GFP_NOFS);
2751
0bec9ef5
JB		 2752 /*
2753 * If the ordered extent had an IOERR or something else went
2754 * wrong we need to return the space for this ordered extent
2755 * back to the allocator.
2756 */
2757 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2758 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2759 btrfs_free_reserved_extent(root, ordered_extent->start,
2760 ordered_extent->disk_len);
2761 }
2762
2763
5fd02043 2764 /*
8bad3c02
LB		 2765 * This needs to be done to make sure anybody waiting knows we are done
2766 * updating everything for this ordered extent.
5fd02043
JB		 2767 */
2768 btrfs_remove_ordered_extent(inode, ordered_extent);
2769
38c227d8
LB		 2770 /* for snapshot-aware defrag */
2771 if (new)
2772 relink_file_extents(new);
2773
e6dcd2dc
CM		 2774 /* once for us */
2775 btrfs_put_ordered_extent(ordered_extent);
2776 /* once for the tree */
2777 btrfs_put_ordered_extent(ordered_extent);
2778
5fd02043
JB		 2779 return ret;
2780}
2781
2782static void finish_ordered_fn(struct btrfs_work *work)
2783{
2784 struct btrfs_ordered_extent *ordered_extent;
2785 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2786 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2787}
2788
b2950863 2789static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2790 struct extent_state *state, int uptodate)
2791{
5fd02043
JB		 2792 struct inode *inode = page->mapping->host;
2793 struct btrfs_root *root = BTRFS_I(inode)->root;
2794 struct btrfs_ordered_extent *ordered_extent = NULL;
2795 struct btrfs_workers *workers;
2796
1abe9b8a 2797 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2798
8b62b72b 2799 ClearPagePrivate2(page);
5fd02043
JB		 2800 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2801 end - start + 1, uptodate))
2802 return 0;
2803
2804 ordered_extent->work.func = finish_ordered_fn;
2805 ordered_extent->work.flags = 0;
2806
83eea1f1 2807 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2808 workers = &root->fs_info->endio_freespace_worker;
2809 else
2810 workers = &root->fs_info->endio_write_workers;
2811 btrfs_queue_worker(workers, &ordered_extent->work);
2812
2813 return 0;
211f90e6
CM		 2814}
2815
d352ac68
CM		 2816/*
2817 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2818 * if there's a match, we allow the bio to finish. If not, the code in
2819 * extent_io.c will try to find good copies for us.
d352ac68 2820 */
b2950863 2821static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2822 struct extent_state *state, int mirror)
07157aac 2823{
4eee4fa4 2824 size_t offset = start - page_offset(page);
07157aac 2825 struct inode *inode = page->mapping->host;
d1310b2e 2826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2827 char *kaddr;
aadfeb6e 2828 u64 private = ~(u32)0;
07157aac 2829 int ret;
ff79f819
CM		 2830 struct btrfs_root *root = BTRFS_I(inode)->root;
2831 u32 csum = ~(u32)0;
c2cf52eb
SK		 2832 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2833 DEFAULT_RATELIMIT_BURST);
d1310b2e 2834
d20f7043
CM		 2835 if (PageChecked(page)) {
2836 ClearPageChecked(page);
2837 goto good;
2838 }
6cbff00f
CH		 2839
2840 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2841 goto good;
17d217fe
YZ		 2842
2843 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2844 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2845 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2846 GFP_NOFS);
b6cda9bc 2847 return 0;
17d217fe 2848 }
d20f7043 2849
c2e639f0 2850 if (state && state->start == start) {
70dec807
CM		 2851 private = state->private;
2852 ret = 0;
2853 } else {
2854 ret = get_state_private(io_tree, start, &private);
2855 }
7ac687d9 2856 kaddr = kmap_atomic(page);
d397712b 2857 if (ret)
07157aac 2858 goto zeroit;
d397712b 2859
b0496686 2860 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2861 btrfs_csum_final(csum, (char *)&csum);
d397712b 2862 if (csum != private)
07157aac 2863 goto zeroit;
d397712b 2864
7ac687d9 2865 kunmap_atomic(kaddr);
d20f7043 2866good:
07157aac
CM		 2867 return 0;
2868
2869zeroit:
c2cf52eb
SK		 2870 if (__ratelimit(&_rs))
2871 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2872 (unsigned long long)btrfs_ino(page->mapping->host),
2873 (unsigned long long)start, csum,
2874 (unsigned long long)private);
db94535d
CM		 2875 memset(kaddr + offset, 1, end - start + 1);
2876 flush_dcache_page(page);
7ac687d9 2877 kunmap_atomic(kaddr);
3b951516
CM		 2878 if (private == 0)
2879 return 0;
7e38326f 2880 return -EIO;
07157aac 2881}
b888db2b 2882
24bbcf04
YZ		 2883struct delayed_iput {
2884 struct list_head list;
2885 struct inode *inode;
2886};
2887
79787eaa
JM		 2888/* JDM: If this is fs-wide, why can't we add a pointer to
2889 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2890void btrfs_add_delayed_iput(struct inode *inode)
2891{
2892 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2893 struct delayed_iput *delayed;
2894
2895 if (atomic_add_unless(&inode->i_count, -1, 1))
2896 return;
2897
2898 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2899 delayed->inode = inode;
2900
2901 spin_lock(&fs_info->delayed_iput_lock);
2902 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2903 spin_unlock(&fs_info->delayed_iput_lock);
2904}
2905
2906void btrfs_run_delayed_iputs(struct btrfs_root *root)
2907{
2908 LIST_HEAD(list);
2909 struct btrfs_fs_info *fs_info = root->fs_info;
2910 struct delayed_iput *delayed;
2911 int empty;
2912
2913 spin_lock(&fs_info->delayed_iput_lock);
2914 empty = list_empty(&fs_info->delayed_iputs);
2915 spin_unlock(&fs_info->delayed_iput_lock);
2916 if (empty)
2917 return;
2918
24bbcf04
YZ		 2919 spin_lock(&fs_info->delayed_iput_lock);
2920 list_splice_init(&fs_info->delayed_iputs, &list);
2921 spin_unlock(&fs_info->delayed_iput_lock);
2922
2923 while (!list_empty(&list)) {
2924 delayed = list_entry(list.next, struct delayed_iput, list);
2925 list_del(&delayed->list);
2926 iput(delayed->inode);
2927 kfree(delayed);
2928 }
24bbcf04
YZ		 2929}
2930
d68fc57b 2931/*
42b2aa86 2932 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2933 * files in the subvolume, it removes orphan item and frees block_rsv
2934 * structure.
2935 */
2936void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2937 struct btrfs_root *root)
2938{
90290e19 2939 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2940 int ret;
2941
8a35d95f 2942 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2943 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2944 return;
2945
90290e19 2946 spin_lock(&root->orphan_lock);
8a35d95f 2947 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2948 spin_unlock(&root->orphan_lock);
2949 return;
2950 }
2951
2952 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2953 spin_unlock(&root->orphan_lock);
2954 return;
2955 }
2956
2957 block_rsv = root->orphan_block_rsv;
2958 root->orphan_block_rsv = NULL;
2959 spin_unlock(&root->orphan_lock);
2960
d68fc57b
YZ		 2961 if (root->orphan_item_inserted &&
2962 btrfs_root_refs(&root->root_item) > 0) {
2963 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2964 root->root_key.objectid);
2965 BUG_ON(ret);
2966 root->orphan_item_inserted = 0;
2967 }
2968
90290e19
JB		 2969 if (block_rsv) {
2970 WARN_ON(block_rsv->size > 0);
2971 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2972 }
2973}
2974
7b128766
JB		 2975/*
2976 * This creates an orphan entry for the given inode in case something goes
2977 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2978 *
2979 * NOTE: caller of this function should reserve 5 units of metadata for
2980 * this function.
7b128766
JB		 2981 */
2982int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2983{
2984 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2985 struct btrfs_block_rsv *block_rsv = NULL;
2986 int reserve = 0;
2987 int insert = 0;
2988 int ret;
7b128766 2989
d68fc57b 2990 if (!root->orphan_block_rsv) {
66d8f3dd 2991 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2992 if (!block_rsv)
2993 return -ENOMEM;
d68fc57b 2994 }
7b128766 2995
d68fc57b
YZ		 2996 spin_lock(&root->orphan_lock);
2997 if (!root->orphan_block_rsv) {
2998 root->orphan_block_rsv = block_rsv;
2999 } else if (block_rsv) {
3000 btrfs_free_block_rsv(root, block_rsv);
3001 block_rsv = NULL;
7b128766 3002 }
7b128766 3003
8a35d95f
JB		 3004 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3005 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3006#if 0
3007 /*
3008 * For proper ENOSPC handling, we should do orphan
3009 * cleanup when mounting. But this introduces backward
3010 * compatibility issue.
3011 */
3012 if (!xchg(&root->orphan_item_inserted, 1))
3013 insert = 2;
3014 else
3015 insert = 1;
3016#endif
3017 insert = 1;
321f0e70 3018 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3019 }
3020
72ac3c0d
JB		 3021 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3022 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3023 reserve = 1;
d68fc57b 3024 spin_unlock(&root->orphan_lock);
7b128766 3025
d68fc57b
YZ		 3026 /* grab metadata reservation from transaction handle */
3027 if (reserve) {
3028 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3029 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3030 }
7b128766 3031
d68fc57b
YZ		 3032 /* insert an orphan item to track this unlinked/truncated file */
3033 if (insert >= 1) {
33345d01 3034 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3035 if (ret && ret != -EEXIST) {
8a35d95f
JB		 3036 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3037 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 3038 btrfs_abort_transaction(trans, root, ret);
3039 return ret;
3040 }
3041 ret = 0;
d68fc57b
YZ		 3042 }
3043
3044 /* insert an orphan item to track subvolume contains orphan files */
3045 if (insert >= 2) {
3046 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3047 root->root_key.objectid);
79787eaa
JM		 3048 if (ret && ret != -EEXIST) {
3049 btrfs_abort_transaction(trans, root, ret);
3050 return ret;
3051 }
d68fc57b
YZ		 3052 }
3053 return 0;
7b128766
JB		 3054}
3055
3056/*
3057 * We have done the truncate/delete so we can go ahead and remove the orphan
3058 * item for this particular inode.
3059 */
48a3b636
ES		 3060static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3061 struct inode *inode)
7b128766
JB		 3062{
3063 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3064 int delete_item = 0;
3065 int release_rsv = 0;
7b128766
JB		 3066 int ret = 0;
3067
d68fc57b 3068 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3069 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3070 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3071 delete_item = 1;
7b128766 3072
72ac3c0d
JB		 3073 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3074 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3075 release_rsv = 1;
d68fc57b 3076 spin_unlock(&root->orphan_lock);
7b128766 3077
d68fc57b 3078 if (trans && delete_item) {
33345d01 3079 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3080 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 3081 }
7b128766 3082
8a35d95f 3083 if (release_rsv) {
d68fc57b 3084 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 3085 atomic_dec(&root->orphan_inodes);
3086 }
7b128766 3087
d68fc57b 3088 return 0;
7b128766
JB		 3089}
3090
3091/*
3092 * this cleans up any orphans that may be left on the list from the last use
3093 * of this root.
3094 */
66b4ffd1 3095int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3096{
3097 struct btrfs_path *path;
3098 struct extent_buffer *leaf;
7b128766
JB		 3099 struct btrfs_key key, found_key;
3100 struct btrfs_trans_handle *trans;
3101 struct inode *inode;
8f6d7f4f 3102 u64 last_objectid = 0;
7b128766
JB		 3103 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3104
d68fc57b 3105 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3106 return 0;
c71bf099
YZ		 3107
3108 path = btrfs_alloc_path();
66b4ffd1
JB		 3109 if (!path) {
3110 ret = -ENOMEM;
3111 goto out;
3112 }
7b128766
JB		 3113 path->reada = -1;
3114
3115 key.objectid = BTRFS_ORPHAN_OBJECTID;
3116 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3117 key.offset = (u64)-1;
3118
7b128766
JB		 3119 while (1) {
3120 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3121 if (ret < 0)
3122 goto out;
7b128766
JB		 3123
3124 /*
3125 * if ret == 0 means we found what we were searching for, which
25985edc 3126 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3127 * find the key and see if we have stuff that matches
3128 */
3129 if (ret > 0) {
66b4ffd1 3130 ret = 0;
7b128766
JB		 3131 if (path->slots[0] == 0)
3132 break;
3133 path->slots[0]--;
3134 }
3135
3136 /* pull out the item */
3137 leaf = path->nodes[0];
7b128766
JB		 3138 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3139
3140 /* make sure the item matches what we want */
3141 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3142 break;
3143 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3144 break;
3145
3146 /* release the path since we're done with it */
b3b4aa74 3147 btrfs_release_path(path);
7b128766
JB		 3148
3149 /*
3150 * this is where we are basically btrfs_lookup, without the
3151 * crossing root thing. we store the inode number in the
3152 * offset of the orphan item.
3153 */
8f6d7f4f
JB		 3154
3155 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3156 btrfs_err(root->fs_info,
3157 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3158 ret = -EINVAL;
3159 goto out;
3160 }
3161
3162 last_objectid = found_key.offset;
3163
5d4f98a2
YZ		 3164 found_key.objectid = found_key.offset;
3165 found_key.type = BTRFS_INODE_ITEM_KEY;
3166 found_key.offset = 0;
73f73415 3167 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 3168 ret = PTR_RET(inode);
3169 if (ret && ret != -ESTALE)
66b4ffd1 3170 goto out;
7b128766 3171
f8e9e0b0
AJ		 3172 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3173 struct btrfs_root *dead_root;
3174 struct btrfs_fs_info *fs_info = root->fs_info;
3175 int is_dead_root = 0;
3176
3177 /*
3178 * this is an orphan in the tree root. Currently these
3179 * could come from 2 sources:
3180 * a) a snapshot deletion in progress
3181 * b) a free space cache inode
3182 * We need to distinguish those two, as the snapshot
3183 * orphan must not get deleted.
3184 * find_dead_roots already ran before us, so if this
3185 * is a snapshot deletion, we should find the root
3186 * in the dead_roots list
3187 */
3188 spin_lock(&fs_info->trans_lock);
3189 list_for_each_entry(dead_root, &fs_info->dead_roots,
3190 root_list) {
3191 if (dead_root->root_key.objectid ==
3192 found_key.objectid) {
3193 is_dead_root = 1;
3194 break;
3195 }
3196 }
3197 spin_unlock(&fs_info->trans_lock);
3198 if (is_dead_root) {
3199 /* prevent this orphan from being found again */
3200 key.offset = found_key.objectid - 1;
3201 continue;
3202 }
3203 }
7b128766 3204 /*
a8c9e576
JB		 3205 * Inode is already gone but the orphan item is still there,
3206 * kill the orphan item.
7b128766 3207 */
a8c9e576
JB		 3208 if (ret == -ESTALE) {
3209 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3210 if (IS_ERR(trans)) {
3211 ret = PTR_ERR(trans);
3212 goto out;
3213 }
c2cf52eb
SK		 3214 btrfs_debug(root->fs_info, "auto deleting %Lu",
3215 found_key.objectid);
a8c9e576
JB		 3216 ret = btrfs_del_orphan_item(trans, root,
3217 found_key.objectid);
79787eaa 3218 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 3219 btrfs_end_transaction(trans, root);
7b128766
JB		 3220 continue;
3221 }
3222
a8c9e576
JB		 3223 /*
3224 * add this inode to the orphan list so btrfs_orphan_del does
3225 * the proper thing when we hit it
3226 */
8a35d95f
JB		 3227 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3228 &BTRFS_I(inode)->runtime_flags);
925396ec 3229 atomic_inc(&root->orphan_inodes);
a8c9e576 3230
7b128766
JB		 3231 /* if we have links, this was a truncate, lets do that */
3232 if (inode->i_nlink) {
a41ad394
JB		 3233 if (!S_ISREG(inode->i_mode)) {
3234 WARN_ON(1);
3235 iput(inode);
3236 continue;
3237 }
7b128766 3238 nr_truncate++;
f3fe820c
JB		 3239
3240 /* 1 for the orphan item deletion. */
3241 trans = btrfs_start_transaction(root, 1);
3242 if (IS_ERR(trans)) {
c69b26b0 3243 iput(inode);
f3fe820c
JB		 3244 ret = PTR_ERR(trans);
3245 goto out;
3246 }
3247 ret = btrfs_orphan_add(trans, inode);
3248 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3249 if (ret) {
3250 iput(inode);
f3fe820c 3251 goto out;
c69b26b0 3252 }
f3fe820c 3253
66b4ffd1 3254 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3255 if (ret)
3256 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3257 } else {
3258 nr_unlink++;
3259 }
3260
3261 /* this will do delete_inode and everything for us */
3262 iput(inode);
66b4ffd1
JB		 3263 if (ret)
3264 goto out;
7b128766 3265 }
3254c876
MX		 3266 /* release the path since we're done with it */
3267 btrfs_release_path(path);
3268
d68fc57b
YZ		 3269 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3270
3271 if (root->orphan_block_rsv)
3272 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3273 (u64)-1);
3274
3275 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3276 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3277 if (!IS_ERR(trans))
3278 btrfs_end_transaction(trans, root);
d68fc57b 3279 }
7b128766
JB		 3280
3281 if (nr_unlink)
4884b476 3282 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3283 if (nr_truncate)
4884b476 3284 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3285
3286out:
3287 if (ret)
c2cf52eb
SK		 3288 btrfs_crit(root->fs_info,
3289 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3290 btrfs_free_path(path);
3291 return ret;
7b128766
JB		 3292}
3293
46a53cca
CM		 3294/*
3295 * very simple check to peek ahead in the leaf looking for xattrs. If we
3296 * don't find any xattrs, we know there can't be any acls.
3297 *
3298 * slot is the slot the inode is in, objectid is the objectid of the inode
3299 */
3300static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3301 int slot, u64 objectid)
3302{
3303 u32 nritems = btrfs_header_nritems(leaf);
3304 struct btrfs_key found_key;
f23b5a59
JB		 3305 static u64 xattr_access = 0;
3306 static u64 xattr_default = 0;
46a53cca
CM		 3307 int scanned = 0;
3308
f23b5a59
JB		 3309 if (!xattr_access) {
3310 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3311 strlen(POSIX_ACL_XATTR_ACCESS));
3312 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3313 strlen(POSIX_ACL_XATTR_DEFAULT));
3314 }
3315
46a53cca
CM		 3316 slot++;
3317 while (slot < nritems) {
3318 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3319
3320 /* we found a different objectid, there must not be acls */
3321 if (found_key.objectid != objectid)
3322 return 0;
3323
3324 /* we found an xattr, assume we've got an acl */
f23b5a59
JB		 3325 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3326 if (found_key.offset == xattr_access ||
3327 found_key.offset == xattr_default)
3328 return 1;
3329 }
46a53cca
CM		 3330
3331 /*
3332 * we found a key greater than an xattr key, there can't
3333 * be any acls later on
3334 */
3335 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3336 return 0;
3337
3338 slot++;
3339 scanned++;
3340
3341 /*
3342 * it goes inode, inode backrefs, xattrs, extents,
3343 * so if there are a ton of hard links to an inode there can
3344 * be a lot of backrefs. Don't waste time searching too hard,
3345 * this is just an optimization
3346 */
3347 if (scanned >= 8)
3348 break;
3349 }
3350 /* we hit the end of the leaf before we found an xattr or
3351 * something larger than an xattr. We have to assume the inode
3352 * has acls
3353 */
3354 return 1;
3355}
3356
d352ac68
CM		 3357/*
3358 * read an inode from the btree into the in-memory inode
3359 */
5d4f98a2 3360static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3361{
3362 struct btrfs_path *path;
5f39d397 3363 struct extent_buffer *leaf;
39279cc3 3364 struct btrfs_inode_item *inode_item;
0b86a832 3365 struct btrfs_timespec *tspec;
39279cc3
CM		 3366 struct btrfs_root *root = BTRFS_I(inode)->root;
3367 struct btrfs_key location;
46a53cca 3368 int maybe_acls;
618e21d5 3369 u32 rdev;
39279cc3 3370 int ret;
2f7e33d4
MX		 3371 bool filled = false;
3372
3373 ret = btrfs_fill_inode(inode, &rdev);
3374 if (!ret)
3375 filled = true;
39279cc3
CM		 3376
3377 path = btrfs_alloc_path();
1748f843
MF		 3378 if (!path)
3379 goto make_bad;
3380
d90c7321 3381 path->leave_spinning = 1;
39279cc3 3382 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3383
39279cc3 3384 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3385 if (ret)
39279cc3 3386 goto make_bad;
39279cc3 3387
5f39d397 3388 leaf = path->nodes[0];
2f7e33d4
MX		 3389
3390 if (filled)
3391 goto cache_acl;
3392
5f39d397
CM		 3393 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3394 struct btrfs_inode_item);
5f39d397 3395 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3396 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3397 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3398 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3399 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3400
3401 tspec = btrfs_inode_atime(inode_item);
3402 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3403 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3404
3405 tspec = btrfs_inode_mtime(inode_item);
3406 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3407 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3408
3409 tspec = btrfs_inode_ctime(inode_item);
3410 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3411 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3412
a76a3cd4 3413 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3414 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3415 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3416
3417 /*
3418 * If we were modified in the current generation and evicted from memory
3419 * and then re-read we need to do a full sync since we don't have any
3420 * idea about which extents were modified before we were evicted from
3421 * cache.
3422 */
3423 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3424 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3425 &BTRFS_I(inode)->runtime_flags);
3426
0c4d2d95 3427 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3428 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3429 inode->i_rdev = 0;
5f39d397
CM		 3430 rdev = btrfs_inode_rdev(leaf, inode_item);
3431
aec7477b 3432 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3433 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3434cache_acl:
46a53cca
CM		 3435 /*
3436 * try to precache a NULL acl entry for files that don't have
3437 * any xattrs or acls
3438 */
33345d01
LZ		 3439 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3440 btrfs_ino(inode));
72c04902
AV		 3441 if (!maybe_acls)
3442 cache_no_acl(inode);
46a53cca 3443
39279cc3 3444 btrfs_free_path(path);
39279cc3 3445
39279cc3 3446 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3447 case S_IFREG:
3448 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3449 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3450 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3451 inode->i_fop = &btrfs_file_operations;
3452 inode->i_op = &btrfs_file_inode_operations;
3453 break;
3454 case S_IFDIR:
3455 inode->i_fop = &btrfs_dir_file_operations;
3456 if (root == root->fs_info->tree_root)
3457 inode->i_op = &btrfs_dir_ro_inode_operations;
3458 else
3459 inode->i_op = &btrfs_dir_inode_operations;
3460 break;
3461 case S_IFLNK:
3462 inode->i_op = &btrfs_symlink_inode_operations;
3463 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3464 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3465 break;
618e21d5 3466 default:
0279b4cd 3467 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3468 init_special_inode(inode, inode->i_mode, rdev);
3469 break;
39279cc3 3470 }
6cbff00f
CH		 3471
3472 btrfs_update_iflags(inode);
39279cc3
CM		 3473 return;
3474
3475make_bad:
39279cc3 3476 btrfs_free_path(path);
39279cc3
CM		 3477 make_bad_inode(inode);
3478}
3479
d352ac68
CM		 3480/*
3481 * given a leaf and an inode, copy the inode fields into the leaf
3482 */
e02119d5
CM		 3483static void fill_inode_item(struct btrfs_trans_handle *trans,
3484 struct extent_buffer *leaf,
5f39d397 3485 struct btrfs_inode_item *item,
39279cc3
CM		 3486 struct inode *inode)
3487{
51fab693
LB		 3488 struct btrfs_map_token token;
3489
3490 btrfs_init_map_token(&token);
5f39d397 3491
51fab693
LB		 3492 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3493 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3494 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3495 &token);
3496 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3497 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3498
51fab693
LB		 3499 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3500 inode->i_atime.tv_sec, &token);
3501 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3502 inode->i_atime.tv_nsec, &token);
5f39d397 3503
51fab693
LB		 3504 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3505 inode->i_mtime.tv_sec, &token);
3506 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3507 inode->i_mtime.tv_nsec, &token);
5f39d397 3508
51fab693
LB		 3509 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3510 inode->i_ctime.tv_sec, &token);
3511 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3512 inode->i_ctime.tv_nsec, &token);
5f39d397 3513
51fab693
LB		 3514 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3515 &token);
3516 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3517 &token);
3518 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3519 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3520 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3521 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3522 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3523}
3524
d352ac68
CM		 3525/*
3526 * copy everything in the in-memory inode into the btree.
3527 */
2115133f 3528static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3529 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3530{
3531 struct btrfs_inode_item *inode_item;
3532 struct btrfs_path *path;
5f39d397 3533 struct extent_buffer *leaf;
39279cc3
CM		 3534 int ret;
3535
3536 path = btrfs_alloc_path();
16cdcec7
MX		 3537 if (!path)
3538 return -ENOMEM;
3539
b9473439 3540 path->leave_spinning = 1;
16cdcec7
MX		 3541 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3542 1);
39279cc3
CM		 3543 if (ret) {
3544 if (ret > 0)
3545 ret = -ENOENT;
3546 goto failed;
3547 }
3548
b4ce94de 3549 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3550 leaf = path->nodes[0];
3551 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3552 struct btrfs_inode_item);
39279cc3 3553
e02119d5 3554 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3555 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3556 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3557 ret = 0;
3558failed:
39279cc3
CM		 3559 btrfs_free_path(path);
3560 return ret;
3561}
3562
2115133f
CM		 3563/*
3564 * copy everything in the in-memory inode into the btree.
3565 */
3566noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3567 struct btrfs_root *root, struct inode *inode)
3568{
3569 int ret;
3570
3571 /*
3572 * If the inode is a free space inode, we can deadlock during commit
3573 * if we put it into the delayed code.
3574 *
3575 * The data relocation inode should also be directly updated
3576 * without delay
3577 */
83eea1f1 3578 if (!btrfs_is_free_space_inode(inode)
2115133f 3579 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3580 btrfs_update_root_times(trans, root);
3581
2115133f
CM		 3582 ret = btrfs_delayed_update_inode(trans, root, inode);
3583 if (!ret)
3584 btrfs_set_inode_last_trans(trans, inode);
3585 return ret;
3586 }
3587
3588 return btrfs_update_inode_item(trans, root, inode);
3589}
3590
be6aef60
JB		 3591noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3592 struct btrfs_root *root,
3593 struct inode *inode)
2115133f
CM		 3594{
3595 int ret;
3596
3597 ret = btrfs_update_inode(trans, root, inode);
3598 if (ret == -ENOSPC)
3599 return btrfs_update_inode_item(trans, root, inode);
3600 return ret;
3601}
3602
d352ac68
CM		 3603/*
3604 * unlink helper that gets used here in inode.c and in the tree logging
3605 * recovery code. It remove a link in a directory with a given name, and
3606 * also drops the back refs in the inode to the directory
3607 */
92986796
AV		 3608static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3609 struct btrfs_root *root,
3610 struct inode *dir, struct inode *inode,
3611 const char *name, int name_len)
39279cc3
CM		 3612{
3613 struct btrfs_path *path;
39279cc3 3614 int ret = 0;
5f39d397 3615 struct extent_buffer *leaf;
39279cc3 3616 struct btrfs_dir_item *di;
5f39d397 3617 struct btrfs_key key;
aec7477b 3618 u64 index;
33345d01
LZ		 3619 u64 ino = btrfs_ino(inode);
3620 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3621
3622 path = btrfs_alloc_path();
54aa1f4d
CM		 3623 if (!path) {
3624 ret = -ENOMEM;
554233a6 3625 goto out;
54aa1f4d
CM		 3626 }
3627
b9473439 3628 path->leave_spinning = 1;
33345d01 3629 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3630 name, name_len, -1);
3631 if (IS_ERR(di)) {
3632 ret = PTR_ERR(di);
3633 goto err;
3634 }
3635 if (!di) {
3636 ret = -ENOENT;
3637 goto err;
3638 }
5f39d397
CM		 3639 leaf = path->nodes[0];
3640 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3641 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3642 if (ret)
3643 goto err;
b3b4aa74 3644 btrfs_release_path(path);
39279cc3 3645
33345d01
LZ		 3646 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3647 dir_ino, &index);
aec7477b 3648 if (ret) {
c2cf52eb
SK		 3649 btrfs_info(root->fs_info,
3650 "failed to delete reference to %.*s, inode %llu parent %llu",
3651 name_len, name,
3652 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 3653 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3654 goto err;
3655 }
3656
16cdcec7 3657 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3658 if (ret) {
3659 btrfs_abort_transaction(trans, root, ret);
39279cc3 3660 goto err;
79787eaa 3661 }
39279cc3 3662
e02119d5 3663 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3664 inode, dir_ino);
79787eaa
JM		 3665 if (ret != 0 && ret != -ENOENT) {
3666 btrfs_abort_transaction(trans, root, ret);
3667 goto err;
3668 }
e02119d5
CM		 3669
3670 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3671 dir, index);
6418c961
CM		 3672 if (ret == -ENOENT)
3673 ret = 0;
d4e3991b
ZB		 3674 else if (ret)
3675 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3676err:
3677 btrfs_free_path(path);
e02119d5
CM		 3678 if (ret)
3679 goto out;
3680
3681 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3682 inode_inc_iversion(inode);
3683 inode_inc_iversion(dir);
e02119d5 3684 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3685 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3686out:
39279cc3
CM		 3687 return ret;
3688}
3689
92986796
AV		 3690int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3691 struct btrfs_root *root,
3692 struct inode *dir, struct inode *inode,
3693 const char *name, int name_len)
3694{
3695 int ret;
3696 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3697 if (!ret) {
3698 btrfs_drop_nlink(inode);
3699 ret = btrfs_update_inode(trans, root, inode);
3700 }
3701 return ret;
3702}
39279cc3 3703
a22285a6
YZ		 3704/*
3705 * helper to start transaction for unlink and rmdir.
3706 *
d52be818
JB		 3707 * unlink and rmdir are special in btrfs, they do not always free space, so
3708 * if we cannot make our reservations the normal way try and see if there is
3709 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3710 * allow the unlink to occur.
a22285a6 3711 */
d52be818 3712static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3713{
39279cc3 3714 struct btrfs_trans_handle *trans;
a22285a6 3715 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3716 int ret;
3717
e70bea5f
JB		 3718 /*
3719 * 1 for the possible orphan item
3720 * 1 for the dir item
3721 * 1 for the dir index
3722 * 1 for the inode ref
e70bea5f
JB		 3723 * 1 for the inode
3724 */
6e137ed3 3725 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3726 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3727 return trans;
4df27c4d 3728
d52be818
JB		 3729 if (PTR_ERR(trans) == -ENOSPC) {
3730 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
a22285a6 3731
d52be818
JB		 3732 trans = btrfs_start_transaction(root, 0);
3733 if (IS_ERR(trans))
3734 return trans;
3735 ret = btrfs_cond_migrate_bytes(root->fs_info,
3736 &root->fs_info->trans_block_rsv,
3737 num_bytes, 5);
3738 if (ret) {
3739 btrfs_end_transaction(trans, root);
3740 return ERR_PTR(ret);
a22285a6 3741 }
5a77d76c 3742 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3743 trans->bytes_reserved = num_bytes;
a22285a6 3744 }
d52be818 3745 return trans;
a22285a6
YZ		 3746}
3747
3748static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3749{
3750 struct btrfs_root *root = BTRFS_I(dir)->root;
3751 struct btrfs_trans_handle *trans;
3752 struct inode *inode = dentry->d_inode;
3753 int ret;
a22285a6 3754
d52be818 3755 trans = __unlink_start_trans(dir);
a22285a6
YZ		 3756 if (IS_ERR(trans))
3757 return PTR_ERR(trans);
5f39d397 3758
12fcfd22
CM		 3759 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3760
e02119d5
CM		 3761 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3762 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3763 if (ret)
3764 goto out;
7b128766 3765
a22285a6 3766 if (inode->i_nlink == 0) {
7b128766 3767 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3768 if (ret)
3769 goto out;
a22285a6 3770 }
7b128766 3771
b532402e 3772out:
d52be818 3773 btrfs_end_transaction(trans, root);
b53d3f5d 3774 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3775 return ret;
3776}
3777
4df27c4d
YZ		 3778int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3779 struct btrfs_root *root,
3780 struct inode *dir, u64 objectid,
3781 const char *name, int name_len)
3782{
3783 struct btrfs_path *path;
3784 struct extent_buffer *leaf;
3785 struct btrfs_dir_item *di;
3786 struct btrfs_key key;
3787 u64 index;
3788 int ret;
33345d01 3789 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3790
3791 path = btrfs_alloc_path();
3792 if (!path)
3793 return -ENOMEM;
3794
33345d01 3795 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3796 name, name_len, -1);
79787eaa
JM		 3797 if (IS_ERR_OR_NULL(di)) {
3798 if (!di)
3799 ret = -ENOENT;
3800 else
3801 ret = PTR_ERR(di);
3802 goto out;
3803 }
4df27c4d
YZ		 3804
3805 leaf = path->nodes[0];
3806 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3807 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3808 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3809 if (ret) {
3810 btrfs_abort_transaction(trans, root, ret);
3811 goto out;
3812 }
b3b4aa74 3813 btrfs_release_path(path);
4df27c4d
YZ		 3814
3815 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3816 objectid, root->root_key.objectid,
33345d01 3817 dir_ino, &index, name, name_len);
4df27c4d 3818 if (ret < 0) {
79787eaa
JM		 3819 if (ret != -ENOENT) {
3820 btrfs_abort_transaction(trans, root, ret);
3821 goto out;
3822 }
33345d01 3823 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3824 name, name_len);
79787eaa
JM		 3825 if (IS_ERR_OR_NULL(di)) {
3826 if (!di)
3827 ret = -ENOENT;
3828 else
3829 ret = PTR_ERR(di);
3830 btrfs_abort_transaction(trans, root, ret);
3831 goto out;
3832 }
4df27c4d
YZ		 3833
3834 leaf = path->nodes[0];
3835 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3836 btrfs_release_path(path);
4df27c4d
YZ		 3837 index = key.offset;
3838 }
945d8962 3839 btrfs_release_path(path);
4df27c4d 3840
16cdcec7 3841 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3842 if (ret) {
3843 btrfs_abort_transaction(trans, root, ret);
3844 goto out;
3845 }
4df27c4d
YZ		 3846
3847 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3848 inode_inc_iversion(dir);
4df27c4d 3849 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3850 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3851 if (ret)
3852 btrfs_abort_transaction(trans, root, ret);
3853out:
71d7aed0 3854 btrfs_free_path(path);
79787eaa 3855 return ret;
4df27c4d
YZ		 3856}
3857
39279cc3
CM		 3858static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3859{
3860 struct inode *inode = dentry->d_inode;
1832a6d5 3861 int err = 0;
39279cc3 3862 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3863 struct btrfs_trans_handle *trans;
39279cc3 3864
b3ae244e 3865 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3866 return -ENOTEMPTY;
b3ae244e
DS		 3867 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3868 return -EPERM;
134d4512 3869
d52be818 3870 trans = __unlink_start_trans(dir);
a22285a6 3871 if (IS_ERR(trans))
5df6a9f6 3872 return PTR_ERR(trans);
5df6a9f6 3873
33345d01 3874 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3875 err = btrfs_unlink_subvol(trans, root, dir,
3876 BTRFS_I(inode)->location.objectid,
3877 dentry->d_name.name,
3878 dentry->d_name.len);
3879 goto out;
3880 }
3881
7b128766
JB		 3882 err = btrfs_orphan_add(trans, inode);
3883 if (err)
4df27c4d 3884 goto out;
7b128766 3885
39279cc3 3886 /* now the directory is empty */
e02119d5
CM		 3887 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3888 dentry->d_name.name, dentry->d_name.len);
d397712b 3889 if (!err)
dbe674a9 3890 btrfs_i_size_write(inode, 0);
4df27c4d 3891out:
d52be818 3892 btrfs_end_transaction(trans, root);
b53d3f5d 3893 btrfs_btree_balance_dirty(root);
3954401f 3894
39279cc3
CM		 3895 return err;
3896}
3897
39279cc3
CM		 3898/*
3899 * this can truncate away extent items, csum items and directory items.
3900 * It starts at a high offset and removes keys until it can't find
d352ac68 3901 * any higher than new_size
39279cc3
CM		 3902 *
3903 * csum items that cross the new i_size are truncated to the new size
3904 * as well.
7b128766
JB		 3905 *
3906 * min_type is the minimum key type to truncate down to. If set to 0, this
3907 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3908 */
8082510e
YZ		 3909int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3910 struct btrfs_root *root,
3911 struct inode *inode,
3912 u64 new_size, u32 min_type)
39279cc3 3913{
39279cc3 3914 struct btrfs_path *path;
5f39d397 3915 struct extent_buffer *leaf;
39279cc3 3916 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3917 struct btrfs_key key;
3918 struct btrfs_key found_key;
39279cc3 3919 u64 extent_start = 0;
db94535d 3920 u64 extent_num_bytes = 0;
5d4f98a2 3921 u64 extent_offset = 0;
39279cc3 3922 u64 item_end = 0;
8082510e 3923 u32 found_type = (u8)-1;
39279cc3
CM		 3924 int found_extent;
3925 int del_item;
85e21bac
CM		 3926 int pending_del_nr = 0;
3927 int pending_del_slot = 0;
179e29e4 3928 int extent_type = -1;
8082510e
YZ		 3929 int ret;
3930 int err = 0;
33345d01 3931 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3932
3933 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3934
0eb0e19c
MF		 3935 path = btrfs_alloc_path();
3936 if (!path)
3937 return -ENOMEM;
3938 path->reada = -1;
3939
5dc562c5
JB		 3940 /*
3941 * We want to drop from the next block forward in case this new size is
3942 * not block aligned since we will be keeping the last block of the
3943 * extent just the way it is.
3944 */
0af3d00b 3945 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 3946 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3947 root->sectorsize), (u64)-1, 0);
8082510e 3948
16cdcec7
MX		 3949 /*
3950 * This function is also used to drop the items in the log tree before
3951 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3952 * it is used to drop the loged items. So we shouldn't kill the delayed
3953 * items.
3954 */
3955 if (min_type == 0 && root == BTRFS_I(inode)->root)
3956 btrfs_kill_delayed_inode_items(inode);
3957
33345d01 3958 key.objectid = ino;
39279cc3 3959 key.offset = (u64)-1;
5f39d397
CM		 3960 key.type = (u8)-1;
3961
85e21bac 3962search_again:
b9473439 3963 path->leave_spinning = 1;
85e21bac 3964 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3965 if (ret < 0) {
3966 err = ret;
3967 goto out;
3968 }
d397712b 3969
85e21bac 3970 if (ret > 0) {
e02119d5
CM		 3971 /* there are no items in the tree for us to truncate, we're
3972 * done
3973 */
8082510e
YZ		 3974 if (path->slots[0] == 0)
3975 goto out;
85e21bac
CM		 3976 path->slots[0]--;
3977 }
3978
d397712b 3979 while (1) {
39279cc3 3980 fi = NULL;
5f39d397
CM		 3981 leaf = path->nodes[0];
3982 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3983 found_type = btrfs_key_type(&found_key);
39279cc3 3984
33345d01 3985 if (found_key.objectid != ino)
39279cc3 3986 break;
5f39d397 3987
85e21bac 3988 if (found_type < min_type)
39279cc3
CM		 3989 break;
3990
5f39d397 3991 item_end = found_key.offset;
39279cc3 3992 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3993 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3994 struct btrfs_file_extent_item);
179e29e4
CM		 3995 extent_type = btrfs_file_extent_type(leaf, fi);
3996 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3997 item_end +=
db94535d 3998 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3999 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4000 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 4001 fi);
39279cc3 4002 }
008630c1 4003 item_end--;
39279cc3 4004 }
8082510e
YZ		 4005 if (found_type > min_type) {
4006 del_item = 1;
4007 } else {
4008 if (item_end < new_size)
b888db2b 4009 break;
8082510e
YZ		 4010 if (found_key.offset >= new_size)
4011 del_item = 1;
4012 else
4013 del_item = 0;
39279cc3 4014 }
39279cc3 4015 found_extent = 0;
39279cc3 4016 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4017 if (found_type != BTRFS_EXTENT_DATA_KEY)
4018 goto delete;
4019
4020 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4021 u64 num_dec;
db94535d 4022 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4023 if (!del_item) {
db94535d
CM		 4024 u64 orig_num_bytes =
4025 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4026 extent_num_bytes = ALIGN(new_size -
4027 found_key.offset,
4028 root->sectorsize);
db94535d
CM		 4029 btrfs_set_file_extent_num_bytes(leaf, fi,
4030 extent_num_bytes);
4031 num_dec = (orig_num_bytes -
9069218d 4032 extent_num_bytes);
e02119d5 4033 if (root->ref_cows && extent_start != 0)
a76a3cd4 4034 inode_sub_bytes(inode, num_dec);
5f39d397 4035 btrfs_mark_buffer_dirty(leaf);
39279cc3 4036 } else {
db94535d
CM		 4037 extent_num_bytes =
4038 btrfs_file_extent_disk_num_bytes(leaf,
4039 fi);
5d4f98a2
YZ		 4040 extent_offset = found_key.offset -
4041 btrfs_file_extent_offset(leaf, fi);
4042
39279cc3 4043 /* FIXME blocksize != 4096 */
9069218d 4044 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4045 if (extent_start != 0) {
4046 found_extent = 1;
e02119d5 4047 if (root->ref_cows)
a76a3cd4 4048 inode_sub_bytes(inode, num_dec);
e02119d5 4049 }
39279cc3 4050 }
9069218d 4051 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4052 /*
4053 * we can't truncate inline items that have had
4054 * special encodings
4055 */
4056 if (!del_item &&
4057 btrfs_file_extent_compression(leaf, fi) == 0 &&
4058 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4059 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4060 u32 size = new_size - found_key.offset;
4061
4062 if (root->ref_cows) {
a76a3cd4
YZ		 4063 inode_sub_bytes(inode, item_end + 1 -
4064 new_size);
e02119d5
CM		 4065 }
4066 size =
4067 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4068 btrfs_truncate_item(root, path, size, 1);
e02119d5 4069 } else if (root->ref_cows) {
a76a3cd4
YZ		 4070 inode_sub_bytes(inode, item_end + 1 -
4071 found_key.offset);
9069218d 4072 }
39279cc3 4073 }
179e29e4 4074delete:
39279cc3 4075 if (del_item) {
85e21bac
CM		 4076 if (!pending_del_nr) {
4077 /* no pending yet, add ourselves */
4078 pending_del_slot = path->slots[0];
4079 pending_del_nr = 1;
4080 } else if (pending_del_nr &&
4081 path->slots[0] + 1 == pending_del_slot) {
4082 /* hop on the pending chunk */
4083 pending_del_nr++;
4084 pending_del_slot = path->slots[0];
4085 } else {
d397712b 4086 BUG();
85e21bac 4087 }
39279cc3
CM		 4088 } else {
4089 break;
4090 }
0af3d00b
JB		 4091 if (found_extent && (root->ref_cows ||
4092 root == root->fs_info->tree_root)) {
b9473439 4093 btrfs_set_path_blocking(path);
39279cc3 4094 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4095 extent_num_bytes, 0,
4096 btrfs_header_owner(leaf),
66d7e7f0 4097 ino, extent_offset, 0);
39279cc3
CM		 4098 BUG_ON(ret);
4099 }
85e21bac 4100
8082510e
YZ		 4101 if (found_type == BTRFS_INODE_ITEM_KEY)
4102 break;
4103
4104 if (path->slots[0] == 0 ||
4105 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4106 if (pending_del_nr) {
4107 ret = btrfs_del_items(trans, root, path,
4108 pending_del_slot,
4109 pending_del_nr);
79787eaa
JM		 4110 if (ret) {
4111 btrfs_abort_transaction(trans,
4112 root, ret);
4113 goto error;
4114 }
8082510e
YZ		 4115 pending_del_nr = 0;
4116 }
b3b4aa74 4117 btrfs_release_path(path);
85e21bac 4118 goto search_again;
8082510e
YZ		 4119 } else {
4120 path->slots[0]--;
85e21bac 4121 }
39279cc3 4122 }
8082510e 4123out:
85e21bac
CM		 4124 if (pending_del_nr) {
4125 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4126 pending_del_nr);
79787eaa
JM		 4127 if (ret)
4128 btrfs_abort_transaction(trans, root, ret);
85e21bac 4129 }
79787eaa 4130error:
39279cc3 4131 btrfs_free_path(path);
8082510e 4132 return err;
39279cc3
CM		 4133}
4134
4135/*
2aaa6655
JB		 4136 * btrfs_truncate_page - read, zero a chunk and write a page
4137 * @inode - inode that we're zeroing
4138 * @from - the offset to start zeroing
4139 * @len - the length to zero, 0 to zero the entire range respective to the
4140 * offset
4141 * @front - zero up to the offset instead of from the offset on
4142 *
4143 * This will find the page for the "from" offset and cow the page and zero the
4144 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4145 */
2aaa6655
JB		 4146int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4147 int front)
39279cc3 4148{
2aaa6655 4149 struct address_space *mapping = inode->i_mapping;
db94535d 4150 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4151 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4152 struct btrfs_ordered_extent *ordered;
2ac55d41 4153 struct extent_state *cached_state = NULL;
e6dcd2dc 4154 char *kaddr;
db94535d 4155 u32 blocksize = root->sectorsize;
39279cc3
CM		 4156 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4157 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4158 struct page *page;
3b16a4e3 4159 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4160 int ret = 0;
a52d9a80 4161 u64 page_start;
e6dcd2dc 4162 u64 page_end;
39279cc3 4163
2aaa6655
JB		 4164 if ((offset & (blocksize - 1)) == 0 &&
4165 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4166 goto out;
0ca1f7ce 4167 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4168 if (ret)
4169 goto out;
39279cc3 4170
211c17f5 4171again:
3b16a4e3 4172 page = find_or_create_page(mapping, index, mask);
5d5e103a 4173 if (!page) {
0ca1f7ce 4174 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4175 ret = -ENOMEM;
39279cc3 4176 goto out;
5d5e103a 4177 }
e6dcd2dc
CM		 4178
4179 page_start = page_offset(page);
4180 page_end = page_start + PAGE_CACHE_SIZE - 1;
4181
39279cc3 4182 if (!PageUptodate(page)) {
9ebefb18 4183 ret = btrfs_readpage(NULL, page);
39279cc3 4184 lock_page(page);
211c17f5
CM		 4185 if (page->mapping != mapping) {
4186 unlock_page(page);
4187 page_cache_release(page);
4188 goto again;
4189 }
39279cc3
CM		 4190 if (!PageUptodate(page)) {
4191 ret = -EIO;
89642229 4192 goto out_unlock;
39279cc3
CM		 4193 }
4194 }
211c17f5 4195 wait_on_page_writeback(page);
e6dcd2dc 4196
d0082371 4197 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4198 set_page_extent_mapped(page);
4199
4200 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4201 if (ordered) {
2ac55d41
JB		 4202 unlock_extent_cached(io_tree, page_start, page_end,
4203 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4204 unlock_page(page);
4205 page_cache_release(page);
eb84ae03 4206 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4207 btrfs_put_ordered_extent(ordered);
4208 goto again;
4209 }
4210
2ac55d41 4211 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4212 EXTENT_DIRTY | EXTENT_DELALLOC |
4213 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4214 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4215
2ac55d41
JB		 4216 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4217 &cached_state);
9ed74f2d 4218 if (ret) {
2ac55d41
JB		 4219 unlock_extent_cached(io_tree, page_start, page_end,
4220 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4221 goto out_unlock;
4222 }
4223
e6dcd2dc 4224 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4225 if (!len)
4226 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4227 kaddr = kmap(page);
2aaa6655
JB		 4228 if (front)
4229 memset(kaddr, 0, offset);
4230 else
4231 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4232 flush_dcache_page(page);
4233 kunmap(page);
4234 }
247e743c 4235 ClearPageChecked(page);
e6dcd2dc 4236 set_page_dirty(page);
2ac55d41
JB		 4237 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4238 GFP_NOFS);
39279cc3 4239
89642229 4240out_unlock:
5d5e103a 4241 if (ret)
0ca1f7ce 4242 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4243 unlock_page(page);
4244 page_cache_release(page);
4245out:
4246 return ret;
4247}
4248
695a0d0d
JB		 4249/*
4250 * This function puts in dummy file extents for the area we're creating a hole
4251 * for. So if we are truncating this file to a larger size we need to insert
4252 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4253 * the range between oldsize and size
4254 */
a41ad394 4255int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4256{
9036c102
YZ		 4257 struct btrfs_trans_handle *trans;
4258 struct btrfs_root *root = BTRFS_I(inode)->root;
4259 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4260 struct extent_map *em = NULL;
2ac55d41 4261 struct extent_state *cached_state = NULL;
5dc562c5 4262 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4263 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4264 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4265 u64 last_byte;
4266 u64 cur_offset;
4267 u64 hole_size;
9ed74f2d 4268 int err = 0;
39279cc3 4269
a71754fc
JB		 4270 /*
4271 * If our size started in the middle of a page we need to zero out the
4272 * rest of the page before we expand the i_size, otherwise we could
4273 * expose stale data.
4274 */
4275 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4276 if (err)
4277 return err;
4278
9036c102
YZ		 4279 if (size <= hole_start)
4280 return 0;
4281
9036c102
YZ		 4282 while (1) {
4283 struct btrfs_ordered_extent *ordered;
4284 btrfs_wait_ordered_range(inode, hole_start,
4285 block_end - hole_start);
2ac55d41 4286 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4287 &cached_state);
9036c102
YZ		 4288 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4289 if (!ordered)
4290 break;
2ac55d41
JB		 4291 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4292 &cached_state, GFP_NOFS);
9036c102
YZ		 4293 btrfs_put_ordered_extent(ordered);
4294 }
39279cc3 4295
9036c102
YZ		 4296 cur_offset = hole_start;
4297 while (1) {
4298 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4299 block_end - cur_offset, 0);
79787eaa
JM		 4300 if (IS_ERR(em)) {
4301 err = PTR_ERR(em);
f2767956 4302 em = NULL;
79787eaa
JM		 4303 break;
4304 }
9036c102 4305 last_byte = min(extent_map_end(em), block_end);
fda2832f 4306 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4307 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4308 struct extent_map *hole_em;
9036c102 4309 hole_size = last_byte - cur_offset;
9ed74f2d 4310
3642320e 4311 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4312 if (IS_ERR(trans)) {
4313 err = PTR_ERR(trans);
9ed74f2d 4314 break;
a22285a6 4315 }
8082510e 4316
5dc562c5
JB		 4317 err = btrfs_drop_extents(trans, root, inode,
4318 cur_offset,
2671485d 4319 cur_offset + hole_size, 1);
5b397377 4320 if (err) {
79787eaa 4321 btrfs_abort_transaction(trans, root, err);
5b397377 4322 btrfs_end_transaction(trans, root);
3893e33b 4323 break;
5b397377 4324 }
8082510e 4325
9036c102 4326 err = btrfs_insert_file_extent(trans, root,
33345d01 4327 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4328 0, hole_size, 0, hole_size,
4329 0, 0, 0);
5b397377 4330 if (err) {
79787eaa 4331 btrfs_abort_transaction(trans, root, err);
5b397377 4332 btrfs_end_transaction(trans, root);
3893e33b 4333 break;
5b397377 4334 }
8082510e 4335
5dc562c5
JB		 4336 btrfs_drop_extent_cache(inode, cur_offset,
4337 cur_offset + hole_size - 1, 0);
4338 hole_em = alloc_extent_map();
4339 if (!hole_em) {
4340 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4341 &BTRFS_I(inode)->runtime_flags);
4342 goto next;
4343 }
4344 hole_em->start = cur_offset;
4345 hole_em->len = hole_size;
4346 hole_em->orig_start = cur_offset;
8082510e 4347
5dc562c5
JB		 4348 hole_em->block_start = EXTENT_MAP_HOLE;
4349 hole_em->block_len = 0;
b4939680 4350 hole_em->orig_block_len = 0;
cc95bef6 4351 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4352 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4353 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4354 hole_em->generation = trans->transid;
8082510e 4355
5dc562c5
JB		 4356 while (1) {
4357 write_lock(&em_tree->lock);
09a2a8f9 4358 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4359 write_unlock(&em_tree->lock);
4360 if (err != -EEXIST)
4361 break;
4362 btrfs_drop_extent_cache(inode, cur_offset,
4363 cur_offset +
4364 hole_size - 1, 0);
4365 }
4366 free_extent_map(hole_em);
4367next:
3642320e 4368 btrfs_update_inode(trans, root, inode);
8082510e 4369 btrfs_end_transaction(trans, root);
9036c102
YZ		 4370 }
4371 free_extent_map(em);
a22285a6 4372 em = NULL;
9036c102 4373 cur_offset = last_byte;
8082510e 4374 if (cur_offset >= block_end)
9036c102
YZ		 4375 break;
4376 }
1832a6d5 4377
a22285a6 4378 free_extent_map(em);
2ac55d41
JB		 4379 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4380 GFP_NOFS);
9036c102
YZ		 4381 return err;
4382}
39279cc3 4383
3972f260 4384static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4385{
f4a2f4c5
MX		 4386 struct btrfs_root *root = BTRFS_I(inode)->root;
4387 struct btrfs_trans_handle *trans;
a41ad394 4388 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4389 loff_t newsize = attr->ia_size;
4390 int mask = attr->ia_valid;
8082510e
YZ		 4391 int ret;
4392
a41ad394 4393 if (newsize == oldsize)
8082510e
YZ		 4394 return 0;
4395
3972f260
ES		 4396 /*
4397 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4398 * special case where we need to update the times despite not having
4399 * these flags set. For all other operations the VFS set these flags
4400 * explicitly if it wants a timestamp update.
4401 */
4402 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4403 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4404
a41ad394 4405 if (newsize > oldsize) {
a41ad394
JB		 4406 truncate_pagecache(inode, oldsize, newsize);
4407 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4408 if (ret)
8082510e 4409 return ret;
8082510e 4410
f4a2f4c5
MX		 4411 trans = btrfs_start_transaction(root, 1);
4412 if (IS_ERR(trans))
4413 return PTR_ERR(trans);
4414
4415 i_size_write(inode, newsize);
4416 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4417 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4418 btrfs_end_transaction(trans, root);
a41ad394 4419 } else {
8082510e 4420
a41ad394
JB		 4421 /*
4422 * We're truncating a file that used to have good data down to
4423 * zero. Make sure it gets into the ordered flush list so that
4424 * any new writes get down to disk quickly.
4425 */
4426 if (newsize == 0)
72ac3c0d
JB		 4427 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4428 &BTRFS_I(inode)->runtime_flags);
8082510e 4429
f3fe820c
JB		 4430 /*
4431 * 1 for the orphan item we're going to add
4432 * 1 for the orphan item deletion.
4433 */
4434 trans = btrfs_start_transaction(root, 2);
4435 if (IS_ERR(trans))
4436 return PTR_ERR(trans);
4437
4438 /*
4439 * We need to do this in case we fail at _any_ point during the
4440 * actual truncate. Once we do the truncate_setsize we could
4441 * invalidate pages which forces any outstanding ordered io to
4442 * be instantly completed which will give us extents that need
4443 * to be truncated. If we fail to get an orphan inode down we
4444 * could have left over extents that were never meant to live,
4445 * so we need to garuntee from this point on that everything
4446 * will be consistent.
4447 */
4448 ret = btrfs_orphan_add(trans, inode);
4449 btrfs_end_transaction(trans, root);
4450 if (ret)
4451 return ret;
4452
a41ad394
JB		 4453 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4454 truncate_setsize(inode, newsize);
2e60a51e
MX		 4455
4456 /* Disable nonlocked read DIO to avoid the end less truncate */
4457 btrfs_inode_block_unlocked_dio(inode);
4458 inode_dio_wait(inode);
4459 btrfs_inode_resume_unlocked_dio(inode);
4460
a41ad394 4461 ret = btrfs_truncate(inode);
f3fe820c
JB		 4462 if (ret && inode->i_nlink)
4463 btrfs_orphan_del(NULL, inode);
8082510e
YZ		 4464 }
4465
a41ad394 4466 return ret;
8082510e
YZ		 4467}
4468
9036c102
YZ		 4469static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4470{
4471 struct inode *inode = dentry->d_inode;
b83cc969 4472 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4473 int err;
39279cc3 4474
b83cc969
LZ		 4475 if (btrfs_root_readonly(root))
4476 return -EROFS;
4477
9036c102
YZ		 4478 err = inode_change_ok(inode, attr);
4479 if (err)
4480 return err;
2bf5a725 4481
5a3f23d5 4482 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4483 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4484 if (err)
4485 return err;
39279cc3 4486 }
9036c102 4487
1025774c
CH		 4488 if (attr->ia_valid) {
4489 setattr_copy(inode, attr);
0c4d2d95 4490 inode_inc_iversion(inode);
22c44fe6 4491 err = btrfs_dirty_inode(inode);
1025774c 4492
22c44fe6 4493 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4494 err = btrfs_acl_chmod(inode);
4495 }
33268eaf 4496
39279cc3
CM		 4497 return err;
4498}
61295eb8 4499
bd555975 4500void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4501{
4502 struct btrfs_trans_handle *trans;
4503 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4504 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4505 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4506 int ret;
4507
1abe9b8a 4508 trace_btrfs_inode_evict(inode);
4509
39279cc3 4510 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 4511 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 4512 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4513 goto no_delete;
4514
39279cc3 4515 if (is_bad_inode(inode)) {
7b128766 4516 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4517 goto no_delete;
4518 }
bd555975 4519 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4520 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4521
c71bf099 4522 if (root->fs_info->log_root_recovering) {
6bf02314 4523 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4524 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4525 goto no_delete;
4526 }
4527
76dda93c
YZ		 4528 if (inode->i_nlink > 0) {
4529 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4530 goto no_delete;
4531 }
4532
0e8c36a9
MX		 4533 ret = btrfs_commit_inode_delayed_inode(inode);
4534 if (ret) {
4535 btrfs_orphan_del(NULL, inode);
4536 goto no_delete;
4537 }
4538
66d8f3dd 4539 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4540 if (!rsv) {
4541 btrfs_orphan_del(NULL, inode);
4542 goto no_delete;
4543 }
4a338542 4544 rsv->size = min_size;
ca7e70f5 4545 rsv->failfast = 1;
726c35fa 4546 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4547
dbe674a9 4548 btrfs_i_size_write(inode, 0);
5f39d397 4549
4289a667 4550 /*
8407aa46
MX		 4551 * This is a bit simpler than btrfs_truncate since we've already
4552 * reserved our space for our orphan item in the unlink, so we just
4553 * need to reserve some slack space in case we add bytes and update
4554 * inode item when doing the truncate.
4289a667 4555 */
8082510e 4556 while (1) {
08e007d2
MX		 4557 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4558 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4559
4560 /*
4561 * Try and steal from the global reserve since we will
4562 * likely not use this space anyway, we want to try as
4563 * hard as possible to get this to work.
4564 */
4565 if (ret)
4566 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4567
d68fc57b 4568 if (ret) {
c2cf52eb
SK		 4569 btrfs_warn(root->fs_info,
4570 "Could not get space for a delete, will truncate on mount %d",
4571 ret);
4289a667
JB		 4572 btrfs_orphan_del(NULL, inode);
4573 btrfs_free_block_rsv(root, rsv);
4574 goto no_delete;
d68fc57b 4575 }
7b128766 4576
0e8c36a9 4577 trans = btrfs_join_transaction(root);
4289a667
JB		 4578 if (IS_ERR(trans)) {
4579 btrfs_orphan_del(NULL, inode);
4580 btrfs_free_block_rsv(root, rsv);
4581 goto no_delete;
d68fc57b 4582 }
7b128766 4583
4289a667
JB		 4584 trans->block_rsv = rsv;
4585
d68fc57b 4586 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4587 if (ret != -ENOSPC)
8082510e 4588 break;
85e21bac 4589
8407aa46 4590 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4591 btrfs_end_transaction(trans, root);
4592 trans = NULL;
b53d3f5d 4593 btrfs_btree_balance_dirty(root);
8082510e 4594 }
5f39d397 4595
4289a667
JB		 4596 btrfs_free_block_rsv(root, rsv);
4597
8082510e 4598 if (ret == 0) {
4289a667 4599 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 4600 ret = btrfs_orphan_del(trans, inode);
4601 BUG_ON(ret);
4602 }
54aa1f4d 4603
4289a667 4604 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4605 if (!(root == root->fs_info->tree_root ||
4606 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4607 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4608
54aa1f4d 4609 btrfs_end_transaction(trans, root);
b53d3f5d 4610 btrfs_btree_balance_dirty(root);
39279cc3 4611no_delete:
89042e5a 4612 btrfs_remove_delayed_node(inode);
dbd5768f 4613 clear_inode(inode);
8082510e 4614 return;
39279cc3
CM		 4615}
4616
4617/*
4618 * this returns the key found in the dir entry in the location pointer.
4619 * If no dir entries were found, location->objectid is 0.
4620 */
4621static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4622 struct btrfs_key *location)
4623{
4624 const char *name = dentry->d_name.name;
4625 int namelen = dentry->d_name.len;
4626 struct btrfs_dir_item *di;
4627 struct btrfs_path *path;
4628 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4629 int ret = 0;
39279cc3
CM		 4630
4631 path = btrfs_alloc_path();
d8926bb3
MF		 4632 if (!path)
4633 return -ENOMEM;
3954401f 4634
33345d01 4635 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4636 namelen, 0);
0d9f7f3e
Y		 4637 if (IS_ERR(di))
4638 ret = PTR_ERR(di);
d397712b 4639
c704005d 4640 if (IS_ERR_OR_NULL(di))
3954401f 4641 goto out_err;
d397712b 4642
5f39d397 4643 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4644out:
39279cc3
CM		 4645 btrfs_free_path(path);
4646 return ret;
3954401f
CM		 4647out_err:
4648 location->objectid = 0;
4649 goto out;
39279cc3
CM		 4650}
4651
4652/*
4653 * when we hit a tree root in a directory, the btrfs part of the inode
4654 * needs to be changed to reflect the root directory of the tree root. This
4655 * is kind of like crossing a mount point.
4656 */
4657static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4658 struct inode *dir,
4659 struct dentry *dentry,
4660 struct btrfs_key *location,
4661 struct btrfs_root **sub_root)
39279cc3 4662{
4df27c4d
YZ		 4663 struct btrfs_path *path;
4664 struct btrfs_root *new_root;
4665 struct btrfs_root_ref *ref;
4666 struct extent_buffer *leaf;
4667 int ret;
4668 int err = 0;
39279cc3 4669
4df27c4d
YZ		 4670 path = btrfs_alloc_path();
4671 if (!path) {
4672 err = -ENOMEM;
4673 goto out;
4674 }
39279cc3 4675
4df27c4d
YZ		 4676 err = -ENOENT;
4677 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4678 BTRFS_I(dir)->root->root_key.objectid,
4679 location->objectid);
4680 if (ret) {
4681 if (ret < 0)
4682 err = ret;
4683 goto out;
4684 }
39279cc3 4685
4df27c4d
YZ		 4686 leaf = path->nodes[0];
4687 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4688 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4689 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4690 goto out;
39279cc3 4691
4df27c4d
YZ		 4692 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4693 (unsigned long)(ref + 1),
4694 dentry->d_name.len);
4695 if (ret)
4696 goto out;
4697
b3b4aa74 4698 btrfs_release_path(path);
4df27c4d
YZ		 4699
4700 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4701 if (IS_ERR(new_root)) {
4702 err = PTR_ERR(new_root);
4703 goto out;
4704 }
4705
4df27c4d
YZ		 4706 *sub_root = new_root;
4707 location->objectid = btrfs_root_dirid(&new_root->root_item);
4708 location->type = BTRFS_INODE_ITEM_KEY;
4709 location->offset = 0;
4710 err = 0;
4711out:
4712 btrfs_free_path(path);
4713 return err;
39279cc3
CM		 4714}
4715
5d4f98a2
YZ		 4716static void inode_tree_add(struct inode *inode)
4717{
4718 struct btrfs_root *root = BTRFS_I(inode)->root;
4719 struct btrfs_inode *entry;
03e860bd
FNP		 4720 struct rb_node **p;
4721 struct rb_node *parent;
33345d01 4722 u64 ino = btrfs_ino(inode);
5d4f98a2 4723
1d3382cb 4724 if (inode_unhashed(inode))
76dda93c 4725 return;
e1409cef
MX		 4726again:
4727 parent = NULL;
5d4f98a2 4728 spin_lock(&root->inode_lock);
e1409cef 4729 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 4730 while (*p) {
4731 parent = *p;
4732 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4733
33345d01 4734 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4735 p = &parent->rb_left;
33345d01 4736 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4737 p = &parent->rb_right;
5d4f98a2
YZ		 4738 else {
4739 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4740 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 4741 rb_erase(parent, &root->inode_tree);
4742 RB_CLEAR_NODE(parent);
4743 spin_unlock(&root->inode_lock);
4744 goto again;
5d4f98a2
YZ		 4745 }
4746 }
4747 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4748 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4749 spin_unlock(&root->inode_lock);
4750}
4751
4752static void inode_tree_del(struct inode *inode)
4753{
4754 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4755 int empty = 0;
5d4f98a2 4756
03e860bd 4757 spin_lock(&root->inode_lock);
5d4f98a2 4758 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4759 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4760 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4761 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4762 }
03e860bd 4763 spin_unlock(&root->inode_lock);
76dda93c 4764
0af3d00b
JB		 4765 /*
4766 * Free space cache has inodes in the tree root, but the tree root has a
4767 * root_refs of 0, so this could end up dropping the tree root as a
4768 * snapshot, so we need the extra !root->fs_info->tree_root check to
4769 * make sure we don't drop it.
4770 */
4771 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4772 root != root->fs_info->tree_root) {
76dda93c
YZ		 4773 synchronize_srcu(&root->fs_info->subvol_srcu);
4774 spin_lock(&root->inode_lock);
4775 empty = RB_EMPTY_ROOT(&root->inode_tree);
4776 spin_unlock(&root->inode_lock);
4777 if (empty)
4778 btrfs_add_dead_root(root);
4779 }
4780}
4781
143bede5 4782void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4783{
4784 struct rb_node *node;
4785 struct rb_node *prev;
4786 struct btrfs_inode *entry;
4787 struct inode *inode;
4788 u64 objectid = 0;
4789
4790 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4791
4792 spin_lock(&root->inode_lock);
4793again:
4794 node = root->inode_tree.rb_node;
4795 prev = NULL;
4796 while (node) {
4797 prev = node;
4798 entry = rb_entry(node, struct btrfs_inode, rb_node);
4799
33345d01 4800 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4801 node = node->rb_left;
33345d01 4802 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4803 node = node->rb_right;
4804 else
4805 break;
4806 }
4807 if (!node) {
4808 while (prev) {
4809 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4810 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4811 node = prev;
4812 break;
4813 }
4814 prev = rb_next(prev);
4815 }
4816 }
4817 while (node) {
4818 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4819 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4820 inode = igrab(&entry->vfs_inode);
4821 if (inode) {
4822 spin_unlock(&root->inode_lock);
4823 if (atomic_read(&inode->i_count) > 1)
4824 d_prune_aliases(inode);
4825 /*
45321ac5 4826 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4827 * the inode cache when its usage count
4828 * hits zero.
4829 */
4830 iput(inode);
4831 cond_resched();
4832 spin_lock(&root->inode_lock);
4833 goto again;
4834 }
4835
4836 if (cond_resched_lock(&root->inode_lock))
4837 goto again;
4838
4839 node = rb_next(node);
4840 }
4841 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4842}
4843
e02119d5
CM		 4844static int btrfs_init_locked_inode(struct inode *inode, void *p)
4845{
4846 struct btrfs_iget_args *args = p;
4847 inode->i_ino = args->ino;
e02119d5 4848 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4849 return 0;
4850}
4851
4852static int btrfs_find_actor(struct inode *inode, void *opaque)
4853{
4854 struct btrfs_iget_args *args = opaque;
33345d01 4855 return args->ino == btrfs_ino(inode) &&
d397712b 4856 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4857}
4858
5d4f98a2
YZ		 4859static struct inode *btrfs_iget_locked(struct super_block *s,
4860 u64 objectid,
4861 struct btrfs_root *root)
39279cc3
CM		 4862{
4863 struct inode *inode;
4864 struct btrfs_iget_args args;
4865 args.ino = objectid;
4866 args.root = root;
4867
4868 inode = iget5_locked(s, objectid, btrfs_find_actor,
4869 btrfs_init_locked_inode,
4870 (void *)&args);
4871 return inode;
4872}
4873
1a54ef8c
BR		 4874/* Get an inode object given its location and corresponding root.
4875 * Returns in *is_new if the inode was read from disk
4876 */
4877struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4878 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4879{
4880 struct inode *inode;
4881
4882 inode = btrfs_iget_locked(s, location->objectid, root);
4883 if (!inode)
5d4f98a2 4884 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4885
4886 if (inode->i_state & I_NEW) {
4887 BTRFS_I(inode)->root = root;
4888 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4889 btrfs_read_locked_inode(inode);
1748f843
MF		 4890 if (!is_bad_inode(inode)) {
4891 inode_tree_add(inode);
4892 unlock_new_inode(inode);
4893 if (new)
4894 *new = 1;
4895 } else {
e0b6d65b
ST		 4896 unlock_new_inode(inode);
4897 iput(inode);
4898 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4899 }
4900 }
4901
1a54ef8c
BR		 4902 return inode;
4903}
4904
4df27c4d
YZ		 4905static struct inode *new_simple_dir(struct super_block *s,
4906 struct btrfs_key *key,
4907 struct btrfs_root *root)
4908{
4909 struct inode *inode = new_inode(s);
4910
4911 if (!inode)
4912 return ERR_PTR(-ENOMEM);
4913
4df27c4d
YZ		 4914 BTRFS_I(inode)->root = root;
4915 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4916 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4917
4918 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4919 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4920 inode->i_fop = &simple_dir_operations;
4921 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4922 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4923
4924 return inode;
4925}
4926
3de4586c 4927struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4928{
d397712b 4929 struct inode *inode;
4df27c4d 4930 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4931 struct btrfs_root *sub_root = root;
4932 struct btrfs_key location;
76dda93c 4933 int index;
b4aff1f8 4934 int ret = 0;
39279cc3
CM		 4935
4936 if (dentry->d_name.len > BTRFS_NAME_LEN)
4937 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4938
39e3c955 4939 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 4940 if (ret < 0)
4941 return ERR_PTR(ret);
5f39d397 4942
4df27c4d
YZ		 4943 if (location.objectid == 0)
4944 return NULL;
4945
4946 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4947 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4948 return inode;
4949 }
4950
4951 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4952
76dda93c 4953 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4954 ret = fixup_tree_root_location(root, dir, dentry,
4955 &location, &sub_root);
4956 if (ret < 0) {
4957 if (ret != -ENOENT)
4958 inode = ERR_PTR(ret);
4959 else
4960 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4961 } else {
73f73415 4962 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4963 }
76dda93c
YZ		 4964 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4965
34d19bad 4966 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4967 down_read(&root->fs_info->cleanup_work_sem);
4968 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4969 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4970 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 4971 if (ret) {
4972 iput(inode);
66b4ffd1 4973 inode = ERR_PTR(ret);
01cd3367 4974 }
c71bf099
YZ		 4975 }
4976
3de4586c
CM		 4977 return inode;
4978}
4979
fe15ce44 4980static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4981{
4982 struct btrfs_root *root;
848cce0d 4983 struct inode *inode = dentry->d_inode;
76dda93c 4984
848cce0d
LZ		 4985 if (!inode && !IS_ROOT(dentry))
4986 inode = dentry->d_parent->d_inode;
76dda93c 4987
848cce0d
LZ		 4988 if (inode) {
4989 root = BTRFS_I(inode)->root;
efefb143
YZ		 4990 if (btrfs_root_refs(&root->root_item) == 0)
4991 return 1;
848cce0d
LZ		 4992
4993 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4994 return 1;
efefb143 4995 }
76dda93c
YZ		 4996 return 0;
4997}
4998
b4aff1f8
JB		 4999static void btrfs_dentry_release(struct dentry *dentry)
5000{
5001 if (dentry->d_fsdata)
5002 kfree(dentry->d_fsdata);
5003}
5004
3de4586c 5005static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5006 unsigned int flags)
3de4586c 5007{
a66e7cc6
JB		 5008 struct dentry *ret;
5009
5010 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 5011 return ret;
39279cc3
CM		 5012}
5013
16cdcec7 5014unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5015 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5016};
5017
cbdf5a24
DW		 5018static int btrfs_real_readdir(struct file *filp, void *dirent,
5019 filldir_t filldir)
39279cc3 5020{
496ad9aa 5021 struct inode *inode = file_inode(filp);
39279cc3
CM		 5022 struct btrfs_root *root = BTRFS_I(inode)->root;
5023 struct btrfs_item *item;
5024 struct btrfs_dir_item *di;
5025 struct btrfs_key key;
5f39d397 5026 struct btrfs_key found_key;
39279cc3 5027 struct btrfs_path *path;
16cdcec7
MX		 5028 struct list_head ins_list;
5029 struct list_head del_list;
39279cc3 5030 int ret;
5f39d397 5031 struct extent_buffer *leaf;
39279cc3 5032 int slot;
39279cc3
CM		 5033 unsigned char d_type;
5034 int over = 0;
5035 u32 di_cur;
5036 u32 di_total;
5037 u32 di_len;
5038 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5039 char tmp_name[32];
5040 char *name_ptr;
5041 int name_len;
16cdcec7 5042 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 5043
5044 /* FIXME, use a real flag for deciding about the key type */
5045 if (root->fs_info->tree_root == root)
5046 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5047
3954401f
CM		 5048 /* special case for "." */
5049 if (filp->f_pos == 0) {
3765fefa
HS		 5050 over = filldir(dirent, ".", 1,
5051 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 5052 if (over)
5053 return 0;
5054 filp->f_pos = 1;
5055 }
3954401f
CM		 5056 /* special case for .., just use the back ref */
5057 if (filp->f_pos == 1) {
5ecc7e5d 5058 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 5059 over = filldir(dirent, "..", 2,
3765fefa 5060 filp->f_pos, pino, DT_DIR);
3954401f 5061 if (over)
49593bfa 5062 return 0;
3954401f
CM		 5063 filp->f_pos = 2;
5064 }
49593bfa 5065 path = btrfs_alloc_path();
16cdcec7
MX		 5066 if (!path)
5067 return -ENOMEM;
ff5714cc 5068
026fd317 5069 path->reada = 1;
49593bfa 5070
16cdcec7
MX		 5071 if (key_type == BTRFS_DIR_INDEX_KEY) {
5072 INIT_LIST_HEAD(&ins_list);
5073 INIT_LIST_HEAD(&del_list);
5074 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5075 }
5076
39279cc3
CM		 5077 btrfs_set_key_type(&key, key_type);
5078 key.offset = filp->f_pos;
33345d01 5079 key.objectid = btrfs_ino(inode);
5f39d397 5080
39279cc3
CM		 5081 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5082 if (ret < 0)
5083 goto err;
49593bfa
DW		 5084
5085 while (1) {
5f39d397 5086 leaf = path->nodes[0];
39279cc3 5087 slot = path->slots[0];
b9e03af0
LZ		 5088 if (slot >= btrfs_header_nritems(leaf)) {
5089 ret = btrfs_next_leaf(root, path);
5090 if (ret < 0)
5091 goto err;
5092 else if (ret > 0)
5093 break;
5094 continue;
39279cc3 5095 }
3de4586c 5096
5f39d397
CM		 5097 item = btrfs_item_nr(leaf, slot);
5098 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5099
5100 if (found_key.objectid != key.objectid)
39279cc3 5101 break;
5f39d397 5102 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5103 break;
5f39d397 5104 if (found_key.offset < filp->f_pos)
b9e03af0 5105 goto next;
16cdcec7
MX		 5106 if (key_type == BTRFS_DIR_INDEX_KEY &&
5107 btrfs_should_delete_dir_index(&del_list,
5108 found_key.offset))
5109 goto next;
5f39d397
CM		 5110
5111 filp->f_pos = found_key.offset;
16cdcec7 5112 is_curr = 1;
49593bfa 5113
39279cc3
CM		 5114 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5115 di_cur = 0;
5f39d397 5116 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5117
5118 while (di_cur < di_total) {
5f39d397
CM		 5119 struct btrfs_key location;
5120
22a94d44
JB		 5121 if (verify_dir_item(root, leaf, di))
5122 break;
5123
5f39d397 5124 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5125 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5126 name_ptr = tmp_name;
5127 } else {
5128 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5129 if (!name_ptr) {
5130 ret = -ENOMEM;
5131 goto err;
5132 }
5f39d397
CM		 5133 }
5134 read_extent_buffer(leaf, name_ptr,
5135 (unsigned long)(di + 1), name_len);
5136
5137 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5138 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5139
fede766f 5140
3de4586c 5141 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5142 * skip it.
5143 *
5144 * In contrast to old kernels, we insert the snapshot's
5145 * dir item and dir index after it has been created, so
5146 * we won't find a reference to our own snapshot. We
5147 * still keep the following code for backward
5148 * compatibility.
3de4586c
CM		 5149 */
5150 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5151 location.objectid == root->root_key.objectid) {
5152 over = 0;
5153 goto skip;
5154 }
5f39d397 5155 over = filldir(dirent, name_ptr, name_len,
49593bfa 5156 found_key.offset, location.objectid,
39279cc3 5157 d_type);
5f39d397 5158
3de4586c 5159skip:
5f39d397
CM		 5160 if (name_ptr != tmp_name)
5161 kfree(name_ptr);
5162
39279cc3
CM		 5163 if (over)
5164 goto nopos;
5103e947 5165 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5166 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5167 di_cur += di_len;
5168 di = (struct btrfs_dir_item *)((char *)di + di_len);
5169 }
b9e03af0
LZ		 5170next:
5171 path->slots[0]++;
39279cc3 5172 }
49593bfa 5173
16cdcec7
MX		 5174 if (key_type == BTRFS_DIR_INDEX_KEY) {
5175 if (is_curr)
5176 filp->f_pos++;
5177 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5178 &ins_list);
5179 if (ret)
5180 goto nopos;
5181 }
5182
49593bfa 5183 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 5184 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 5185 /*
5186 * 32-bit glibc will use getdents64, but then strtol -
5187 * so the last number we can serve is this.
5188 */
5189 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 5190 else
5191 filp->f_pos++;
39279cc3
CM		 5192nopos:
5193 ret = 0;
5194err:
16cdcec7
MX		 5195 if (key_type == BTRFS_DIR_INDEX_KEY)
5196 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5197 btrfs_free_path(path);
39279cc3
CM		 5198 return ret;
5199}
5200
a9185b41 5201int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5202{
5203 struct btrfs_root *root = BTRFS_I(inode)->root;
5204 struct btrfs_trans_handle *trans;
5205 int ret = 0;
0af3d00b 5206 bool nolock = false;
39279cc3 5207
72ac3c0d 5208 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5209 return 0;
5210
83eea1f1 5211 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5212 nolock = true;
0af3d00b 5213
a9185b41 5214 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5215 if (nolock)
7a7eaa40 5216 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5217 else
7a7eaa40 5218 trans = btrfs_join_transaction(root);
3612b495
TI		 5219 if (IS_ERR(trans))
5220 return PTR_ERR(trans);
a698d075 5221 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5222 }
5223 return ret;
5224}
5225
5226/*
54aa1f4d 5227 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5228 * inode changes. But, it is most likely to find the inode in cache.
5229 * FIXME, needs more benchmarking...there are no reasons other than performance
5230 * to keep or drop this code.
5231 */
48a3b636 5232static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5233{
5234 struct btrfs_root *root = BTRFS_I(inode)->root;
5235 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5236 int ret;
5237
72ac3c0d 5238 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5239 return 0;
39279cc3 5240
7a7eaa40 5241 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5242 if (IS_ERR(trans))
5243 return PTR_ERR(trans);
8929ecfa
YZ		 5244
5245 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5246 if (ret && ret == -ENOSPC) {
5247 /* whoops, lets try again with the full transaction */
5248 btrfs_end_transaction(trans, root);
5249 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5250 if (IS_ERR(trans))
5251 return PTR_ERR(trans);
8929ecfa 5252
94b60442 5253 ret = btrfs_update_inode(trans, root, inode);
94b60442 5254 }
39279cc3 5255 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5256 if (BTRFS_I(inode)->delayed_node)
5257 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5258
5259 return ret;
5260}
5261
5262/*
5263 * This is a copy of file_update_time. We need this so we can return error on
5264 * ENOSPC for updating the inode in the case of file write and mmap writes.
5265 */
e41f941a
JB		 5266static int btrfs_update_time(struct inode *inode, struct timespec *now,
5267 int flags)
22c44fe6 5268{
2bc55652
AB		 5269 struct btrfs_root *root = BTRFS_I(inode)->root;
5270
5271 if (btrfs_root_readonly(root))
5272 return -EROFS;
5273
e41f941a 5274 if (flags & S_VERSION)
22c44fe6 5275 inode_inc_iversion(inode);
e41f941a
JB		 5276 if (flags & S_CTIME)
5277 inode->i_ctime = *now;
5278 if (flags & S_MTIME)
5279 inode->i_mtime = *now;
5280 if (flags & S_ATIME)
5281 inode->i_atime = *now;
5282 return btrfs_dirty_inode(inode);
39279cc3
CM		 5283}
5284
d352ac68
CM		 5285/*
5286 * find the highest existing sequence number in a directory
5287 * and then set the in-memory index_cnt variable to reflect
5288 * free sequence numbers
5289 */
aec7477b
JB		 5290static int btrfs_set_inode_index_count(struct inode *inode)
5291{
5292 struct btrfs_root *root = BTRFS_I(inode)->root;
5293 struct btrfs_key key, found_key;
5294 struct btrfs_path *path;
5295 struct extent_buffer *leaf;
5296 int ret;
5297
33345d01 5298 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5299 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5300 key.offset = (u64)-1;
5301
5302 path = btrfs_alloc_path();
5303 if (!path)
5304 return -ENOMEM;
5305
5306 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5307 if (ret < 0)
5308 goto out;
5309 /* FIXME: we should be able to handle this */
5310 if (ret == 0)
5311 goto out;
5312 ret = 0;
5313
5314 /*
5315 * MAGIC NUMBER EXPLANATION:
5316 * since we search a directory based on f_pos we have to start at 2
5317 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5318 * else has to start at 2
5319 */
5320 if (path->slots[0] == 0) {
5321 BTRFS_I(inode)->index_cnt = 2;
5322 goto out;
5323 }
5324
5325 path->slots[0]--;
5326
5327 leaf = path->nodes[0];
5328 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5329
33345d01 5330 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5331 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5332 BTRFS_I(inode)->index_cnt = 2;
5333 goto out;
5334 }
5335
5336 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5337out:
5338 btrfs_free_path(path);
5339 return ret;
5340}
5341
d352ac68
CM		 5342/*
5343 * helper to find a free sequence number in a given directory. This current
5344 * code is very simple, later versions will do smarter things in the btree
5345 */
3de4586c 5346int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5347{
5348 int ret = 0;
5349
5350 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5351 ret = btrfs_inode_delayed_dir_index_count(dir);
5352 if (ret) {
5353 ret = btrfs_set_inode_index_count(dir);
5354 if (ret)
5355 return ret;
5356 }
aec7477b
JB		 5357 }
5358
00e4e6b3 5359 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5360 BTRFS_I(dir)->index_cnt++;
5361
5362 return ret;
5363}
5364
39279cc3
CM		 5365static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5366 struct btrfs_root *root,
aec7477b 5367 struct inode *dir,
9c58309d 5368 const char *name, int name_len,
175a4eb7
AV		 5369 u64 ref_objectid, u64 objectid,
5370 umode_t mode, u64 *index)
39279cc3
CM		 5371{
5372 struct inode *inode;
5f39d397 5373 struct btrfs_inode_item *inode_item;
39279cc3 5374 struct btrfs_key *location;
5f39d397 5375 struct btrfs_path *path;
9c58309d
CM		 5376 struct btrfs_inode_ref *ref;
5377 struct btrfs_key key[2];
5378 u32 sizes[2];
5379 unsigned long ptr;
39279cc3
CM		 5380 int ret;
5381 int owner;
5382
5f39d397 5383 path = btrfs_alloc_path();
d8926bb3
MF		 5384 if (!path)
5385 return ERR_PTR(-ENOMEM);
5f39d397 5386
39279cc3 5387 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5388 if (!inode) {
5389 btrfs_free_path(path);
39279cc3 5390 return ERR_PTR(-ENOMEM);
8fb27640 5391 }
39279cc3 5392
581bb050
LZ		 5393 /*
5394 * we have to initialize this early, so we can reclaim the inode
5395 * number if we fail afterwards in this function.
5396 */
5397 inode->i_ino = objectid;
5398
aec7477b 5399 if (dir) {
1abe9b8a 5400 trace_btrfs_inode_request(dir);
5401
3de4586c 5402 ret = btrfs_set_inode_index(dir, index);
09771430 5403 if (ret) {
8fb27640 5404 btrfs_free_path(path);
09771430 5405 iput(inode);
aec7477b 5406 return ERR_PTR(ret);
09771430 5407 }
aec7477b
JB		 5408 }
5409 /*
5410 * index_cnt is ignored for everything but a dir,
5411 * btrfs_get_inode_index_count has an explanation for the magic
5412 * number
5413 */
5414 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5415 BTRFS_I(inode)->root = root;
e02119d5 5416 BTRFS_I(inode)->generation = trans->transid;
76195853 5417 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5418
5dc562c5
JB		 5419 /*
5420 * We could have gotten an inode number from somebody who was fsynced
5421 * and then removed in this same transaction, so let's just set full
5422 * sync since it will be a full sync anyway and this will blow away the
5423 * old info in the log.
5424 */
5425 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5426
569254b0 5427 if (S_ISDIR(mode))
39279cc3
CM		 5428 owner = 0;
5429 else
5430 owner = 1;
9c58309d
CM		 5431
5432 key[0].objectid = objectid;
5433 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5434 key[0].offset = 0;
5435
f186373f
MF		 5436 /*
5437 * Start new inodes with an inode_ref. This is slightly more
5438 * efficient for small numbers of hard links since they will
5439 * be packed into one item. Extended refs will kick in if we
5440 * add more hard links than can fit in the ref item.
5441 */
9c58309d
CM		 5442 key[1].objectid = objectid;
5443 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5444 key[1].offset = ref_objectid;
5445
5446 sizes[0] = sizeof(struct btrfs_inode_item);
5447 sizes[1] = name_len + sizeof(*ref);
5448
b9473439 5449 path->leave_spinning = 1;
9c58309d
CM		 5450 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5451 if (ret != 0)
5f39d397
CM		 5452 goto fail;
5453
ecc11fab 5454 inode_init_owner(inode, dir, mode);
a76a3cd4 5455 inode_set_bytes(inode, 0);
39279cc3 5456 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5457 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5458 struct btrfs_inode_item);
293f7e07
LZ		 5459 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5460 sizeof(*inode_item));
e02119d5 5461 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5462
5463 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5464 struct btrfs_inode_ref);
5465 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5466 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5467 ptr = (unsigned long)(ref + 1);
5468 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5469
5f39d397
CM		 5470 btrfs_mark_buffer_dirty(path->nodes[0]);
5471 btrfs_free_path(path);
5472
39279cc3
CM		 5473 location = &BTRFS_I(inode)->location;
5474 location->objectid = objectid;
39279cc3
CM		 5475 location->offset = 0;
5476 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5477
6cbff00f
CH		 5478 btrfs_inherit_iflags(inode, dir);
5479
569254b0 5480 if (S_ISREG(mode)) {
94272164
CM		 5481 if (btrfs_test_opt(root, NODATASUM))
5482 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5483 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5484 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5485 BTRFS_INODE_NODATASUM;
94272164
CM		 5486 }
5487
39279cc3 5488 insert_inode_hash(inode);
5d4f98a2 5489 inode_tree_add(inode);
1abe9b8a 5490
5491 trace_btrfs_inode_new(inode);
1973f0fa 5492 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5493
8ea05e3a
AB		 5494 btrfs_update_root_times(trans, root);
5495
39279cc3 5496 return inode;
5f39d397 5497fail:
aec7477b
JB		 5498 if (dir)
5499 BTRFS_I(dir)->index_cnt--;
5f39d397 5500 btrfs_free_path(path);
09771430 5501 iput(inode);
5f39d397 5502 return ERR_PTR(ret);
39279cc3
CM		 5503}
5504
5505static inline u8 btrfs_inode_type(struct inode *inode)
5506{
5507 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5508}
5509
d352ac68
CM		 5510/*
5511 * utility function to add 'inode' into 'parent_inode' with
5512 * a give name and a given sequence number.
5513 * if 'add_backref' is true, also insert a backref from the
5514 * inode to the parent directory.
5515 */
e02119d5
CM		 5516int btrfs_add_link(struct btrfs_trans_handle *trans,
5517 struct inode *parent_inode, struct inode *inode,
5518 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5519{
4df27c4d 5520 int ret = 0;
39279cc3 5521 struct btrfs_key key;
e02119d5 5522 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5523 u64 ino = btrfs_ino(inode);
5524 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5525
33345d01 5526 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5527 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5528 } else {
33345d01 5529 key.objectid = ino;
4df27c4d
YZ		 5530 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5531 key.offset = 0;
5532 }
5533
33345d01 5534 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5535 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5536 key.objectid, root->root_key.objectid,
33345d01 5537 parent_ino, index, name, name_len);
4df27c4d 5538 } else if (add_backref) {
33345d01
LZ		 5539 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5540 parent_ino, index);
4df27c4d 5541 }
39279cc3 5542
79787eaa
JM		 5543 /* Nothing to clean up yet */
5544 if (ret)
5545 return ret;
4df27c4d 5546
79787eaa
JM		 5547 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5548 parent_inode, &key,
5549 btrfs_inode_type(inode), index);
9c52057c 5550 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5551 goto fail_dir_item;
5552 else if (ret) {
5553 btrfs_abort_transaction(trans, root, ret);
5554 return ret;
39279cc3 5555 }
79787eaa
JM		 5556
5557 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5558 name_len * 2);
0c4d2d95 5559 inode_inc_iversion(parent_inode);
79787eaa
JM		 5560 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5561 ret = btrfs_update_inode(trans, root, parent_inode);
5562 if (ret)
5563 btrfs_abort_transaction(trans, root, ret);
39279cc3 5564 return ret;
fe66a05a
CM		 5565
5566fail_dir_item:
5567 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5568 u64 local_index;
5569 int err;
5570 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5571 key.objectid, root->root_key.objectid,
5572 parent_ino, &local_index, name, name_len);
5573
5574 } else if (add_backref) {
5575 u64 local_index;
5576 int err;
5577
5578 err = btrfs_del_inode_ref(trans, root, name, name_len,
5579 ino, parent_ino, &local_index);
5580 }
5581 return ret;
39279cc3
CM		 5582}
5583
5584static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5585 struct inode *dir, struct dentry *dentry,
5586 struct inode *inode, int backref, u64 index)
39279cc3 5587{
a1b075d2
JB		 5588 int err = btrfs_add_link(trans, dir, inode,
5589 dentry->d_name.name, dentry->d_name.len,
5590 backref, index);
39279cc3
CM		 5591 if (err > 0)
5592 err = -EEXIST;
5593 return err;
5594}
5595
618e21d5 5596static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5597 umode_t mode, dev_t rdev)
618e21d5
JB		 5598{
5599 struct btrfs_trans_handle *trans;
5600 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5601 struct inode *inode = NULL;
618e21d5
JB		 5602 int err;
5603 int drop_inode = 0;
5604 u64 objectid;
00e4e6b3 5605 u64 index = 0;
618e21d5
JB		 5606
5607 if (!new_valid_dev(rdev))
5608 return -EINVAL;
5609
9ed74f2d
JB		 5610 /*
5611 * 2 for inode item and ref
5612 * 2 for dir items
5613 * 1 for xattr if selinux is on
5614 */
a22285a6
YZ		 5615 trans = btrfs_start_transaction(root, 5);
5616 if (IS_ERR(trans))
5617 return PTR_ERR(trans);
1832a6d5 5618
581bb050
LZ		 5619 err = btrfs_find_free_ino(root, &objectid);
5620 if (err)
5621 goto out_unlock;
5622
aec7477b 5623 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5624 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5625 mode, &index);
7cf96da3
TI		 5626 if (IS_ERR(inode)) {
5627 err = PTR_ERR(inode);
618e21d5 5628 goto out_unlock;
7cf96da3 5629 }
618e21d5 5630
2a7dba39 5631 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5632 if (err) {
5633 drop_inode = 1;
5634 goto out_unlock;
5635 }
5636
ad19db71
CS		 5637 /*
5638 * If the active LSM wants to access the inode during
5639 * d_instantiate it needs these. Smack checks to see
5640 * if the filesystem supports xattrs by looking at the
5641 * ops vector.
5642 */
5643
5644 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5645 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5646 if (err)
5647 drop_inode = 1;
5648 else {
618e21d5 5649 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5650 btrfs_update_inode(trans, root, inode);
08c422c2 5651 d_instantiate(dentry, inode);
618e21d5 5652 }
618e21d5 5653out_unlock:
7ad85bb7 5654 btrfs_end_transaction(trans, root);
b53d3f5d 5655 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5656 if (drop_inode) {
5657 inode_dec_link_count(inode);
5658 iput(inode);
5659 }
618e21d5
JB		 5660 return err;
5661}
5662
39279cc3 5663static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5664 umode_t mode, bool excl)
39279cc3
CM		 5665{
5666 struct btrfs_trans_handle *trans;
5667 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5668 struct inode *inode = NULL;
43baa579 5669 int drop_inode_on_err = 0;
a22285a6 5670 int err;
39279cc3 5671 u64 objectid;
00e4e6b3 5672 u64 index = 0;
39279cc3 5673
9ed74f2d
JB		 5674 /*
5675 * 2 for inode item and ref
5676 * 2 for dir items
5677 * 1 for xattr if selinux is on
5678 */
a22285a6
YZ		 5679 trans = btrfs_start_transaction(root, 5);
5680 if (IS_ERR(trans))
5681 return PTR_ERR(trans);
9ed74f2d 5682
581bb050
LZ		 5683 err = btrfs_find_free_ino(root, &objectid);
5684 if (err)
5685 goto out_unlock;
5686
aec7477b 5687 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5688 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5689 mode, &index);
7cf96da3
TI		 5690 if (IS_ERR(inode)) {
5691 err = PTR_ERR(inode);
39279cc3 5692 goto out_unlock;
7cf96da3 5693 }
43baa579 5694 drop_inode_on_err = 1;
39279cc3 5695
2a7dba39 5696 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5697 if (err)
33268eaf 5698 goto out_unlock;
33268eaf 5699
9185aa58
FB		 5700 err = btrfs_update_inode(trans, root, inode);
5701 if (err)
5702 goto out_unlock;
5703
ad19db71
CS		 5704 /*
5705 * If the active LSM wants to access the inode during
5706 * d_instantiate it needs these. Smack checks to see
5707 * if the filesystem supports xattrs by looking at the
5708 * ops vector.
5709 */
5710 inode->i_fop = &btrfs_file_operations;
5711 inode->i_op = &btrfs_file_inode_operations;
5712
a1b075d2 5713 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5714 if (err)
43baa579
FB		 5715 goto out_unlock;
5716
5717 inode->i_mapping->a_ops = &btrfs_aops;
5718 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5719 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5720 d_instantiate(dentry, inode);
5721
39279cc3 5722out_unlock:
7ad85bb7 5723 btrfs_end_transaction(trans, root);
43baa579 5724 if (err && drop_inode_on_err) {
39279cc3
CM		 5725 inode_dec_link_count(inode);
5726 iput(inode);
5727 }
b53d3f5d 5728 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5729 return err;
5730}
5731
5732static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5733 struct dentry *dentry)
5734{
5735 struct btrfs_trans_handle *trans;
5736 struct btrfs_root *root = BTRFS_I(dir)->root;
5737 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5738 u64 index;
39279cc3
CM		 5739 int err;
5740 int drop_inode = 0;
5741
4a8be425
TH		 5742 /* do not allow sys_link's with other subvols of the same device */
5743 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5744 return -EXDEV;
4a8be425 5745
f186373f 5746 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5747 return -EMLINK;
4a8be425 5748
3de4586c 5749 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5750 if (err)
5751 goto fail;
5752
a22285a6 5753 /*
7e6b6465 5754 * 2 items for inode and inode ref
a22285a6 5755 * 2 items for dir items
7e6b6465 5756 * 1 item for parent inode
a22285a6 5757 */
7e6b6465 5758 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5759 if (IS_ERR(trans)) {
5760 err = PTR_ERR(trans);
5761 goto fail;
5762 }
5f39d397 5763
3153495d 5764 btrfs_inc_nlink(inode);
0c4d2d95 5765 inode_inc_iversion(inode);
3153495d 5766 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5767 ihold(inode);
e9976151 5768 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5769
a1b075d2 5770 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5771
a5719521 5772 if (err) {
54aa1f4d 5773 drop_inode = 1;
a5719521 5774 } else {
10d9f309 5775 struct dentry *parent = dentry->d_parent;
a5719521 5776 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5777 if (err)
5778 goto fail;
08c422c2 5779 d_instantiate(dentry, inode);
6a912213 5780 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5781 }
39279cc3 5782
7ad85bb7 5783 btrfs_end_transaction(trans, root);
1832a6d5 5784fail:
39279cc3
CM		 5785 if (drop_inode) {
5786 inode_dec_link_count(inode);
5787 iput(inode);
5788 }
b53d3f5d 5789 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5790 return err;
5791}
5792
18bb1db3 5793static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5794{
b9d86667 5795 struct inode *inode = NULL;
39279cc3
CM		 5796 struct btrfs_trans_handle *trans;
5797 struct btrfs_root *root = BTRFS_I(dir)->root;
5798 int err = 0;
5799 int drop_on_err = 0;
b9d86667 5800 u64 objectid = 0;
00e4e6b3 5801 u64 index = 0;
39279cc3 5802
9ed74f2d
JB		 5803 /*
5804 * 2 items for inode and ref
5805 * 2 items for dir items
5806 * 1 for xattr if selinux is on
5807 */
a22285a6
YZ		 5808 trans = btrfs_start_transaction(root, 5);
5809 if (IS_ERR(trans))
5810 return PTR_ERR(trans);
39279cc3 5811
581bb050
LZ		 5812 err = btrfs_find_free_ino(root, &objectid);
5813 if (err)
5814 goto out_fail;
5815
aec7477b 5816 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5817 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5818 S_IFDIR | mode, &index);
39279cc3
CM		 5819 if (IS_ERR(inode)) {
5820 err = PTR_ERR(inode);
5821 goto out_fail;
5822 }
5f39d397 5823
39279cc3 5824 drop_on_err = 1;
33268eaf 5825
2a7dba39 5826 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5827 if (err)
5828 goto out_fail;
5829
39279cc3
CM		 5830 inode->i_op = &btrfs_dir_inode_operations;
5831 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5832
dbe674a9 5833 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5834 err = btrfs_update_inode(trans, root, inode);
5835 if (err)
5836 goto out_fail;
5f39d397 5837
a1b075d2
JB		 5838 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5839 dentry->d_name.len, 0, index);
39279cc3
CM		 5840 if (err)
5841 goto out_fail;
5f39d397 5842
39279cc3
CM		 5843 d_instantiate(dentry, inode);
5844 drop_on_err = 0;
39279cc3
CM		 5845
5846out_fail:
7ad85bb7 5847 btrfs_end_transaction(trans, root);
39279cc3
CM		 5848 if (drop_on_err)
5849 iput(inode);
b53d3f5d 5850 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5851 return err;
5852}
5853
d352ac68
CM		 5854/* helper for btfs_get_extent. Given an existing extent in the tree,
5855 * and an extent that you want to insert, deal with overlap and insert
5856 * the new extent into the tree.
5857 */
3b951516
CM		 5858static int merge_extent_mapping(struct extent_map_tree *em_tree,
5859 struct extent_map *existing,
e6dcd2dc
CM		 5860 struct extent_map *em,
5861 u64 map_start, u64 map_len)
3b951516
CM		 5862{
5863 u64 start_diff;
3b951516 5864
e6dcd2dc
CM		 5865 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5866 start_diff = map_start - em->start;
5867 em->start = map_start;
5868 em->len = map_len;
c8b97818
CM		 5869 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5870 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5871 em->block_start += start_diff;
c8b97818
CM		 5872 em->block_len -= start_diff;
5873 }
09a2a8f9 5874 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 5875}
5876
c8b97818
CM		 5877static noinline int uncompress_inline(struct btrfs_path *path,
5878 struct inode *inode, struct page *page,
5879 size_t pg_offset, u64 extent_offset,
5880 struct btrfs_file_extent_item *item)
5881{
5882 int ret;
5883 struct extent_buffer *leaf = path->nodes[0];
5884 char *tmp;
5885 size_t max_size;
5886 unsigned long inline_size;
5887 unsigned long ptr;
261507a0 5888 int compress_type;
c8b97818
CM		 5889
5890 WARN_ON(pg_offset != 0);
261507a0 5891 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5892 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5893 inline_size = btrfs_file_extent_inline_item_len(leaf,
5894 btrfs_item_nr(leaf, path->slots[0]));
5895 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5896 if (!tmp)
5897 return -ENOMEM;
c8b97818
CM		 5898 ptr = btrfs_file_extent_inline_start(item);
5899
5900 read_extent_buffer(leaf, tmp, ptr, inline_size);
5901
5b050f04 5902 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5903 ret = btrfs_decompress(compress_type, tmp, page,
5904 extent_offset, inline_size, max_size);
c8b97818 5905 if (ret) {
7ac687d9 5906 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5907 unsigned long copy_size = min_t(u64,
5908 PAGE_CACHE_SIZE - pg_offset,
5909 max_size - extent_offset);
5910 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5911 kunmap_atomic(kaddr);
c8b97818
CM		 5912 }
5913 kfree(tmp);
5914 return 0;
5915}
5916
d352ac68
CM		 5917/*
5918 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5919 * the ugly parts come from merging extents from the disk with the in-ram
5920 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5921 * where the in-ram extents might be locked pending data=ordered completion.
5922 *
5923 * This also copies inline extents directly into the page.
5924 */
d397712b 5925
a52d9a80 5926struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5927 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5928 int create)
5929{
5930 int ret;
5931 int err = 0;
db94535d 5932 u64 bytenr;
a52d9a80
CM		 5933 u64 extent_start = 0;
5934 u64 extent_end = 0;
33345d01 5935 u64 objectid = btrfs_ino(inode);
a52d9a80 5936 u32 found_type;
f421950f 5937 struct btrfs_path *path = NULL;
a52d9a80
CM		 5938 struct btrfs_root *root = BTRFS_I(inode)->root;
5939 struct btrfs_file_extent_item *item;
5f39d397
CM		 5940 struct extent_buffer *leaf;
5941 struct btrfs_key found_key;
a52d9a80
CM		 5942 struct extent_map *em = NULL;
5943 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5944 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5945 struct btrfs_trans_handle *trans = NULL;
261507a0 5946 int compress_type;
a52d9a80 5947
a52d9a80 5948again:
890871be 5949 read_lock(&em_tree->lock);
d1310b2e 5950 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5951 if (em)
5952 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5953 read_unlock(&em_tree->lock);
d1310b2e 5954
a52d9a80 5955 if (em) {
e1c4b745
CM		 5956 if (em->start > start || em->start + em->len <= start)
5957 free_extent_map(em);
5958 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5959 free_extent_map(em);
5960 else
5961 goto out;
a52d9a80 5962 }
172ddd60 5963 em = alloc_extent_map();
a52d9a80 5964 if (!em) {
d1310b2e
CM		 5965 err = -ENOMEM;
5966 goto out;
a52d9a80 5967 }
e6dcd2dc 5968 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5969 em->start = EXTENT_MAP_HOLE;
445a6944 5970 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5971 em->len = (u64)-1;
c8b97818 5972 em->block_len = (u64)-1;
f421950f
CM		 5973
5974 if (!path) {
5975 path = btrfs_alloc_path();
026fd317
JB		 5976 if (!path) {
5977 err = -ENOMEM;
5978 goto out;
5979 }
5980 /*
5981 * Chances are we'll be called again, so go ahead and do
5982 * readahead
5983 */
5984 path->reada = 1;
f421950f
CM		 5985 }
5986
179e29e4
CM		 5987 ret = btrfs_lookup_file_extent(trans, root, path,
5988 objectid, start, trans != NULL);
a52d9a80
CM		 5989 if (ret < 0) {
5990 err = ret;
5991 goto out;
5992 }
5993
5994 if (ret != 0) {
5995 if (path->slots[0] == 0)
5996 goto not_found;
5997 path->slots[0]--;
5998 }
5999
5f39d397
CM		 6000 leaf = path->nodes[0];
6001 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6002 struct btrfs_file_extent_item);
a52d9a80 6003 /* are we inside the extent that was found? */
5f39d397
CM		 6004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6005 found_type = btrfs_key_type(&found_key);
6006 if (found_key.objectid != objectid ||
a52d9a80
CM		 6007 found_type != BTRFS_EXTENT_DATA_KEY) {
6008 goto not_found;
6009 }
6010
5f39d397
CM		 6011 found_type = btrfs_file_extent_type(leaf, item);
6012 extent_start = found_key.offset;
261507a0 6013 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 6014 if (found_type == BTRFS_FILE_EXTENT_REG ||
6015 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6016 extent_end = extent_start +
db94535d 6017 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6018 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6019 size_t size;
6020 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 6021 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102
YZ		 6022 }
6023
6024 if (start >= extent_end) {
6025 path->slots[0]++;
6026 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6027 ret = btrfs_next_leaf(root, path);
6028 if (ret < 0) {
6029 err = ret;
6030 goto out;
a52d9a80 6031 }
9036c102
YZ		 6032 if (ret > 0)
6033 goto not_found;
6034 leaf = path->nodes[0];
a52d9a80 6035 }
9036c102
YZ		 6036 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6037 if (found_key.objectid != objectid ||
6038 found_key.type != BTRFS_EXTENT_DATA_KEY)
6039 goto not_found;
6040 if (start + len <= found_key.offset)
6041 goto not_found;
6042 em->start = start;
70c8a91c 6043 em->orig_start = start;
9036c102
YZ		 6044 em->len = found_key.offset - start;
6045 goto not_found_em;
6046 }
6047
cc95bef6 6048 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6049 if (found_type == BTRFS_FILE_EXTENT_REG ||
6050 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6051 em->start = extent_start;
6052 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6053 em->orig_start = extent_start -
6054 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6055 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6056 item);
db94535d
CM		 6057 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6058 if (bytenr == 0) {
5f39d397 6059 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6060 goto insert;
6061 }
261507a0 6062 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6063 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6064 em->compress_type = compress_type;
c8b97818 6065 em->block_start = bytenr;
b4939680 6066 em->block_len = em->orig_block_len;
c8b97818
CM		 6067 } else {
6068 bytenr += btrfs_file_extent_offset(leaf, item);
6069 em->block_start = bytenr;
6070 em->block_len = em->len;
d899e052
YZ		 6071 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6072 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6073 }
a52d9a80
CM		 6074 goto insert;
6075 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6076 unsigned long ptr;
a52d9a80 6077 char *map;
3326d1b0
CM		 6078 size_t size;
6079 size_t extent_offset;
6080 size_t copy_size;
a52d9a80 6081
689f9346 6082 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6083 if (!page || create) {
689f9346 6084 em->start = extent_start;
9036c102 6085 em->len = extent_end - extent_start;
689f9346
Y		 6086 goto out;
6087 }
5f39d397 6088
9036c102
YZ		 6089 size = btrfs_file_extent_inline_len(leaf, item);
6090 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6091 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6092 size - extent_offset);
3326d1b0 6093 em->start = extent_start + extent_offset;
fda2832f 6094 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6095 em->orig_block_len = em->len;
70c8a91c 6096 em->orig_start = em->start;
261507a0 6097 if (compress_type) {
c8b97818 6098 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6099 em->compress_type = compress_type;
6100 }
689f9346 6101 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6102 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6103 if (btrfs_file_extent_compression(leaf, item) !=
6104 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6105 ret = uncompress_inline(path, inode, page,
6106 pg_offset,
6107 extent_offset, item);
79787eaa 6108 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6109 } else {
6110 map = kmap(page);
6111 read_extent_buffer(leaf, map + pg_offset, ptr,
6112 copy_size);
93c82d57
CM		 6113 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6114 memset(map + pg_offset + copy_size, 0,
6115 PAGE_CACHE_SIZE - pg_offset -
6116 copy_size);
6117 }
c8b97818
CM		 6118 kunmap(page);
6119 }
179e29e4
CM		 6120 flush_dcache_page(page);
6121 } else if (create && PageUptodate(page)) {
6bf7e080 6122 BUG();
179e29e4
CM		 6123 if (!trans) {
6124 kunmap(page);
6125 free_extent_map(em);
6126 em = NULL;
ff5714cc 6127
b3b4aa74 6128 btrfs_release_path(path);
7a7eaa40 6129 trans = btrfs_join_transaction(root);
ff5714cc 6130
3612b495
TI		 6131 if (IS_ERR(trans))
6132 return ERR_CAST(trans);
179e29e4
CM		 6133 goto again;
6134 }
c8b97818 6135 map = kmap(page);
70dec807 6136 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6137 copy_size);
c8b97818 6138 kunmap(page);
179e29e4 6139 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6140 }
d1310b2e 6141 set_extent_uptodate(io_tree, em->start,
507903b8 6142 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6143 goto insert;
6144 } else {
31b1a2bd 6145 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6146 }
6147not_found:
6148 em->start = start;
70c8a91c 6149 em->orig_start = start;
d1310b2e 6150 em->len = len;
a52d9a80 6151not_found_em:
5f39d397 6152 em->block_start = EXTENT_MAP_HOLE;
9036c102 6153 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6154insert:
b3b4aa74 6155 btrfs_release_path(path);
d1310b2e 6156 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb
SK		 6157 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6158 (unsigned long long)em->start,
6159 (unsigned long long)em->len,
6160 (unsigned long long)start,
6161 (unsigned long long)len);
a52d9a80
CM		 6162 err = -EIO;
6163 goto out;
6164 }
d1310b2e
CM		 6165
6166 err = 0;
890871be 6167 write_lock(&em_tree->lock);
09a2a8f9 6168 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6169 /* it is possible that someone inserted the extent into the tree
6170 * while we had the lock dropped. It is also possible that
6171 * an overlapping map exists in the tree
6172 */
a52d9a80 6173 if (ret == -EEXIST) {
3b951516 6174 struct extent_map *existing;
e6dcd2dc
CM		 6175
6176 ret = 0;
6177
3b951516 6178 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6179 if (existing && (existing->start > start ||
6180 existing->start + existing->len <= start)) {
6181 free_extent_map(existing);
6182 existing = NULL;
6183 }
3b951516
CM		 6184 if (!existing) {
6185 existing = lookup_extent_mapping(em_tree, em->start,
6186 em->len);
6187 if (existing) {
6188 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6189 em, start,
6190 root->sectorsize);
3b951516
CM		 6191 free_extent_map(existing);
6192 if (err) {
6193 free_extent_map(em);
6194 em = NULL;
6195 }
6196 } else {
6197 err = -EIO;
3b951516
CM		 6198 free_extent_map(em);
6199 em = NULL;
6200 }
6201 } else {
6202 free_extent_map(em);
6203 em = existing;
e6dcd2dc 6204 err = 0;
a52d9a80 6205 }
a52d9a80 6206 }
890871be 6207 write_unlock(&em_tree->lock);
a52d9a80 6208out:
1abe9b8a 6209
f0bd95ea
TI		 6210 if (em)
6211 trace_btrfs_get_extent(root, em);
1abe9b8a 6212
f421950f
CM		 6213 if (path)
6214 btrfs_free_path(path);
a52d9a80
CM		 6215 if (trans) {
6216 ret = btrfs_end_transaction(trans, root);
d397712b 6217 if (!err)
a52d9a80
CM		 6218 err = ret;
6219 }
a52d9a80
CM		 6220 if (err) {
6221 free_extent_map(em);
a52d9a80
CM		 6222 return ERR_PTR(err);
6223 }
79787eaa 6224 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6225 return em;
6226}
6227
ec29ed5b
CM		 6228struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6229 size_t pg_offset, u64 start, u64 len,
6230 int create)
6231{
6232 struct extent_map *em;
6233 struct extent_map *hole_em = NULL;
6234 u64 range_start = start;
6235 u64 end;
6236 u64 found;
6237 u64 found_end;
6238 int err = 0;
6239
6240 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6241 if (IS_ERR(em))
6242 return em;
6243 if (em) {
6244 /*
f9e4fb53
LB		 6245 * if our em maps to
6246 * - a hole or
6247 * - a pre-alloc extent,
6248 * there might actually be delalloc bytes behind it.
ec29ed5b 6249 */
f9e4fb53
LB		 6250 if (em->block_start != EXTENT_MAP_HOLE &&
6251 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6252 return em;
6253 else
6254 hole_em = em;
6255 }
6256
6257 /* check to see if we've wrapped (len == -1 or similar) */
6258 end = start + len;
6259 if (end < start)
6260 end = (u64)-1;
6261 else
6262 end -= 1;
6263
6264 em = NULL;
6265
6266 /* ok, we didn't find anything, lets look for delalloc */
6267 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6268 end, len, EXTENT_DELALLOC, 1);
6269 found_end = range_start + found;
6270 if (found_end < range_start)
6271 found_end = (u64)-1;
6272
6273 /*
6274 * we didn't find anything useful, return
6275 * the original results from get_extent()
6276 */
6277 if (range_start > end || found_end <= start) {
6278 em = hole_em;
6279 hole_em = NULL;
6280 goto out;
6281 }
6282
6283 /* adjust the range_start to make sure it doesn't
6284 * go backwards from the start they passed in
6285 */
6286 range_start = max(start,range_start);
6287 found = found_end - range_start;
6288
6289 if (found > 0) {
6290 u64 hole_start = start;
6291 u64 hole_len = len;
6292
172ddd60 6293 em = alloc_extent_map();
ec29ed5b
CM		 6294 if (!em) {
6295 err = -ENOMEM;
6296 goto out;
6297 }
6298 /*
6299 * when btrfs_get_extent can't find anything it
6300 * returns one huge hole
6301 *
6302 * make sure what it found really fits our range, and
6303 * adjust to make sure it is based on the start from
6304 * the caller
6305 */
6306 if (hole_em) {
6307 u64 calc_end = extent_map_end(hole_em);
6308
6309 if (calc_end <= start || (hole_em->start > end)) {
6310 free_extent_map(hole_em);
6311 hole_em = NULL;
6312 } else {
6313 hole_start = max(hole_em->start, start);
6314 hole_len = calc_end - hole_start;
6315 }
6316 }
6317 em->bdev = NULL;
6318 if (hole_em && range_start > hole_start) {
6319 /* our hole starts before our delalloc, so we
6320 * have to return just the parts of the hole
6321 * that go until the delalloc starts
6322 */
6323 em->len = min(hole_len,
6324 range_start - hole_start);
6325 em->start = hole_start;
6326 em->orig_start = hole_start;
6327 /*
6328 * don't adjust block start at all,
6329 * it is fixed at EXTENT_MAP_HOLE
6330 */
6331 em->block_start = hole_em->block_start;
6332 em->block_len = hole_len;
f9e4fb53
LB		 6333 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6334 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6335 } else {
6336 em->start = range_start;
6337 em->len = found;
6338 em->orig_start = range_start;
6339 em->block_start = EXTENT_MAP_DELALLOC;
6340 em->block_len = found;
6341 }
6342 } else if (hole_em) {
6343 return hole_em;
6344 }
6345out:
6346
6347 free_extent_map(hole_em);
6348 if (err) {
6349 free_extent_map(em);
6350 return ERR_PTR(err);
6351 }
6352 return em;
6353}
6354
4b46fce2
JB		 6355static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6356 u64 start, u64 len)
6357{
6358 struct btrfs_root *root = BTRFS_I(inode)->root;
6359 struct btrfs_trans_handle *trans;
70c8a91c 6360 struct extent_map *em;
4b46fce2
JB		 6361 struct btrfs_key ins;
6362 u64 alloc_hint;
6363 int ret;
4b46fce2 6364
7a7eaa40 6365 trans = btrfs_join_transaction(root);
3612b495
TI		 6366 if (IS_ERR(trans))
6367 return ERR_CAST(trans);
4b46fce2
JB		 6368
6369 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6370
6371 alloc_hint = get_extent_allocation_hint(inode, start, len);
6372 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 6373 alloc_hint, &ins, 1);
4b46fce2
JB		 6374 if (ret) {
6375 em = ERR_PTR(ret);
6376 goto out;
6377 }
6378
70c8a91c 6379 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6380 ins.offset, ins.offset, ins.offset, 0);
70c8a91c
JB		 6381 if (IS_ERR(em))
6382 goto out;
4b46fce2
JB		 6383
6384 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6385 ins.offset, ins.offset, 0);
6386 if (ret) {
6387 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6388 em = ERR_PTR(ret);
6389 }
6390out:
6391 btrfs_end_transaction(trans, root);
6392 return em;
6393}
6394
46bfbb5c
CM		 6395/*
6396 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6397 * block must be cow'd
6398 */
6399static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
eb384b55
JB		 6400 struct inode *inode, u64 offset, u64 *len,
6401 u64 *orig_start, u64 *orig_block_len,
6402 u64 *ram_bytes)
46bfbb5c
CM		 6403{
6404 struct btrfs_path *path;
6405 int ret;
6406 struct extent_buffer *leaf;
6407 struct btrfs_root *root = BTRFS_I(inode)->root;
6408 struct btrfs_file_extent_item *fi;
6409 struct btrfs_key key;
6410 u64 disk_bytenr;
6411 u64 backref_offset;
6412 u64 extent_end;
6413 u64 num_bytes;
6414 int slot;
6415 int found_type;
6416
6417 path = btrfs_alloc_path();
6418 if (!path)
6419 return -ENOMEM;
6420
33345d01 6421 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6422 offset, 0);
6423 if (ret < 0)
6424 goto out;
6425
6426 slot = path->slots[0];
6427 if (ret == 1) {
6428 if (slot == 0) {
6429 /* can't find the item, must cow */
6430 ret = 0;
6431 goto out;
6432 }
6433 slot--;
6434 }
6435 ret = 0;
6436 leaf = path->nodes[0];
6437 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6438 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6439 key.type != BTRFS_EXTENT_DATA_KEY) {
6440 /* not our file or wrong item type, must cow */
6441 goto out;
6442 }
6443
6444 if (key.offset > offset) {
6445 /* Wrong offset, must cow */
6446 goto out;
6447 }
6448
6449 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6450 found_type = btrfs_file_extent_type(leaf, fi);
6451 if (found_type != BTRFS_FILE_EXTENT_REG &&
6452 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6453 /* not a regular extent, must cow */
6454 goto out;
6455 }
6456 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6457 backref_offset = btrfs_file_extent_offset(leaf, fi);
6458
eb384b55
JB		 6459 *orig_start = key.offset - backref_offset;
6460 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6461 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6462
46bfbb5c 6463 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
eb384b55 6464 if (extent_end < offset + *len) {
46bfbb5c
CM		 6465 /* extent doesn't include our full range, must cow */
6466 goto out;
6467 }
6468
6469 if (btrfs_extent_readonly(root, disk_bytenr))
6470 goto out;
6471
6472 /*
6473 * look for other files referencing this extent, if we
6474 * find any we must cow
6475 */
33345d01 6476 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 6477 key.offset - backref_offset, disk_bytenr))
6478 goto out;
6479
6480 /*
6481 * adjust disk_bytenr and num_bytes to cover just the bytes
6482 * in this extent we are about to write. If there
6483 * are any csums in that range we have to cow in order
6484 * to keep the csums correct
6485 */
6486 disk_bytenr += backref_offset;
6487 disk_bytenr += offset - key.offset;
eb384b55 6488 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6489 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6490 goto out;
6491 /*
6492 * all of the above have passed, it is safe to overwrite this extent
6493 * without cow
6494 */
eb384b55 6495 *len = num_bytes;
46bfbb5c
CM		 6496 ret = 1;
6497out:
6498 btrfs_free_path(path);
6499 return ret;
6500}
6501
eb838e73
JB		 6502static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6503 struct extent_state **cached_state, int writing)
6504{
6505 struct btrfs_ordered_extent *ordered;
6506 int ret = 0;
6507
6508 while (1) {
6509 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6510 0, cached_state);
6511 /*
6512 * We're concerned with the entire range that we're going to be
6513 * doing DIO to, so we need to make sure theres no ordered
6514 * extents in this range.
6515 */
6516 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6517 lockend - lockstart + 1);
6518
6519 /*
6520 * We need to make sure there are no buffered pages in this
6521 * range either, we could have raced between the invalidate in
6522 * generic_file_direct_write and locking the extent. The
6523 * invalidate needs to happen so that reads after a write do not
6524 * get stale data.
6525 */
6526 if (!ordered && (!writing ||
6527 !test_range_bit(&BTRFS_I(inode)->io_tree,
6528 lockstart, lockend, EXTENT_UPTODATE, 0,
6529 *cached_state)))
6530 break;
6531
6532 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6533 cached_state, GFP_NOFS);
6534
6535 if (ordered) {
6536 btrfs_start_ordered_extent(inode, ordered, 1);
6537 btrfs_put_ordered_extent(ordered);
6538 } else {
6539 /* Screw you mmap */
6540 ret = filemap_write_and_wait_range(inode->i_mapping,
6541 lockstart,
6542 lockend);
6543 if (ret)
6544 break;
6545
6546 /*
6547 * If we found a page that couldn't be invalidated just
6548 * fall back to buffered.
6549 */
6550 ret = invalidate_inode_pages2_range(inode->i_mapping,
6551 lockstart >> PAGE_CACHE_SHIFT,
6552 lockend >> PAGE_CACHE_SHIFT);
6553 if (ret)
6554 break;
6555 }
6556
6557 cond_resched();
6558 }
6559
6560 return ret;
6561}
6562
69ffb543
JB		 6563static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6564 u64 len, u64 orig_start,
6565 u64 block_start, u64 block_len,
cc95bef6
JB		 6566 u64 orig_block_len, u64 ram_bytes,
6567 int type)
69ffb543
JB		 6568{
6569 struct extent_map_tree *em_tree;
6570 struct extent_map *em;
6571 struct btrfs_root *root = BTRFS_I(inode)->root;
6572 int ret;
6573
6574 em_tree = &BTRFS_I(inode)->extent_tree;
6575 em = alloc_extent_map();
6576 if (!em)
6577 return ERR_PTR(-ENOMEM);
6578
6579 em->start = start;
6580 em->orig_start = orig_start;
2ab28f32
JB		 6581 em->mod_start = start;
6582 em->mod_len = len;
69ffb543
JB		 6583 em->len = len;
6584 em->block_len = block_len;
6585 em->block_start = block_start;
6586 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6587 em->orig_block_len = orig_block_len;
cc95bef6 6588 em->ram_bytes = ram_bytes;
70c8a91c 6589 em->generation = -1;
69ffb543
JB		 6590 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6591 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6592 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6593
6594 do {
6595 btrfs_drop_extent_cache(inode, em->start,
6596 em->start + em->len - 1, 0);
6597 write_lock(&em_tree->lock);
09a2a8f9 6598 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6599 write_unlock(&em_tree->lock);
6600 } while (ret == -EEXIST);
6601
6602 if (ret) {
6603 free_extent_map(em);
6604 return ERR_PTR(ret);
6605 }
6606
6607 return em;
6608}
6609
6610
4b46fce2
JB		 6611static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6612 struct buffer_head *bh_result, int create)
6613{
6614 struct extent_map *em;
6615 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6616 struct extent_state *cached_state = NULL;
4b46fce2 6617 u64 start = iblock << inode->i_blkbits;
eb838e73 6618 u64 lockstart, lockend;
4b46fce2 6619 u64 len = bh_result->b_size;
46bfbb5c 6620 struct btrfs_trans_handle *trans;
eb838e73 6621 int unlock_bits = EXTENT_LOCKED;
0934856d 6622 int ret = 0;
eb838e73 6623
172a5049 6624 if (create)
eb838e73 6625 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6626 else
c329861d 6627 len = min_t(u64, len, root->sectorsize);
eb838e73 6628
c329861d
JB		 6629 lockstart = start;
6630 lockend = start + len - 1;
6631
eb838e73
JB		 6632 /*
6633 * If this errors out it's because we couldn't invalidate pagecache for
6634 * this range and we need to fallback to buffered.
6635 */
6636 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6637 return -ENOTBLK;
6638
4b46fce2 6639 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6640 if (IS_ERR(em)) {
6641 ret = PTR_ERR(em);
6642 goto unlock_err;
6643 }
4b46fce2
JB		 6644
6645 /*
6646 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6647 * io. INLINE is special, and we could probably kludge it in here, but
6648 * it's still buffered so for safety lets just fall back to the generic
6649 * buffered path.
6650 *
6651 * For COMPRESSED we _have_ to read the entire extent in so we can
6652 * decompress it, so there will be buffering required no matter what we
6653 * do, so go ahead and fallback to buffered.
6654 *
6655 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6656 * to buffered IO. Don't blame me, this is the price we pay for using
6657 * the generic code.
6658 */
6659 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6660 em->block_start == EXTENT_MAP_INLINE) {
6661 free_extent_map(em);
eb838e73
JB		 6662 ret = -ENOTBLK;
6663 goto unlock_err;
4b46fce2
JB		 6664 }
6665
6666 /* Just a good old fashioned hole, return */
6667 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6668 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6669 free_extent_map(em);
eb838e73 6670 goto unlock_err;
4b46fce2
JB		 6671 }
6672
6673 /*
6674 * We don't allocate a new extent in the following cases
6675 *
6676 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6677 * existing extent.
6678 * 2) The extent is marked as PREALLOC. We're good to go here and can
6679 * just use the extent.
6680 *
6681 */
46bfbb5c 6682 if (!create) {
eb838e73
JB		 6683 len = min(len, em->len - (start - em->start));
6684 lockstart = start + len;
6685 goto unlock;
46bfbb5c 6686 }
4b46fce2
JB		 6687
6688 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6689 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6690 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6691 int type;
6692 int ret;
eb384b55 6693 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6694
6695 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6696 type = BTRFS_ORDERED_PREALLOC;
6697 else
6698 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6699 len = min(len, em->len - (start - em->start));
4b46fce2 6700 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 6701
6702 /*
6703 * we're not going to log anything, but we do need
6704 * to make sure the current transaction stays open
6705 * while we look for nocow cross refs
6706 */
7a7eaa40 6707 trans = btrfs_join_transaction(root);
3612b495 6708 if (IS_ERR(trans))
46bfbb5c
CM		 6709 goto must_cow;
6710
eb384b55
JB		 6711 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6712 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6713 if (type == BTRFS_ORDERED_PREALLOC) {
6714 free_extent_map(em);
6715 em = create_pinned_em(inode, start, len,
6716 orig_start,
b4939680 6717 block_start, len,
cc95bef6
JB		 6718 orig_block_len,
6719 ram_bytes, type);
69ffb543
JB		 6720 if (IS_ERR(em)) {
6721 btrfs_end_transaction(trans, root);
6722 goto unlock_err;
6723 }
6724 }
6725
46bfbb5c
CM		 6726 ret = btrfs_add_ordered_extent_dio(inode, start,
6727 block_start, len, len, type);
6728 btrfs_end_transaction(trans, root);
6729 if (ret) {
6730 free_extent_map(em);
eb838e73 6731 goto unlock_err;
46bfbb5c
CM		 6732 }
6733 goto unlock;
4b46fce2 6734 }
46bfbb5c 6735 btrfs_end_transaction(trans, root);
4b46fce2 6736 }
46bfbb5c
CM		 6737must_cow:
6738 /*
6739 * this will cow the extent, reset the len in case we changed
6740 * it above
6741 */
6742 len = bh_result->b_size;
70c8a91c
JB		 6743 free_extent_map(em);
6744 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6745 if (IS_ERR(em)) {
6746 ret = PTR_ERR(em);
6747 goto unlock_err;
6748 }
46bfbb5c
CM		 6749 len = min(len, em->len - (start - em->start));
6750unlock:
4b46fce2
JB		 6751 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6752 inode->i_blkbits;
46bfbb5c 6753 bh_result->b_size = len;
4b46fce2
JB		 6754 bh_result->b_bdev = em->bdev;
6755 set_buffer_mapped(bh_result);
c3473e83
JB		 6756 if (create) {
6757 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6758 set_buffer_new(bh_result);
6759
6760 /*
6761 * Need to update the i_size under the extent lock so buffered
6762 * readers will get the updated i_size when we unlock.
6763 */
6764 if (start + len > i_size_read(inode))
6765 i_size_write(inode, start + len);
0934856d 6766
172a5049
MX		 6767 spin_lock(&BTRFS_I(inode)->lock);
6768 BTRFS_I(inode)->outstanding_extents++;
6769 spin_unlock(&BTRFS_I(inode)->lock);
6770
0934856d
MX		 6771 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6772 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6773 &cached_state, GFP_NOFS);
6774 BUG_ON(ret);
c3473e83 6775 }
4b46fce2 6776
eb838e73
JB		 6777 /*
6778 * In the case of write we need to clear and unlock the entire range,
6779 * in the case of read we need to unlock only the end area that we
6780 * aren't using if there is any left over space.
6781 */
24c03fa5 6782 if (lockstart < lockend) {
0934856d
MX		 6783 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6784 lockend, unlock_bits, 1, 0,
6785 &cached_state, GFP_NOFS);
24c03fa5 6786 } else {
eb838e73 6787 free_extent_state(cached_state);
24c03fa5 6788 }
eb838e73 6789
4b46fce2
JB		 6790 free_extent_map(em);
6791
6792 return 0;
eb838e73
JB		 6793
6794unlock_err:
eb838e73
JB		 6795 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6796 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6797 return ret;
4b46fce2
JB		 6798}
6799
6800struct btrfs_dio_private {
6801 struct inode *inode;
6802 u64 logical_offset;
6803 u64 disk_bytenr;
6804 u64 bytes;
4b46fce2 6805 void *private;
e65e1535
MX		 6806
6807 /* number of bios pending for this dio */
6808 atomic_t pending_bios;
6809
6810 /* IO errors */
6811 int errors;
6812
9be3395b 6813 /* orig_bio is our btrfs_io_bio */
e65e1535 6814 struct bio *orig_bio;
9be3395b
CM		 6815
6816 /* dio_bio came from fs/direct-io.c */
6817 struct bio *dio_bio;
4b46fce2
JB		 6818};
6819
6820static void btrfs_endio_direct_read(struct bio *bio, int err)
6821{
e65e1535 6822 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6823 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6824 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6825 struct inode *inode = dip->inode;
c2cf52eb 6826 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 6827 struct bio *dio_bio;
4b46fce2 6828 u64 start;
4b46fce2
JB		 6829
6830 start = dip->logical_offset;
6831 do {
6832 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6833 struct page *page = bvec->bv_page;
6834 char *kaddr;
6835 u32 csum = ~(u32)0;
c329861d 6836 u64 private = ~(u32)0;
4b46fce2
JB		 6837 unsigned long flags;
6838
c329861d
JB		 6839 if (get_state_private(&BTRFS_I(inode)->io_tree,
6840 start, &private))
6841 goto failed;
4b46fce2 6842 local_irq_save(flags);
7ac687d9 6843 kaddr = kmap_atomic(page);
b0496686 6844 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6845 csum, bvec->bv_len);
6846 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6847 kunmap_atomic(kaddr);
4b46fce2
JB		 6848 local_irq_restore(flags);
6849
6850 flush_dcache_page(bvec->bv_page);
c329861d
JB		 6851 if (csum != private) {
6852failed:
c2cf52eb
SK		 6853 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6854 (unsigned long long)btrfs_ino(inode),
6855 (unsigned long long)start,
6856 csum, (unsigned)private);
4b46fce2
JB		 6857 err = -EIO;
6858 }
6859 }
6860
6861 start += bvec->bv_len;
4b46fce2
JB		 6862 bvec++;
6863 } while (bvec <= bvec_end);
6864
6865 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6866 dip->logical_offset + dip->bytes - 1);
9be3395b 6867 dio_bio = dip->dio_bio;
4b46fce2 6868
4b46fce2 6869 kfree(dip);
c0da7aa1
JB		 6870
6871 /* If we had a csum failure make sure to clear the uptodate flag */
6872 if (err)
9be3395b
CM		 6873 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6874 dio_end_io(dio_bio, err);
6875 bio_put(bio);
4b46fce2
JB		 6876}
6877
6878static void btrfs_endio_direct_write(struct bio *bio, int err)
6879{
6880 struct btrfs_dio_private *dip = bio->bi_private;
6881 struct inode *inode = dip->inode;
6882 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6883 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 6884 u64 ordered_offset = dip->logical_offset;
6885 u64 ordered_bytes = dip->bytes;
9be3395b 6886 struct bio *dio_bio;
4b46fce2
JB		 6887 int ret;
6888
6889 if (err)
6890 goto out_done;
163cf09c
CM		 6891again:
6892 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6893 &ordered_offset,
5fd02043 6894 ordered_bytes, !err);
4b46fce2 6895 if (!ret)
163cf09c 6896 goto out_test;
4b46fce2 6897
5fd02043
JB		 6898 ordered->work.func = finish_ordered_fn;
6899 ordered->work.flags = 0;
6900 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6901 &ordered->work);
163cf09c
CM		 6902out_test:
6903 /*
6904 * our bio might span multiple ordered extents. If we haven't
6905 * completed the accounting for the whole dio, go back and try again
6906 */
6907 if (ordered_offset < dip->logical_offset + dip->bytes) {
6908 ordered_bytes = dip->logical_offset + dip->bytes -
6909 ordered_offset;
5fd02043 6910 ordered = NULL;
163cf09c
CM		 6911 goto again;
6912 }
4b46fce2 6913out_done:
9be3395b 6914 dio_bio = dip->dio_bio;
4b46fce2 6915
4b46fce2 6916 kfree(dip);
c0da7aa1
JB		 6917
6918 /* If we had an error make sure to clear the uptodate flag */
6919 if (err)
9be3395b
CM		 6920 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6921 dio_end_io(dio_bio, err);
6922 bio_put(bio);
4b46fce2
JB		 6923}
6924
eaf25d93
CM		 6925static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6926 struct bio *bio, int mirror_num,
6927 unsigned long bio_flags, u64 offset)
6928{
6929 int ret;
6930 struct btrfs_root *root = BTRFS_I(inode)->root;
6931 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6932 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6933 return 0;
6934}
6935
e65e1535
MX		 6936static void btrfs_end_dio_bio(struct bio *bio, int err)
6937{
6938 struct btrfs_dio_private *dip = bio->bi_private;
6939
6940 if (err) {
33345d01 6941 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6942 "sector %#Lx len %u err no %d\n",
33345d01 6943 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6944 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6945 dip->errors = 1;
6946
6947 /*
6948 * before atomic variable goto zero, we must make sure
6949 * dip->errors is perceived to be set.
6950 */
6951 smp_mb__before_atomic_dec();
6952 }
6953
6954 /* if there are more bios still pending for this dio, just exit */
6955 if (!atomic_dec_and_test(&dip->pending_bios))
6956 goto out;
6957
9be3395b 6958 if (dip->errors) {
e65e1535 6959 bio_io_error(dip->orig_bio);
9be3395b
CM		 6960 } else {
6961 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 6962 bio_endio(dip->orig_bio, 0);
6963 }
6964out:
6965 bio_put(bio);
6966}
6967
6968static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6969 u64 first_sector, gfp_t gfp_flags)
6970{
6971 int nr_vecs = bio_get_nr_vecs(bdev);
6972 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6973}
6974
6975static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6976 int rw, u64 file_offset, int skip_sum,
c329861d 6977 int async_submit)
e65e1535
MX		 6978{
6979 int write = rw & REQ_WRITE;
6980 struct btrfs_root *root = BTRFS_I(inode)->root;
6981 int ret;
6982
b812ce28
JB		 6983 if (async_submit)
6984 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6985
e65e1535 6986 bio_get(bio);
5fd02043
JB		 6987
6988 if (!write) {
6989 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6990 if (ret)
6991 goto err;
6992 }
e65e1535 6993
1ae39938
JB		 6994 if (skip_sum)
6995 goto map;
6996
6997 if (write && async_submit) {
e65e1535
MX		 6998 ret = btrfs_wq_submit_bio(root->fs_info,
6999 inode, rw, bio, 0, 0,
7000 file_offset,
7001 __btrfs_submit_bio_start_direct_io,
7002 __btrfs_submit_bio_done);
7003 goto err;
1ae39938
JB		 7004 } else if (write) {
7005 /*
7006 * If we aren't doing async submit, calculate the csum of the
7007 * bio now.
7008 */
7009 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7010 if (ret)
7011 goto err;
c2db1073 7012 } else if (!skip_sum) {
c329861d 7013 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI		 7014 if (ret)
7015 goto err;
7016 }
e65e1535 7017
1ae39938
JB		 7018map:
7019 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7020err:
7021 bio_put(bio);
7022 return ret;
7023}
7024
7025static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7026 int skip_sum)
7027{
7028 struct inode *inode = dip->inode;
7029 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7030 struct bio *bio;
7031 struct bio *orig_bio = dip->orig_bio;
7032 struct bio_vec *bvec = orig_bio->bi_io_vec;
7033 u64 start_sector = orig_bio->bi_sector;
7034 u64 file_offset = dip->logical_offset;
7035 u64 submit_len = 0;
7036 u64 map_length;
7037 int nr_pages = 0;
e65e1535 7038 int ret = 0;
1ae39938 7039 int async_submit = 0;
e65e1535 7040
e65e1535 7041 map_length = orig_bio->bi_size;
53b381b3 7042 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 7043 &map_length, NULL, 0);
7044 if (ret) {
64728bbb 7045 bio_put(orig_bio);
e65e1535
MX		 7046 return -EIO;
7047 }
02f57c7a
JB		 7048 if (map_length >= orig_bio->bi_size) {
7049 bio = orig_bio;
7050 goto submit;
7051 }
7052
53b381b3
DW		 7053 /* async crcs make it difficult to collect full stripe writes. */
7054 if (btrfs_get_alloc_profile(root, 1) &
7055 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7056 async_submit = 0;
7057 else
7058 async_submit = 1;
7059
02f57c7a
JB		 7060 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7061 if (!bio)
7062 return -ENOMEM;
7063 bio->bi_private = dip;
7064 bio->bi_end_io = btrfs_end_dio_bio;
7065 atomic_inc(&dip->pending_bios);
7066
e65e1535
MX		 7067 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7068 if (unlikely(map_length < submit_len + bvec->bv_len ||
7069 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7070 bvec->bv_offset) < bvec->bv_len)) {
7071 /*
7072 * inc the count before we submit the bio so
7073 * we know the end IO handler won't happen before
7074 * we inc the count. Otherwise, the dip might get freed
7075 * before we're done setting it up
7076 */
7077 atomic_inc(&dip->pending_bios);
7078 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7079 file_offset, skip_sum,
c329861d 7080 async_submit);
e65e1535
MX		 7081 if (ret) {
7082 bio_put(bio);
7083 atomic_dec(&dip->pending_bios);
7084 goto out_err;
7085 }
7086
e65e1535
MX		 7087 start_sector += submit_len >> 9;
7088 file_offset += submit_len;
7089
7090 submit_len = 0;
7091 nr_pages = 0;
7092
7093 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7094 start_sector, GFP_NOFS);
7095 if (!bio)
7096 goto out_err;
7097 bio->bi_private = dip;
7098 bio->bi_end_io = btrfs_end_dio_bio;
7099
7100 map_length = orig_bio->bi_size;
53b381b3 7101 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7102 start_sector << 9,
e65e1535
MX		 7103 &map_length, NULL, 0);
7104 if (ret) {
7105 bio_put(bio);
7106 goto out_err;
7107 }
7108 } else {
7109 submit_len += bvec->bv_len;
7110 nr_pages ++;
7111 bvec++;
7112 }
7113 }
7114
02f57c7a 7115submit:
e65e1535 7116 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7117 async_submit);
e65e1535
MX		 7118 if (!ret)
7119 return 0;
7120
7121 bio_put(bio);
7122out_err:
7123 dip->errors = 1;
7124 /*
7125 * before atomic variable goto zero, we must
7126 * make sure dip->errors is perceived to be set.
7127 */
7128 smp_mb__before_atomic_dec();
7129 if (atomic_dec_and_test(&dip->pending_bios))
7130 bio_io_error(dip->orig_bio);
7131
7132 /* bio_end_io() will handle error, so we needn't return it */
7133 return 0;
7134}
7135
9be3395b
CM		 7136static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7137 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7138{
7139 struct btrfs_root *root = BTRFS_I(inode)->root;
7140 struct btrfs_dio_private *dip;
9be3395b
CM		 7141 struct bio_vec *bvec = dio_bio->bi_io_vec;
7142 struct bio *io_bio;
4b46fce2 7143 int skip_sum;
7b6d91da 7144 int write = rw & REQ_WRITE;
4b46fce2
JB		 7145 int ret = 0;
7146
7147 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7148
9be3395b
CM		 7149 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7150
7151 if (!io_bio) {
7152 ret = -ENOMEM;
7153 goto free_ordered;
7154 }
7155
4b46fce2
JB		 7156 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7157 if (!dip) {
7158 ret = -ENOMEM;
9be3395b 7159 goto free_io_bio;
4b46fce2 7160 }
4b46fce2 7161
9be3395b
CM		 7162 dip->private = dio_bio->bi_private;
7163 io_bio->bi_private = dio_bio->bi_private;
4b46fce2
JB		 7164 dip->inode = inode;
7165 dip->logical_offset = file_offset;
7166
4b46fce2
JB		 7167 dip->bytes = 0;
7168 do {
7169 dip->bytes += bvec->bv_len;
7170 bvec++;
9be3395b 7171 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
4b46fce2 7172
9be3395b
CM		 7173 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7174 io_bio->bi_private = dip;
e65e1535 7175 dip->errors = 0;
9be3395b
CM		 7176 dip->orig_bio = io_bio;
7177 dip->dio_bio = dio_bio;
e65e1535 7178 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7179
7180 if (write)
9be3395b 7181 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7182 else
9be3395b 7183 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7184
e65e1535
MX		 7185 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7186 if (!ret)
eaf25d93 7187 return;
9be3395b
CM		 7188
7189free_io_bio:
7190 bio_put(io_bio);
7191
4b46fce2
JB		 7192free_ordered:
7193 /*
7194 * If this is a write, we need to clean up the reserved space and kill
7195 * the ordered extent.
7196 */
7197 if (write) {
7198 struct btrfs_ordered_extent *ordered;
955256f2 7199 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7200 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7201 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7202 btrfs_free_reserved_extent(root, ordered->start,
7203 ordered->disk_len);
7204 btrfs_put_ordered_extent(ordered);
7205 btrfs_put_ordered_extent(ordered);
7206 }
9be3395b 7207 bio_endio(dio_bio, ret);
4b46fce2
JB		 7208}
7209
5a5f79b5
CM		 7210static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7211 const struct iovec *iov, loff_t offset,
7212 unsigned long nr_segs)
7213{
7214 int seg;
a1b75f7d 7215 int i;
5a5f79b5
CM		 7216 size_t size;
7217 unsigned long addr;
7218 unsigned blocksize_mask = root->sectorsize - 1;
7219 ssize_t retval = -EINVAL;
7220 loff_t end = offset;
7221
7222 if (offset & blocksize_mask)
7223 goto out;
7224
7225 /* Check the memory alignment. Blocks cannot straddle pages */
7226 for (seg = 0; seg < nr_segs; seg++) {
7227 addr = (unsigned long)iov[seg].iov_base;
7228 size = iov[seg].iov_len;
7229 end += size;
a1b75f7d 7230 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7231 goto out;
a1b75f7d
JB		 7232
7233 /* If this is a write we don't need to check anymore */
7234 if (rw & WRITE)
7235 continue;
7236
7237 /*
7238 * Check to make sure we don't have duplicate iov_base's in this
7239 * iovec, if so return EINVAL, otherwise we'll get csum errors
7240 * when reading back.
7241 */
7242 for (i = seg + 1; i < nr_segs; i++) {
7243 if (iov[seg].iov_base == iov[i].iov_base)
7244 goto out;
7245 }
5a5f79b5
CM		 7246 }
7247 retval = 0;
7248out:
7249 return retval;
7250}
eb838e73 7251
16432985
CM		 7252static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7253 const struct iovec *iov, loff_t offset,
7254 unsigned long nr_segs)
7255{
4b46fce2
JB		 7256 struct file *file = iocb->ki_filp;
7257 struct inode *inode = file->f_mapping->host;
0934856d 7258 size_t count = 0;
2e60a51e 7259 int flags = 0;
38851cc1
MX		 7260 bool wakeup = true;
7261 bool relock = false;
0934856d 7262 ssize_t ret;
4b46fce2 7263
5a5f79b5 7264 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7265 offset, nr_segs))
5a5f79b5 7266 return 0;
3f7c579c 7267
38851cc1
MX		 7268 atomic_inc(&inode->i_dio_count);
7269 smp_mb__after_atomic_inc();
7270
0934856d
MX		 7271 if (rw & WRITE) {
7272 count = iov_length(iov, nr_segs);
38851cc1
MX		 7273 /*
7274 * If the write DIO is beyond the EOF, we need update
7275 * the isize, but it is protected by i_mutex. So we can
7276 * not unlock the i_mutex at this case.
7277 */
7278 if (offset + count <= inode->i_size) {
7279 mutex_unlock(&inode->i_mutex);
7280 relock = true;
7281 }
0934856d
MX		 7282 ret = btrfs_delalloc_reserve_space(inode, count);
7283 if (ret)
38851cc1
MX		 7284 goto out;
7285 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7286 &BTRFS_I(inode)->runtime_flags))) {
7287 inode_dio_done(inode);
7288 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7289 wakeup = false;
0934856d
MX		 7290 }
7291
7292 ret = __blockdev_direct_IO(rw, iocb, inode,
7293 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7294 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7295 btrfs_submit_direct, flags);
0934856d
MX		 7296 if (rw & WRITE) {
7297 if (ret < 0 && ret != -EIOCBQUEUED)
7298 btrfs_delalloc_release_space(inode, count);
172a5049 7299 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7300 btrfs_delalloc_release_space(inode,
7301 count - (size_t)ret);
172a5049
MX		 7302 else
7303 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7304 }
38851cc1 7305out:
2e60a51e
MX		 7306 if (wakeup)
7307 inode_dio_done(inode);
38851cc1
MX		 7308 if (relock)
7309 mutex_lock(&inode->i_mutex);
0934856d
MX		 7310
7311 return ret;
16432985
CM		 7312}
7313
05dadc09
TI		 7314#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7315
1506fcc8
YS		 7316static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7317 __u64 start, __u64 len)
7318{
05dadc09
TI		 7319 int ret;
7320
7321 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7322 if (ret)
7323 return ret;
7324
ec29ed5b 7325 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7326}
7327
a52d9a80 7328int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7329{
d1310b2e
CM		 7330 struct extent_io_tree *tree;
7331 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7332 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7333}
1832a6d5 7334
a52d9a80 7335static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7336{
d1310b2e 7337 struct extent_io_tree *tree;
b888db2b
CM		 7338
7339
7340 if (current->flags & PF_MEMALLOC) {
7341 redirty_page_for_writepage(wbc, page);
7342 unlock_page(page);
7343 return 0;
7344 }
d1310b2e 7345 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7346 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7347}
7348
48a3b636
ES		 7349static int btrfs_writepages(struct address_space *mapping,
7350 struct writeback_control *wbc)
b293f02e 7351{
d1310b2e 7352 struct extent_io_tree *tree;
771ed689 7353
d1310b2e 7354 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7355 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7356}
7357
3ab2fb5a
CM		 7358static int
7359btrfs_readpages(struct file *file, struct address_space *mapping,
7360 struct list_head *pages, unsigned nr_pages)
7361{
d1310b2e
CM		 7362 struct extent_io_tree *tree;
7363 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7364 return extent_readpages(tree, mapping, pages, nr_pages,
7365 btrfs_get_extent);
7366}
e6dcd2dc 7367static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7368{
d1310b2e
CM		 7369 struct extent_io_tree *tree;
7370 struct extent_map_tree *map;
a52d9a80 7371 int ret;
8c2383c3 7372
d1310b2e
CM		 7373 tree = &BTRFS_I(page->mapping->host)->io_tree;
7374 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7375 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7376 if (ret == 1) {
7377 ClearPagePrivate(page);
7378 set_page_private(page, 0);
7379 page_cache_release(page);
39279cc3 7380 }
a52d9a80 7381 return ret;
39279cc3
CM		 7382}
7383
e6dcd2dc
CM		 7384static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7385{
98509cfc
CM		 7386 if (PageWriteback(page) || PageDirty(page))
7387 return 0;
b335b003 7388 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7389}
7390
a52d9a80 7391static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 7392{
5fd02043 7393 struct inode *inode = page->mapping->host;
d1310b2e 7394 struct extent_io_tree *tree;
e6dcd2dc 7395 struct btrfs_ordered_extent *ordered;
2ac55d41 7396 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7397 u64 page_start = page_offset(page);
7398 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7399
8b62b72b
CM		 7400 /*
7401 * we have the page locked, so new writeback can't start,
7402 * and the dirty bit won't be cleared while we are here.
7403 *
7404 * Wait for IO on this page so that we can safely clear
7405 * the PagePrivate2 bit and do ordered accounting
7406 */
e6dcd2dc 7407 wait_on_page_writeback(page);
8b62b72b 7408
5fd02043 7409 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7410 if (offset) {
7411 btrfs_releasepage(page, GFP_NOFS);
7412 return;
7413 }
d0082371 7414 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7415 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7416 if (ordered) {
eb84ae03
CM		 7417 /*
7418 * IO on this page will never be started, so we need
7419 * to account for any ordered extents now
7420 */
e6dcd2dc
CM		 7421 clear_extent_bit(tree, page_start, page_end,
7422 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7423 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7424 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7425 /*
7426 * whoever cleared the private bit is responsible
7427 * for the finish_ordered_io
7428 */
5fd02043
JB		 7429 if (TestClearPagePrivate2(page) &&
7430 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7431 PAGE_CACHE_SIZE, 1)) {
7432 btrfs_finish_ordered_io(ordered);
8b62b72b 7433 }
e6dcd2dc 7434 btrfs_put_ordered_extent(ordered);
2ac55d41 7435 cached_state = NULL;
d0082371 7436 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7437 }
7438 clear_extent_bit(tree, page_start, page_end,
32c00aff 7439 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7440 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7441 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7442 __btrfs_releasepage(page, GFP_NOFS);
7443
4a096752 7444 ClearPageChecked(page);
9ad6b7bc 7445 if (PagePrivate(page)) {
9ad6b7bc
CM		 7446 ClearPagePrivate(page);
7447 set_page_private(page, 0);
7448 page_cache_release(page);
7449 }
39279cc3
CM		 7450}
7451
9ebefb18
CM		 7452/*
7453 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7454 * called from a page fault handler when a page is first dirtied. Hence we must
7455 * be careful to check for EOF conditions here. We set the page up correctly
7456 * for a written page which means we get ENOSPC checking when writing into
7457 * holes and correct delalloc and unwritten extent mapping on filesystems that
7458 * support these features.
7459 *
7460 * We are not allowed to take the i_mutex here so we have to play games to
7461 * protect against truncate races as the page could now be beyond EOF. Because
7462 * vmtruncate() writes the inode size before removing pages, once we have the
7463 * page lock we can determine safely if the page is beyond EOF. If it is not
7464 * beyond EOF, then the page is guaranteed safe against truncation until we
7465 * unlock the page.
7466 */
c2ec175c 7467int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7468{
c2ec175c 7469 struct page *page = vmf->page;
496ad9aa 7470 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7471 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7472 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7473 struct btrfs_ordered_extent *ordered;
2ac55d41 7474 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7475 char *kaddr;
7476 unsigned long zero_start;
9ebefb18 7477 loff_t size;
1832a6d5 7478 int ret;
9998eb70 7479 int reserved = 0;
a52d9a80 7480 u64 page_start;
e6dcd2dc 7481 u64 page_end;
9ebefb18 7482
b2b5ef5c 7483 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7484 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7485 if (!ret) {
e41f941a 7486 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7487 reserved = 1;
7488 }
56a76f82
NP		 7489 if (ret) {
7490 if (ret == -ENOMEM)
7491 ret = VM_FAULT_OOM;
7492 else /* -ENOSPC, -EIO, etc */
7493 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7494 if (reserved)
7495 goto out;
7496 goto out_noreserve;
56a76f82 7497 }
1832a6d5 7498
56a76f82 7499 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7500again:
9ebefb18 7501 lock_page(page);
9ebefb18 7502 size = i_size_read(inode);
e6dcd2dc
CM		 7503 page_start = page_offset(page);
7504 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7505
9ebefb18 7506 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7507 (page_start >= size)) {
9ebefb18
CM		 7508 /* page got truncated out from underneath us */
7509 goto out_unlock;
7510 }
e6dcd2dc
CM		 7511 wait_on_page_writeback(page);
7512
d0082371 7513 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7514 set_page_extent_mapped(page);
7515
eb84ae03
CM		 7516 /*
7517 * we can't set the delalloc bits if there are pending ordered
7518 * extents. Drop our locks and wait for them to finish
7519 */
e6dcd2dc
CM		 7520 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7521 if (ordered) {
2ac55d41
JB		 7522 unlock_extent_cached(io_tree, page_start, page_end,
7523 &cached_state, GFP_NOFS);
e6dcd2dc 7524 unlock_page(page);
eb84ae03 7525 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7526 btrfs_put_ordered_extent(ordered);
7527 goto again;
7528 }
7529
fbf19087
JB		 7530 /*
7531 * XXX - page_mkwrite gets called every time the page is dirtied, even
7532 * if it was already dirty, so for space accounting reasons we need to
7533 * clear any delalloc bits for the range we are fixing to save. There
7534 * is probably a better way to do this, but for now keep consistent with
7535 * prepare_pages in the normal write path.
7536 */
2ac55d41 7537 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7538 EXTENT_DIRTY | EXTENT_DELALLOC |
7539 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7540 0, 0, &cached_state, GFP_NOFS);
fbf19087 7541
2ac55d41
JB		 7542 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7543 &cached_state);
9ed74f2d 7544 if (ret) {
2ac55d41
JB		 7545 unlock_extent_cached(io_tree, page_start, page_end,
7546 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7547 ret = VM_FAULT_SIGBUS;
7548 goto out_unlock;
7549 }
e6dcd2dc 7550 ret = 0;
9ebefb18
CM		 7551
7552 /* page is wholly or partially inside EOF */
a52d9a80 7553 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7554 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7555 else
e6dcd2dc 7556 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7557
e6dcd2dc
CM		 7558 if (zero_start != PAGE_CACHE_SIZE) {
7559 kaddr = kmap(page);
7560 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7561 flush_dcache_page(page);
7562 kunmap(page);
7563 }
247e743c 7564 ClearPageChecked(page);
e6dcd2dc 7565 set_page_dirty(page);
50a9b214 7566 SetPageUptodate(page);
5a3f23d5 7567
257c62e1
CM		 7568 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7569 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7570 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7571
2ac55d41 7572 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7573
7574out_unlock:
b2b5ef5c
JK		 7575 if (!ret) {
7576 sb_end_pagefault(inode->i_sb);
50a9b214 7577 return VM_FAULT_LOCKED;
b2b5ef5c 7578 }
9ebefb18 7579 unlock_page(page);
1832a6d5 7580out:
ec39e180 7581 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7582out_noreserve:
b2b5ef5c 7583 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7584 return ret;
7585}
7586
a41ad394 7587static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7588{
7589 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7590 struct btrfs_block_rsv *rsv;
a71754fc 7591 int ret = 0;
3893e33b 7592 int err = 0;
39279cc3 7593 struct btrfs_trans_handle *trans;
dbe674a9 7594 u64 mask = root->sectorsize - 1;
07127184 7595 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7596
4a096752 7597 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 7598 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 7599
fcb80c2a
JB		 7600 /*
7601 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7602 * 3 things going on here
7603 *
7604 * 1) We need to reserve space for our orphan item and the space to
7605 * delete our orphan item. Lord knows we don't want to have a dangling
7606 * orphan item because we didn't reserve space to remove it.
7607 *
7608 * 2) We need to reserve space to update our inode.
7609 *
7610 * 3) We need to have something to cache all the space that is going to
7611 * be free'd up by the truncate operation, but also have some slack
7612 * space reserved in case it uses space during the truncate (thank you
7613 * very much snapshotting).
7614 *
7615 * And we need these to all be seperate. The fact is we can use alot of
7616 * space doing the truncate, and we have no earthly idea how much space
7617 * we will use, so we need the truncate reservation to be seperate so it
7618 * doesn't end up using space reserved for updating the inode or
7619 * removing the orphan item. We also need to be able to stop the
7620 * transaction and start a new one, which means we need to be able to
7621 * update the inode several times, and we have no idea of knowing how
7622 * many times that will be, so we can't just reserve 1 item for the
7623 * entirety of the opration, so that has to be done seperately as well.
7624 * Then there is the orphan item, which does indeed need to be held on
7625 * to for the whole operation, and we need nobody to touch this reserved
7626 * space except the orphan code.
7627 *
7628 * So that leaves us with
7629 *
7630 * 1) root->orphan_block_rsv - for the orphan deletion.
7631 * 2) rsv - for the truncate reservation, which we will steal from the
7632 * transaction reservation.
7633 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7634 * updating the inode.
7635 */
66d8f3dd 7636 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 7637 if (!rsv)
7638 return -ENOMEM;
4a338542 7639 rsv->size = min_size;
ca7e70f5 7640 rsv->failfast = 1;
f0cd846e 7641
907cbceb 7642 /*
07127184 7643 * 1 for the truncate slack space
907cbceb
JB		 7644 * 1 for updating the inode.
7645 */
f3fe820c 7646 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 7647 if (IS_ERR(trans)) {
7648 err = PTR_ERR(trans);
7649 goto out;
7650 }
f0cd846e 7651
907cbceb
JB		 7652 /* Migrate the slack space for the truncate to our reserve */
7653 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7654 min_size);
fcb80c2a 7655 BUG_ON(ret);
f0cd846e 7656
5a3f23d5
CM		 7657 /*
7658 * setattr is responsible for setting the ordered_data_close flag,
7659 * but that is only tested during the last file release. That
7660 * could happen well after the next commit, leaving a great big
7661 * window where new writes may get lost if someone chooses to write
7662 * to this file after truncating to zero
7663 *
7664 * The inode doesn't have any dirty data here, and so if we commit
7665 * this is a noop. If someone immediately starts writing to the inode
7666 * it is very likely we'll catch some of their writes in this
7667 * transaction, and the commit will find this file on the ordered
7668 * data list with good things to send down.
7669 *
7670 * This is a best effort solution, there is still a window where
7671 * using truncate to replace the contents of the file will
7672 * end up with a zero length file after a crash.
7673 */
72ac3c0d
JB		 7674 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7675 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 7676 btrfs_add_ordered_operation(trans, root, inode);
7677
5dc562c5
JB		 7678 /*
7679 * So if we truncate and then write and fsync we normally would just
7680 * write the extents that changed, which is a problem if we need to
7681 * first truncate that entire inode. So set this flag so we write out
7682 * all of the extents in the inode to the sync log so we're completely
7683 * safe.
7684 */
7685 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7686 trans->block_rsv = rsv;
907cbceb 7687
8082510e
YZ		 7688 while (1) {
7689 ret = btrfs_truncate_inode_items(trans, root, inode,
7690 inode->i_size,
7691 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7692 if (ret != -ENOSPC) {
3893e33b 7693 err = ret;
8082510e 7694 break;
3893e33b 7695 }
39279cc3 7696
fcb80c2a 7697 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7698 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 7699 if (ret) {
7700 err = ret;
7701 break;
7702 }
ca7e70f5 7703
8082510e 7704 btrfs_end_transaction(trans, root);
b53d3f5d 7705 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 7706
7707 trans = btrfs_start_transaction(root, 2);
7708 if (IS_ERR(trans)) {
7709 ret = err = PTR_ERR(trans);
7710 trans = NULL;
7711 break;
7712 }
7713
7714 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7715 rsv, min_size);
7716 BUG_ON(ret); /* shouldn't happen */
7717 trans->block_rsv = rsv;
8082510e
YZ		 7718 }
7719
7720 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7721 trans->block_rsv = root->orphan_block_rsv;
8082510e 7722 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7723 if (ret)
7724 err = ret;
8082510e
YZ		 7725 }
7726
917c16b2
CM		 7727 if (trans) {
7728 trans->block_rsv = &root->fs_info->trans_block_rsv;
7729 ret = btrfs_update_inode(trans, root, inode);
7730 if (ret && !err)
7731 err = ret;
7b128766 7732
7ad85bb7 7733 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7734 btrfs_btree_balance_dirty(root);
917c16b2 7735 }
fcb80c2a
JB		 7736
7737out:
7738 btrfs_free_block_rsv(root, rsv);
7739
3893e33b
JB		 7740 if (ret && !err)
7741 err = ret;
a41ad394 7742
3893e33b 7743 return err;
39279cc3
CM		 7744}
7745
d352ac68
CM		 7746/*
7747 * create a new subvolume directory/inode (helper for the ioctl).
7748 */
d2fb3437 7749int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7750 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7751{
39279cc3 7752 struct inode *inode;
76dda93c 7753 int err;
00e4e6b3 7754 u64 index = 0;
39279cc3 7755
12fc9d09
FA		 7756 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7757 new_dirid, new_dirid,
7758 S_IFDIR | (~current_umask() & S_IRWXUGO),
7759 &index);
54aa1f4d 7760 if (IS_ERR(inode))
f46b5a66 7761 return PTR_ERR(inode);
39279cc3
CM		 7762 inode->i_op = &btrfs_dir_inode_operations;
7763 inode->i_fop = &btrfs_dir_file_operations;
7764
bfe86848 7765 set_nlink(inode, 1);
dbe674a9 7766 btrfs_i_size_write(inode, 0);
3b96362c 7767
76dda93c 7768 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7769
76dda93c 7770 iput(inode);
ce598979 7771 return err;
39279cc3
CM		 7772}
7773
39279cc3
CM		 7774struct inode *btrfs_alloc_inode(struct super_block *sb)
7775{
7776 struct btrfs_inode *ei;
2ead6ae7 7777 struct inode *inode;
39279cc3
CM		 7778
7779 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7780 if (!ei)
7781 return NULL;
2ead6ae7
YZ		 7782
7783 ei->root = NULL;
2ead6ae7 7784 ei->generation = 0;
15ee9bc7 7785 ei->last_trans = 0;
257c62e1 7786 ei->last_sub_trans = 0;
e02119d5 7787 ei->logged_trans = 0;
2ead6ae7 7788 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7789 ei->disk_i_size = 0;
7790 ei->flags = 0;
7709cde3 7791 ei->csum_bytes = 0;
2ead6ae7
YZ		 7792 ei->index_cnt = (u64)-1;
7793 ei->last_unlink_trans = 0;
46d8bc34 7794 ei->last_log_commit = 0;
2ead6ae7 7795
9e0baf60
JB		 7796 spin_lock_init(&ei->lock);
7797 ei->outstanding_extents = 0;
7798 ei->reserved_extents = 0;
2ead6ae7 7799
72ac3c0d 7800 ei->runtime_flags = 0;
261507a0 7801 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7802
16cdcec7
MX		 7803 ei->delayed_node = NULL;
7804
2ead6ae7 7805 inode = &ei->vfs_inode;
a8067e02 7806 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7807 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7808 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7809 ei->io_tree.track_uptodate = 1;
7810 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7811 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7812 mutex_init(&ei->log_mutex);
f248679e 7813 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7814 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7815 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7816 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7817 RB_CLEAR_NODE(&ei->rb_node);
7818
7819 return inode;
39279cc3
CM		 7820}
7821
fa0d7e3d
NP		 7822static void btrfs_i_callback(struct rcu_head *head)
7823{
7824 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7825 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7826}
7827
39279cc3
CM		 7828void btrfs_destroy_inode(struct inode *inode)
7829{
e6dcd2dc 7830 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7831 struct btrfs_root *root = BTRFS_I(inode)->root;
7832
b3d9b7a3 7833 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7834 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7835 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7836 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7837 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7838 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7839
a6dbd429
JB		 7840 /*
7841 * This can happen where we create an inode, but somebody else also
7842 * created the same inode and we need to destroy the one we already
7843 * created.
7844 */
7845 if (!root)
7846 goto free;
7847
5a3f23d5
CM		 7848 /*
7849 * Make sure we're properly removed from the ordered operation
7850 * lists.
7851 */
7852 smp_mb();
7853 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
199c2a9c 7854 spin_lock(&root->fs_info->ordered_root_lock);
5a3f23d5 7855 list_del_init(&BTRFS_I(inode)->ordered_operations);
199c2a9c 7856 spin_unlock(&root->fs_info->ordered_root_lock);
5a3f23d5
CM		 7857 }
7858
8a35d95f
JB		 7859 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7860 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb
SK		 7861 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7862 (unsigned long long)btrfs_ino(inode));
8a35d95f 7863 atomic_dec(&root->orphan_inodes);
7b128766 7864 }
7b128766 7865
d397712b 7866 while (1) {
e6dcd2dc
CM		 7867 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7868 if (!ordered)
7869 break;
7870 else {
c2cf52eb
SK		 7871 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7872 (unsigned long long)ordered->file_offset,
7873 (unsigned long long)ordered->len);
e6dcd2dc
CM		 7874 btrfs_remove_ordered_extent(inode, ordered);
7875 btrfs_put_ordered_extent(ordered);
7876 btrfs_put_ordered_extent(ordered);
7877 }
7878 }
5d4f98a2 7879 inode_tree_del(inode);
5b21f2ed 7880 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7881free:
fa0d7e3d 7882 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7883}
7884
45321ac5 7885int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7886{
7887 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7888
6379ef9f
NA		 7889 if (root == NULL)
7890 return 1;
7891
fa6ac876 7892 /* the snap/subvol tree is on deleting */
0af3d00b 7893 if (btrfs_root_refs(&root->root_item) == 0 &&
fa6ac876 7894 root != root->fs_info->tree_root)
45321ac5 7895 return 1;
76dda93c 7896 else
45321ac5 7897 return generic_drop_inode(inode);
76dda93c
YZ		 7898}
7899
0ee0fda0 7900static void init_once(void *foo)
39279cc3
CM		 7901{
7902 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7903
7904 inode_init_once(&ei->vfs_inode);
7905}
7906
7907void btrfs_destroy_cachep(void)
7908{
8c0a8537
KS		 7909 /*
7910 * Make sure all delayed rcu free inodes are flushed before we
7911 * destroy cache.
7912 */
7913 rcu_barrier();
39279cc3
CM		 7914 if (btrfs_inode_cachep)
7915 kmem_cache_destroy(btrfs_inode_cachep);
7916 if (btrfs_trans_handle_cachep)
7917 kmem_cache_destroy(btrfs_trans_handle_cachep);
7918 if (btrfs_transaction_cachep)
7919 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7920 if (btrfs_path_cachep)
7921 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7922 if (btrfs_free_space_cachep)
7923 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 7924 if (btrfs_delalloc_work_cachep)
7925 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 7926}
7927
7928int btrfs_init_cachep(void)
7929{
837e1972 7930 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 7931 sizeof(struct btrfs_inode), 0,
7932 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7933 if (!btrfs_inode_cachep)
7934 goto fail;
9601e3f6 7935
837e1972 7936 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 7937 sizeof(struct btrfs_trans_handle), 0,
7938 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7939 if (!btrfs_trans_handle_cachep)
7940 goto fail;
9601e3f6 7941
837e1972 7942 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 7943 sizeof(struct btrfs_transaction), 0,
7944 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7945 if (!btrfs_transaction_cachep)
7946 goto fail;
9601e3f6 7947
837e1972 7948 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 7949 sizeof(struct btrfs_path), 0,
7950 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7951 if (!btrfs_path_cachep)
7952 goto fail;
9601e3f6 7953
837e1972 7954 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 7955 sizeof(struct btrfs_free_space), 0,
7956 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7957 if (!btrfs_free_space_cachep)
7958 goto fail;
7959
8ccf6f19
MX		 7960 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7961 sizeof(struct btrfs_delalloc_work), 0,
7962 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7963 NULL);
7964 if (!btrfs_delalloc_work_cachep)
7965 goto fail;
7966
39279cc3
CM		 7967 return 0;
7968fail:
7969 btrfs_destroy_cachep();
7970 return -ENOMEM;
7971}
7972
7973static int btrfs_getattr(struct vfsmount *mnt,
7974 struct dentry *dentry, struct kstat *stat)
7975{
df0af1a5 7976 u64 delalloc_bytes;
39279cc3 7977 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7978 u32 blocksize = inode->i_sb->s_blocksize;
7979
39279cc3 7980 generic_fillattr(inode, stat);
0ee5dc67 7981 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7982 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 7983
7984 spin_lock(&BTRFS_I(inode)->lock);
7985 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7986 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 7987 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 7988 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7989 return 0;
7990}
7991
d397712b
CM		 7992static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7993 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7994{
7995 struct btrfs_trans_handle *trans;
7996 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7997 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7998 struct inode *new_inode = new_dentry->d_inode;
7999 struct inode *old_inode = old_dentry->d_inode;
8000 struct timespec ctime = CURRENT_TIME;
00e4e6b3 8001 u64 index = 0;
4df27c4d 8002 u64 root_objectid;
39279cc3 8003 int ret;
33345d01 8004 u64 old_ino = btrfs_ino(old_inode);
39279cc3 8005
33345d01 8006 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 8007 return -EPERM;
8008
4df27c4d 8009 /* we only allow rename subvolume link between subvolumes */
33345d01 8010 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8011 return -EXDEV;
8012
33345d01
LZ		 8013 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8014 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8015 return -ENOTEMPTY;
5f39d397 8016
4df27c4d
YZ		 8017 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8018 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8019 return -ENOTEMPTY;
9c52057c
CM		 8020
8021
8022 /* check for collisions, even if the name isn't there */
8023 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8024 new_dentry->d_name.name,
8025 new_dentry->d_name.len);
8026
8027 if (ret) {
8028 if (ret == -EEXIST) {
8029 /* we shouldn't get
8030 * eexist without a new_inode */
8031 if (!new_inode) {
8032 WARN_ON(1);
8033 return ret;
8034 }
8035 } else {
8036 /* maybe -EOVERFLOW */
8037 return ret;
8038 }
8039 }
8040 ret = 0;
8041
5a3f23d5
CM		 8042 /*
8043 * we're using rename to replace one file with another.
8044 * and the replacement file is large. Start IO on it now so
8045 * we don't add too much work to the end of the transaction
8046 */
4baf8c92 8047 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8048 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8049 filemap_flush(old_inode->i_mapping);
8050
76dda93c 8051 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8052 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8053 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8054 /*
8055 * We want to reserve the absolute worst case amount of items. So if
8056 * both inodes are subvols and we need to unlink them then that would
8057 * require 4 item modifications, but if they are both normal inodes it
8058 * would require 5 item modifications, so we'll assume their normal
8059 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8060 * should cover the worst case number of items we'll modify.
8061 */
6e137ed3 8062 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8063 if (IS_ERR(trans)) {
8064 ret = PTR_ERR(trans);
8065 goto out_notrans;
8066 }
76dda93c 8067
4df27c4d
YZ		 8068 if (dest != root)
8069 btrfs_record_root_in_trans(trans, dest);
5f39d397 8070
a5719521
YZ		 8071 ret = btrfs_set_inode_index(new_dir, &index);
8072 if (ret)
8073 goto out_fail;
5a3f23d5 8074
33345d01 8075 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8076 /* force full log commit if subvolume involved. */
8077 root->fs_info->last_trans_log_full_commit = trans->transid;
8078 } else {
a5719521
YZ		 8079 ret = btrfs_insert_inode_ref(trans, dest,
8080 new_dentry->d_name.name,
8081 new_dentry->d_name.len,
33345d01
LZ		 8082 old_ino,
8083 btrfs_ino(new_dir), index);
a5719521
YZ		 8084 if (ret)
8085 goto out_fail;
4df27c4d
YZ		 8086 /*
8087 * this is an ugly little race, but the rename is required
8088 * to make sure that if we crash, the inode is either at the
8089 * old name or the new one. pinning the log transaction lets
8090 * us make sure we don't allow a log commit to come in after
8091 * we unlink the name but before we add the new name back in.
8092 */
8093 btrfs_pin_log_trans(root);
8094 }
5a3f23d5
CM		 8095 /*
8096 * make sure the inode gets flushed if it is replacing
8097 * something.
8098 */
33345d01 8099 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8100 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8101
0c4d2d95
JB		 8102 inode_inc_iversion(old_dir);
8103 inode_inc_iversion(new_dir);
8104 inode_inc_iversion(old_inode);
39279cc3
CM		 8105 old_dir->i_ctime = old_dir->i_mtime = ctime;
8106 new_dir->i_ctime = new_dir->i_mtime = ctime;
8107 old_inode->i_ctime = ctime;
5f39d397 8108
12fcfd22
CM		 8109 if (old_dentry->d_parent != new_dentry->d_parent)
8110 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8111
33345d01 8112 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8113 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8114 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8115 old_dentry->d_name.name,
8116 old_dentry->d_name.len);
8117 } else {
92986796
AV		 8118 ret = __btrfs_unlink_inode(trans, root, old_dir,
8119 old_dentry->d_inode,
8120 old_dentry->d_name.name,
8121 old_dentry->d_name.len);
8122 if (!ret)
8123 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8124 }
79787eaa
JM		 8125 if (ret) {
8126 btrfs_abort_transaction(trans, root, ret);
8127 goto out_fail;
8128 }
39279cc3
CM		 8129
8130 if (new_inode) {
0c4d2d95 8131 inode_inc_iversion(new_inode);
39279cc3 8132 new_inode->i_ctime = CURRENT_TIME;
33345d01 8133 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8134 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8135 root_objectid = BTRFS_I(new_inode)->location.objectid;
8136 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8137 root_objectid,
8138 new_dentry->d_name.name,
8139 new_dentry->d_name.len);
8140 BUG_ON(new_inode->i_nlink == 0);
8141 } else {
8142 ret = btrfs_unlink_inode(trans, dest, new_dir,
8143 new_dentry->d_inode,
8144 new_dentry->d_name.name,
8145 new_dentry->d_name.len);
8146 }
79787eaa 8147 if (!ret && new_inode->i_nlink == 0) {
e02119d5 8148 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 8149 BUG_ON(ret);
7b128766 8150 }
79787eaa
JM		 8151 if (ret) {
8152 btrfs_abort_transaction(trans, root, ret);
8153 goto out_fail;
8154 }
39279cc3 8155 }
aec7477b 8156
4df27c4d
YZ		 8157 ret = btrfs_add_link(trans, new_dir, old_inode,
8158 new_dentry->d_name.name,
a5719521 8159 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8160 if (ret) {
8161 btrfs_abort_transaction(trans, root, ret);
8162 goto out_fail;
8163 }
39279cc3 8164
33345d01 8165 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8166 struct dentry *parent = new_dentry->d_parent;
6a912213 8167 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8168 btrfs_end_log_trans(root);
8169 }
39279cc3 8170out_fail:
7ad85bb7 8171 btrfs_end_transaction(trans, root);
b44c59a8 8172out_notrans:
33345d01 8173 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8174 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8175
39279cc3
CM		 8176 return ret;
8177}
8178
8ccf6f19
MX		 8179static void btrfs_run_delalloc_work(struct btrfs_work *work)
8180{
8181 struct btrfs_delalloc_work *delalloc_work;
8182
8183 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8184 work);
8185 if (delalloc_work->wait)
8186 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8187 else
8188 filemap_flush(delalloc_work->inode->i_mapping);
8189
8190 if (delalloc_work->delay_iput)
8191 btrfs_add_delayed_iput(delalloc_work->inode);
8192 else
8193 iput(delalloc_work->inode);
8194 complete(&delalloc_work->completion);
8195}
8196
8197struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8198 int wait, int delay_iput)
8199{
8200 struct btrfs_delalloc_work *work;
8201
8202 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8203 if (!work)
8204 return NULL;
8205
8206 init_completion(&work->completion);
8207 INIT_LIST_HEAD(&work->list);
8208 work->inode = inode;
8209 work->wait = wait;
8210 work->delay_iput = delay_iput;
8211 work->work.func = btrfs_run_delalloc_work;
8212
8213 return work;
8214}
8215
8216void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8217{
8218 wait_for_completion(&work->completion);
8219 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8220}
8221
d352ac68
CM		 8222/*
8223 * some fairly slow code that needs optimization. This walks the list
8224 * of all the inodes with pending delalloc and forces them to disk.
8225 */
eb73c1b7 8226static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8227{
ea8c2819 8228 struct btrfs_inode *binode;
5b21f2ed 8229 struct inode *inode;
8ccf6f19
MX		 8230 struct btrfs_delalloc_work *work, *next;
8231 struct list_head works;
1eafa6c7 8232 struct list_head splice;
8ccf6f19 8233 int ret = 0;
ea8c2819 8234
8ccf6f19 8235 INIT_LIST_HEAD(&works);
1eafa6c7 8236 INIT_LIST_HEAD(&splice);
63607cc8 8237
eb73c1b7
MX		 8238 spin_lock(&root->delalloc_lock);
8239 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8240 while (!list_empty(&splice)) {
8241 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8242 delalloc_inodes);
1eafa6c7 8243
eb73c1b7
MX		 8244 list_move_tail(&binode->delalloc_inodes,
8245 &root->delalloc_inodes);
5b21f2ed 8246 inode = igrab(&binode->vfs_inode);
df0af1a5 8247 if (!inode) {
eb73c1b7 8248 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8249 continue;
df0af1a5 8250 }
eb73c1b7 8251 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8252
8253 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8254 if (unlikely(!work)) {
8255 ret = -ENOMEM;
8256 goto out;
5b21f2ed 8257 }
1eafa6c7
MX		 8258 list_add_tail(&work->list, &works);
8259 btrfs_queue_worker(&root->fs_info->flush_workers,
8260 &work->work);
8261
5b21f2ed 8262 cond_resched();
eb73c1b7 8263 spin_lock(&root->delalloc_lock);
ea8c2819 8264 }
eb73c1b7 8265 spin_unlock(&root->delalloc_lock);
8c8bee1d 8266
1eafa6c7
MX		 8267 list_for_each_entry_safe(work, next, &works, list) {
8268 list_del_init(&work->list);
8269 btrfs_wait_and_free_delalloc_work(work);
8270 }
eb73c1b7
MX		 8271 return 0;
8272out:
8273 list_for_each_entry_safe(work, next, &works, list) {
8274 list_del_init(&work->list);
8275 btrfs_wait_and_free_delalloc_work(work);
8276 }
8277
8278 if (!list_empty_careful(&splice)) {
8279 spin_lock(&root->delalloc_lock);
8280 list_splice_tail(&splice, &root->delalloc_inodes);
8281 spin_unlock(&root->delalloc_lock);
8282 }
8283 return ret;
8284}
1eafa6c7 8285
eb73c1b7
MX		 8286int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8287{
8288 int ret;
8289
8290 if (root->fs_info->sb->s_flags & MS_RDONLY)
8291 return -EROFS;
8292
8293 ret = __start_delalloc_inodes(root, delay_iput);
8294 /*
8295 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8296 * we have to make sure the IO is actually started and that
8297 * ordered extents get created before we return
8298 */
8299 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8300 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8301 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8302 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8303 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8304 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8305 }
8306 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8307 return ret;
8308}
8309
8310int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8311 int delay_iput)
8312{
8313 struct btrfs_root *root;
8314 struct list_head splice;
8315 int ret;
8316
8317 if (fs_info->sb->s_flags & MS_RDONLY)
8318 return -EROFS;
8319
8320 INIT_LIST_HEAD(&splice);
8321
8322 spin_lock(&fs_info->delalloc_root_lock);
8323 list_splice_init(&fs_info->delalloc_roots, &splice);
8324 while (!list_empty(&splice)) {
8325 root = list_first_entry(&splice, struct btrfs_root,
8326 delalloc_root);
8327 root = btrfs_grab_fs_root(root);
8328 BUG_ON(!root);
8329 list_move_tail(&root->delalloc_root,
8330 &fs_info->delalloc_roots);
8331 spin_unlock(&fs_info->delalloc_root_lock);
8332
8333 ret = __start_delalloc_inodes(root, delay_iput);
8334 btrfs_put_fs_root(root);
8335 if (ret)
8336 goto out;
8337
8338 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8339 }
eb73c1b7 8340 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8341
eb73c1b7
MX		 8342 atomic_inc(&fs_info->async_submit_draining);
8343 while (atomic_read(&fs_info->nr_async_submits) ||
8344 atomic_read(&fs_info->async_delalloc_pages)) {
8345 wait_event(fs_info->async_submit_wait,
8346 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8347 atomic_read(&fs_info->async_delalloc_pages) == 0));
8348 }
8349 atomic_dec(&fs_info->async_submit_draining);
8350 return 0;
8351out:
1eafa6c7 8352 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8353 spin_lock(&fs_info->delalloc_root_lock);
8354 list_splice_tail(&splice, &fs_info->delalloc_roots);
8355 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8356 }
8ccf6f19 8357 return ret;
ea8c2819
CM		 8358}
8359
39279cc3
CM		 8360static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8361 const char *symname)
8362{
8363 struct btrfs_trans_handle *trans;
8364 struct btrfs_root *root = BTRFS_I(dir)->root;
8365 struct btrfs_path *path;
8366 struct btrfs_key key;
1832a6d5 8367 struct inode *inode = NULL;
39279cc3
CM		 8368 int err;
8369 int drop_inode = 0;
8370 u64 objectid;
00e4e6b3 8371 u64 index = 0 ;
39279cc3
CM		 8372 int name_len;
8373 int datasize;
5f39d397 8374 unsigned long ptr;
39279cc3 8375 struct btrfs_file_extent_item *ei;
5f39d397 8376 struct extent_buffer *leaf;
39279cc3
CM		 8377
8378 name_len = strlen(symname) + 1;
8379 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8380 return -ENAMETOOLONG;
1832a6d5 8381
9ed74f2d
JB		 8382 /*
8383 * 2 items for inode item and ref
8384 * 2 items for dir items
8385 * 1 item for xattr if selinux is on
8386 */
a22285a6
YZ		 8387 trans = btrfs_start_transaction(root, 5);
8388 if (IS_ERR(trans))
8389 return PTR_ERR(trans);
1832a6d5 8390
581bb050
LZ		 8391 err = btrfs_find_free_ino(root, &objectid);
8392 if (err)
8393 goto out_unlock;
8394
aec7477b 8395 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8396 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8397 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8398 if (IS_ERR(inode)) {
8399 err = PTR_ERR(inode);
39279cc3 8400 goto out_unlock;
7cf96da3 8401 }
39279cc3 8402
2a7dba39 8403 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8404 if (err) {
8405 drop_inode = 1;
8406 goto out_unlock;
8407 }
8408
ad19db71
CS		 8409 /*
8410 * If the active LSM wants to access the inode during
8411 * d_instantiate it needs these. Smack checks to see
8412 * if the filesystem supports xattrs by looking at the
8413 * ops vector.
8414 */
8415 inode->i_fop = &btrfs_file_operations;
8416 inode->i_op = &btrfs_file_inode_operations;
8417
a1b075d2 8418 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8419 if (err)
8420 drop_inode = 1;
8421 else {
8422 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8423 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8424 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8425 }
39279cc3
CM		 8426 if (drop_inode)
8427 goto out_unlock;
8428
8429 path = btrfs_alloc_path();
d8926bb3
MF		 8430 if (!path) {
8431 err = -ENOMEM;
8432 drop_inode = 1;
8433 goto out_unlock;
8434 }
33345d01 8435 key.objectid = btrfs_ino(inode);
39279cc3 8436 key.offset = 0;
39279cc3
CM		 8437 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8438 datasize = btrfs_file_extent_calc_inline_size(name_len);
8439 err = btrfs_insert_empty_item(trans, root, path, &key,
8440 datasize);
54aa1f4d
CM		 8441 if (err) {
8442 drop_inode = 1;
b0839166 8443 btrfs_free_path(path);
54aa1f4d
CM		 8444 goto out_unlock;
8445 }
5f39d397
CM		 8446 leaf = path->nodes[0];
8447 ei = btrfs_item_ptr(leaf, path->slots[0],
8448 struct btrfs_file_extent_item);
8449 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8450 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8451 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8452 btrfs_set_file_extent_encryption(leaf, ei, 0);
8453 btrfs_set_file_extent_compression(leaf, ei, 0);
8454 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8455 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8456
39279cc3 8457 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8458 write_extent_buffer(leaf, symname, ptr, name_len);
8459 btrfs_mark_buffer_dirty(leaf);
39279cc3 8460 btrfs_free_path(path);
5f39d397 8461
39279cc3
CM		 8462 inode->i_op = &btrfs_symlink_inode_operations;
8463 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8464 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8465 inode_set_bytes(inode, name_len);
dbe674a9 8466 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 8467 err = btrfs_update_inode(trans, root, inode);
8468 if (err)
8469 drop_inode = 1;
39279cc3
CM		 8470
8471out_unlock:
08c422c2
AV		 8472 if (!err)
8473 d_instantiate(dentry, inode);
7ad85bb7 8474 btrfs_end_transaction(trans, root);
39279cc3
CM		 8475 if (drop_inode) {
8476 inode_dec_link_count(inode);
8477 iput(inode);
8478 }
b53d3f5d 8479 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8480 return err;
8481}
16432985 8482
0af3d00b
JB		 8483static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8484 u64 start, u64 num_bytes, u64 min_size,
8485 loff_t actual_len, u64 *alloc_hint,
8486 struct btrfs_trans_handle *trans)
d899e052 8487{
5dc562c5
JB		 8488 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8489 struct extent_map *em;
d899e052
YZ		 8490 struct btrfs_root *root = BTRFS_I(inode)->root;
8491 struct btrfs_key ins;
d899e052 8492 u64 cur_offset = start;
55a61d1d 8493 u64 i_size;
154ea289 8494 u64 cur_bytes;
d899e052 8495 int ret = 0;
0af3d00b 8496 bool own_trans = true;
d899e052 8497
0af3d00b
JB		 8498 if (trans)
8499 own_trans = false;
d899e052 8500 while (num_bytes > 0) {
0af3d00b
JB		 8501 if (own_trans) {
8502 trans = btrfs_start_transaction(root, 3);
8503 if (IS_ERR(trans)) {
8504 ret = PTR_ERR(trans);
8505 break;
8506 }
5a303d5d
YZ		 8507 }
8508
154ea289
CM		 8509 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8510 cur_bytes = max(cur_bytes, min_size);
8511 ret = btrfs_reserve_extent(trans, root, cur_bytes,
24542bf7 8512 min_size, 0, *alloc_hint, &ins, 1);
5a303d5d 8513 if (ret) {
0af3d00b
JB		 8514 if (own_trans)
8515 btrfs_end_transaction(trans, root);
a22285a6 8516 break;
d899e052 8517 }
5a303d5d 8518
d899e052
YZ		 8519 ret = insert_reserved_file_extent(trans, inode,
8520 cur_offset, ins.objectid,
8521 ins.offset, ins.offset,
920bbbfb 8522 ins.offset, 0, 0, 0,
d899e052 8523 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 8524 if (ret) {
8525 btrfs_abort_transaction(trans, root, ret);
8526 if (own_trans)
8527 btrfs_end_transaction(trans, root);
8528 break;
8529 }
a1ed835e
CM		 8530 btrfs_drop_extent_cache(inode, cur_offset,
8531 cur_offset + ins.offset -1, 0);
5a303d5d 8532
5dc562c5
JB		 8533 em = alloc_extent_map();
8534 if (!em) {
8535 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8536 &BTRFS_I(inode)->runtime_flags);
8537 goto next;
8538 }
8539
8540 em->start = cur_offset;
8541 em->orig_start = cur_offset;
8542 em->len = ins.offset;
8543 em->block_start = ins.objectid;
8544 em->block_len = ins.offset;
b4939680 8545 em->orig_block_len = ins.offset;
cc95bef6 8546 em->ram_bytes = ins.offset;
5dc562c5
JB		 8547 em->bdev = root->fs_info->fs_devices->latest_bdev;
8548 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8549 em->generation = trans->transid;
8550
8551 while (1) {
8552 write_lock(&em_tree->lock);
09a2a8f9 8553 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8554 write_unlock(&em_tree->lock);
8555 if (ret != -EEXIST)
8556 break;
8557 btrfs_drop_extent_cache(inode, cur_offset,
8558 cur_offset + ins.offset - 1,
8559 0);
8560 }
8561 free_extent_map(em);
8562next:
d899e052
YZ		 8563 num_bytes -= ins.offset;
8564 cur_offset += ins.offset;
efa56464 8565 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8566
0c4d2d95 8567 inode_inc_iversion(inode);
d899e052 8568 inode->i_ctime = CURRENT_TIME;
6cbff00f 8569 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8570 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8571 (actual_len > inode->i_size) &&
8572 (cur_offset > inode->i_size)) {
d1ea6a61 8573 if (cur_offset > actual_len)
55a61d1d 8574 i_size = actual_len;
d1ea6a61 8575 else
55a61d1d
JB		 8576 i_size = cur_offset;
8577 i_size_write(inode, i_size);
8578 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8579 }
8580
d899e052 8581 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8582
8583 if (ret) {
8584 btrfs_abort_transaction(trans, root, ret);
8585 if (own_trans)
8586 btrfs_end_transaction(trans, root);
8587 break;
8588 }
d899e052 8589
0af3d00b
JB		 8590 if (own_trans)
8591 btrfs_end_transaction(trans, root);
5a303d5d 8592 }
d899e052
YZ		 8593 return ret;
8594}
8595
0af3d00b
JB		 8596int btrfs_prealloc_file_range(struct inode *inode, int mode,
8597 u64 start, u64 num_bytes, u64 min_size,
8598 loff_t actual_len, u64 *alloc_hint)
8599{
8600 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8601 min_size, actual_len, alloc_hint,
8602 NULL);
8603}
8604
8605int btrfs_prealloc_file_range_trans(struct inode *inode,
8606 struct btrfs_trans_handle *trans, int mode,
8607 u64 start, u64 num_bytes, u64 min_size,
8608 loff_t actual_len, u64 *alloc_hint)
8609{
8610 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8611 min_size, actual_len, alloc_hint, trans);
8612}
8613
e6dcd2dc
CM		 8614static int btrfs_set_page_dirty(struct page *page)
8615{
e6dcd2dc
CM		 8616 return __set_page_dirty_nobuffers(page);
8617}
8618
10556cb2 8619static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8620{
b83cc969 8621 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8622 umode_t mode = inode->i_mode;
b83cc969 8623
cb6db4e5
JM		 8624 if (mask & MAY_WRITE &&
8625 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8626 if (btrfs_root_readonly(root))
8627 return -EROFS;
8628 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8629 return -EACCES;
8630 }
2830ba7f 8631 return generic_permission(inode, mask);
fdebe2bd 8632}
39279cc3 8633
6e1d5dcc 8634static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8635 .getattr = btrfs_getattr,
39279cc3
CM		 8636 .lookup = btrfs_lookup,
8637 .create = btrfs_create,
8638 .unlink = btrfs_unlink,
8639 .link = btrfs_link,
8640 .mkdir = btrfs_mkdir,
8641 .rmdir = btrfs_rmdir,
8642 .rename = btrfs_rename,
8643 .symlink = btrfs_symlink,
8644 .setattr = btrfs_setattr,
618e21d5 8645 .mknod = btrfs_mknod,
95819c05
CH		 8646 .setxattr = btrfs_setxattr,
8647 .getxattr = btrfs_getxattr,
5103e947 8648 .listxattr = btrfs_listxattr,
95819c05 8649 .removexattr = btrfs_removexattr,
fdebe2bd 8650 .permission = btrfs_permission,
4e34e719 8651 .get_acl = btrfs_get_acl,
39279cc3 8652};
6e1d5dcc 8653static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8654 .lookup = btrfs_lookup,
fdebe2bd 8655 .permission = btrfs_permission,
4e34e719 8656 .get_acl = btrfs_get_acl,
39279cc3 8657};
76dda93c 8658
828c0950 8659static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8660 .llseek = generic_file_llseek,
8661 .read = generic_read_dir,
cbdf5a24 8662 .readdir = btrfs_real_readdir,
34287aa3 8663 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8664#ifdef CONFIG_COMPAT
34287aa3 8665 .compat_ioctl = btrfs_ioctl,
39279cc3 8666#endif
6bf13c0c 8667 .release = btrfs_release_file,
e02119d5 8668 .fsync = btrfs_sync_file,
39279cc3
CM		 8669};
8670
d1310b2e 8671static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8672 .fill_delalloc = run_delalloc_range,
065631f6 8673 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8674 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8675 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8676 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8677 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8678 .set_bit_hook = btrfs_set_bit_hook,
8679 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8680 .merge_extent_hook = btrfs_merge_extent_hook,
8681 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8682};
8683
35054394
CM		 8684/*
8685 * btrfs doesn't support the bmap operation because swapfiles
8686 * use bmap to make a mapping of extents in the file. They assume
8687 * these extents won't change over the life of the file and they
8688 * use the bmap result to do IO directly to the drive.
8689 *
8690 * the btrfs bmap call would return logical addresses that aren't
8691 * suitable for IO and they also will change frequently as COW
8692 * operations happen. So, swapfile + btrfs == corruption.
8693 *
8694 * For now we're avoiding this by dropping bmap.
8695 */
7f09410b 8696static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8697 .readpage = btrfs_readpage,
8698 .writepage = btrfs_writepage,
b293f02e 8699 .writepages = btrfs_writepages,
3ab2fb5a 8700 .readpages = btrfs_readpages,
16432985 8701 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8702 .invalidatepage = btrfs_invalidatepage,
8703 .releasepage = btrfs_releasepage,
e6dcd2dc 8704 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8705 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8706};
8707
7f09410b 8708static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8709 .readpage = btrfs_readpage,
8710 .writepage = btrfs_writepage,
2bf5a725
CM		 8711 .invalidatepage = btrfs_invalidatepage,
8712 .releasepage = btrfs_releasepage,
39279cc3
CM		 8713};
8714
6e1d5dcc 8715static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8716 .getattr = btrfs_getattr,
8717 .setattr = btrfs_setattr,
95819c05
CH		 8718 .setxattr = btrfs_setxattr,
8719 .getxattr = btrfs_getxattr,
5103e947 8720 .listxattr = btrfs_listxattr,
95819c05 8721 .removexattr = btrfs_removexattr,
fdebe2bd 8722 .permission = btrfs_permission,
1506fcc8 8723 .fiemap = btrfs_fiemap,
4e34e719 8724 .get_acl = btrfs_get_acl,
e41f941a 8725 .update_time = btrfs_update_time,
39279cc3 8726};
6e1d5dcc 8727static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8728 .getattr = btrfs_getattr,
8729 .setattr = btrfs_setattr,
fdebe2bd 8730 .permission = btrfs_permission,
95819c05
CH		 8731 .setxattr = btrfs_setxattr,
8732 .getxattr = btrfs_getxattr,
33268eaf 8733 .listxattr = btrfs_listxattr,
95819c05 8734 .removexattr = btrfs_removexattr,
4e34e719 8735 .get_acl = btrfs_get_acl,
e41f941a 8736 .update_time = btrfs_update_time,
618e21d5 8737};
6e1d5dcc 8738static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8739 .readlink = generic_readlink,
8740 .follow_link = page_follow_link_light,
8741 .put_link = page_put_link,
f209561a 8742 .getattr = btrfs_getattr,
22c44fe6 8743 .setattr = btrfs_setattr,
fdebe2bd 8744 .permission = btrfs_permission,
0279b4cd
JO		 8745 .setxattr = btrfs_setxattr,
8746 .getxattr = btrfs_getxattr,
8747 .listxattr = btrfs_listxattr,
8748 .removexattr = btrfs_removexattr,
4e34e719 8749 .get_acl = btrfs_get_acl,
e41f941a 8750 .update_time = btrfs_update_time,
39279cc3 8751};
76dda93c 8752
82d339d9 8753const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8754 .d_delete = btrfs_dentry_delete,
b4aff1f8 8755 .d_release = btrfs_dentry_release,
76dda93c 8756};
Linux 6.x block layer and io_uring tree(s)