projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix num_workers_starting bug and other bugs in async thread
[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>
4b4e25f2 41#include "compat.h"
39279cc3
CM		 42#include "ctree.h"
43#include "disk-io.h"
44#include "transaction.h"
45#include "btrfs_inode.h"
46#include "ioctl.h"
47#include "print-tree.h"
e6dcd2dc 48#include "ordered-data.h"
95819c05 49#include "xattr.h"
e02119d5 50#include "tree-log.h"
4a54c8c1 51#include "volumes.h"
c8b97818 52#include "compression.h"
b4ce94de 53#include "locking.h"
dc89e982 54#include "free-space-cache.h"
581bb050 55#include "inode-map.h"
39279cc3
CM		 56
57struct btrfs_iget_args {
58 u64 ino;
59 struct btrfs_root *root;
60};
61
6e1d5dcc
AD		 62static const struct inode_operations btrfs_dir_inode_operations;
63static const struct inode_operations btrfs_symlink_inode_operations;
64static const struct inode_operations btrfs_dir_ro_inode_operations;
65static const struct inode_operations btrfs_special_inode_operations;
66static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 67static const struct address_space_operations btrfs_aops;
68static const struct address_space_operations btrfs_symlink_aops;
828c0950 69static const struct file_operations btrfs_dir_file_operations;
d1310b2e 70static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 71
72static struct kmem_cache *btrfs_inode_cachep;
73struct kmem_cache *btrfs_trans_handle_cachep;
74struct kmem_cache *btrfs_transaction_cachep;
39279cc3 75struct kmem_cache *btrfs_path_cachep;
dc89e982 76struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 77
78#define S_SHIFT 12
79static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87};
88
a41ad394
JB		 89static int btrfs_setsize(struct inode *inode, loff_t newsize);
90static int btrfs_truncate(struct inode *inode);
c8b97818 91static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 92static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
2115133f
CM		 96static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct inode *inode);
7b128766 98
f34f57a3 99static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 100 struct inode *inode, struct inode *dir,
101 const struct qstr *qstr)
0279b4cd
JO		 102{
103 int err;
104
f34f57a3 105 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 106 if (!err)
2a7dba39 107 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 108 return err;
109}
110
c8b97818
CM		 111/*
112 * this does all the hard work for inserting an inline extent into
113 * the btree. The caller should have done a btrfs_drop_extents so that
114 * no overlapping inline items exist in the btree
115 */
d397712b 116static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 117 struct btrfs_root *root, struct inode *inode,
118 u64 start, size_t size, size_t compressed_size,
fe3f566c 119 int compress_type,
c8b97818
CM		 120 struct page **compressed_pages)
121{
122 struct btrfs_key key;
123 struct btrfs_path *path;
124 struct extent_buffer *leaf;
125 struct page *page = NULL;
126 char *kaddr;
127 unsigned long ptr;
128 struct btrfs_file_extent_item *ei;
129 int err = 0;
130 int ret;
131 size_t cur_size = size;
132 size_t datasize;
133 unsigned long offset;
c8b97818 134
fe3f566c 135 if (compressed_size && compressed_pages)
c8b97818 136 cur_size = compressed_size;
c8b97818 137
d397712b
CM		 138 path = btrfs_alloc_path();
139 if (!path)
c8b97818
CM		 140 return -ENOMEM;
141
b9473439 142 path->leave_spinning = 1;
c8b97818 143
33345d01 144 key.objectid = btrfs_ino(inode);
c8b97818
CM		 145 key.offset = start;
146 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 147 datasize = btrfs_file_extent_calc_inline_size(cur_size);
148
149 inode_add_bytes(inode, size);
150 ret = btrfs_insert_empty_item(trans, root, path, &key,
151 datasize);
152 BUG_ON(ret);
153 if (ret) {
154 err = ret;
c8b97818
CM		 155 goto fail;
156 }
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
166
261507a0 167 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 168 struct page *cpage;
169 int i = 0;
d397712b 170 while (compressed_size > 0) {
c8b97818 171 cpage = compressed_pages[i];
5b050f04 172 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 173 PAGE_CACHE_SIZE);
174
b9473439 175 kaddr = kmap_atomic(cpage, KM_USER0);
c8b97818 176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
b9473439 177 kunmap_atomic(kaddr, KM_USER0);
c8b97818
CM		 178
179 i++;
180 ptr += cur_size;
181 compressed_size -= cur_size;
182 }
183 btrfs_set_file_extent_compression(leaf, ei,
261507a0 184 compress_type);
c8b97818
CM		 185 } else {
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page, KM_USER0);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr, KM_USER0);
193 page_cache_release(page);
194 }
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
197
c2167754
YZ		 198 /*
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
202 *
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
206 */
c8b97818
CM		 207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 btrfs_update_inode(trans, root, inode);
c2167754 209
c8b97818
CM		 210 return 0;
211fail:
212 btrfs_free_path(path);
213 return err;
214}
215
216
217/*
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
221 */
7f366cfe 222static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
fe3f566c 225 size_t compressed_size, int compress_type,
c8b97818
CM		 226 struct page **compressed_pages)
227{
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
233 u64 hint_byte;
234 u64 data_len = inline_len;
235 int ret;
236
237 if (compressed_size)
238 data_len = compressed_size;
239
240 if (start > 0 ||
70b99e69 241 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243 (!compressed_size &&
244 (actual_end & (root->sectorsize - 1)) == 0) ||
245 end + 1 < isize ||
246 data_len > root->fs_info->max_inline) {
247 return 1;
248 }
249
920bbbfb 250 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
a1ed835e 251 &hint_byte, 1);
c8b97818
CM		 252 BUG_ON(ret);
253
254 if (isize > actual_end)
255 inline_len = min_t(u64, isize, actual_end);
256 ret = insert_inline_extent(trans, root, inode, start,
257 inline_len, compressed_size,
fe3f566c 258 compress_type, compressed_pages);
c8b97818 259 BUG_ON(ret);
0ca1f7ce 260 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 261 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 262 return 0;
263}
264
771ed689
CM		 265struct async_extent {
266 u64 start;
267 u64 ram_size;
268 u64 compressed_size;
269 struct page **pages;
270 unsigned long nr_pages;
261507a0 271 int compress_type;
771ed689
CM		 272 struct list_head list;
273};
274
275struct async_cow {
276 struct inode *inode;
277 struct btrfs_root *root;
278 struct page *locked_page;
279 u64 start;
280 u64 end;
281 struct list_head extents;
282 struct btrfs_work work;
283};
284
285static noinline int add_async_extent(struct async_cow *cow,
286 u64 start, u64 ram_size,
287 u64 compressed_size,
288 struct page **pages,
261507a0
LZ		 289 unsigned long nr_pages,
290 int compress_type)
771ed689
CM		 291{
292 struct async_extent *async_extent;
293
294 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
dac97e51 295 BUG_ON(!async_extent);
771ed689
CM		 296 async_extent->start = start;
297 async_extent->ram_size = ram_size;
298 async_extent->compressed_size = compressed_size;
299 async_extent->pages = pages;
300 async_extent->nr_pages = nr_pages;
261507a0 301 async_extent->compress_type = compress_type;
771ed689
CM		 302 list_add_tail(&async_extent->list, &cow->extents);
303 return 0;
304}
305
d352ac68 306/*
771ed689
CM		 307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
c8b97818 310 *
771ed689
CM		 311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
c8b97818 316 *
771ed689
CM		 317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
d352ac68 321 */
771ed689
CM		 322static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
324 u64 start, u64 end,
325 struct async_cow *async_cow,
326 int *num_added)
b888db2b
CM		 327{
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
db94535d 330 u64 num_bytes;
db94535d 331 u64 blocksize = root->sectorsize;
c8b97818 332 u64 actual_end;
42dc7bab 333 u64 isize = i_size_read(inode);
e6dcd2dc 334 int ret = 0;
c8b97818
CM		 335 struct page **pages = NULL;
336 unsigned long nr_pages;
337 unsigned long nr_pages_ret = 0;
338 unsigned long total_compressed = 0;
339 unsigned long total_in = 0;
340 unsigned long max_compressed = 128 * 1024;
771ed689 341 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 342 int i;
343 int will_compress;
261507a0 344 int compress_type = root->fs_info->compress_type;
b888db2b 345
4cb5300b
CM		 346 /* if this is a small write inside eof, kick off a defragbot */
347 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
348 btrfs_add_inode_defrag(NULL, inode);
349
42dc7bab 350 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 351again:
352 will_compress = 0;
353 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
354 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 355
f03d9301
CM		 356 /*
357 * we don't want to send crud past the end of i_size through
358 * compression, that's just a waste of CPU time. So, if the
359 * end of the file is before the start of our current
360 * requested range of bytes, we bail out to the uncompressed
361 * cleanup code that can deal with all of this.
362 *
363 * It isn't really the fastest way to fix things, but this is a
364 * very uncommon corner.
365 */
366 if (actual_end <= start)
367 goto cleanup_and_bail_uncompressed;
368
c8b97818
CM		 369 total_compressed = actual_end - start;
370
371 /* we want to make sure that amount of ram required to uncompress
372 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 373 * of a compressed extent to 128k. This is a crucial number
374 * because it also controls how easily we can spread reads across
375 * cpus for decompression.
376 *
377 * We also want to make sure the amount of IO required to do
378 * a random read is reasonably small, so we limit the size of
379 * a compressed extent to 128k.
c8b97818
CM		 380 */
381 total_compressed = min(total_compressed, max_uncompressed);
db94535d 382 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 383 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 384 total_in = 0;
385 ret = 0;
db94535d 386
771ed689
CM		 387 /*
388 * we do compression for mount -o compress and when the
389 * inode has not been flagged as nocompress. This flag can
390 * change at any time if we discover bad compression ratios.
c8b97818 391 */
6cbff00f 392 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 393 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 394 (BTRFS_I(inode)->force_compress) ||
395 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 396 WARN_ON(pages);
cfbc246e 397 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 398 if (!pages) {
399 /* just bail out to the uncompressed code */
400 goto cont;
401 }
c8b97818 402
261507a0
LZ		 403 if (BTRFS_I(inode)->force_compress)
404 compress_type = BTRFS_I(inode)->force_compress;
405
406 ret = btrfs_compress_pages(compress_type,
407 inode->i_mapping, start,
408 total_compressed, pages,
409 nr_pages, &nr_pages_ret,
410 &total_in,
411 &total_compressed,
412 max_compressed);
c8b97818
CM		 413
414 if (!ret) {
415 unsigned long offset = total_compressed &
416 (PAGE_CACHE_SIZE - 1);
417 struct page *page = pages[nr_pages_ret - 1];
418 char *kaddr;
419
420 /* zero the tail end of the last page, we might be
421 * sending it down to disk
422 */
423 if (offset) {
424 kaddr = kmap_atomic(page, KM_USER0);
425 memset(kaddr + offset, 0,
426 PAGE_CACHE_SIZE - offset);
427 kunmap_atomic(kaddr, KM_USER0);
428 }
429 will_compress = 1;
430 }
431 }
560f7d75 432cont:
c8b97818 433 if (start == 0) {
7a7eaa40 434 trans = btrfs_join_transaction(root);
3612b495 435 BUG_ON(IS_ERR(trans));
0ca1f7ce 436 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 437
c8b97818 438 /* lets try to make an inline extent */
771ed689 439 if (ret || total_in < (actual_end - start)) {
c8b97818 440 /* we didn't compress the entire range, try
771ed689 441 * to make an uncompressed inline extent.
c8b97818
CM		 442 */
443 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 444 start, end, 0, 0, NULL);
c8b97818 445 } else {
771ed689 446 /* try making a compressed inline extent */
c8b97818
CM		 447 ret = cow_file_range_inline(trans, root, inode,
448 start, end,
fe3f566c
LZ		 449 total_compressed,
450 compress_type, pages);
c8b97818
CM		 451 }
452 if (ret == 0) {
771ed689
CM		 453 /*
454 * inline extent creation worked, we don't need
455 * to create any more async work items. Unlock
456 * and free up our temp pages.
457 */
c8b97818 458 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 459 &BTRFS_I(inode)->io_tree,
460 start, end, NULL,
461 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 462 EXTENT_CLEAR_DELALLOC |
a791e35e 463 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 464
465 btrfs_end_transaction(trans, root);
c8b97818
CM		 466 goto free_pages_out;
467 }
c2167754 468 btrfs_end_transaction(trans, root);
c8b97818
CM		 469 }
470
471 if (will_compress) {
472 /*
473 * we aren't doing an inline extent round the compressed size
474 * up to a block size boundary so the allocator does sane
475 * things
476 */
477 total_compressed = (total_compressed + blocksize - 1) &
478 ~(blocksize - 1);
479
480 /*
481 * one last check to make sure the compression is really a
482 * win, compare the page count read with the blocks on disk
483 */
484 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
485 ~(PAGE_CACHE_SIZE - 1);
486 if (total_compressed >= total_in) {
487 will_compress = 0;
488 } else {
c8b97818
CM		 489 num_bytes = total_in;
490 }
491 }
492 if (!will_compress && pages) {
493 /*
494 * the compression code ran but failed to make things smaller,
495 * free any pages it allocated and our page pointer array
496 */
497 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 498 WARN_ON(pages[i]->mapping);
c8b97818
CM		 499 page_cache_release(pages[i]);
500 }
501 kfree(pages);
502 pages = NULL;
503 total_compressed = 0;
504 nr_pages_ret = 0;
505
506 /* flag the file so we don't compress in the future */
1e701a32
CM		 507 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
508 !(BTRFS_I(inode)->force_compress)) {
a555f810 509 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 510 }
c8b97818 511 }
771ed689
CM		 512 if (will_compress) {
513 *num_added += 1;
c8b97818 514
771ed689
CM		 515 /* the async work queues will take care of doing actual
516 * allocation on disk for these compressed pages,
517 * and will submit them to the elevator.
518 */
519 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 520 total_compressed, pages, nr_pages_ret,
521 compress_type);
179e29e4 522
24ae6365 523 if (start + num_bytes < end) {
771ed689
CM		 524 start += num_bytes;
525 pages = NULL;
526 cond_resched();
527 goto again;
528 }
529 } else {
f03d9301 530cleanup_and_bail_uncompressed:
771ed689
CM		 531 /*
532 * No compression, but we still need to write the pages in
533 * the file we've been given so far. redirty the locked
534 * page if it corresponds to our extent and set things up
535 * for the async work queue to run cow_file_range to do
536 * the normal delalloc dance
537 */
538 if (page_offset(locked_page) >= start &&
539 page_offset(locked_page) <= end) {
540 __set_page_dirty_nobuffers(locked_page);
541 /* unlocked later on in the async handlers */
542 }
261507a0
LZ		 543 add_async_extent(async_cow, start, end - start + 1,
544 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 545 *num_added += 1;
546 }
3b951516 547
771ed689
CM		 548out:
549 return 0;
550
551free_pages_out:
552 for (i = 0; i < nr_pages_ret; i++) {
553 WARN_ON(pages[i]->mapping);
554 page_cache_release(pages[i]);
555 }
d397712b 556 kfree(pages);
771ed689
CM		 557
558 goto out;
559}
560
561/*
562 * phase two of compressed writeback. This is the ordered portion
563 * of the code, which only gets called in the order the work was
564 * queued. We walk all the async extents created by compress_file_range
565 * and send them down to the disk.
566 */
567static noinline int submit_compressed_extents(struct inode *inode,
568 struct async_cow *async_cow)
569{
570 struct async_extent *async_extent;
571 u64 alloc_hint = 0;
572 struct btrfs_trans_handle *trans;
573 struct btrfs_key ins;
574 struct extent_map *em;
575 struct btrfs_root *root = BTRFS_I(inode)->root;
576 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
577 struct extent_io_tree *io_tree;
f5a84ee3 578 int ret = 0;
771ed689
CM		 579
580 if (list_empty(&async_cow->extents))
581 return 0;
582
771ed689 583
d397712b 584 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 585 async_extent = list_entry(async_cow->extents.next,
586 struct async_extent, list);
587 list_del(&async_extent->list);
c8b97818 588
771ed689
CM		 589 io_tree = &BTRFS_I(inode)->io_tree;
590
f5a84ee3 591retry:
771ed689
CM		 592 /* did the compression code fall back to uncompressed IO? */
593 if (!async_extent->pages) {
594 int page_started = 0;
595 unsigned long nr_written = 0;
596
597 lock_extent(io_tree, async_extent->start,
2ac55d41
JB		 598 async_extent->start +
599 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 600
601 /* allocate blocks */
f5a84ee3
JB		 602 ret = cow_file_range(inode, async_cow->locked_page,
603 async_extent->start,
604 async_extent->start +
605 async_extent->ram_size - 1,
606 &page_started, &nr_written, 0);
771ed689
CM		 607
608 /*
609 * if page_started, cow_file_range inserted an
610 * inline extent and took care of all the unlocking
611 * and IO for us. Otherwise, we need to submit
612 * all those pages down to the drive.
613 */
f5a84ee3 614 if (!page_started && !ret)
771ed689
CM		 615 extent_write_locked_range(io_tree,
616 inode, async_extent->start,
d397712b 617 async_extent->start +
771ed689
CM		 618 async_extent->ram_size - 1,
619 btrfs_get_extent,
620 WB_SYNC_ALL);
621 kfree(async_extent);
622 cond_resched();
623 continue;
624 }
625
626 lock_extent(io_tree, async_extent->start,
627 async_extent->start + async_extent->ram_size - 1,
628 GFP_NOFS);
771ed689 629
7a7eaa40 630 trans = btrfs_join_transaction(root);
3612b495 631 BUG_ON(IS_ERR(trans));
74b21075 632 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689
CM		 633 ret = btrfs_reserve_extent(trans, root,
634 async_extent->compressed_size,
635 async_extent->compressed_size,
636 0, alloc_hint,
637 (u64)-1, &ins, 1);
c2167754
YZ		 638 btrfs_end_transaction(trans, root);
639
f5a84ee3
JB		 640 if (ret) {
641 int i;
642 for (i = 0; i < async_extent->nr_pages; i++) {
643 WARN_ON(async_extent->pages[i]->mapping);
644 page_cache_release(async_extent->pages[i]);
645 }
646 kfree(async_extent->pages);
647 async_extent->nr_pages = 0;
648 async_extent->pages = NULL;
649 unlock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1, GFP_NOFS);
652 goto retry;
653 }
654
c2167754
YZ		 655 /*
656 * here we're doing allocation and writeback of the
657 * compressed pages
658 */
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
662
172ddd60 663 em = alloc_extent_map();
c26a9203 664 BUG_ON(!em);
771ed689
CM		 665 em->start = async_extent->start;
666 em->len = async_extent->ram_size;
445a6944 667 em->orig_start = em->start;
c8b97818 668
771ed689
CM		 669 em->block_start = ins.objectid;
670 em->block_len = ins.offset;
671 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 672 em->compress_type = async_extent->compress_type;
771ed689
CM		 673 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
675
d397712b 676 while (1) {
890871be 677 write_lock(&em_tree->lock);
771ed689 678 ret = add_extent_mapping(em_tree, em);
890871be 679 write_unlock(&em_tree->lock);
771ed689
CM		 680 if (ret != -EEXIST) {
681 free_extent_map(em);
682 break;
683 }
684 btrfs_drop_extent_cache(inode, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1, 0);
687 }
688
261507a0
LZ		 689 ret = btrfs_add_ordered_extent_compress(inode,
690 async_extent->start,
691 ins.objectid,
692 async_extent->ram_size,
693 ins.offset,
694 BTRFS_ORDERED_COMPRESSED,
695 async_extent->compress_type);
771ed689
CM		 696 BUG_ON(ret);
697
771ed689
CM		 698 /*
699 * clear dirty, set writeback and unlock the pages.
700 */
701 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 702 &BTRFS_I(inode)->io_tree,
703 async_extent->start,
704 async_extent->start +
705 async_extent->ram_size - 1,
706 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707 EXTENT_CLEAR_UNLOCK |
a3429ab7 708 EXTENT_CLEAR_DELALLOC |
a791e35e 709 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 710
711 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 712 async_extent->start,
713 async_extent->ram_size,
714 ins.objectid,
715 ins.offset, async_extent->pages,
716 async_extent->nr_pages);
771ed689
CM		 717
718 BUG_ON(ret);
771ed689
CM		 719 alloc_hint = ins.objectid + ins.offset;
720 kfree(async_extent);
721 cond_resched();
722 }
723
771ed689
CM		 724 return 0;
725}
726
4b46fce2
JB		 727static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
728 u64 num_bytes)
729{
730 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731 struct extent_map *em;
732 u64 alloc_hint = 0;
733
734 read_lock(&em_tree->lock);
735 em = search_extent_mapping(em_tree, start, num_bytes);
736 if (em) {
737 /*
738 * if block start isn't an actual block number then find the
739 * first block in this inode and use that as a hint. If that
740 * block is also bogus then just don't worry about it.
741 */
742 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
743 free_extent_map(em);
744 em = search_extent_mapping(em_tree, 0, 0);
745 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746 alloc_hint = em->block_start;
747 if (em)
748 free_extent_map(em);
749 } else {
750 alloc_hint = em->block_start;
751 free_extent_map(em);
752 }
753 }
754 read_unlock(&em_tree->lock);
755
756 return alloc_hint;
757}
758
771ed689
CM		 759/*
760 * when extent_io.c finds a delayed allocation range in the file,
761 * the call backs end up in this code. The basic idea is to
762 * allocate extents on disk for the range, and create ordered data structs
763 * in ram to track those extents.
764 *
765 * locked_page is the page that writepage had locked already. We use
766 * it to make sure we don't do extra locks or unlocks.
767 *
768 * *page_started is set to one if we unlock locked_page and do everything
769 * required to start IO on it. It may be clean and already done with
770 * IO when we return.
771 */
772static noinline int cow_file_range(struct inode *inode,
773 struct page *locked_page,
774 u64 start, u64 end, int *page_started,
775 unsigned long *nr_written,
776 int unlock)
777{
778 struct btrfs_root *root = BTRFS_I(inode)->root;
779 struct btrfs_trans_handle *trans;
780 u64 alloc_hint = 0;
781 u64 num_bytes;
782 unsigned long ram_size;
783 u64 disk_num_bytes;
784 u64 cur_alloc_size;
785 u64 blocksize = root->sectorsize;
771ed689
CM		 786 struct btrfs_key ins;
787 struct extent_map *em;
788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
789 int ret = 0;
790
2cf8572d 791 BUG_ON(btrfs_is_free_space_inode(root, inode));
7a7eaa40 792 trans = btrfs_join_transaction(root);
3612b495 793 BUG_ON(IS_ERR(trans));
0ca1f7ce 794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 795
771ed689
CM		 796 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797 num_bytes = max(blocksize, num_bytes);
798 disk_num_bytes = num_bytes;
799 ret = 0;
800
4cb5300b
CM		 801 /* if this is a small write inside eof, kick off defrag */
802 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803 btrfs_add_inode_defrag(trans, inode);
804
771ed689
CM		 805 if (start == 0) {
806 /* lets try to make an inline extent */
807 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 808 start, end, 0, 0, NULL);
771ed689
CM		 809 if (ret == 0) {
810 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 811 &BTRFS_I(inode)->io_tree,
812 start, end, NULL,
813 EXTENT_CLEAR_UNLOCK_PAGE |
814 EXTENT_CLEAR_UNLOCK |
815 EXTENT_CLEAR_DELALLOC |
816 EXTENT_CLEAR_DIRTY |
817 EXTENT_SET_WRITEBACK |
818 EXTENT_END_WRITEBACK);
c2167754 819
771ed689
CM		 820 *nr_written = *nr_written +
821 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
822 *page_started = 1;
823 ret = 0;
824 goto out;
825 }
826 }
827
828 BUG_ON(disk_num_bytes >
6c41761f 829 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 830
4b46fce2 831 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 832 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
833
d397712b 834 while (disk_num_bytes > 0) {
a791e35e
CM		 835 unsigned long op;
836
287a0ab9 837 cur_alloc_size = disk_num_bytes;
e6dcd2dc 838 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 839 root->sectorsize, 0, alloc_hint,
e6dcd2dc 840 (u64)-1, &ins, 1);
d397712b
CM		 841 BUG_ON(ret);
842
172ddd60 843 em = alloc_extent_map();
c26a9203 844 BUG_ON(!em);
e6dcd2dc 845 em->start = start;
445a6944 846 em->orig_start = em->start;
771ed689
CM		 847 ram_size = ins.offset;
848 em->len = ins.offset;
c8b97818 849
e6dcd2dc 850 em->block_start = ins.objectid;
c8b97818 851 em->block_len = ins.offset;
e6dcd2dc 852 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 853 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 854
d397712b 855 while (1) {
890871be 856 write_lock(&em_tree->lock);
e6dcd2dc 857 ret = add_extent_mapping(em_tree, em);
890871be 858 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 859 if (ret != -EEXIST) {
860 free_extent_map(em);
861 break;
862 }
863 btrfs_drop_extent_cache(inode, start,
c8b97818 864 start + ram_size - 1, 0);
e6dcd2dc
CM		 865 }
866
98d20f67 867 cur_alloc_size = ins.offset;
e6dcd2dc 868 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 869 ram_size, cur_alloc_size, 0);
e6dcd2dc 870 BUG_ON(ret);
c8b97818 871
17d217fe
YZ		 872 if (root->root_key.objectid ==
873 BTRFS_DATA_RELOC_TREE_OBJECTID) {
874 ret = btrfs_reloc_clone_csums(inode, start,
875 cur_alloc_size);
876 BUG_ON(ret);
877 }
878
d397712b 879 if (disk_num_bytes < cur_alloc_size)
3b951516 880 break;
d397712b 881
c8b97818
CM		 882 /* we're not doing compressed IO, don't unlock the first
883 * page (which the caller expects to stay locked), don't
884 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 885 *
886 * Do set the Private2 bit so we know this page was properly
887 * setup for writepage
c8b97818 888 */
a791e35e
CM		 889 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
891 EXTENT_SET_PRIVATE2;
892
c8b97818
CM		 893 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894 start, start + ram_size - 1,
a791e35e 895 locked_page, op);
c8b97818 896 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 897 num_bytes -= cur_alloc_size;
898 alloc_hint = ins.objectid + ins.offset;
899 start += cur_alloc_size;
b888db2b 900 }
b888db2b 901out:
771ed689 902 ret = 0;
b888db2b 903 btrfs_end_transaction(trans, root);
c8b97818 904
be20aa9d 905 return ret;
771ed689 906}
c8b97818 907
771ed689
CM		 908/*
909 * work queue call back to started compression on a file and pages
910 */
911static noinline void async_cow_start(struct btrfs_work *work)
912{
913 struct async_cow *async_cow;
914 int num_added = 0;
915 async_cow = container_of(work, struct async_cow, work);
916
917 compress_file_range(async_cow->inode, async_cow->locked_page,
918 async_cow->start, async_cow->end, async_cow,
919 &num_added);
920 if (num_added == 0)
921 async_cow->inode = NULL;
922}
923
924/*
925 * work queue call back to submit previously compressed pages
926 */
927static noinline void async_cow_submit(struct btrfs_work *work)
928{
929 struct async_cow *async_cow;
930 struct btrfs_root *root;
931 unsigned long nr_pages;
932
933 async_cow = container_of(work, struct async_cow, work);
934
935 root = async_cow->root;
936 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
937 PAGE_CACHE_SHIFT;
938
939 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
940
941 if (atomic_read(&root->fs_info->async_delalloc_pages) <
942 5 * 1042 * 1024 &&
943 waitqueue_active(&root->fs_info->async_submit_wait))
944 wake_up(&root->fs_info->async_submit_wait);
945
d397712b 946 if (async_cow->inode)
771ed689 947 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 948}
c8b97818 949
771ed689
CM		 950static noinline void async_cow_free(struct btrfs_work *work)
951{
952 struct async_cow *async_cow;
953 async_cow = container_of(work, struct async_cow, work);
954 kfree(async_cow);
955}
956
957static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958 u64 start, u64 end, int *page_started,
959 unsigned long *nr_written)
960{
961 struct async_cow *async_cow;
962 struct btrfs_root *root = BTRFS_I(inode)->root;
963 unsigned long nr_pages;
964 u64 cur_end;
965 int limit = 10 * 1024 * 1042;
966
a3429ab7
CM		 967 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968 1, 0, NULL, GFP_NOFS);
d397712b 969 while (start < end) {
771ed689 970 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
8d413713 971 BUG_ON(!async_cow);
771ed689
CM		 972 async_cow->inode = inode;
973 async_cow->root = root;
974 async_cow->locked_page = locked_page;
975 async_cow->start = start;
976
6cbff00f 977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 978 cur_end = end;
979 else
980 cur_end = min(end, start + 512 * 1024 - 1);
981
982 async_cow->end = cur_end;
983 INIT_LIST_HEAD(&async_cow->extents);
984
985 async_cow->work.func = async_cow_start;
986 async_cow->work.ordered_func = async_cow_submit;
987 async_cow->work.ordered_free = async_cow_free;
988 async_cow->work.flags = 0;
989
771ed689
CM		 990 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
991 PAGE_CACHE_SHIFT;
992 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
993
994 btrfs_queue_worker(&root->fs_info->delalloc_workers,
995 &async_cow->work);
996
997 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998 wait_event(root->fs_info->async_submit_wait,
999 (atomic_read(&root->fs_info->async_delalloc_pages) <
1000 limit));
1001 }
1002
d397712b 1003 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1004 atomic_read(&root->fs_info->async_delalloc_pages)) {
1005 wait_event(root->fs_info->async_submit_wait,
1006 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1007 0));
1008 }
1009
1010 *nr_written += nr_pages;
1011 start = cur_end + 1;
1012 }
1013 *page_started = 1;
1014 return 0;
be20aa9d
CM		 1015}
1016
d397712b 1017static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1018 u64 bytenr, u64 num_bytes)
1019{
1020 int ret;
1021 struct btrfs_ordered_sum *sums;
1022 LIST_HEAD(list);
1023
07d400a6 1024 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1025 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1026 if (ret == 0 && list_empty(&list))
1027 return 0;
1028
1029 while (!list_empty(&list)) {
1030 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031 list_del(&sums->list);
1032 kfree(sums);
1033 }
1034 return 1;
1035}
1036
d352ac68
CM		 1037/*
1038 * when nowcow writeback call back. This checks for snapshots or COW copies
1039 * of the extents that exist in the file, and COWs the file as required.
1040 *
1041 * If no cow copies or snapshots exist, we write directly to the existing
1042 * blocks on disk
1043 */
7f366cfe
CM		 1044static noinline int run_delalloc_nocow(struct inode *inode,
1045 struct page *locked_page,
771ed689
CM		 1046 u64 start, u64 end, int *page_started, int force,
1047 unsigned long *nr_written)
be20aa9d 1048{
be20aa9d 1049 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1050 struct btrfs_trans_handle *trans;
be20aa9d 1051 struct extent_buffer *leaf;
be20aa9d 1052 struct btrfs_path *path;
80ff3856 1053 struct btrfs_file_extent_item *fi;
be20aa9d 1054 struct btrfs_key found_key;
80ff3856
YZ		 1055 u64 cow_start;
1056 u64 cur_offset;
1057 u64 extent_end;
5d4f98a2 1058 u64 extent_offset;
80ff3856
YZ		 1059 u64 disk_bytenr;
1060 u64 num_bytes;
1061 int extent_type;
1062 int ret;
d899e052 1063 int type;
80ff3856
YZ		 1064 int nocow;
1065 int check_prev = 1;
82d5902d 1066 bool nolock;
33345d01 1067 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1068
1069 path = btrfs_alloc_path();
d8926bb3
MF		 1070 if (!path)
1071 return -ENOMEM;
82d5902d 1072
2cf8572d 1073 nolock = btrfs_is_free_space_inode(root, inode);
82d5902d
LZ		 1074
1075 if (nolock)
7a7eaa40 1076 trans = btrfs_join_transaction_nolock(root);
82d5902d 1077 else
7a7eaa40 1078 trans = btrfs_join_transaction(root);
ff5714cc 1079
3612b495 1080 BUG_ON(IS_ERR(trans));
74b21075 1081 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1082
80ff3856
YZ		 1083 cow_start = (u64)-1;
1084 cur_offset = start;
1085 while (1) {
33345d01 1086 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856
YZ		 1087 cur_offset, 0);
1088 BUG_ON(ret < 0);
1089 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0] - 1);
33345d01 1093 if (found_key.objectid == ino &&
80ff3856
YZ		 1094 found_key.type == BTRFS_EXTENT_DATA_KEY)
1095 path->slots[0]--;
1096 }
1097 check_prev = 0;
1098next_slot:
1099 leaf = path->nodes[0];
1100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101 ret = btrfs_next_leaf(root, path);
1102 if (ret < 0)
1103 BUG_ON(1);
1104 if (ret > 0)
1105 break;
1106 leaf = path->nodes[0];
1107 }
be20aa9d 1108
80ff3856
YZ		 1109 nocow = 0;
1110 disk_bytenr = 0;
17d217fe 1111 num_bytes = 0;
80ff3856
YZ		 1112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1113
33345d01 1114 if (found_key.objectid > ino ||
80ff3856
YZ		 1115 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116 found_key.offset > end)
1117 break;
1118
1119 if (found_key.offset > cur_offset) {
1120 extent_end = found_key.offset;
e9061e21 1121 extent_type = 0;
80ff3856
YZ		 1122 goto out_check;
1123 }
1124
1125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 extent_type = btrfs_file_extent_type(leaf, fi);
1128
d899e052
YZ		 1129 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1131 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1132 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1133 extent_end = found_key.offset +
1134 btrfs_file_extent_num_bytes(leaf, fi);
1135 if (extent_end <= start) {
1136 path->slots[0]++;
1137 goto next_slot;
1138 }
17d217fe
YZ		 1139 if (disk_bytenr == 0)
1140 goto out_check;
80ff3856
YZ		 1141 if (btrfs_file_extent_compression(leaf, fi) ||
1142 btrfs_file_extent_encryption(leaf, fi) ||
1143 btrfs_file_extent_other_encoding(leaf, fi))
1144 goto out_check;
d899e052
YZ		 1145 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1146 goto out_check;
d2fb3437 1147 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1148 goto out_check;
33345d01 1149 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1150 found_key.offset -
1151 extent_offset, disk_bytenr))
17d217fe 1152 goto out_check;
5d4f98a2 1153 disk_bytenr += extent_offset;
17d217fe
YZ		 1154 disk_bytenr += cur_offset - found_key.offset;
1155 num_bytes = min(end + 1, extent_end) - cur_offset;
1156 /*
1157 * force cow if csum exists in the range.
1158 * this ensure that csum for a given extent are
1159 * either valid or do not exist.
1160 */
1161 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162 goto out_check;
80ff3856
YZ		 1163 nocow = 1;
1164 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165 extent_end = found_key.offset +
1166 btrfs_file_extent_inline_len(leaf, fi);
1167 extent_end = ALIGN(extent_end, root->sectorsize);
1168 } else {
1169 BUG_ON(1);
1170 }
1171out_check:
1172 if (extent_end <= start) {
1173 path->slots[0]++;
1174 goto next_slot;
1175 }
1176 if (!nocow) {
1177 if (cow_start == (u64)-1)
1178 cow_start = cur_offset;
1179 cur_offset = extent_end;
1180 if (cur_offset > end)
1181 break;
1182 path->slots[0]++;
1183 goto next_slot;
7ea394f1
YZ		 1184 }
1185
b3b4aa74 1186 btrfs_release_path(path);
80ff3856
YZ		 1187 if (cow_start != (u64)-1) {
1188 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1189 found_key.offset - 1, page_started,
1190 nr_written, 1);
80ff3856
YZ		 1191 BUG_ON(ret);
1192 cow_start = (u64)-1;
7ea394f1 1193 }
80ff3856 1194
d899e052
YZ		 1195 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196 struct extent_map *em;
1197 struct extent_map_tree *em_tree;
1198 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1199 em = alloc_extent_map();
c26a9203 1200 BUG_ON(!em);
d899e052 1201 em->start = cur_offset;
445a6944 1202 em->orig_start = em->start;
d899e052
YZ		 1203 em->len = num_bytes;
1204 em->block_len = num_bytes;
1205 em->block_start = disk_bytenr;
1206 em->bdev = root->fs_info->fs_devices->latest_bdev;
1207 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1208 while (1) {
890871be 1209 write_lock(&em_tree->lock);
d899e052 1210 ret = add_extent_mapping(em_tree, em);
890871be 1211 write_unlock(&em_tree->lock);
d899e052
YZ		 1212 if (ret != -EEXIST) {
1213 free_extent_map(em);
1214 break;
1215 }
1216 btrfs_drop_extent_cache(inode, em->start,
1217 em->start + em->len - 1, 0);
1218 }
1219 type = BTRFS_ORDERED_PREALLOC;
1220 } else {
1221 type = BTRFS_ORDERED_NOCOW;
1222 }
80ff3856
YZ		 1223
1224 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1225 num_bytes, num_bytes, type);
1226 BUG_ON(ret);
771ed689 1227
efa56464
YZ		 1228 if (root->root_key.objectid ==
1229 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1231 num_bytes);
1232 BUG_ON(ret);
1233 }
1234
d899e052 1235 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1236 cur_offset, cur_offset + num_bytes - 1,
1237 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1240 cur_offset = extent_end;
1241 if (cur_offset > end)
1242 break;
be20aa9d 1243 }
b3b4aa74 1244 btrfs_release_path(path);
80ff3856
YZ		 1245
1246 if (cur_offset <= end && cow_start == (u64)-1)
1247 cow_start = cur_offset;
1248 if (cow_start != (u64)-1) {
1249 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1250 page_started, nr_written, 1);
80ff3856
YZ		 1251 BUG_ON(ret);
1252 }
1253
0cb59c99
JB		 1254 if (nolock) {
1255 ret = btrfs_end_transaction_nolock(trans, root);
1256 BUG_ON(ret);
1257 } else {
1258 ret = btrfs_end_transaction(trans, root);
1259 BUG_ON(ret);
1260 }
7ea394f1 1261 btrfs_free_path(path);
80ff3856 1262 return 0;
be20aa9d
CM		 1263}
1264
d352ac68
CM		 1265/*
1266 * extent_io.c call back to do delayed allocation processing
1267 */
c8b97818 1268static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1269 u64 start, u64 end, int *page_started,
1270 unsigned long *nr_written)
be20aa9d 1271{
be20aa9d 1272 int ret;
7f366cfe 1273 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1274
6cbff00f 1275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
c8b97818 1276 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1277 page_started, 1, nr_written);
6cbff00f 1278 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
d899e052 1279 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1280 page_started, 0, nr_written);
1e701a32 1281 else if (!btrfs_test_opt(root, COMPRESS) &&
75e7cb7f
LB		 1282 !(BTRFS_I(inode)->force_compress) &&
1283 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
7f366cfe
CM		 1284 ret = cow_file_range(inode, locked_page, start, end,
1285 page_started, nr_written, 1);
be20aa9d 1286 else
771ed689 1287 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1288 page_started, nr_written);
b888db2b
CM		 1289 return ret;
1290}
1291
1bf85046
JM		 1292static void btrfs_split_extent_hook(struct inode *inode,
1293 struct extent_state *orig, u64 split)
9ed74f2d 1294{
0ca1f7ce 1295 /* not delalloc, ignore it */
9ed74f2d 1296 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1297 return;
9ed74f2d 1298
9e0baf60
JB		 1299 spin_lock(&BTRFS_I(inode)->lock);
1300 BTRFS_I(inode)->outstanding_extents++;
1301 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1302}
1303
1304/*
1305 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306 * extents so we can keep track of new extents that are just merged onto old
1307 * extents, such as when we are doing sequential writes, so we can properly
1308 * account for the metadata space we'll need.
1309 */
1bf85046
JM		 1310static void btrfs_merge_extent_hook(struct inode *inode,
1311 struct extent_state *new,
1312 struct extent_state *other)
9ed74f2d 1313{
9ed74f2d
JB		 1314 /* not delalloc, ignore it */
1315 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1316 return;
9ed74f2d 1317
9e0baf60
JB		 1318 spin_lock(&BTRFS_I(inode)->lock);
1319 BTRFS_I(inode)->outstanding_extents--;
1320 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1321}
1322
d352ac68
CM		 1323/*
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1327 */
1bf85046
JM		 1328static void btrfs_set_bit_hook(struct inode *inode,
1329 struct extent_state *state, int *bits)
291d673e 1330{
9ed74f2d 1331
75eff68e
CM		 1332 /*
1333 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1334 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1335 * bit, which is only set or cleared with irqs on
1336 */
0ca1f7ce 1337 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1338 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1339 u64 len = state->end + 1 - state->start;
2cf8572d 1340 bool do_list = !btrfs_is_free_space_inode(root, inode);
9ed74f2d 1341
9e0baf60 1342 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1343 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1344 } else {
1345 spin_lock(&BTRFS_I(inode)->lock);
1346 BTRFS_I(inode)->outstanding_extents++;
1347 spin_unlock(&BTRFS_I(inode)->lock);
1348 }
287a0ab9 1349
75eff68e 1350 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1351 BTRFS_I(inode)->delalloc_bytes += len;
1352 root->fs_info->delalloc_bytes += len;
0cb59c99 1353 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
ea8c2819
CM		 1354 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355 &root->fs_info->delalloc_inodes);
1356 }
75eff68e 1357 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1358 }
291d673e
CM		 1359}
1360
d352ac68
CM		 1361/*
1362 * extent_io.c clear_bit_hook, see set_bit_hook for why
1363 */
1bf85046
JM		 1364static void btrfs_clear_bit_hook(struct inode *inode,
1365 struct extent_state *state, int *bits)
291d673e 1366{
75eff68e
CM		 1367 /*
1368 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1369 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1370 * bit, which is only set or cleared with irqs on
1371 */
0ca1f7ce 1372 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1373 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1374 u64 len = state->end + 1 - state->start;
2cf8572d 1375 bool do_list = !btrfs_is_free_space_inode(root, inode);
bcbfce8a 1376
9e0baf60 1377 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1378 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1379 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380 spin_lock(&BTRFS_I(inode)->lock);
1381 BTRFS_I(inode)->outstanding_extents--;
1382 spin_unlock(&BTRFS_I(inode)->lock);
1383 }
0ca1f7ce
YZ		 1384
1385 if (*bits & EXTENT_DO_ACCOUNTING)
1386 btrfs_delalloc_release_metadata(inode, len);
1387
0cb59c99
JB		 1388 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1389 && do_list)
0ca1f7ce 1390 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1391
75eff68e 1392 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1393 root->fs_info->delalloc_bytes -= len;
1394 BTRFS_I(inode)->delalloc_bytes -= len;
1395
0cb59c99 1396 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
ea8c2819
CM		 1397 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1399 }
75eff68e 1400 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1401 }
291d673e
CM		 1402}
1403
d352ac68
CM		 1404/*
1405 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406 * we don't create bios that span stripes or chunks
1407 */
239b14b3 1408int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1409 size_t size, struct bio *bio,
1410 unsigned long bio_flags)
239b14b3
CM		 1411{
1412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413 struct btrfs_mapping_tree *map_tree;
a62b9401 1414 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1415 u64 length = 0;
1416 u64 map_length;
239b14b3
CM		 1417 int ret;
1418
771ed689
CM		 1419 if (bio_flags & EXTENT_BIO_COMPRESSED)
1420 return 0;
1421
f2d8d74d 1422 length = bio->bi_size;
239b14b3
CM		 1423 map_tree = &root->fs_info->mapping_tree;
1424 map_length = length;
cea9e445 1425 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1426 &map_length, NULL, 0);
cea9e445 1427
d397712b 1428 if (map_length < length + size)
239b14b3 1429 return 1;
411fc6bc 1430 return ret;
239b14b3
CM		 1431}
1432
d352ac68
CM		 1433/*
1434 * in order to insert checksums into the metadata in large chunks,
1435 * we wait until bio submission time. All the pages in the bio are
1436 * checksummed and sums are attached onto the ordered extent record.
1437 *
1438 * At IO completion time the cums attached on the ordered extent record
1439 * are inserted into the btree
1440 */
d397712b
CM		 1441static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442 struct bio *bio, int mirror_num,
eaf25d93
CM		 1443 unsigned long bio_flags,
1444 u64 bio_offset)
065631f6 1445{
065631f6 1446 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1447 int ret = 0;
e015640f 1448
d20f7043 1449 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1450 BUG_ON(ret);
4a69a410
CM		 1451 return 0;
1452}
e015640f 1453
4a69a410
CM		 1454/*
1455 * in order to insert checksums into the metadata in large chunks,
1456 * we wait until bio submission time. All the pages in the bio are
1457 * checksummed and sums are attached onto the ordered extent record.
1458 *
1459 * At IO completion time the cums attached on the ordered extent record
1460 * are inserted into the btree
1461 */
b2950863 1462static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1463 int mirror_num, unsigned long bio_flags,
1464 u64 bio_offset)
4a69a410
CM		 1465{
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1467 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1468}
1469
d352ac68 1470/*
cad321ad
CM		 1471 * extent_io.c submission hook. This does the right thing for csum calculation
1472 * on write, or reading the csums from the tree before a read
d352ac68 1473 */
b2950863 1474static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1475 int mirror_num, unsigned long bio_flags,
1476 u64 bio_offset)
44b8bd7e
CM		 1477{
1478 struct btrfs_root *root = BTRFS_I(inode)->root;
1479 int ret = 0;
19b9bdb0 1480 int skip_sum;
44b8bd7e 1481
6cbff00f 1482 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1483
2cf8572d 1484 if (btrfs_is_free_space_inode(root, inode))
0cb59c99
JB		 1485 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1486 else
1487 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
e6dcd2dc 1488 BUG_ON(ret);
065631f6 1489
7b6d91da 1490 if (!(rw & REQ_WRITE)) {
d20f7043 1491 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1492 return btrfs_submit_compressed_read(inode, bio,
1493 mirror_num, bio_flags);
c2db1073
TI		 1494 } else if (!skip_sum) {
1495 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1496 if (ret)
1497 return ret;
1498 }
4d1b5fb4 1499 goto mapit;
19b9bdb0 1500 } else if (!skip_sum) {
17d217fe
YZ		 1501 /* csum items have already been cloned */
1502 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1503 goto mapit;
19b9bdb0
CM		 1504 /* we're doing a write, do the async checksumming */
1505 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1506 inode, rw, bio, mirror_num,
eaf25d93
CM		 1507 bio_flags, bio_offset,
1508 __btrfs_submit_bio_start,
4a69a410 1509 __btrfs_submit_bio_done);
19b9bdb0
CM		 1510 }
1511
0b86a832 1512mapit:
8b712842 1513 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1514}
6885f308 1515
d352ac68
CM		 1516/*
1517 * given a list of ordered sums record them in the inode. This happens
1518 * at IO completion time based on sums calculated at bio submission time.
1519 */
ba1da2f4 1520static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1521 struct inode *inode, u64 file_offset,
1522 struct list_head *list)
1523{
e6dcd2dc
CM		 1524 struct btrfs_ordered_sum *sum;
1525
c6e30871 1526 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1527 btrfs_csum_file_blocks(trans,
1528 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1529 }
1530 return 0;
1531}
1532
2ac55d41
JB		 1533int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534 struct extent_state **cached_state)
ea8c2819 1535{
d397712b 1536 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1537 WARN_ON(1);
ea8c2819 1538 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1539 cached_state, GFP_NOFS);
ea8c2819
CM		 1540}
1541
d352ac68 1542/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1543struct btrfs_writepage_fixup {
1544 struct page *page;
1545 struct btrfs_work work;
1546};
1547
b2950863 1548static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1549{
1550 struct btrfs_writepage_fixup *fixup;
1551 struct btrfs_ordered_extent *ordered;
2ac55d41 1552 struct extent_state *cached_state = NULL;
247e743c
CM		 1553 struct page *page;
1554 struct inode *inode;
1555 u64 page_start;
1556 u64 page_end;
1557
1558 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1559 page = fixup->page;
4a096752 1560again:
247e743c
CM		 1561 lock_page(page);
1562 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1563 ClearPageChecked(page);
1564 goto out_page;
1565 }
1566
1567 inode = page->mapping->host;
1568 page_start = page_offset(page);
1569 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1570
2ac55d41
JB		 1571 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1572 &cached_state, GFP_NOFS);
4a096752
CM		 1573
1574 /* already ordered? We're done */
8b62b72b 1575 if (PagePrivate2(page))
247e743c 1576 goto out;
4a096752
CM		 1577
1578 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1579 if (ordered) {
2ac55d41
JB		 1580 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1581 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1582 unlock_page(page);
1583 btrfs_start_ordered_extent(inode, ordered, 1);
1584 goto again;
1585 }
247e743c 1586
0ca1f7ce 1587 BUG();
2ac55d41 1588 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c
CM		 1589 ClearPageChecked(page);
1590out:
2ac55d41
JB		 1591 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1592 &cached_state, GFP_NOFS);
247e743c
CM		 1593out_page:
1594 unlock_page(page);
1595 page_cache_release(page);
b897abec 1596 kfree(fixup);
247e743c
CM		 1597}
1598
1599/*
1600 * There are a few paths in the higher layers of the kernel that directly
1601 * set the page dirty bit without asking the filesystem if it is a
1602 * good idea. This causes problems because we want to make sure COW
1603 * properly happens and the data=ordered rules are followed.
1604 *
c8b97818 1605 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1606 * hasn't been properly setup for IO. We kick off an async process
1607 * to fix it up. The async helper will wait for ordered extents, set
1608 * the delalloc bit and make it safe to write the page.
1609 */
b2950863 1610static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1611{
1612 struct inode *inode = page->mapping->host;
1613 struct btrfs_writepage_fixup *fixup;
1614 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1615
8b62b72b
CM		 1616 /* this page is properly in the ordered list */
1617 if (TestClearPagePrivate2(page))
247e743c
CM		 1618 return 0;
1619
1620 if (PageChecked(page))
1621 return -EAGAIN;
1622
1623 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1624 if (!fixup)
1625 return -EAGAIN;
f421950f 1626
247e743c
CM		 1627 SetPageChecked(page);
1628 page_cache_get(page);
1629 fixup->work.func = btrfs_writepage_fixup_worker;
1630 fixup->page = page;
1631 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1632 return -EAGAIN;
1633}
1634
d899e052
YZ		 1635static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1636 struct inode *inode, u64 file_pos,
1637 u64 disk_bytenr, u64 disk_num_bytes,
1638 u64 num_bytes, u64 ram_bytes,
1639 u8 compression, u8 encryption,
1640 u16 other_encoding, int extent_type)
1641{
1642 struct btrfs_root *root = BTRFS_I(inode)->root;
1643 struct btrfs_file_extent_item *fi;
1644 struct btrfs_path *path;
1645 struct extent_buffer *leaf;
1646 struct btrfs_key ins;
1647 u64 hint;
1648 int ret;
1649
1650 path = btrfs_alloc_path();
d8926bb3
MF		 1651 if (!path)
1652 return -ENOMEM;
d899e052 1653
b9473439 1654 path->leave_spinning = 1;
a1ed835e
CM		 1655
1656 /*
1657 * we may be replacing one extent in the tree with another.
1658 * The new extent is pinned in the extent map, and we don't want
1659 * to drop it from the cache until it is completely in the btree.
1660 *
1661 * So, tell btrfs_drop_extents to leave this extent in the cache.
1662 * the caller is expected to unpin it and allow it to be merged
1663 * with the others.
1664 */
920bbbfb
YZ		 1665 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1666 &hint, 0);
d899e052
YZ		 1667 BUG_ON(ret);
1668
33345d01 1669 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1670 ins.offset = file_pos;
1671 ins.type = BTRFS_EXTENT_DATA_KEY;
1672 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1673 BUG_ON(ret);
1674 leaf = path->nodes[0];
1675 fi = btrfs_item_ptr(leaf, path->slots[0],
1676 struct btrfs_file_extent_item);
1677 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1678 btrfs_set_file_extent_type(leaf, fi, extent_type);
1679 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1680 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1681 btrfs_set_file_extent_offset(leaf, fi, 0);
1682 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1683 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1684 btrfs_set_file_extent_compression(leaf, fi, compression);
1685 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1686 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1687
1688 btrfs_unlock_up_safe(path, 1);
1689 btrfs_set_lock_blocking(leaf);
1690
d899e052
YZ		 1691 btrfs_mark_buffer_dirty(leaf);
1692
1693 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1694
1695 ins.objectid = disk_bytenr;
1696 ins.offset = disk_num_bytes;
1697 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1698 ret = btrfs_alloc_reserved_file_extent(trans, root,
1699 root->root_key.objectid,
33345d01 1700 btrfs_ino(inode), file_pos, &ins);
d899e052 1701 BUG_ON(ret);
d899e052 1702 btrfs_free_path(path);
b9473439 1703
d899e052
YZ		 1704 return 0;
1705}
1706
5d13a98f
CM		 1707/*
1708 * helper function for btrfs_finish_ordered_io, this
1709 * just reads in some of the csum leaves to prime them into ram
1710 * before we start the transaction. It limits the amount of btree
1711 * reads required while inside the transaction.
1712 */
d352ac68
CM		 1713/* as ordered data IO finishes, this gets called so we can finish
1714 * an ordered extent if the range of bytes in the file it covers are
1715 * fully written.
1716 */
211f90e6 1717static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1718{
e6dcd2dc 1719 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1720 struct btrfs_trans_handle *trans = NULL;
5d13a98f 1721 struct btrfs_ordered_extent *ordered_extent = NULL;
e6dcd2dc 1722 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1723 struct extent_state *cached_state = NULL;
261507a0 1724 int compress_type = 0;
e6dcd2dc 1725 int ret;
82d5902d 1726 bool nolock;
e6dcd2dc 1727
5a1a3df1
JB		 1728 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1729 end - start + 1);
ba1da2f4 1730 if (!ret)
e6dcd2dc 1731 return 0;
e6dcd2dc 1732 BUG_ON(!ordered_extent);
efd049fb 1733
2cf8572d 1734 nolock = btrfs_is_free_space_inode(root, inode);
0cb59c99 1735
c2167754
YZ		 1736 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1737 BUG_ON(!list_empty(&ordered_extent->list));
1738 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1739 if (!ret) {
0cb59c99 1740 if (nolock)
7a7eaa40 1741 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1742 else
7a7eaa40 1743 trans = btrfs_join_transaction(root);
3612b495 1744 BUG_ON(IS_ERR(trans));
0ca1f7ce 1745 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2115133f 1746 ret = btrfs_update_inode_fallback(trans, root, inode);
c2167754 1747 BUG_ON(ret);
c2167754
YZ		 1748 }
1749 goto out;
1750 }
e6dcd2dc 1751
2ac55d41
JB		 1752 lock_extent_bits(io_tree, ordered_extent->file_offset,
1753 ordered_extent->file_offset + ordered_extent->len - 1,
1754 0, &cached_state, GFP_NOFS);
e6dcd2dc 1755
0cb59c99 1756 if (nolock)
7a7eaa40 1757 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1758 else
7a7eaa40 1759 trans = btrfs_join_transaction(root);
3612b495 1760 BUG_ON(IS_ERR(trans));
0ca1f7ce 1761 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1762
c8b97818 1763 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1764 compress_type = ordered_extent->compress_type;
d899e052 1765 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1766 BUG_ON(compress_type);
920bbbfb 1767 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1768 ordered_extent->file_offset,
1769 ordered_extent->file_offset +
1770 ordered_extent->len);
1771 BUG_ON(ret);
1772 } else {
0af3d00b 1773 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1774 ret = insert_reserved_file_extent(trans, inode,
1775 ordered_extent->file_offset,
1776 ordered_extent->start,
1777 ordered_extent->disk_len,
1778 ordered_extent->len,
1779 ordered_extent->len,
261507a0 1780 compress_type, 0, 0,
d899e052 1781 BTRFS_FILE_EXTENT_REG);
a1ed835e
CM		 1782 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1783 ordered_extent->file_offset,
1784 ordered_extent->len);
d899e052
YZ		 1785 BUG_ON(ret);
1786 }
2ac55d41
JB		 1787 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1788 ordered_extent->file_offset +
1789 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1790
e6dcd2dc
CM		 1791 add_pending_csums(trans, inode, ordered_extent->file_offset,
1792 &ordered_extent->list);
1793
1ef30be1 1794 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1795 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2115133f 1796 ret = btrfs_update_inode_fallback(trans, root, inode);
1ef30be1
JB		 1797 BUG_ON(ret);
1798 }
1799 ret = 0;
c2167754 1800out:
5b0e95bf
JB		 1801 if (root != root->fs_info->tree_root)
1802 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1803 if (trans) {
1804 if (nolock)
0cb59c99 1805 btrfs_end_transaction_nolock(trans, root);
5b0e95bf 1806 else
0cb59c99
JB		 1807 btrfs_end_transaction(trans, root);
1808 }
1809
e6dcd2dc
CM		 1810 /* once for us */
1811 btrfs_put_ordered_extent(ordered_extent);
1812 /* once for the tree */
1813 btrfs_put_ordered_extent(ordered_extent);
1814
e6dcd2dc
CM		 1815 return 0;
1816}
1817
b2950863 1818static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1819 struct extent_state *state, int uptodate)
1820{
1abe9b8a 1821 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1822
8b62b72b 1823 ClearPagePrivate2(page);
211f90e6
CM		 1824 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1825}
1826
d352ac68
CM		 1827/*
1828 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 1829 * if there's a match, we allow the bio to finish. If not, the code in
1830 * extent_io.c will try to find good copies for us.
d352ac68 1831 */
b2950863 1832static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1833 struct extent_state *state)
07157aac 1834{
35ebb934 1835 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1836 struct inode *inode = page->mapping->host;
d1310b2e 1837 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1838 char *kaddr;
aadfeb6e 1839 u64 private = ~(u32)0;
07157aac 1840 int ret;
ff79f819
CM		 1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 u32 csum = ~(u32)0;
d1310b2e 1843
d20f7043
CM		 1844 if (PageChecked(page)) {
1845 ClearPageChecked(page);
1846 goto good;
1847 }
6cbff00f
CH		 1848
1849 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 1850 goto good;
17d217fe
YZ		 1851
1852 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 1853 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 1854 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1855 GFP_NOFS);
b6cda9bc 1856 return 0;
17d217fe 1857 }
d20f7043 1858
c2e639f0 1859 if (state && state->start == start) {
70dec807
CM		 1860 private = state->private;
1861 ret = 0;
1862 } else {
1863 ret = get_state_private(io_tree, start, &private);
1864 }
9ab86c8e 1865 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1866 if (ret)
07157aac 1867 goto zeroit;
d397712b 1868
ff79f819
CM		 1869 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1870 btrfs_csum_final(csum, (char *)&csum);
d397712b 1871 if (csum != private)
07157aac 1872 goto zeroit;
d397712b 1873
9ab86c8e 1874 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1875good:
07157aac
CM		 1876 return 0;
1877
1878zeroit:
945d8962 1879 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 1880 "private %llu\n",
1881 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 1882 (unsigned long long)start, csum,
1883 (unsigned long long)private);
db94535d
CM		 1884 memset(kaddr + offset, 1, end - start + 1);
1885 flush_dcache_page(page);
9ab86c8e 1886 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1887 if (private == 0)
1888 return 0;
7e38326f 1889 return -EIO;
07157aac 1890}
b888db2b 1891
24bbcf04
YZ		 1892struct delayed_iput {
1893 struct list_head list;
1894 struct inode *inode;
1895};
1896
1897void btrfs_add_delayed_iput(struct inode *inode)
1898{
1899 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1900 struct delayed_iput *delayed;
1901
1902 if (atomic_add_unless(&inode->i_count, -1, 1))
1903 return;
1904
1905 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1906 delayed->inode = inode;
1907
1908 spin_lock(&fs_info->delayed_iput_lock);
1909 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1910 spin_unlock(&fs_info->delayed_iput_lock);
1911}
1912
1913void btrfs_run_delayed_iputs(struct btrfs_root *root)
1914{
1915 LIST_HEAD(list);
1916 struct btrfs_fs_info *fs_info = root->fs_info;
1917 struct delayed_iput *delayed;
1918 int empty;
1919
1920 spin_lock(&fs_info->delayed_iput_lock);
1921 empty = list_empty(&fs_info->delayed_iputs);
1922 spin_unlock(&fs_info->delayed_iput_lock);
1923 if (empty)
1924 return;
1925
1926 down_read(&root->fs_info->cleanup_work_sem);
1927 spin_lock(&fs_info->delayed_iput_lock);
1928 list_splice_init(&fs_info->delayed_iputs, &list);
1929 spin_unlock(&fs_info->delayed_iput_lock);
1930
1931 while (!list_empty(&list)) {
1932 delayed = list_entry(list.next, struct delayed_iput, list);
1933 list_del(&delayed->list);
1934 iput(delayed->inode);
1935 kfree(delayed);
1936 }
1937 up_read(&root->fs_info->cleanup_work_sem);
1938}
1939
d68fc57b
YZ		 1940enum btrfs_orphan_cleanup_state {
1941 ORPHAN_CLEANUP_STARTED = 1,
1942 ORPHAN_CLEANUP_DONE = 2,
1943};
1944
1945/*
1946 * This is called in transaction commmit time. If there are no orphan
1947 * files in the subvolume, it removes orphan item and frees block_rsv
1948 * structure.
1949 */
1950void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root)
1952{
1953 int ret;
1954
1955 if (!list_empty(&root->orphan_list) ||
1956 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1957 return;
1958
1959 if (root->orphan_item_inserted &&
1960 btrfs_root_refs(&root->root_item) > 0) {
1961 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1962 root->root_key.objectid);
1963 BUG_ON(ret);
1964 root->orphan_item_inserted = 0;
1965 }
1966
1967 if (root->orphan_block_rsv) {
1968 WARN_ON(root->orphan_block_rsv->size > 0);
1969 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1970 root->orphan_block_rsv = NULL;
1971 }
1972}
1973
7b128766
JB		 1974/*
1975 * This creates an orphan entry for the given inode in case something goes
1976 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 1977 *
1978 * NOTE: caller of this function should reserve 5 units of metadata for
1979 * this function.
7b128766
JB		 1980 */
1981int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1982{
1983 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 1984 struct btrfs_block_rsv *block_rsv = NULL;
1985 int reserve = 0;
1986 int insert = 0;
1987 int ret;
7b128766 1988
d68fc57b
YZ		 1989 if (!root->orphan_block_rsv) {
1990 block_rsv = btrfs_alloc_block_rsv(root);
b532402e
TI		 1991 if (!block_rsv)
1992 return -ENOMEM;
d68fc57b 1993 }
7b128766 1994
d68fc57b
YZ		 1995 spin_lock(&root->orphan_lock);
1996 if (!root->orphan_block_rsv) {
1997 root->orphan_block_rsv = block_rsv;
1998 } else if (block_rsv) {
1999 btrfs_free_block_rsv(root, block_rsv);
2000 block_rsv = NULL;
7b128766 2001 }
7b128766 2002
d68fc57b
YZ		 2003 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2004 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2005#if 0
2006 /*
2007 * For proper ENOSPC handling, we should do orphan
2008 * cleanup when mounting. But this introduces backward
2009 * compatibility issue.
2010 */
2011 if (!xchg(&root->orphan_item_inserted, 1))
2012 insert = 2;
2013 else
2014 insert = 1;
2015#endif
2016 insert = 1;
7b128766
JB		 2017 }
2018
d68fc57b
YZ		 2019 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2020 BTRFS_I(inode)->orphan_meta_reserved = 1;
2021 reserve = 1;
2022 }
2023 spin_unlock(&root->orphan_lock);
7b128766 2024
d68fc57b
YZ		 2025 /* grab metadata reservation from transaction handle */
2026 if (reserve) {
2027 ret = btrfs_orphan_reserve_metadata(trans, inode);
2028 BUG_ON(ret);
2029 }
7b128766 2030
d68fc57b
YZ		 2031 /* insert an orphan item to track this unlinked/truncated file */
2032 if (insert >= 1) {
33345d01 2033 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2034 BUG_ON(ret);
2035 }
2036
2037 /* insert an orphan item to track subvolume contains orphan files */
2038 if (insert >= 2) {
2039 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2040 root->root_key.objectid);
2041 BUG_ON(ret);
2042 }
2043 return 0;
7b128766
JB		 2044}
2045
2046/*
2047 * We have done the truncate/delete so we can go ahead and remove the orphan
2048 * item for this particular inode.
2049 */
2050int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2051{
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2053 int delete_item = 0;
2054 int release_rsv = 0;
7b128766
JB		 2055 int ret = 0;
2056
d68fc57b
YZ		 2057 spin_lock(&root->orphan_lock);
2058 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2059 list_del_init(&BTRFS_I(inode)->i_orphan);
2060 delete_item = 1;
7b128766
JB		 2061 }
2062
d68fc57b
YZ		 2063 if (BTRFS_I(inode)->orphan_meta_reserved) {
2064 BTRFS_I(inode)->orphan_meta_reserved = 0;
2065 release_rsv = 1;
7b128766 2066 }
d68fc57b 2067 spin_unlock(&root->orphan_lock);
7b128766 2068
d68fc57b 2069 if (trans && delete_item) {
33345d01 2070 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2071 BUG_ON(ret);
2072 }
7b128766 2073
d68fc57b
YZ		 2074 if (release_rsv)
2075 btrfs_orphan_release_metadata(inode);
7b128766 2076
d68fc57b 2077 return 0;
7b128766
JB		 2078}
2079
2080/*
2081 * this cleans up any orphans that may be left on the list from the last use
2082 * of this root.
2083 */
66b4ffd1 2084int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2085{
2086 struct btrfs_path *path;
2087 struct extent_buffer *leaf;
7b128766
JB		 2088 struct btrfs_key key, found_key;
2089 struct btrfs_trans_handle *trans;
2090 struct inode *inode;
8f6d7f4f 2091 u64 last_objectid = 0;
7b128766
JB		 2092 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2093
d68fc57b 2094 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2095 return 0;
c71bf099
YZ		 2096
2097 path = btrfs_alloc_path();
66b4ffd1
JB		 2098 if (!path) {
2099 ret = -ENOMEM;
2100 goto out;
2101 }
7b128766
JB		 2102 path->reada = -1;
2103
2104 key.objectid = BTRFS_ORPHAN_OBJECTID;
2105 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2106 key.offset = (u64)-1;
2107
7b128766
JB		 2108 while (1) {
2109 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2110 if (ret < 0)
2111 goto out;
7b128766
JB		 2112
2113 /*
2114 * if ret == 0 means we found what we were searching for, which
25985edc 2115 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2116 * find the key and see if we have stuff that matches
2117 */
2118 if (ret > 0) {
66b4ffd1 2119 ret = 0;
7b128766
JB		 2120 if (path->slots[0] == 0)
2121 break;
2122 path->slots[0]--;
2123 }
2124
2125 /* pull out the item */
2126 leaf = path->nodes[0];
7b128766
JB		 2127 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2128
2129 /* make sure the item matches what we want */
2130 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2131 break;
2132 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2133 break;
2134
2135 /* release the path since we're done with it */
b3b4aa74 2136 btrfs_release_path(path);
7b128766
JB		 2137
2138 /*
2139 * this is where we are basically btrfs_lookup, without the
2140 * crossing root thing. we store the inode number in the
2141 * offset of the orphan item.
2142 */
8f6d7f4f
JB		 2143
2144 if (found_key.offset == last_objectid) {
2145 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2146 "stopping orphan cleanup\n");
2147 ret = -EINVAL;
2148 goto out;
2149 }
2150
2151 last_objectid = found_key.offset;
2152
5d4f98a2
YZ		 2153 found_key.objectid = found_key.offset;
2154 found_key.type = BTRFS_INODE_ITEM_KEY;
2155 found_key.offset = 0;
73f73415 2156 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2157 ret = PTR_RET(inode);
2158 if (ret && ret != -ESTALE)
66b4ffd1 2159 goto out;
7b128766
JB		 2160
2161 /*
a8c9e576
JB		 2162 * Inode is already gone but the orphan item is still there,
2163 * kill the orphan item.
7b128766 2164 */
a8c9e576
JB		 2165 if (ret == -ESTALE) {
2166 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2167 if (IS_ERR(trans)) {
2168 ret = PTR_ERR(trans);
2169 goto out;
2170 }
a8c9e576
JB		 2171 ret = btrfs_del_orphan_item(trans, root,
2172 found_key.objectid);
2173 BUG_ON(ret);
5b21f2ed 2174 btrfs_end_transaction(trans, root);
7b128766
JB		 2175 continue;
2176 }
2177
a8c9e576
JB		 2178 /*
2179 * add this inode to the orphan list so btrfs_orphan_del does
2180 * the proper thing when we hit it
2181 */
2182 spin_lock(&root->orphan_lock);
2183 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2184 spin_unlock(&root->orphan_lock);
2185
7b128766
JB		 2186 /* if we have links, this was a truncate, lets do that */
2187 if (inode->i_nlink) {
a41ad394
JB		 2188 if (!S_ISREG(inode->i_mode)) {
2189 WARN_ON(1);
2190 iput(inode);
2191 continue;
2192 }
7b128766 2193 nr_truncate++;
66b4ffd1 2194 ret = btrfs_truncate(inode);
7b128766
JB		 2195 } else {
2196 nr_unlink++;
2197 }
2198
2199 /* this will do delete_inode and everything for us */
2200 iput(inode);
66b4ffd1
JB		 2201 if (ret)
2202 goto out;
7b128766 2203 }
3254c876
MX		 2204 /* release the path since we're done with it */
2205 btrfs_release_path(path);
2206
d68fc57b
YZ		 2207 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2208
2209 if (root->orphan_block_rsv)
2210 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2211 (u64)-1);
2212
2213 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2214 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2215 if (!IS_ERR(trans))
2216 btrfs_end_transaction(trans, root);
d68fc57b 2217 }
7b128766
JB		 2218
2219 if (nr_unlink)
2220 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2221 if (nr_truncate)
2222 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2223
2224out:
2225 if (ret)
2226 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2227 btrfs_free_path(path);
2228 return ret;
7b128766
JB		 2229}
2230
46a53cca
CM		 2231/*
2232 * very simple check to peek ahead in the leaf looking for xattrs. If we
2233 * don't find any xattrs, we know there can't be any acls.
2234 *
2235 * slot is the slot the inode is in, objectid is the objectid of the inode
2236 */
2237static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2238 int slot, u64 objectid)
2239{
2240 u32 nritems = btrfs_header_nritems(leaf);
2241 struct btrfs_key found_key;
2242 int scanned = 0;
2243
2244 slot++;
2245 while (slot < nritems) {
2246 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2247
2248 /* we found a different objectid, there must not be acls */
2249 if (found_key.objectid != objectid)
2250 return 0;
2251
2252 /* we found an xattr, assume we've got an acl */
2253 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2254 return 1;
2255
2256 /*
2257 * we found a key greater than an xattr key, there can't
2258 * be any acls later on
2259 */
2260 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2261 return 0;
2262
2263 slot++;
2264 scanned++;
2265
2266 /*
2267 * it goes inode, inode backrefs, xattrs, extents,
2268 * so if there are a ton of hard links to an inode there can
2269 * be a lot of backrefs. Don't waste time searching too hard,
2270 * this is just an optimization
2271 */
2272 if (scanned >= 8)
2273 break;
2274 }
2275 /* we hit the end of the leaf before we found an xattr or
2276 * something larger than an xattr. We have to assume the inode
2277 * has acls
2278 */
2279 return 1;
2280}
2281
d352ac68
CM		 2282/*
2283 * read an inode from the btree into the in-memory inode
2284 */
5d4f98a2 2285static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2286{
2287 struct btrfs_path *path;
5f39d397 2288 struct extent_buffer *leaf;
39279cc3 2289 struct btrfs_inode_item *inode_item;
0b86a832 2290 struct btrfs_timespec *tspec;
39279cc3
CM		 2291 struct btrfs_root *root = BTRFS_I(inode)->root;
2292 struct btrfs_key location;
46a53cca 2293 int maybe_acls;
618e21d5 2294 u32 rdev;
39279cc3 2295 int ret;
2f7e33d4
MX		 2296 bool filled = false;
2297
2298 ret = btrfs_fill_inode(inode, &rdev);
2299 if (!ret)
2300 filled = true;
39279cc3
CM		 2301
2302 path = btrfs_alloc_path();
1748f843
MF		 2303 if (!path)
2304 goto make_bad;
2305
d90c7321 2306 path->leave_spinning = 1;
39279cc3 2307 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2308
39279cc3 2309 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2310 if (ret)
39279cc3 2311 goto make_bad;
39279cc3 2312
5f39d397 2313 leaf = path->nodes[0];
2f7e33d4
MX		 2314
2315 if (filled)
2316 goto cache_acl;
2317
5f39d397
CM		 2318 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2319 struct btrfs_inode_item);
5f39d397
CM		 2320 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2321 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2322 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2323 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2324 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2325
2326 tspec = btrfs_inode_atime(inode_item);
2327 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2328 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2329
2330 tspec = btrfs_inode_mtime(inode_item);
2331 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2332 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2333
2334 tspec = btrfs_inode_ctime(inode_item);
2335 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2336 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2337
a76a3cd4 2338 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2339 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2340 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2341 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2342 inode->i_rdev = 0;
5f39d397
CM		 2343 rdev = btrfs_inode_rdev(leaf, inode_item);
2344
aec7477b 2345 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2346 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2347cache_acl:
46a53cca
CM		 2348 /*
2349 * try to precache a NULL acl entry for files that don't have
2350 * any xattrs or acls
2351 */
33345d01
LZ		 2352 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2353 btrfs_ino(inode));
72c04902
AV		 2354 if (!maybe_acls)
2355 cache_no_acl(inode);
46a53cca 2356
39279cc3 2357 btrfs_free_path(path);
39279cc3 2358
39279cc3 2359 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2360 case S_IFREG:
2361 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2362 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2363 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2364 inode->i_fop = &btrfs_file_operations;
2365 inode->i_op = &btrfs_file_inode_operations;
2366 break;
2367 case S_IFDIR:
2368 inode->i_fop = &btrfs_dir_file_operations;
2369 if (root == root->fs_info->tree_root)
2370 inode->i_op = &btrfs_dir_ro_inode_operations;
2371 else
2372 inode->i_op = &btrfs_dir_inode_operations;
2373 break;
2374 case S_IFLNK:
2375 inode->i_op = &btrfs_symlink_inode_operations;
2376 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2377 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2378 break;
618e21d5 2379 default:
0279b4cd 2380 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2381 init_special_inode(inode, inode->i_mode, rdev);
2382 break;
39279cc3 2383 }
6cbff00f
CH		 2384
2385 btrfs_update_iflags(inode);
39279cc3
CM		 2386 return;
2387
2388make_bad:
39279cc3 2389 btrfs_free_path(path);
39279cc3
CM		 2390 make_bad_inode(inode);
2391}
2392
d352ac68
CM		 2393/*
2394 * given a leaf and an inode, copy the inode fields into the leaf
2395 */
e02119d5
CM		 2396static void fill_inode_item(struct btrfs_trans_handle *trans,
2397 struct extent_buffer *leaf,
5f39d397 2398 struct btrfs_inode_item *item,
39279cc3
CM		 2399 struct inode *inode)
2400{
5f39d397
CM		 2401 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2402 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2403 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2404 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2405 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2406
2407 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2408 inode->i_atime.tv_sec);
2409 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2410 inode->i_atime.tv_nsec);
2411
2412 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2413 inode->i_mtime.tv_sec);
2414 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2415 inode->i_mtime.tv_nsec);
2416
2417 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2418 inode->i_ctime.tv_sec);
2419 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2420 inode->i_ctime.tv_nsec);
2421
a76a3cd4 2422 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2423 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2424 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2425 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2426 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2427 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2428 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2429}
2430
d352ac68
CM		 2431/*
2432 * copy everything in the in-memory inode into the btree.
2433 */
2115133f 2434static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2435 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2436{
2437 struct btrfs_inode_item *inode_item;
2438 struct btrfs_path *path;
5f39d397 2439 struct extent_buffer *leaf;
39279cc3
CM		 2440 int ret;
2441
2442 path = btrfs_alloc_path();
16cdcec7
MX		 2443 if (!path)
2444 return -ENOMEM;
2445
b9473439 2446 path->leave_spinning = 1;
16cdcec7
MX		 2447 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2448 1);
39279cc3
CM		 2449 if (ret) {
2450 if (ret > 0)
2451 ret = -ENOENT;
2452 goto failed;
2453 }
2454
b4ce94de 2455 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2456 leaf = path->nodes[0];
2457 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2458 struct btrfs_inode_item);
39279cc3 2459
e02119d5 2460 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2461 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2462 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2463 ret = 0;
2464failed:
39279cc3
CM		 2465 btrfs_free_path(path);
2466 return ret;
2467}
2468
2115133f
CM		 2469/*
2470 * copy everything in the in-memory inode into the btree.
2471 */
2472noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root, struct inode *inode)
2474{
2475 int ret;
2476
2477 /*
2478 * If the inode is a free space inode, we can deadlock during commit
2479 * if we put it into the delayed code.
2480 *
2481 * The data relocation inode should also be directly updated
2482 * without delay
2483 */
2484 if (!btrfs_is_free_space_inode(root, inode)
2485 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2486 ret = btrfs_delayed_update_inode(trans, root, inode);
2487 if (!ret)
2488 btrfs_set_inode_last_trans(trans, inode);
2489 return ret;
2490 }
2491
2492 return btrfs_update_inode_item(trans, root, inode);
2493}
2494
2495static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2496 struct btrfs_root *root, struct inode *inode)
2497{
2498 int ret;
2499
2500 ret = btrfs_update_inode(trans, root, inode);
2501 if (ret == -ENOSPC)
2502 return btrfs_update_inode_item(trans, root, inode);
2503 return ret;
2504}
2505
d352ac68
CM		 2506/*
2507 * unlink helper that gets used here in inode.c and in the tree logging
2508 * recovery code. It remove a link in a directory with a given name, and
2509 * also drops the back refs in the inode to the directory
2510 */
92986796
AV		 2511static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2512 struct btrfs_root *root,
2513 struct inode *dir, struct inode *inode,
2514 const char *name, int name_len)
39279cc3
CM		 2515{
2516 struct btrfs_path *path;
39279cc3 2517 int ret = 0;
5f39d397 2518 struct extent_buffer *leaf;
39279cc3 2519 struct btrfs_dir_item *di;
5f39d397 2520 struct btrfs_key key;
aec7477b 2521 u64 index;
33345d01
LZ		 2522 u64 ino = btrfs_ino(inode);
2523 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2524
2525 path = btrfs_alloc_path();
54aa1f4d
CM		 2526 if (!path) {
2527 ret = -ENOMEM;
554233a6 2528 goto out;
54aa1f4d
CM		 2529 }
2530
b9473439 2531 path->leave_spinning = 1;
33345d01 2532 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2533 name, name_len, -1);
2534 if (IS_ERR(di)) {
2535 ret = PTR_ERR(di);
2536 goto err;
2537 }
2538 if (!di) {
2539 ret = -ENOENT;
2540 goto err;
2541 }
5f39d397
CM		 2542 leaf = path->nodes[0];
2543 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2544 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2545 if (ret)
2546 goto err;
b3b4aa74 2547 btrfs_release_path(path);
39279cc3 2548
33345d01
LZ		 2549 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2550 dir_ino, &index);
aec7477b 2551 if (ret) {
d397712b 2552 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2553 "inode %llu parent %llu\n", name_len, name,
2554 (unsigned long long)ino, (unsigned long long)dir_ino);
aec7477b
JB		 2555 goto err;
2556 }
2557
16cdcec7
MX		 2558 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2559 if (ret)
39279cc3 2560 goto err;
39279cc3 2561
e02119d5 2562 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2563 inode, dir_ino);
49eb7e46 2564 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2565
2566 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2567 dir, index);
6418c961
CM		 2568 if (ret == -ENOENT)
2569 ret = 0;
39279cc3
CM		 2570err:
2571 btrfs_free_path(path);
e02119d5
CM		 2572 if (ret)
2573 goto out;
2574
2575 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2576 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2577 btrfs_update_inode(trans, root, dir);
e02119d5 2578out:
39279cc3
CM		 2579 return ret;
2580}
2581
92986796
AV		 2582int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2583 struct btrfs_root *root,
2584 struct inode *dir, struct inode *inode,
2585 const char *name, int name_len)
2586{
2587 int ret;
2588 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2589 if (!ret) {
2590 btrfs_drop_nlink(inode);
2591 ret = btrfs_update_inode(trans, root, inode);
2592 }
2593 return ret;
2594}
2595
2596
a22285a6
YZ		 2597/* helper to check if there is any shared block in the path */
2598static int check_path_shared(struct btrfs_root *root,
2599 struct btrfs_path *path)
39279cc3 2600{
a22285a6
YZ		 2601 struct extent_buffer *eb;
2602 int level;
0e4dcbef 2603 u64 refs = 1;
5df6a9f6 2604
a22285a6 2605 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2606 int ret;
2607
a22285a6
YZ		 2608 if (!path->nodes[level])
2609 break;
2610 eb = path->nodes[level];
2611 if (!btrfs_block_can_be_shared(root, eb))
2612 continue;
2613 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2614 &refs, NULL);
2615 if (refs > 1)
2616 return 1;
5df6a9f6 2617 }
dedefd72 2618 return 0;
39279cc3
CM		 2619}
2620
a22285a6
YZ		 2621/*
2622 * helper to start transaction for unlink and rmdir.
2623 *
2624 * unlink and rmdir are special in btrfs, they do not always free space.
2625 * so in enospc case, we should make sure they will free space before
2626 * allowing them to use the global metadata reservation.
2627 */
2628static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2629 struct dentry *dentry)
4df27c4d 2630{
39279cc3 2631 struct btrfs_trans_handle *trans;
a22285a6 2632 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2633 struct btrfs_path *path;
a22285a6 2634 struct btrfs_inode_ref *ref;
4df27c4d 2635 struct btrfs_dir_item *di;
7b128766 2636 struct inode *inode = dentry->d_inode;
4df27c4d 2637 u64 index;
a22285a6
YZ		 2638 int check_link = 1;
2639 int err = -ENOSPC;
4df27c4d 2640 int ret;
33345d01
LZ		 2641 u64 ino = btrfs_ino(inode);
2642 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2643
e70bea5f
JB		 2644 /*
2645 * 1 for the possible orphan item
2646 * 1 for the dir item
2647 * 1 for the dir index
2648 * 1 for the inode ref
2649 * 1 for the inode ref in the tree log
2650 * 2 for the dir entries in the log
2651 * 1 for the inode
2652 */
2653 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2654 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2655 return trans;
4df27c4d 2656
33345d01 2657 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2658 return ERR_PTR(-ENOSPC);
4df27c4d 2659
a22285a6
YZ		 2660 /* check if there is someone else holds reference */
2661 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2662 return ERR_PTR(-ENOSPC);
4df27c4d 2663
a22285a6
YZ		 2664 if (atomic_read(&inode->i_count) > 2)
2665 return ERR_PTR(-ENOSPC);
4df27c4d 2666
a22285a6
YZ		 2667 if (xchg(&root->fs_info->enospc_unlink, 1))
2668 return ERR_PTR(-ENOSPC);
2669
2670 path = btrfs_alloc_path();
2671 if (!path) {
2672 root->fs_info->enospc_unlink = 0;
2673 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2674 }
2675
3880a1b4
JB		 2676 /* 1 for the orphan item */
2677 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2678 if (IS_ERR(trans)) {
a22285a6
YZ		 2679 btrfs_free_path(path);
2680 root->fs_info->enospc_unlink = 0;
2681 return trans;
2682 }
4df27c4d 2683
a22285a6
YZ		 2684 path->skip_locking = 1;
2685 path->search_commit_root = 1;
4df27c4d 2686
a22285a6
YZ		 2687 ret = btrfs_lookup_inode(trans, root, path,
2688 &BTRFS_I(dir)->location, 0);
2689 if (ret < 0) {
2690 err = ret;
2691 goto out;
2692 }
2693 if (ret == 0) {
2694 if (check_path_shared(root, path))
2695 goto out;
2696 } else {
2697 check_link = 0;
5df6a9f6 2698 }
b3b4aa74 2699 btrfs_release_path(path);
a22285a6
YZ		 2700
2701 ret = btrfs_lookup_inode(trans, root, path,
2702 &BTRFS_I(inode)->location, 0);
2703 if (ret < 0) {
2704 err = ret;
2705 goto out;
2706 }
2707 if (ret == 0) {
2708 if (check_path_shared(root, path))
2709 goto out;
2710 } else {
2711 check_link = 0;
2712 }
b3b4aa74 2713 btrfs_release_path(path);
a22285a6
YZ		 2714
2715 if (ret == 0 && S_ISREG(inode->i_mode)) {
2716 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2717 ino, (u64)-1, 0);
a22285a6
YZ		 2718 if (ret < 0) {
2719 err = ret;
2720 goto out;
2721 }
2722 BUG_ON(ret == 0);
2723 if (check_path_shared(root, path))
2724 goto out;
b3b4aa74 2725 btrfs_release_path(path);
a22285a6
YZ		 2726 }
2727
2728 if (!check_link) {
2729 err = 0;
2730 goto out;
2731 }
2732
33345d01 2733 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2734 dentry->d_name.name, dentry->d_name.len, 0);
2735 if (IS_ERR(di)) {
2736 err = PTR_ERR(di);
2737 goto out;
2738 }
2739 if (di) {
2740 if (check_path_shared(root, path))
2741 goto out;
2742 } else {
2743 err = 0;
2744 goto out;
2745 }
b3b4aa74 2746 btrfs_release_path(path);
a22285a6
YZ		 2747
2748 ref = btrfs_lookup_inode_ref(trans, root, path,
2749 dentry->d_name.name, dentry->d_name.len,
33345d01 2750 ino, dir_ino, 0);
a22285a6
YZ		 2751 if (IS_ERR(ref)) {
2752 err = PTR_ERR(ref);
2753 goto out;
2754 }
2755 BUG_ON(!ref);
2756 if (check_path_shared(root, path))
2757 goto out;
2758 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 2759 btrfs_release_path(path);
a22285a6 2760
16cdcec7
MX		 2761 /*
2762 * This is a commit root search, if we can lookup inode item and other
2763 * relative items in the commit root, it means the transaction of
2764 * dir/file creation has been committed, and the dir index item that we
2765 * delay to insert has also been inserted into the commit root. So
2766 * we needn't worry about the delayed insertion of the dir index item
2767 * here.
2768 */
33345d01 2769 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 2770 dentry->d_name.name, dentry->d_name.len, 0);
2771 if (IS_ERR(di)) {
2772 err = PTR_ERR(di);
2773 goto out;
2774 }
2775 BUG_ON(ret == -ENOENT);
2776 if (check_path_shared(root, path))
2777 goto out;
2778
2779 err = 0;
2780out:
2781 btrfs_free_path(path);
3880a1b4
JB		 2782 /* Migrate the orphan reservation over */
2783 if (!err)
2784 err = btrfs_block_rsv_migrate(trans->block_rsv,
2785 &root->fs_info->global_block_rsv,
5a77d76c 2786 trans->bytes_reserved);
3880a1b4 2787
a22285a6
YZ		 2788 if (err) {
2789 btrfs_end_transaction(trans, root);
2790 root->fs_info->enospc_unlink = 0;
2791 return ERR_PTR(err);
2792 }
2793
2794 trans->block_rsv = &root->fs_info->global_block_rsv;
2795 return trans;
2796}
2797
2798static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2799 struct btrfs_root *root)
2800{
2801 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
5a77d76c
JB		 2802 btrfs_block_rsv_release(root, trans->block_rsv,
2803 trans->bytes_reserved);
2804 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 2805 BUG_ON(!root->fs_info->enospc_unlink);
2806 root->fs_info->enospc_unlink = 0;
2807 }
2808 btrfs_end_transaction_throttle(trans, root);
2809}
2810
2811static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2812{
2813 struct btrfs_root *root = BTRFS_I(dir)->root;
2814 struct btrfs_trans_handle *trans;
2815 struct inode *inode = dentry->d_inode;
2816 int ret;
2817 unsigned long nr = 0;
2818
2819 trans = __unlink_start_trans(dir, dentry);
2820 if (IS_ERR(trans))
2821 return PTR_ERR(trans);
5f39d397 2822
12fcfd22
CM		 2823 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2824
e02119d5
CM		 2825 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2826 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 2827 if (ret)
2828 goto out;
7b128766 2829
a22285a6 2830 if (inode->i_nlink == 0) {
7b128766 2831 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 2832 if (ret)
2833 goto out;
a22285a6 2834 }
7b128766 2835
b532402e 2836out:
d3c2fdcf 2837 nr = trans->blocks_used;
a22285a6 2838 __unlink_end_trans(trans, root);
d3c2fdcf 2839 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2840 return ret;
2841}
2842
4df27c4d
YZ		 2843int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root,
2845 struct inode *dir, u64 objectid,
2846 const char *name, int name_len)
2847{
2848 struct btrfs_path *path;
2849 struct extent_buffer *leaf;
2850 struct btrfs_dir_item *di;
2851 struct btrfs_key key;
2852 u64 index;
2853 int ret;
33345d01 2854 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 2855
2856 path = btrfs_alloc_path();
2857 if (!path)
2858 return -ENOMEM;
2859
33345d01 2860 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 2861 name, name_len, -1);
c704005d 2862 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2863
2864 leaf = path->nodes[0];
2865 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2866 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2867 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2868 BUG_ON(ret);
b3b4aa74 2869 btrfs_release_path(path);
4df27c4d
YZ		 2870
2871 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2872 objectid, root->root_key.objectid,
33345d01 2873 dir_ino, &index, name, name_len);
4df27c4d
YZ		 2874 if (ret < 0) {
2875 BUG_ON(ret != -ENOENT);
33345d01 2876 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 2877 name, name_len);
c704005d 2878 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2879
2880 leaf = path->nodes[0];
2881 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2882 btrfs_release_path(path);
4df27c4d
YZ		 2883 index = key.offset;
2884 }
945d8962 2885 btrfs_release_path(path);
4df27c4d 2886
16cdcec7 2887 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4df27c4d 2888 BUG_ON(ret);
4df27c4d
YZ		 2889
2890 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2891 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2892 ret = btrfs_update_inode(trans, root, dir);
2893 BUG_ON(ret);
4df27c4d 2894
71d7aed0 2895 btrfs_free_path(path);
4df27c4d
YZ		 2896 return 0;
2897}
2898
39279cc3
CM		 2899static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2900{
2901 struct inode *inode = dentry->d_inode;
1832a6d5 2902 int err = 0;
39279cc3 2903 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2904 struct btrfs_trans_handle *trans;
1832a6d5 2905 unsigned long nr = 0;
39279cc3 2906
3394e160 2907 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 2908 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 2909 return -ENOTEMPTY;
2910
a22285a6
YZ		 2911 trans = __unlink_start_trans(dir, dentry);
2912 if (IS_ERR(trans))
5df6a9f6 2913 return PTR_ERR(trans);
5df6a9f6 2914
33345d01 2915 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 2916 err = btrfs_unlink_subvol(trans, root, dir,
2917 BTRFS_I(inode)->location.objectid,
2918 dentry->d_name.name,
2919 dentry->d_name.len);
2920 goto out;
2921 }
2922
7b128766
JB		 2923 err = btrfs_orphan_add(trans, inode);
2924 if (err)
4df27c4d 2925 goto out;
7b128766 2926
39279cc3 2927 /* now the directory is empty */
e02119d5
CM		 2928 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2929 dentry->d_name.name, dentry->d_name.len);
d397712b 2930 if (!err)
dbe674a9 2931 btrfs_i_size_write(inode, 0);
4df27c4d 2932out:
d3c2fdcf 2933 nr = trans->blocks_used;
a22285a6 2934 __unlink_end_trans(trans, root);
d3c2fdcf 2935 btrfs_btree_balance_dirty(root, nr);
3954401f 2936
39279cc3
CM		 2937 return err;
2938}
2939
39279cc3
CM		 2940/*
2941 * this can truncate away extent items, csum items and directory items.
2942 * It starts at a high offset and removes keys until it can't find
d352ac68 2943 * any higher than new_size
39279cc3
CM		 2944 *
2945 * csum items that cross the new i_size are truncated to the new size
2946 * as well.
7b128766
JB		 2947 *
2948 * min_type is the minimum key type to truncate down to. If set to 0, this
2949 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2950 */
8082510e
YZ		 2951int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2952 struct btrfs_root *root,
2953 struct inode *inode,
2954 u64 new_size, u32 min_type)
39279cc3 2955{
39279cc3 2956 struct btrfs_path *path;
5f39d397 2957 struct extent_buffer *leaf;
39279cc3 2958 struct btrfs_file_extent_item *fi;
8082510e
YZ		 2959 struct btrfs_key key;
2960 struct btrfs_key found_key;
39279cc3 2961 u64 extent_start = 0;
db94535d 2962 u64 extent_num_bytes = 0;
5d4f98a2 2963 u64 extent_offset = 0;
39279cc3 2964 u64 item_end = 0;
8082510e
YZ		 2965 u64 mask = root->sectorsize - 1;
2966 u32 found_type = (u8)-1;
39279cc3
CM		 2967 int found_extent;
2968 int del_item;
85e21bac
CM		 2969 int pending_del_nr = 0;
2970 int pending_del_slot = 0;
179e29e4 2971 int extent_type = -1;
771ed689 2972 int encoding;
8082510e
YZ		 2973 int ret;
2974 int err = 0;
33345d01 2975 u64 ino = btrfs_ino(inode);
8082510e
YZ		 2976
2977 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 2978
0eb0e19c
MF		 2979 path = btrfs_alloc_path();
2980 if (!path)
2981 return -ENOMEM;
2982 path->reada = -1;
2983
0af3d00b 2984 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 2985 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 2986
16cdcec7
MX		 2987 /*
2988 * This function is also used to drop the items in the log tree before
2989 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2990 * it is used to drop the loged items. So we shouldn't kill the delayed
2991 * items.
2992 */
2993 if (min_type == 0 && root == BTRFS_I(inode)->root)
2994 btrfs_kill_delayed_inode_items(inode);
2995
33345d01 2996 key.objectid = ino;
39279cc3 2997 key.offset = (u64)-1;
5f39d397
CM		 2998 key.type = (u8)-1;
2999
85e21bac 3000search_again:
b9473439 3001 path->leave_spinning = 1;
85e21bac 3002 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3003 if (ret < 0) {
3004 err = ret;
3005 goto out;
3006 }
d397712b 3007
85e21bac 3008 if (ret > 0) {
e02119d5
CM		 3009 /* there are no items in the tree for us to truncate, we're
3010 * done
3011 */
8082510e
YZ		 3012 if (path->slots[0] == 0)
3013 goto out;
85e21bac
CM		 3014 path->slots[0]--;
3015 }
3016
d397712b 3017 while (1) {
39279cc3 3018 fi = NULL;
5f39d397
CM		 3019 leaf = path->nodes[0];
3020 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3021 found_type = btrfs_key_type(&found_key);
771ed689 3022 encoding = 0;
39279cc3 3023
33345d01 3024 if (found_key.objectid != ino)
39279cc3 3025 break;
5f39d397 3026
85e21bac 3027 if (found_type < min_type)
39279cc3
CM		 3028 break;
3029
5f39d397 3030 item_end = found_key.offset;
39279cc3 3031 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3032 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3033 struct btrfs_file_extent_item);
179e29e4 3034 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 3035 encoding = btrfs_file_extent_compression(leaf, fi);
3036 encoding |= btrfs_file_extent_encryption(leaf, fi);
3037 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3038
179e29e4 3039 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3040 item_end +=
db94535d 3041 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3042 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3043 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3044 fi);
39279cc3 3045 }
008630c1 3046 item_end--;
39279cc3 3047 }
8082510e
YZ		 3048 if (found_type > min_type) {
3049 del_item = 1;
3050 } else {
3051 if (item_end < new_size)
b888db2b 3052 break;
8082510e
YZ		 3053 if (found_key.offset >= new_size)
3054 del_item = 1;
3055 else
3056 del_item = 0;
39279cc3 3057 }
39279cc3 3058 found_extent = 0;
39279cc3 3059 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3060 if (found_type != BTRFS_EXTENT_DATA_KEY)
3061 goto delete;
3062
3063 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3064 u64 num_dec;
db94535d 3065 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 3066 if (!del_item && !encoding) {
db94535d
CM		 3067 u64 orig_num_bytes =
3068 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3069 extent_num_bytes = new_size -
5f39d397 3070 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3071 extent_num_bytes = extent_num_bytes &
3072 ~((u64)root->sectorsize - 1);
db94535d
CM		 3073 btrfs_set_file_extent_num_bytes(leaf, fi,
3074 extent_num_bytes);
3075 num_dec = (orig_num_bytes -
9069218d 3076 extent_num_bytes);
e02119d5 3077 if (root->ref_cows && extent_start != 0)
a76a3cd4 3078 inode_sub_bytes(inode, num_dec);
5f39d397 3079 btrfs_mark_buffer_dirty(leaf);
39279cc3 3080 } else {
db94535d
CM		 3081 extent_num_bytes =
3082 btrfs_file_extent_disk_num_bytes(leaf,
3083 fi);
5d4f98a2
YZ		 3084 extent_offset = found_key.offset -
3085 btrfs_file_extent_offset(leaf, fi);
3086
39279cc3 3087 /* FIXME blocksize != 4096 */
9069218d 3088 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3089 if (extent_start != 0) {
3090 found_extent = 1;
e02119d5 3091 if (root->ref_cows)
a76a3cd4 3092 inode_sub_bytes(inode, num_dec);
e02119d5 3093 }
39279cc3 3094 }
9069218d 3095 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3096 /*
3097 * we can't truncate inline items that have had
3098 * special encodings
3099 */
3100 if (!del_item &&
3101 btrfs_file_extent_compression(leaf, fi) == 0 &&
3102 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3103 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3104 u32 size = new_size - found_key.offset;
3105
3106 if (root->ref_cows) {
a76a3cd4
YZ		 3107 inode_sub_bytes(inode, item_end + 1 -
3108 new_size);
e02119d5
CM		 3109 }
3110 size =
3111 btrfs_file_extent_calc_inline_size(size);
9069218d 3112 ret = btrfs_truncate_item(trans, root, path,
e02119d5 3113 size, 1);
e02119d5 3114 } else if (root->ref_cows) {
a76a3cd4
YZ		 3115 inode_sub_bytes(inode, item_end + 1 -
3116 found_key.offset);
9069218d 3117 }
39279cc3 3118 }
179e29e4 3119delete:
39279cc3 3120 if (del_item) {
85e21bac
CM		 3121 if (!pending_del_nr) {
3122 /* no pending yet, add ourselves */
3123 pending_del_slot = path->slots[0];
3124 pending_del_nr = 1;
3125 } else if (pending_del_nr &&
3126 path->slots[0] + 1 == pending_del_slot) {
3127 /* hop on the pending chunk */
3128 pending_del_nr++;
3129 pending_del_slot = path->slots[0];
3130 } else {
d397712b 3131 BUG();
85e21bac 3132 }
39279cc3
CM		 3133 } else {
3134 break;
3135 }
0af3d00b
JB		 3136 if (found_extent && (root->ref_cows ||
3137 root == root->fs_info->tree_root)) {
b9473439 3138 btrfs_set_path_blocking(path);
39279cc3 3139 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3140 extent_num_bytes, 0,
3141 btrfs_header_owner(leaf),
33345d01 3142 ino, extent_offset);
39279cc3
CM		 3143 BUG_ON(ret);
3144 }
85e21bac 3145
8082510e
YZ		 3146 if (found_type == BTRFS_INODE_ITEM_KEY)
3147 break;
3148
3149 if (path->slots[0] == 0 ||
3150 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3151 if (root->ref_cows &&
3152 BTRFS_I(inode)->location.objectid !=
3153 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3154 err = -EAGAIN;
3155 goto out;
3156 }
3157 if (pending_del_nr) {
3158 ret = btrfs_del_items(trans, root, path,
3159 pending_del_slot,
3160 pending_del_nr);
3161 BUG_ON(ret);
3162 pending_del_nr = 0;
3163 }
b3b4aa74 3164 btrfs_release_path(path);
85e21bac 3165 goto search_again;
8082510e
YZ		 3166 } else {
3167 path->slots[0]--;
85e21bac 3168 }
39279cc3 3169 }
8082510e 3170out:
85e21bac
CM		 3171 if (pending_del_nr) {
3172 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3173 pending_del_nr);
d68fc57b 3174 BUG_ON(ret);
85e21bac 3175 }
39279cc3 3176 btrfs_free_path(path);
8082510e 3177 return err;
39279cc3
CM		 3178}
3179
3180/*
3181 * taken from block_truncate_page, but does cow as it zeros out
3182 * any bytes left in the last page in the file.
3183 */
3184static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3185{
3186 struct inode *inode = mapping->host;
db94535d 3187 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3188 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3189 struct btrfs_ordered_extent *ordered;
2ac55d41 3190 struct extent_state *cached_state = NULL;
e6dcd2dc 3191 char *kaddr;
db94535d 3192 u32 blocksize = root->sectorsize;
39279cc3
CM		 3193 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3194 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3195 struct page *page;
3b16a4e3 3196 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3197 int ret = 0;
a52d9a80 3198 u64 page_start;
e6dcd2dc 3199 u64 page_end;
39279cc3
CM		 3200
3201 if ((offset & (blocksize - 1)) == 0)
3202 goto out;
0ca1f7ce 3203 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3204 if (ret)
3205 goto out;
39279cc3
CM		 3206
3207 ret = -ENOMEM;
211c17f5 3208again:
3b16a4e3 3209 page = find_or_create_page(mapping, index, mask);
5d5e103a 3210 if (!page) {
0ca1f7ce 3211 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3212 goto out;
5d5e103a 3213 }
e6dcd2dc
CM		 3214
3215 page_start = page_offset(page);
3216 page_end = page_start + PAGE_CACHE_SIZE - 1;
3217
39279cc3 3218 if (!PageUptodate(page)) {
9ebefb18 3219 ret = btrfs_readpage(NULL, page);
39279cc3 3220 lock_page(page);
211c17f5
CM		 3221 if (page->mapping != mapping) {
3222 unlock_page(page);
3223 page_cache_release(page);
3224 goto again;
3225 }
39279cc3
CM		 3226 if (!PageUptodate(page)) {
3227 ret = -EIO;
89642229 3228 goto out_unlock;
39279cc3
CM		 3229 }
3230 }
211c17f5 3231 wait_on_page_writeback(page);
e6dcd2dc 3232
2ac55d41
JB		 3233 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3234 GFP_NOFS);
e6dcd2dc
CM		 3235 set_page_extent_mapped(page);
3236
3237 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3238 if (ordered) {
2ac55d41
JB		 3239 unlock_extent_cached(io_tree, page_start, page_end,
3240 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3241 unlock_page(page);
3242 page_cache_release(page);
eb84ae03 3243 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3244 btrfs_put_ordered_extent(ordered);
3245 goto again;
3246 }
3247
2ac55d41 3248 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3249 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3250 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3251
2ac55d41
JB		 3252 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3253 &cached_state);
9ed74f2d 3254 if (ret) {
2ac55d41
JB		 3255 unlock_extent_cached(io_tree, page_start, page_end,
3256 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3257 goto out_unlock;
3258 }
3259
e6dcd2dc
CM		 3260 ret = 0;
3261 if (offset != PAGE_CACHE_SIZE) {
3262 kaddr = kmap(page);
3263 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3264 flush_dcache_page(page);
3265 kunmap(page);
3266 }
247e743c 3267 ClearPageChecked(page);
e6dcd2dc 3268 set_page_dirty(page);
2ac55d41
JB		 3269 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3270 GFP_NOFS);
39279cc3 3271
89642229 3272out_unlock:
5d5e103a 3273 if (ret)
0ca1f7ce 3274 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3275 unlock_page(page);
3276 page_cache_release(page);
3277out:
3278 return ret;
3279}
3280
695a0d0d
JB		 3281/*
3282 * This function puts in dummy file extents for the area we're creating a hole
3283 * for. So if we are truncating this file to a larger size we need to insert
3284 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3285 * the range between oldsize and size
3286 */
a41ad394 3287int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3288{
9036c102
YZ		 3289 struct btrfs_trans_handle *trans;
3290 struct btrfs_root *root = BTRFS_I(inode)->root;
3291 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3292 struct extent_map *em = NULL;
2ac55d41 3293 struct extent_state *cached_state = NULL;
9036c102 3294 u64 mask = root->sectorsize - 1;
a41ad394 3295 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3296 u64 block_end = (size + mask) & ~mask;
3297 u64 last_byte;
3298 u64 cur_offset;
3299 u64 hole_size;
9ed74f2d 3300 int err = 0;
39279cc3 3301
9036c102
YZ		 3302 if (size <= hole_start)
3303 return 0;
3304
9036c102
YZ		 3305 while (1) {
3306 struct btrfs_ordered_extent *ordered;
3307 btrfs_wait_ordered_range(inode, hole_start,
3308 block_end - hole_start);
2ac55d41
JB		 3309 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3310 &cached_state, GFP_NOFS);
9036c102
YZ		 3311 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3312 if (!ordered)
3313 break;
2ac55d41
JB		 3314 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3315 &cached_state, GFP_NOFS);
9036c102
YZ		 3316 btrfs_put_ordered_extent(ordered);
3317 }
39279cc3 3318
9036c102
YZ		 3319 cur_offset = hole_start;
3320 while (1) {
3321 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3322 block_end - cur_offset, 0);
c704005d 3323 BUG_ON(IS_ERR_OR_NULL(em));
9036c102
YZ		 3324 last_byte = min(extent_map_end(em), block_end);
3325 last_byte = (last_byte + mask) & ~mask;
8082510e 3326 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3327 u64 hint_byte = 0;
9036c102 3328 hole_size = last_byte - cur_offset;
9ed74f2d 3329
a22285a6
YZ		 3330 trans = btrfs_start_transaction(root, 2);
3331 if (IS_ERR(trans)) {
3332 err = PTR_ERR(trans);
9ed74f2d 3333 break;
a22285a6 3334 }
8082510e
YZ		 3335
3336 err = btrfs_drop_extents(trans, inode, cur_offset,
3337 cur_offset + hole_size,
3338 &hint_byte, 1);
5b397377
MX		 3339 if (err) {
3340 btrfs_end_transaction(trans, root);
3893e33b 3341 break;
5b397377 3342 }
8082510e 3343
9036c102 3344 err = btrfs_insert_file_extent(trans, root,
33345d01 3345 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3346 0, hole_size, 0, hole_size,
3347 0, 0, 0);
5b397377
MX		 3348 if (err) {
3349 btrfs_end_transaction(trans, root);
3893e33b 3350 break;
5b397377 3351 }
8082510e 3352
9036c102
YZ		 3353 btrfs_drop_extent_cache(inode, hole_start,
3354 last_byte - 1, 0);
8082510e
YZ		 3355
3356 btrfs_end_transaction(trans, root);
9036c102
YZ		 3357 }
3358 free_extent_map(em);
a22285a6 3359 em = NULL;
9036c102 3360 cur_offset = last_byte;
8082510e 3361 if (cur_offset >= block_end)
9036c102
YZ		 3362 break;
3363 }
1832a6d5 3364
a22285a6 3365 free_extent_map(em);
2ac55d41
JB		 3366 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3367 GFP_NOFS);
9036c102
YZ		 3368 return err;
3369}
39279cc3 3370
a41ad394 3371static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3372{
a41ad394 3373 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3374 int ret;
3375
a41ad394 3376 if (newsize == oldsize)
8082510e
YZ		 3377 return 0;
3378
a41ad394
JB		 3379 if (newsize > oldsize) {
3380 i_size_write(inode, newsize);
3381 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3382 truncate_pagecache(inode, oldsize, newsize);
3383 ret = btrfs_cont_expand(inode, oldsize, newsize);
8082510e 3384 if (ret) {
a41ad394 3385 btrfs_setsize(inode, oldsize);
8082510e
YZ		 3386 return ret;
3387 }
3388
930f028a 3389 mark_inode_dirty(inode);
a41ad394 3390 } else {
8082510e 3391
a41ad394
JB		 3392 /*
3393 * We're truncating a file that used to have good data down to
3394 * zero. Make sure it gets into the ordered flush list so that
3395 * any new writes get down to disk quickly.
3396 */
3397 if (newsize == 0)
3398 BTRFS_I(inode)->ordered_data_close = 1;
8082510e 3399
a41ad394
JB		 3400 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3401 truncate_setsize(inode, newsize);
3402 ret = btrfs_truncate(inode);
8082510e
YZ		 3403 }
3404
a41ad394 3405 return ret;
8082510e
YZ		 3406}
3407
9036c102
YZ		 3408static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3409{
3410 struct inode *inode = dentry->d_inode;
b83cc969 3411 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3412 int err;
39279cc3 3413
b83cc969
LZ		 3414 if (btrfs_root_readonly(root))
3415 return -EROFS;
3416
9036c102
YZ		 3417 err = inode_change_ok(inode, attr);
3418 if (err)
3419 return err;
2bf5a725 3420
5a3f23d5 3421 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3422 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3423 if (err)
3424 return err;
39279cc3 3425 }
9036c102 3426
1025774c
CH		 3427 if (attr->ia_valid) {
3428 setattr_copy(inode, attr);
3429 mark_inode_dirty(inode);
3430
3431 if (attr->ia_valid & ATTR_MODE)
3432 err = btrfs_acl_chmod(inode);
3433 }
33268eaf 3434
39279cc3
CM		 3435 return err;
3436}
61295eb8 3437
bd555975 3438void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3439{
3440 struct btrfs_trans_handle *trans;
3441 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3442 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3443 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3444 unsigned long nr;
39279cc3
CM		 3445 int ret;
3446
1abe9b8a 3447 trace_btrfs_inode_evict(inode);
3448
39279cc3 3449 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3450 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
2cf8572d 3451 btrfs_is_free_space_inode(root, inode)))
bd555975
AV		 3452 goto no_delete;
3453
39279cc3 3454 if (is_bad_inode(inode)) {
7b128766 3455 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3456 goto no_delete;
3457 }
bd555975 3458 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3459 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3460
c71bf099
YZ		 3461 if (root->fs_info->log_root_recovering) {
3462 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3463 goto no_delete;
3464 }
3465
76dda93c
YZ		 3466 if (inode->i_nlink > 0) {
3467 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3468 goto no_delete;
3469 }
3470
4289a667
JB		 3471 rsv = btrfs_alloc_block_rsv(root);
3472 if (!rsv) {
3473 btrfs_orphan_del(NULL, inode);
3474 goto no_delete;
3475 }
4a338542 3476 rsv->size = min_size;
726c35fa 3477 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3478
dbe674a9 3479 btrfs_i_size_write(inode, 0);
5f39d397 3480
4289a667
JB		 3481 /*
3482 * This is a bit simpler than btrfs_truncate since
3483 *
3484 * 1) We've already reserved our space for our orphan item in the
3485 * unlink.
3486 * 2) We're going to delete the inode item, so we don't need to update
3487 * it at all.
3488 *
3489 * So we just need to reserve some slack space in case we add bytes when
3490 * doing the truncate.
3491 */
8082510e 3492 while (1) {
aa38a711 3493 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
726c35fa
JB		 3494
3495 /*
3496 * Try and steal from the global reserve since we will
3497 * likely not use this space anyway, we want to try as
3498 * hard as possible to get this to work.
3499 */
3500 if (ret)
3501 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3502
d68fc57b 3503 if (ret) {
4289a667 3504 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3505 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3506 btrfs_orphan_del(NULL, inode);
3507 btrfs_free_block_rsv(root, rsv);
3508 goto no_delete;
3509 }
3510
3511 trans = btrfs_start_transaction(root, 0);
3512 if (IS_ERR(trans)) {
3513 btrfs_orphan_del(NULL, inode);
3514 btrfs_free_block_rsv(root, rsv);
3515 goto no_delete;
d68fc57b 3516 }
7b128766 3517
4289a667
JB		 3518 trans->block_rsv = rsv;
3519
d68fc57b 3520 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3521 if (ret != -EAGAIN)
3522 break;
85e21bac 3523
8082510e
YZ		 3524 nr = trans->blocks_used;
3525 btrfs_end_transaction(trans, root);
3526 trans = NULL;
3527 btrfs_btree_balance_dirty(root, nr);
3528 }
5f39d397 3529
4289a667
JB		 3530 btrfs_free_block_rsv(root, rsv);
3531
8082510e 3532 if (ret == 0) {
4289a667 3533 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3534 ret = btrfs_orphan_del(trans, inode);
3535 BUG_ON(ret);
3536 }
54aa1f4d 3537
4289a667 3538 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3539 if (!(root == root->fs_info->tree_root ||
3540 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3541 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3542
d3c2fdcf 3543 nr = trans->blocks_used;
54aa1f4d 3544 btrfs_end_transaction(trans, root);
d3c2fdcf 3545 btrfs_btree_balance_dirty(root, nr);
39279cc3 3546no_delete:
bd555975 3547 end_writeback(inode);
8082510e 3548 return;
39279cc3
CM		 3549}
3550
3551/*
3552 * this returns the key found in the dir entry in the location pointer.
3553 * If no dir entries were found, location->objectid is 0.
3554 */
3555static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3556 struct btrfs_key *location)
3557{
3558 const char *name = dentry->d_name.name;
3559 int namelen = dentry->d_name.len;
3560 struct btrfs_dir_item *di;
3561 struct btrfs_path *path;
3562 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3563 int ret = 0;
39279cc3
CM		 3564
3565 path = btrfs_alloc_path();
d8926bb3
MF		 3566 if (!path)
3567 return -ENOMEM;
3954401f 3568
33345d01 3569 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3570 namelen, 0);
0d9f7f3e
Y		 3571 if (IS_ERR(di))
3572 ret = PTR_ERR(di);
d397712b 3573
c704005d 3574 if (IS_ERR_OR_NULL(di))
3954401f 3575 goto out_err;
d397712b 3576
5f39d397 3577 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3578out:
39279cc3
CM		 3579 btrfs_free_path(path);
3580 return ret;
3954401f
CM		 3581out_err:
3582 location->objectid = 0;
3583 goto out;
39279cc3
CM		 3584}
3585
3586/*
3587 * when we hit a tree root in a directory, the btrfs part of the inode
3588 * needs to be changed to reflect the root directory of the tree root. This
3589 * is kind of like crossing a mount point.
3590 */
3591static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3592 struct inode *dir,
3593 struct dentry *dentry,
3594 struct btrfs_key *location,
3595 struct btrfs_root **sub_root)
39279cc3 3596{
4df27c4d
YZ		 3597 struct btrfs_path *path;
3598 struct btrfs_root *new_root;
3599 struct btrfs_root_ref *ref;
3600 struct extent_buffer *leaf;
3601 int ret;
3602 int err = 0;
39279cc3 3603
4df27c4d
YZ		 3604 path = btrfs_alloc_path();
3605 if (!path) {
3606 err = -ENOMEM;
3607 goto out;
3608 }
39279cc3 3609
4df27c4d
YZ		 3610 err = -ENOENT;
3611 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3612 BTRFS_I(dir)->root->root_key.objectid,
3613 location->objectid);
3614 if (ret) {
3615 if (ret < 0)
3616 err = ret;
3617 goto out;
3618 }
39279cc3 3619
4df27c4d
YZ		 3620 leaf = path->nodes[0];
3621 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3622 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3623 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3624 goto out;
39279cc3 3625
4df27c4d
YZ		 3626 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3627 (unsigned long)(ref + 1),
3628 dentry->d_name.len);
3629 if (ret)
3630 goto out;
3631
b3b4aa74 3632 btrfs_release_path(path);
4df27c4d
YZ		 3633
3634 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3635 if (IS_ERR(new_root)) {
3636 err = PTR_ERR(new_root);
3637 goto out;
3638 }
3639
3640 if (btrfs_root_refs(&new_root->root_item) == 0) {
3641 err = -ENOENT;
3642 goto out;
3643 }
3644
3645 *sub_root = new_root;
3646 location->objectid = btrfs_root_dirid(&new_root->root_item);
3647 location->type = BTRFS_INODE_ITEM_KEY;
3648 location->offset = 0;
3649 err = 0;
3650out:
3651 btrfs_free_path(path);
3652 return err;
39279cc3
CM		 3653}
3654
5d4f98a2
YZ		 3655static void inode_tree_add(struct inode *inode)
3656{
3657 struct btrfs_root *root = BTRFS_I(inode)->root;
3658 struct btrfs_inode *entry;
03e860bd
FNP		 3659 struct rb_node **p;
3660 struct rb_node *parent;
33345d01 3661 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3662again:
3663 p = &root->inode_tree.rb_node;
3664 parent = NULL;
5d4f98a2 3665
1d3382cb 3666 if (inode_unhashed(inode))
76dda93c
YZ		 3667 return;
3668
5d4f98a2
YZ		 3669 spin_lock(&root->inode_lock);
3670 while (*p) {
3671 parent = *p;
3672 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3673
33345d01 3674 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3675 p = &parent->rb_left;
33345d01 3676 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3677 p = &parent->rb_right;
5d4f98a2
YZ		 3678 else {
3679 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3680 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3681 rb_erase(parent, &root->inode_tree);
3682 RB_CLEAR_NODE(parent);
3683 spin_unlock(&root->inode_lock);
3684 goto again;
5d4f98a2
YZ		 3685 }
3686 }
3687 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3688 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3689 spin_unlock(&root->inode_lock);
3690}
3691
3692static void inode_tree_del(struct inode *inode)
3693{
3694 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3695 int empty = 0;
5d4f98a2 3696
03e860bd 3697 spin_lock(&root->inode_lock);
5d4f98a2 3698 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3699 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3700 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3701 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3702 }
03e860bd 3703 spin_unlock(&root->inode_lock);
76dda93c 3704
0af3d00b
JB		 3705 /*
3706 * Free space cache has inodes in the tree root, but the tree root has a
3707 * root_refs of 0, so this could end up dropping the tree root as a
3708 * snapshot, so we need the extra !root->fs_info->tree_root check to
3709 * make sure we don't drop it.
3710 */
3711 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3712 root != root->fs_info->tree_root) {
76dda93c
YZ		 3713 synchronize_srcu(&root->fs_info->subvol_srcu);
3714 spin_lock(&root->inode_lock);
3715 empty = RB_EMPTY_ROOT(&root->inode_tree);
3716 spin_unlock(&root->inode_lock);
3717 if (empty)
3718 btrfs_add_dead_root(root);
3719 }
3720}
3721
3722int btrfs_invalidate_inodes(struct btrfs_root *root)
3723{
3724 struct rb_node *node;
3725 struct rb_node *prev;
3726 struct btrfs_inode *entry;
3727 struct inode *inode;
3728 u64 objectid = 0;
3729
3730 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3731
3732 spin_lock(&root->inode_lock);
3733again:
3734 node = root->inode_tree.rb_node;
3735 prev = NULL;
3736 while (node) {
3737 prev = node;
3738 entry = rb_entry(node, struct btrfs_inode, rb_node);
3739
33345d01 3740 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 3741 node = node->rb_left;
33345d01 3742 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 3743 node = node->rb_right;
3744 else
3745 break;
3746 }
3747 if (!node) {
3748 while (prev) {
3749 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 3750 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 3751 node = prev;
3752 break;
3753 }
3754 prev = rb_next(prev);
3755 }
3756 }
3757 while (node) {
3758 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 3759 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 3760 inode = igrab(&entry->vfs_inode);
3761 if (inode) {
3762 spin_unlock(&root->inode_lock);
3763 if (atomic_read(&inode->i_count) > 1)
3764 d_prune_aliases(inode);
3765 /*
45321ac5 3766 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 3767 * the inode cache when its usage count
3768 * hits zero.
3769 */
3770 iput(inode);
3771 cond_resched();
3772 spin_lock(&root->inode_lock);
3773 goto again;
3774 }
3775
3776 if (cond_resched_lock(&root->inode_lock))
3777 goto again;
3778
3779 node = rb_next(node);
3780 }
3781 spin_unlock(&root->inode_lock);
3782 return 0;
5d4f98a2
YZ		 3783}
3784
e02119d5
CM		 3785static int btrfs_init_locked_inode(struct inode *inode, void *p)
3786{
3787 struct btrfs_iget_args *args = p;
3788 inode->i_ino = args->ino;
e02119d5 3789 BTRFS_I(inode)->root = args->root;
6a63209f 3790 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3791 return 0;
3792}
3793
3794static int btrfs_find_actor(struct inode *inode, void *opaque)
3795{
3796 struct btrfs_iget_args *args = opaque;
33345d01 3797 return args->ino == btrfs_ino(inode) &&
d397712b 3798 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3799}
3800
5d4f98a2
YZ		 3801static struct inode *btrfs_iget_locked(struct super_block *s,
3802 u64 objectid,
3803 struct btrfs_root *root)
39279cc3
CM		 3804{
3805 struct inode *inode;
3806 struct btrfs_iget_args args;
3807 args.ino = objectid;
3808 args.root = root;
3809
3810 inode = iget5_locked(s, objectid, btrfs_find_actor,
3811 btrfs_init_locked_inode,
3812 (void *)&args);
3813 return inode;
3814}
3815
1a54ef8c
BR		 3816/* Get an inode object given its location and corresponding root.
3817 * Returns in *is_new if the inode was read from disk
3818 */
3819struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 3820 struct btrfs_root *root, int *new)
1a54ef8c
BR		 3821{
3822 struct inode *inode;
3823
3824 inode = btrfs_iget_locked(s, location->objectid, root);
3825 if (!inode)
5d4f98a2 3826 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3827
3828 if (inode->i_state & I_NEW) {
3829 BTRFS_I(inode)->root = root;
3830 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3831 btrfs_read_locked_inode(inode);
1748f843
MF		 3832 if (!is_bad_inode(inode)) {
3833 inode_tree_add(inode);
3834 unlock_new_inode(inode);
3835 if (new)
3836 *new = 1;
3837 } else {
e0b6d65b
ST		 3838 unlock_new_inode(inode);
3839 iput(inode);
3840 inode = ERR_PTR(-ESTALE);
1748f843
MF		 3841 }
3842 }
3843
1a54ef8c
BR		 3844 return inode;
3845}
3846
4df27c4d
YZ		 3847static struct inode *new_simple_dir(struct super_block *s,
3848 struct btrfs_key *key,
3849 struct btrfs_root *root)
3850{
3851 struct inode *inode = new_inode(s);
3852
3853 if (!inode)
3854 return ERR_PTR(-ENOMEM);
3855
4df27c4d
YZ		 3856 BTRFS_I(inode)->root = root;
3857 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3858 BTRFS_I(inode)->dummy_inode = 1;
3859
3860 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3861 inode->i_op = &simple_dir_inode_operations;
3862 inode->i_fop = &simple_dir_operations;
3863 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3864 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3865
3866 return inode;
3867}
3868
3de4586c 3869struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3870{
d397712b 3871 struct inode *inode;
4df27c4d 3872 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 3873 struct btrfs_root *sub_root = root;
3874 struct btrfs_key location;
76dda93c 3875 int index;
b4aff1f8 3876 int ret = 0;
39279cc3
CM		 3877
3878 if (dentry->d_name.len > BTRFS_NAME_LEN)
3879 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3880
b4aff1f8
JB		 3881 if (unlikely(d_need_lookup(dentry))) {
3882 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3883 kfree(dentry->d_fsdata);
3884 dentry->d_fsdata = NULL;
a66e7cc6
JB		 3885 /* This thing is hashed, drop it for now */
3886 d_drop(dentry);
b4aff1f8
JB		 3887 } else {
3888 ret = btrfs_inode_by_name(dir, dentry, &location);
3889 }
5f39d397 3890
39279cc3
CM		 3891 if (ret < 0)
3892 return ERR_PTR(ret);
5f39d397 3893
4df27c4d
YZ		 3894 if (location.objectid == 0)
3895 return NULL;
3896
3897 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 3898 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 3899 return inode;
3900 }
3901
3902 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3903
76dda93c 3904 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 3905 ret = fixup_tree_root_location(root, dir, dentry,
3906 &location, &sub_root);
3907 if (ret < 0) {
3908 if (ret != -ENOENT)
3909 inode = ERR_PTR(ret);
3910 else
3911 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3912 } else {
73f73415 3913 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 3914 }
76dda93c
YZ		 3915 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3916
34d19bad 3917 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 3918 down_read(&root->fs_info->cleanup_work_sem);
3919 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 3920 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 3921 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 3922 if (ret)
3923 inode = ERR_PTR(ret);
c71bf099
YZ		 3924 }
3925
3de4586c
CM		 3926 return inode;
3927}
3928
fe15ce44 3929static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 3930{
3931 struct btrfs_root *root;
3932
efefb143
YZ		 3933 if (!dentry->d_inode && !IS_ROOT(dentry))
3934 dentry = dentry->d_parent;
76dda93c 3935
efefb143
YZ		 3936 if (dentry->d_inode) {
3937 root = BTRFS_I(dentry->d_inode)->root;
3938 if (btrfs_root_refs(&root->root_item) == 0)
3939 return 1;
3940 }
76dda93c
YZ		 3941 return 0;
3942}
3943
b4aff1f8
JB		 3944static void btrfs_dentry_release(struct dentry *dentry)
3945{
3946 if (dentry->d_fsdata)
3947 kfree(dentry->d_fsdata);
3948}
3949
3de4586c
CM		 3950static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3951 struct nameidata *nd)
3952{
a66e7cc6
JB		 3953 struct dentry *ret;
3954
3955 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3956 if (unlikely(d_need_lookup(dentry))) {
3957 spin_lock(&dentry->d_lock);
3958 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3959 spin_unlock(&dentry->d_lock);
3960 }
3961 return ret;
39279cc3
CM		 3962}
3963
16cdcec7 3964unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 3965 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3966};
3967
cbdf5a24
DW		 3968static int btrfs_real_readdir(struct file *filp, void *dirent,
3969 filldir_t filldir)
39279cc3 3970{
6da6abae 3971 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3972 struct btrfs_root *root = BTRFS_I(inode)->root;
3973 struct btrfs_item *item;
3974 struct btrfs_dir_item *di;
3975 struct btrfs_key key;
5f39d397 3976 struct btrfs_key found_key;
39279cc3 3977 struct btrfs_path *path;
16cdcec7
MX		 3978 struct list_head ins_list;
3979 struct list_head del_list;
b4aff1f8 3980 struct qstr q;
39279cc3 3981 int ret;
5f39d397 3982 struct extent_buffer *leaf;
39279cc3 3983 int slot;
39279cc3
CM		 3984 unsigned char d_type;
3985 int over = 0;
3986 u32 di_cur;
3987 u32 di_total;
3988 u32 di_len;
3989 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3990 char tmp_name[32];
3991 char *name_ptr;
3992 int name_len;
16cdcec7 3993 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 3994
3995 /* FIXME, use a real flag for deciding about the key type */
3996 if (root->fs_info->tree_root == root)
3997 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3998
3954401f
CM		 3999 /* special case for "." */
4000 if (filp->f_pos == 0) {
3765fefa
HS		 4001 over = filldir(dirent, ".", 1,
4002 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4003 if (over)
4004 return 0;
4005 filp->f_pos = 1;
4006 }
3954401f
CM		 4007 /* special case for .., just use the back ref */
4008 if (filp->f_pos == 1) {
5ecc7e5d 4009 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4010 over = filldir(dirent, "..", 2,
3765fefa 4011 filp->f_pos, pino, DT_DIR);
3954401f 4012 if (over)
49593bfa 4013 return 0;
3954401f
CM		 4014 filp->f_pos = 2;
4015 }
49593bfa 4016 path = btrfs_alloc_path();
16cdcec7
MX		 4017 if (!path)
4018 return -ENOMEM;
ff5714cc 4019
026fd317 4020 path->reada = 1;
49593bfa 4021
16cdcec7
MX		 4022 if (key_type == BTRFS_DIR_INDEX_KEY) {
4023 INIT_LIST_HEAD(&ins_list);
4024 INIT_LIST_HEAD(&del_list);
4025 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4026 }
4027
39279cc3
CM		 4028 btrfs_set_key_type(&key, key_type);
4029 key.offset = filp->f_pos;
33345d01 4030 key.objectid = btrfs_ino(inode);
5f39d397 4031
39279cc3
CM		 4032 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4033 if (ret < 0)
4034 goto err;
49593bfa
DW		 4035
4036 while (1) {
5f39d397 4037 leaf = path->nodes[0];
39279cc3 4038 slot = path->slots[0];
b9e03af0
LZ		 4039 if (slot >= btrfs_header_nritems(leaf)) {
4040 ret = btrfs_next_leaf(root, path);
4041 if (ret < 0)
4042 goto err;
4043 else if (ret > 0)
4044 break;
4045 continue;
39279cc3 4046 }
3de4586c 4047
5f39d397
CM		 4048 item = btrfs_item_nr(leaf, slot);
4049 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4050
4051 if (found_key.objectid != key.objectid)
39279cc3 4052 break;
5f39d397 4053 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4054 break;
5f39d397 4055 if (found_key.offset < filp->f_pos)
b9e03af0 4056 goto next;
16cdcec7
MX		 4057 if (key_type == BTRFS_DIR_INDEX_KEY &&
4058 btrfs_should_delete_dir_index(&del_list,
4059 found_key.offset))
4060 goto next;
5f39d397
CM		 4061
4062 filp->f_pos = found_key.offset;
16cdcec7 4063 is_curr = 1;
49593bfa 4064
39279cc3
CM		 4065 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4066 di_cur = 0;
5f39d397 4067 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4068
4069 while (di_cur < di_total) {
5f39d397 4070 struct btrfs_key location;
b4aff1f8 4071 struct dentry *tmp;
5f39d397 4072
22a94d44
JB		 4073 if (verify_dir_item(root, leaf, di))
4074 break;
4075
5f39d397 4076 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4077 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4078 name_ptr = tmp_name;
4079 } else {
4080 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4081 if (!name_ptr) {
4082 ret = -ENOMEM;
4083 goto err;
4084 }
5f39d397
CM		 4085 }
4086 read_extent_buffer(leaf, name_ptr,
4087 (unsigned long)(di + 1), name_len);
4088
4089 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4090 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4091
b4aff1f8
JB		 4092 q.name = name_ptr;
4093 q.len = name_len;
4094 q.hash = full_name_hash(q.name, q.len);
4095 tmp = d_lookup(filp->f_dentry, &q);
4096 if (!tmp) {
4097 struct btrfs_key *newkey;
4098
4099 newkey = kzalloc(sizeof(struct btrfs_key),
4100 GFP_NOFS);
4101 if (!newkey)
4102 goto no_dentry;
4103 tmp = d_alloc(filp->f_dentry, &q);
4104 if (!tmp) {
4105 kfree(newkey);
4106 dput(tmp);
4107 goto no_dentry;
4108 }
4109 memcpy(newkey, &location,
4110 sizeof(struct btrfs_key));
4111 tmp->d_fsdata = newkey;
4112 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4113 d_rehash(tmp);
4114 dput(tmp);
4115 } else {
4116 dput(tmp);
4117 }
4118no_dentry:
3de4586c
CM		 4119 /* is this a reference to our own snapshot? If so
4120 * skip it
4121 */
4122 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4123 location.objectid == root->root_key.objectid) {
4124 over = 0;
4125 goto skip;
4126 }
5f39d397 4127 over = filldir(dirent, name_ptr, name_len,
49593bfa 4128 found_key.offset, location.objectid,
39279cc3 4129 d_type);
5f39d397 4130
3de4586c 4131skip:
5f39d397
CM		 4132 if (name_ptr != tmp_name)
4133 kfree(name_ptr);
4134
39279cc3
CM		 4135 if (over)
4136 goto nopos;
5103e947 4137 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4138 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4139 di_cur += di_len;
4140 di = (struct btrfs_dir_item *)((char *)di + di_len);
4141 }
b9e03af0
LZ		 4142next:
4143 path->slots[0]++;
39279cc3 4144 }
49593bfa 4145
16cdcec7
MX		 4146 if (key_type == BTRFS_DIR_INDEX_KEY) {
4147 if (is_curr)
4148 filp->f_pos++;
4149 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4150 &ins_list);
4151 if (ret)
4152 goto nopos;
4153 }
4154
49593bfa 4155 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4156 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4157 /*
4158 * 32-bit glibc will use getdents64, but then strtol -
4159 * so the last number we can serve is this.
4160 */
4161 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4162 else
4163 filp->f_pos++;
39279cc3
CM		 4164nopos:
4165 ret = 0;
4166err:
16cdcec7
MX		 4167 if (key_type == BTRFS_DIR_INDEX_KEY)
4168 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4169 btrfs_free_path(path);
39279cc3
CM		 4170 return ret;
4171}
4172
a9185b41 4173int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4174{
4175 struct btrfs_root *root = BTRFS_I(inode)->root;
4176 struct btrfs_trans_handle *trans;
4177 int ret = 0;
0af3d00b 4178 bool nolock = false;
39279cc3 4179
8929ecfa 4180 if (BTRFS_I(inode)->dummy_inode)
4ca8b41e
CM		 4181 return 0;
4182
2cf8572d 4183 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
82d5902d 4184 nolock = true;
0af3d00b 4185
a9185b41 4186 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4187 if (nolock)
7a7eaa40 4188 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4189 else
7a7eaa40 4190 trans = btrfs_join_transaction(root);
3612b495
TI		 4191 if (IS_ERR(trans))
4192 return PTR_ERR(trans);
0af3d00b
JB		 4193 if (nolock)
4194 ret = btrfs_end_transaction_nolock(trans, root);
4195 else
4196 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4197 }
4198 return ret;
4199}
4200
4201/*
54aa1f4d 4202 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4203 * inode changes. But, it is most likely to find the inode in cache.
4204 * FIXME, needs more benchmarking...there are no reasons other than performance
4205 * to keep or drop this code.
4206 */
aa385729 4207void btrfs_dirty_inode(struct inode *inode, int flags)
39279cc3
CM		 4208{
4209 struct btrfs_root *root = BTRFS_I(inode)->root;
4210 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4211 int ret;
4212
4213 if (BTRFS_I(inode)->dummy_inode)
4214 return;
39279cc3 4215
7a7eaa40 4216 trans = btrfs_join_transaction(root);
3612b495 4217 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 4218
4219 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4220 if (ret && ret == -ENOSPC) {
4221 /* whoops, lets try again with the full transaction */
4222 btrfs_end_transaction(trans, root);
4223 trans = btrfs_start_transaction(root, 1);
9aeead73 4224 if (IS_ERR(trans)) {
7a36ddec 4225 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4226 "dirty inode %llu error %ld\n",
4227 (unsigned long long)btrfs_ino(inode),
4228 PTR_ERR(trans));
9aeead73
CM		 4229 return;
4230 }
8929ecfa 4231
94b60442
CM		 4232 ret = btrfs_update_inode(trans, root, inode);
4233 if (ret) {
7a36ddec 4234 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4235 "dirty inode %llu error %d\n",
4236 (unsigned long long)btrfs_ino(inode),
4237 ret);
94b60442
CM		 4238 }
4239 }
39279cc3 4240 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4241 if (BTRFS_I(inode)->delayed_node)
4242 btrfs_balance_delayed_items(root);
39279cc3
CM		 4243}
4244
d352ac68
CM		 4245/*
4246 * find the highest existing sequence number in a directory
4247 * and then set the in-memory index_cnt variable to reflect
4248 * free sequence numbers
4249 */
aec7477b
JB		 4250static int btrfs_set_inode_index_count(struct inode *inode)
4251{
4252 struct btrfs_root *root = BTRFS_I(inode)->root;
4253 struct btrfs_key key, found_key;
4254 struct btrfs_path *path;
4255 struct extent_buffer *leaf;
4256 int ret;
4257
33345d01 4258 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4259 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4260 key.offset = (u64)-1;
4261
4262 path = btrfs_alloc_path();
4263 if (!path)
4264 return -ENOMEM;
4265
4266 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4267 if (ret < 0)
4268 goto out;
4269 /* FIXME: we should be able to handle this */
4270 if (ret == 0)
4271 goto out;
4272 ret = 0;
4273
4274 /*
4275 * MAGIC NUMBER EXPLANATION:
4276 * since we search a directory based on f_pos we have to start at 2
4277 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4278 * else has to start at 2
4279 */
4280 if (path->slots[0] == 0) {
4281 BTRFS_I(inode)->index_cnt = 2;
4282 goto out;
4283 }
4284
4285 path->slots[0]--;
4286
4287 leaf = path->nodes[0];
4288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4289
33345d01 4290 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4291 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4292 BTRFS_I(inode)->index_cnt = 2;
4293 goto out;
4294 }
4295
4296 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4297out:
4298 btrfs_free_path(path);
4299 return ret;
4300}
4301
d352ac68
CM		 4302/*
4303 * helper to find a free sequence number in a given directory. This current
4304 * code is very simple, later versions will do smarter things in the btree
4305 */
3de4586c 4306int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4307{
4308 int ret = 0;
4309
4310 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4311 ret = btrfs_inode_delayed_dir_index_count(dir);
4312 if (ret) {
4313 ret = btrfs_set_inode_index_count(dir);
4314 if (ret)
4315 return ret;
4316 }
aec7477b
JB		 4317 }
4318
00e4e6b3 4319 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4320 BTRFS_I(dir)->index_cnt++;
4321
4322 return ret;
4323}
4324
39279cc3
CM		 4325static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4326 struct btrfs_root *root,
aec7477b 4327 struct inode *dir,
9c58309d 4328 const char *name, int name_len,
d82a6f1d
JB		 4329 u64 ref_objectid, u64 objectid, int mode,
4330 u64 *index)
39279cc3
CM		 4331{
4332 struct inode *inode;
5f39d397 4333 struct btrfs_inode_item *inode_item;
39279cc3 4334 struct btrfs_key *location;
5f39d397 4335 struct btrfs_path *path;
9c58309d
CM		 4336 struct btrfs_inode_ref *ref;
4337 struct btrfs_key key[2];
4338 u32 sizes[2];
4339 unsigned long ptr;
39279cc3
CM		 4340 int ret;
4341 int owner;
4342
5f39d397 4343 path = btrfs_alloc_path();
d8926bb3
MF		 4344 if (!path)
4345 return ERR_PTR(-ENOMEM);
5f39d397 4346
39279cc3 4347 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4348 if (!inode) {
4349 btrfs_free_path(path);
39279cc3 4350 return ERR_PTR(-ENOMEM);
8fb27640 4351 }
39279cc3 4352
581bb050
LZ		 4353 /*
4354 * we have to initialize this early, so we can reclaim the inode
4355 * number if we fail afterwards in this function.
4356 */
4357 inode->i_ino = objectid;
4358
aec7477b 4359 if (dir) {
1abe9b8a 4360 trace_btrfs_inode_request(dir);
4361
3de4586c 4362 ret = btrfs_set_inode_index(dir, index);
09771430 4363 if (ret) {
8fb27640 4364 btrfs_free_path(path);
09771430 4365 iput(inode);
aec7477b 4366 return ERR_PTR(ret);
09771430 4367 }
aec7477b
JB		 4368 }
4369 /*
4370 * index_cnt is ignored for everything but a dir,
4371 * btrfs_get_inode_index_count has an explanation for the magic
4372 * number
4373 */
4374 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4375 BTRFS_I(inode)->root = root;
e02119d5 4376 BTRFS_I(inode)->generation = trans->transid;
76195853 4377 inode->i_generation = BTRFS_I(inode)->generation;
6a63209f 4378 btrfs_set_inode_space_info(root, inode);
b888db2b 4379
569254b0 4380 if (S_ISDIR(mode))
39279cc3
CM		 4381 owner = 0;
4382 else
4383 owner = 1;
9c58309d
CM		 4384
4385 key[0].objectid = objectid;
4386 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4387 key[0].offset = 0;
4388
4389 key[1].objectid = objectid;
4390 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4391 key[1].offset = ref_objectid;
4392
4393 sizes[0] = sizeof(struct btrfs_inode_item);
4394 sizes[1] = name_len + sizeof(*ref);
4395
b9473439 4396 path->leave_spinning = 1;
9c58309d
CM		 4397 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4398 if (ret != 0)
5f39d397
CM		 4399 goto fail;
4400
ecc11fab 4401 inode_init_owner(inode, dir, mode);
a76a3cd4 4402 inode_set_bytes(inode, 0);
39279cc3 4403 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4404 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4405 struct btrfs_inode_item);
e02119d5 4406 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4407
4408 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4409 struct btrfs_inode_ref);
4410 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4411 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4412 ptr = (unsigned long)(ref + 1);
4413 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4414
5f39d397
CM		 4415 btrfs_mark_buffer_dirty(path->nodes[0]);
4416 btrfs_free_path(path);
4417
39279cc3
CM		 4418 location = &BTRFS_I(inode)->location;
4419 location->objectid = objectid;
39279cc3
CM		 4420 location->offset = 0;
4421 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4422
6cbff00f
CH		 4423 btrfs_inherit_iflags(inode, dir);
4424
569254b0 4425 if (S_ISREG(mode)) {
94272164
CM		 4426 if (btrfs_test_opt(root, NODATASUM))
4427 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4428 if (btrfs_test_opt(root, NODATACOW) ||
4429 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4430 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4431 }
4432
39279cc3 4433 insert_inode_hash(inode);
5d4f98a2 4434 inode_tree_add(inode);
1abe9b8a 4435
4436 trace_btrfs_inode_new(inode);
1973f0fa 4437 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4438
39279cc3 4439 return inode;
5f39d397 4440fail:
aec7477b
JB		 4441 if (dir)
4442 BTRFS_I(dir)->index_cnt--;
5f39d397 4443 btrfs_free_path(path);
09771430 4444 iput(inode);
5f39d397 4445 return ERR_PTR(ret);
39279cc3
CM		 4446}
4447
4448static inline u8 btrfs_inode_type(struct inode *inode)
4449{
4450 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4451}
4452
d352ac68
CM		 4453/*
4454 * utility function to add 'inode' into 'parent_inode' with
4455 * a give name and a given sequence number.
4456 * if 'add_backref' is true, also insert a backref from the
4457 * inode to the parent directory.
4458 */
e02119d5
CM		 4459int btrfs_add_link(struct btrfs_trans_handle *trans,
4460 struct inode *parent_inode, struct inode *inode,
4461 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4462{
4df27c4d 4463 int ret = 0;
39279cc3 4464 struct btrfs_key key;
e02119d5 4465 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4466 u64 ino = btrfs_ino(inode);
4467 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4468
33345d01 4469 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4470 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4471 } else {
33345d01 4472 key.objectid = ino;
4df27c4d
YZ		 4473 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4474 key.offset = 0;
4475 }
4476
33345d01 4477 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4478 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4479 key.objectid, root->root_key.objectid,
33345d01 4480 parent_ino, index, name, name_len);
4df27c4d 4481 } else if (add_backref) {
33345d01
LZ		 4482 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4483 parent_ino, index);
4df27c4d 4484 }
39279cc3 4485
39279cc3 4486 if (ret == 0) {
4df27c4d 4487 ret = btrfs_insert_dir_item(trans, root, name, name_len,
16cdcec7 4488 parent_inode, &key,
4df27c4d
YZ		 4489 btrfs_inode_type(inode), index);
4490 BUG_ON(ret);
4491
dbe674a9 4492 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 4493 name_len * 2);
79c44584 4494 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 4495 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 4496 }
4497 return ret;
4498}
4499
4500static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4501 struct inode *dir, struct dentry *dentry,
4502 struct inode *inode, int backref, u64 index)
39279cc3 4503{
a1b075d2
JB		 4504 int err = btrfs_add_link(trans, dir, inode,
4505 dentry->d_name.name, dentry->d_name.len,
4506 backref, index);
39279cc3
CM		 4507 if (!err) {
4508 d_instantiate(dentry, inode);
4509 return 0;
4510 }
4511 if (err > 0)
4512 err = -EEXIST;
4513 return err;
4514}
4515
618e21d5
JB		 4516static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4517 int mode, dev_t rdev)
4518{
4519 struct btrfs_trans_handle *trans;
4520 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4521 struct inode *inode = NULL;
618e21d5
JB		 4522 int err;
4523 int drop_inode = 0;
4524 u64 objectid;
1832a6d5 4525 unsigned long nr = 0;
00e4e6b3 4526 u64 index = 0;
618e21d5
JB		 4527
4528 if (!new_valid_dev(rdev))
4529 return -EINVAL;
4530
9ed74f2d
JB		 4531 /*
4532 * 2 for inode item and ref
4533 * 2 for dir items
4534 * 1 for xattr if selinux is on
4535 */
a22285a6
YZ		 4536 trans = btrfs_start_transaction(root, 5);
4537 if (IS_ERR(trans))
4538 return PTR_ERR(trans);
1832a6d5 4539
581bb050
LZ		 4540 err = btrfs_find_free_ino(root, &objectid);
4541 if (err)
4542 goto out_unlock;
4543
aec7477b 4544 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4545 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4546 mode, &index);
7cf96da3
TI		 4547 if (IS_ERR(inode)) {
4548 err = PTR_ERR(inode);
618e21d5 4549 goto out_unlock;
7cf96da3 4550 }
618e21d5 4551
2a7dba39 4552 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4553 if (err) {
4554 drop_inode = 1;
4555 goto out_unlock;
4556 }
4557
a1b075d2 4558 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4559 if (err)
4560 drop_inode = 1;
4561 else {
4562 inode->i_op = &btrfs_special_inode_operations;
4563 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4564 btrfs_update_inode(trans, root, inode);
618e21d5 4565 }
618e21d5 4566out_unlock:
d3c2fdcf 4567 nr = trans->blocks_used;
89ce8a63 4568 btrfs_end_transaction_throttle(trans, root);
a22285a6 4569 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4570 if (drop_inode) {
4571 inode_dec_link_count(inode);
4572 iput(inode);
4573 }
618e21d5
JB		 4574 return err;
4575}
4576
39279cc3
CM		 4577static int btrfs_create(struct inode *dir, struct dentry *dentry,
4578 int mode, struct nameidata *nd)
4579{
4580 struct btrfs_trans_handle *trans;
4581 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4582 struct inode *inode = NULL;
39279cc3 4583 int drop_inode = 0;
a22285a6 4584 int err;
1832a6d5 4585 unsigned long nr = 0;
39279cc3 4586 u64 objectid;
00e4e6b3 4587 u64 index = 0;
39279cc3 4588
9ed74f2d
JB		 4589 /*
4590 * 2 for inode item and ref
4591 * 2 for dir items
4592 * 1 for xattr if selinux is on
4593 */
a22285a6
YZ		 4594 trans = btrfs_start_transaction(root, 5);
4595 if (IS_ERR(trans))
4596 return PTR_ERR(trans);
9ed74f2d 4597
581bb050
LZ		 4598 err = btrfs_find_free_ino(root, &objectid);
4599 if (err)
4600 goto out_unlock;
4601
aec7477b 4602 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4603 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4604 mode, &index);
7cf96da3
TI		 4605 if (IS_ERR(inode)) {
4606 err = PTR_ERR(inode);
39279cc3 4607 goto out_unlock;
7cf96da3 4608 }
39279cc3 4609
2a7dba39 4610 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4611 if (err) {
4612 drop_inode = 1;
4613 goto out_unlock;
4614 }
4615
a1b075d2 4616 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4617 if (err)
4618 drop_inode = 1;
4619 else {
4620 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4621 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4622 inode->i_fop = &btrfs_file_operations;
4623 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4624 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 4625 }
39279cc3 4626out_unlock:
d3c2fdcf 4627 nr = trans->blocks_used;
ab78c84d 4628 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4629 if (drop_inode) {
4630 inode_dec_link_count(inode);
4631 iput(inode);
4632 }
d3c2fdcf 4633 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4634 return err;
4635}
4636
4637static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4638 struct dentry *dentry)
4639{
4640 struct btrfs_trans_handle *trans;
4641 struct btrfs_root *root = BTRFS_I(dir)->root;
4642 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4643 u64 index;
1832a6d5 4644 unsigned long nr = 0;
39279cc3
CM		 4645 int err;
4646 int drop_inode = 0;
4647
4a8be425
TH		 4648 /* do not allow sys_link's with other subvols of the same device */
4649 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4650 return -EXDEV;
4a8be425 4651
c055e99e
AV		 4652 if (inode->i_nlink == ~0U)
4653 return -EMLINK;
4a8be425 4654
3de4586c 4655 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4656 if (err)
4657 goto fail;
4658
a22285a6 4659 /*
7e6b6465 4660 * 2 items for inode and inode ref
a22285a6 4661 * 2 items for dir items
7e6b6465 4662 * 1 item for parent inode
a22285a6 4663 */
7e6b6465 4664 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4665 if (IS_ERR(trans)) {
4666 err = PTR_ERR(trans);
4667 goto fail;
4668 }
5f39d397 4669
3153495d
MX		 4670 btrfs_inc_nlink(inode);
4671 inode->i_ctime = CURRENT_TIME;
7de9c6ee 4672 ihold(inode);
aec7477b 4673
a1b075d2 4674 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 4675
a5719521 4676 if (err) {
54aa1f4d 4677 drop_inode = 1;
a5719521 4678 } else {
10d9f309 4679 struct dentry *parent = dentry->d_parent;
a5719521
YZ		 4680 err = btrfs_update_inode(trans, root, inode);
4681 BUG_ON(err);
6a912213 4682 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 4683 }
39279cc3 4684
d3c2fdcf 4685 nr = trans->blocks_used;
ab78c84d 4686 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4687fail:
39279cc3
CM		 4688 if (drop_inode) {
4689 inode_dec_link_count(inode);
4690 iput(inode);
4691 }
d3c2fdcf 4692 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4693 return err;
4694}
4695
39279cc3
CM		 4696static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4697{
b9d86667 4698 struct inode *inode = NULL;
39279cc3
CM		 4699 struct btrfs_trans_handle *trans;
4700 struct btrfs_root *root = BTRFS_I(dir)->root;
4701 int err = 0;
4702 int drop_on_err = 0;
b9d86667 4703 u64 objectid = 0;
00e4e6b3 4704 u64 index = 0;
d3c2fdcf 4705 unsigned long nr = 1;
39279cc3 4706
9ed74f2d
JB		 4707 /*
4708 * 2 items for inode and ref
4709 * 2 items for dir items
4710 * 1 for xattr if selinux is on
4711 */
a22285a6
YZ		 4712 trans = btrfs_start_transaction(root, 5);
4713 if (IS_ERR(trans))
4714 return PTR_ERR(trans);
39279cc3 4715
581bb050
LZ		 4716 err = btrfs_find_free_ino(root, &objectid);
4717 if (err)
4718 goto out_fail;
4719
aec7477b 4720 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4721 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4722 S_IFDIR | mode, &index);
39279cc3
CM		 4723 if (IS_ERR(inode)) {
4724 err = PTR_ERR(inode);
4725 goto out_fail;
4726 }
5f39d397 4727
39279cc3 4728 drop_on_err = 1;
33268eaf 4729
2a7dba39 4730 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4731 if (err)
4732 goto out_fail;
4733
39279cc3
CM		 4734 inode->i_op = &btrfs_dir_inode_operations;
4735 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 4736
dbe674a9 4737 btrfs_i_size_write(inode, 0);
39279cc3
CM		 4738 err = btrfs_update_inode(trans, root, inode);
4739 if (err)
4740 goto out_fail;
5f39d397 4741
a1b075d2
JB		 4742 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4743 dentry->d_name.len, 0, index);
39279cc3
CM		 4744 if (err)
4745 goto out_fail;
5f39d397 4746
39279cc3
CM		 4747 d_instantiate(dentry, inode);
4748 drop_on_err = 0;
39279cc3
CM		 4749
4750out_fail:
d3c2fdcf 4751 nr = trans->blocks_used;
ab78c84d 4752 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4753 if (drop_on_err)
4754 iput(inode);
d3c2fdcf 4755 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4756 return err;
4757}
4758
d352ac68
CM		 4759/* helper for btfs_get_extent. Given an existing extent in the tree,
4760 * and an extent that you want to insert, deal with overlap and insert
4761 * the new extent into the tree.
4762 */
3b951516
CM		 4763static int merge_extent_mapping(struct extent_map_tree *em_tree,
4764 struct extent_map *existing,
e6dcd2dc
CM		 4765 struct extent_map *em,
4766 u64 map_start, u64 map_len)
3b951516
CM		 4767{
4768 u64 start_diff;
3b951516 4769
e6dcd2dc
CM		 4770 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4771 start_diff = map_start - em->start;
4772 em->start = map_start;
4773 em->len = map_len;
c8b97818
CM		 4774 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4775 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 4776 em->block_start += start_diff;
c8b97818
CM		 4777 em->block_len -= start_diff;
4778 }
e6dcd2dc 4779 return add_extent_mapping(em_tree, em);
3b951516
CM		 4780}
4781
c8b97818
CM		 4782static noinline int uncompress_inline(struct btrfs_path *path,
4783 struct inode *inode, struct page *page,
4784 size_t pg_offset, u64 extent_offset,
4785 struct btrfs_file_extent_item *item)
4786{
4787 int ret;
4788 struct extent_buffer *leaf = path->nodes[0];
4789 char *tmp;
4790 size_t max_size;
4791 unsigned long inline_size;
4792 unsigned long ptr;
261507a0 4793 int compress_type;
c8b97818
CM		 4794
4795 WARN_ON(pg_offset != 0);
261507a0 4796 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 4797 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4798 inline_size = btrfs_file_extent_inline_item_len(leaf,
4799 btrfs_item_nr(leaf, path->slots[0]));
4800 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 4801 if (!tmp)
4802 return -ENOMEM;
c8b97818
CM		 4803 ptr = btrfs_file_extent_inline_start(item);
4804
4805 read_extent_buffer(leaf, tmp, ptr, inline_size);
4806
5b050f04 4807 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 4808 ret = btrfs_decompress(compress_type, tmp, page,
4809 extent_offset, inline_size, max_size);
c8b97818
CM		 4810 if (ret) {
4811 char *kaddr = kmap_atomic(page, KM_USER0);
4812 unsigned long copy_size = min_t(u64,
4813 PAGE_CACHE_SIZE - pg_offset,
4814 max_size - extent_offset);
4815 memset(kaddr + pg_offset, 0, copy_size);
4816 kunmap_atomic(kaddr, KM_USER0);
4817 }
4818 kfree(tmp);
4819 return 0;
4820}
4821
d352ac68
CM		 4822/*
4823 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4824 * the ugly parts come from merging extents from the disk with the in-ram
4825 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4826 * where the in-ram extents might be locked pending data=ordered completion.
4827 *
4828 * This also copies inline extents directly into the page.
4829 */
d397712b 4830
a52d9a80 4831struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4832 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4833 int create)
4834{
4835 int ret;
4836 int err = 0;
db94535d 4837 u64 bytenr;
a52d9a80
CM		 4838 u64 extent_start = 0;
4839 u64 extent_end = 0;
33345d01 4840 u64 objectid = btrfs_ino(inode);
a52d9a80 4841 u32 found_type;
f421950f 4842 struct btrfs_path *path = NULL;
a52d9a80
CM		 4843 struct btrfs_root *root = BTRFS_I(inode)->root;
4844 struct btrfs_file_extent_item *item;
5f39d397
CM		 4845 struct extent_buffer *leaf;
4846 struct btrfs_key found_key;
a52d9a80
CM		 4847 struct extent_map *em = NULL;
4848 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4849 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4850 struct btrfs_trans_handle *trans = NULL;
261507a0 4851 int compress_type;
a52d9a80 4852
a52d9a80 4853again:
890871be 4854 read_lock(&em_tree->lock);
d1310b2e 4855 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4856 if (em)
4857 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 4858 read_unlock(&em_tree->lock);
d1310b2e 4859
a52d9a80 4860 if (em) {
e1c4b745
CM		 4861 if (em->start > start || em->start + em->len <= start)
4862 free_extent_map(em);
4863 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4864 free_extent_map(em);
4865 else
4866 goto out;
a52d9a80 4867 }
172ddd60 4868 em = alloc_extent_map();
a52d9a80 4869 if (!em) {
d1310b2e
CM		 4870 err = -ENOMEM;
4871 goto out;
a52d9a80 4872 }
e6dcd2dc 4873 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 4874 em->start = EXTENT_MAP_HOLE;
445a6944 4875 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 4876 em->len = (u64)-1;
c8b97818 4877 em->block_len = (u64)-1;
f421950f
CM		 4878
4879 if (!path) {
4880 path = btrfs_alloc_path();
026fd317
JB		 4881 if (!path) {
4882 err = -ENOMEM;
4883 goto out;
4884 }
4885 /*
4886 * Chances are we'll be called again, so go ahead and do
4887 * readahead
4888 */
4889 path->reada = 1;
f421950f
CM		 4890 }
4891
179e29e4
CM		 4892 ret = btrfs_lookup_file_extent(trans, root, path,
4893 objectid, start, trans != NULL);
a52d9a80
CM		 4894 if (ret < 0) {
4895 err = ret;
4896 goto out;
4897 }
4898
4899 if (ret != 0) {
4900 if (path->slots[0] == 0)
4901 goto not_found;
4902 path->slots[0]--;
4903 }
4904
5f39d397
CM		 4905 leaf = path->nodes[0];
4906 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 4907 struct btrfs_file_extent_item);
a52d9a80 4908 /* are we inside the extent that was found? */
5f39d397
CM		 4909 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4910 found_type = btrfs_key_type(&found_key);
4911 if (found_key.objectid != objectid ||
a52d9a80
CM		 4912 found_type != BTRFS_EXTENT_DATA_KEY) {
4913 goto not_found;
4914 }
4915
5f39d397
CM		 4916 found_type = btrfs_file_extent_type(leaf, item);
4917 extent_start = found_key.offset;
261507a0 4918 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 4919 if (found_type == BTRFS_FILE_EXTENT_REG ||
4920 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 4921 extent_end = extent_start +
db94535d 4922 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 4923 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4924 size_t size;
4925 size = btrfs_file_extent_inline_len(leaf, item);
4926 extent_end = (extent_start + size + root->sectorsize - 1) &
4927 ~((u64)root->sectorsize - 1);
4928 }
4929
4930 if (start >= extent_end) {
4931 path->slots[0]++;
4932 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4933 ret = btrfs_next_leaf(root, path);
4934 if (ret < 0) {
4935 err = ret;
4936 goto out;
a52d9a80 4937 }
9036c102
YZ		 4938 if (ret > 0)
4939 goto not_found;
4940 leaf = path->nodes[0];
a52d9a80 4941 }
9036c102
YZ		 4942 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4943 if (found_key.objectid != objectid ||
4944 found_key.type != BTRFS_EXTENT_DATA_KEY)
4945 goto not_found;
4946 if (start + len <= found_key.offset)
4947 goto not_found;
4948 em->start = start;
4949 em->len = found_key.offset - start;
4950 goto not_found_em;
4951 }
4952
d899e052
YZ		 4953 if (found_type == BTRFS_FILE_EXTENT_REG ||
4954 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 4955 em->start = extent_start;
4956 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 4957 em->orig_start = extent_start -
4958 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 4959 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4960 if (bytenr == 0) {
5f39d397 4961 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4962 goto insert;
4963 }
261507a0 4964 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 4965 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 4966 em->compress_type = compress_type;
c8b97818
CM		 4967 em->block_start = bytenr;
4968 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4969 item);
4970 } else {
4971 bytenr += btrfs_file_extent_offset(leaf, item);
4972 em->block_start = bytenr;
4973 em->block_len = em->len;
d899e052
YZ		 4974 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4975 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4976 }
a52d9a80
CM		 4977 goto insert;
4978 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4979 unsigned long ptr;
a52d9a80 4980 char *map;
3326d1b0
CM		 4981 size_t size;
4982 size_t extent_offset;
4983 size_t copy_size;
a52d9a80 4984
689f9346 4985 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4986 if (!page || create) {
689f9346 4987 em->start = extent_start;
9036c102 4988 em->len = extent_end - extent_start;
689f9346
Y		 4989 goto out;
4990 }
5f39d397 4991
9036c102
YZ		 4992 size = btrfs_file_extent_inline_len(leaf, item);
4993 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4994 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4995 size - extent_offset);
3326d1b0 4996 em->start = extent_start + extent_offset;
70dec807
CM		 4997 em->len = (copy_size + root->sectorsize - 1) &
4998 ~((u64)root->sectorsize - 1);
ff5b7ee3 4999 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5000 if (compress_type) {
c8b97818 5001 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5002 em->compress_type = compress_type;
5003 }
689f9346 5004 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5005 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5006 if (btrfs_file_extent_compression(leaf, item) !=
5007 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5008 ret = uncompress_inline(path, inode, page,
5009 pg_offset,
5010 extent_offset, item);
5011 BUG_ON(ret);
5012 } else {
5013 map = kmap(page);
5014 read_extent_buffer(leaf, map + pg_offset, ptr,
5015 copy_size);
93c82d57
CM		 5016 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5017 memset(map + pg_offset + copy_size, 0,
5018 PAGE_CACHE_SIZE - pg_offset -
5019 copy_size);
5020 }
c8b97818
CM		 5021 kunmap(page);
5022 }
179e29e4
CM		 5023 flush_dcache_page(page);
5024 } else if (create && PageUptodate(page)) {
0ca1f7ce 5025 WARN_ON(1);
179e29e4
CM		 5026 if (!trans) {
5027 kunmap(page);
5028 free_extent_map(em);
5029 em = NULL;
ff5714cc 5030
b3b4aa74 5031 btrfs_release_path(path);
7a7eaa40 5032 trans = btrfs_join_transaction(root);
ff5714cc 5033
3612b495
TI		 5034 if (IS_ERR(trans))
5035 return ERR_CAST(trans);
179e29e4
CM		 5036 goto again;
5037 }
c8b97818 5038 map = kmap(page);
70dec807 5039 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5040 copy_size);
c8b97818 5041 kunmap(page);
179e29e4 5042 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5043 }
d1310b2e 5044 set_extent_uptodate(io_tree, em->start,
507903b8 5045 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5046 goto insert;
5047 } else {
d397712b 5048 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5049 WARN_ON(1);
5050 }
5051not_found:
5052 em->start = start;
d1310b2e 5053 em->len = len;
a52d9a80 5054not_found_em:
5f39d397 5055 em->block_start = EXTENT_MAP_HOLE;
9036c102 5056 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5057insert:
b3b4aa74 5058 btrfs_release_path(path);
d1310b2e 5059 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5060 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5061 "[%llu %llu]\n", (unsigned long long)em->start,
5062 (unsigned long long)em->len,
5063 (unsigned long long)start,
5064 (unsigned long long)len);
a52d9a80
CM		 5065 err = -EIO;
5066 goto out;
5067 }
d1310b2e
CM		 5068
5069 err = 0;
890871be 5070 write_lock(&em_tree->lock);
a52d9a80 5071 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5072 /* it is possible that someone inserted the extent into the tree
5073 * while we had the lock dropped. It is also possible that
5074 * an overlapping map exists in the tree
5075 */
a52d9a80 5076 if (ret == -EEXIST) {
3b951516 5077 struct extent_map *existing;
e6dcd2dc
CM		 5078
5079 ret = 0;
5080
3b951516 5081 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5082 if (existing && (existing->start > start ||
5083 existing->start + existing->len <= start)) {
5084 free_extent_map(existing);
5085 existing = NULL;
5086 }
3b951516
CM		 5087 if (!existing) {
5088 existing = lookup_extent_mapping(em_tree, em->start,
5089 em->len);
5090 if (existing) {
5091 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5092 em, start,
5093 root->sectorsize);
3b951516
CM		 5094 free_extent_map(existing);
5095 if (err) {
5096 free_extent_map(em);
5097 em = NULL;
5098 }
5099 } else {
5100 err = -EIO;
3b951516
CM		 5101 free_extent_map(em);
5102 em = NULL;
5103 }
5104 } else {
5105 free_extent_map(em);
5106 em = existing;
e6dcd2dc 5107 err = 0;
a52d9a80 5108 }
a52d9a80 5109 }
890871be 5110 write_unlock(&em_tree->lock);
a52d9a80 5111out:
1abe9b8a 5112
5113 trace_btrfs_get_extent(root, em);
5114
f421950f
CM		 5115 if (path)
5116 btrfs_free_path(path);
a52d9a80
CM		 5117 if (trans) {
5118 ret = btrfs_end_transaction(trans, root);
d397712b 5119 if (!err)
a52d9a80
CM		 5120 err = ret;
5121 }
a52d9a80
CM		 5122 if (err) {
5123 free_extent_map(em);
a52d9a80
CM		 5124 return ERR_PTR(err);
5125 }
5126 return em;
5127}
5128
ec29ed5b
CM		 5129struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5130 size_t pg_offset, u64 start, u64 len,
5131 int create)
5132{
5133 struct extent_map *em;
5134 struct extent_map *hole_em = NULL;
5135 u64 range_start = start;
5136 u64 end;
5137 u64 found;
5138 u64 found_end;
5139 int err = 0;
5140
5141 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5142 if (IS_ERR(em))
5143 return em;
5144 if (em) {
5145 /*
5146 * if our em maps to a hole, there might
5147 * actually be delalloc bytes behind it
5148 */
5149 if (em->block_start != EXTENT_MAP_HOLE)
5150 return em;
5151 else
5152 hole_em = em;
5153 }
5154
5155 /* check to see if we've wrapped (len == -1 or similar) */
5156 end = start + len;
5157 if (end < start)
5158 end = (u64)-1;
5159 else
5160 end -= 1;
5161
5162 em = NULL;
5163
5164 /* ok, we didn't find anything, lets look for delalloc */
5165 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5166 end, len, EXTENT_DELALLOC, 1);
5167 found_end = range_start + found;
5168 if (found_end < range_start)
5169 found_end = (u64)-1;
5170
5171 /*
5172 * we didn't find anything useful, return
5173 * the original results from get_extent()
5174 */
5175 if (range_start > end || found_end <= start) {
5176 em = hole_em;
5177 hole_em = NULL;
5178 goto out;
5179 }
5180
5181 /* adjust the range_start to make sure it doesn't
5182 * go backwards from the start they passed in
5183 */
5184 range_start = max(start,range_start);
5185 found = found_end - range_start;
5186
5187 if (found > 0) {
5188 u64 hole_start = start;
5189 u64 hole_len = len;
5190
172ddd60 5191 em = alloc_extent_map();
ec29ed5b
CM		 5192 if (!em) {
5193 err = -ENOMEM;
5194 goto out;
5195 }
5196 /*
5197 * when btrfs_get_extent can't find anything it
5198 * returns one huge hole
5199 *
5200 * make sure what it found really fits our range, and
5201 * adjust to make sure it is based on the start from
5202 * the caller
5203 */
5204 if (hole_em) {
5205 u64 calc_end = extent_map_end(hole_em);
5206
5207 if (calc_end <= start || (hole_em->start > end)) {
5208 free_extent_map(hole_em);
5209 hole_em = NULL;
5210 } else {
5211 hole_start = max(hole_em->start, start);
5212 hole_len = calc_end - hole_start;
5213 }
5214 }
5215 em->bdev = NULL;
5216 if (hole_em && range_start > hole_start) {
5217 /* our hole starts before our delalloc, so we
5218 * have to return just the parts of the hole
5219 * that go until the delalloc starts
5220 */
5221 em->len = min(hole_len,
5222 range_start - hole_start);
5223 em->start = hole_start;
5224 em->orig_start = hole_start;
5225 /*
5226 * don't adjust block start at all,
5227 * it is fixed at EXTENT_MAP_HOLE
5228 */
5229 em->block_start = hole_em->block_start;
5230 em->block_len = hole_len;
5231 } else {
5232 em->start = range_start;
5233 em->len = found;
5234 em->orig_start = range_start;
5235 em->block_start = EXTENT_MAP_DELALLOC;
5236 em->block_len = found;
5237 }
5238 } else if (hole_em) {
5239 return hole_em;
5240 }
5241out:
5242
5243 free_extent_map(hole_em);
5244 if (err) {
5245 free_extent_map(em);
5246 return ERR_PTR(err);
5247 }
5248 return em;
5249}
5250
4b46fce2 5251static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5252 struct extent_map *em,
4b46fce2
JB		 5253 u64 start, u64 len)
5254{
5255 struct btrfs_root *root = BTRFS_I(inode)->root;
5256 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5257 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5258 struct btrfs_key ins;
5259 u64 alloc_hint;
5260 int ret;
16d299ac 5261 bool insert = false;
4b46fce2 5262
16d299ac
JB		 5263 /*
5264 * Ok if the extent map we looked up is a hole and is for the exact
5265 * range we want, there is no reason to allocate a new one, however if
5266 * it is not right then we need to free this one and drop the cache for
5267 * our range.
5268 */
5269 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5270 em->len != len) {
5271 free_extent_map(em);
5272 em = NULL;
5273 insert = true;
5274 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5275 }
4b46fce2 5276
7a7eaa40 5277 trans = btrfs_join_transaction(root);
3612b495
TI		 5278 if (IS_ERR(trans))
5279 return ERR_CAST(trans);
4b46fce2 5280
4cb5300b
CM		 5281 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5282 btrfs_add_inode_defrag(trans, inode);
5283
4b46fce2
JB		 5284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5285
5286 alloc_hint = get_extent_allocation_hint(inode, start, len);
5287 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5288 alloc_hint, (u64)-1, &ins, 1);
5289 if (ret) {
5290 em = ERR_PTR(ret);
5291 goto out;
5292 }
5293
4b46fce2 5294 if (!em) {
172ddd60 5295 em = alloc_extent_map();
16d299ac
JB		 5296 if (!em) {
5297 em = ERR_PTR(-ENOMEM);
5298 goto out;
5299 }
4b46fce2
JB		 5300 }
5301
5302 em->start = start;
5303 em->orig_start = em->start;
5304 em->len = ins.offset;
5305
5306 em->block_start = ins.objectid;
5307 em->block_len = ins.offset;
5308 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5309
5310 /*
5311 * We need to do this because if we're using the original em we searched
5312 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5313 */
5314 em->flags = 0;
4b46fce2
JB		 5315 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5316
16d299ac 5317 while (insert) {
4b46fce2
JB		 5318 write_lock(&em_tree->lock);
5319 ret = add_extent_mapping(em_tree, em);
5320 write_unlock(&em_tree->lock);
5321 if (ret != -EEXIST)
5322 break;
5323 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5324 }
5325
5326 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5327 ins.offset, ins.offset, 0);
5328 if (ret) {
5329 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5330 em = ERR_PTR(ret);
5331 }
5332out:
5333 btrfs_end_transaction(trans, root);
5334 return em;
5335}
5336
46bfbb5c
CM		 5337/*
5338 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5339 * block must be cow'd
5340 */
5341static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5342 struct inode *inode, u64 offset, u64 len)
5343{
5344 struct btrfs_path *path;
5345 int ret;
5346 struct extent_buffer *leaf;
5347 struct btrfs_root *root = BTRFS_I(inode)->root;
5348 struct btrfs_file_extent_item *fi;
5349 struct btrfs_key key;
5350 u64 disk_bytenr;
5351 u64 backref_offset;
5352 u64 extent_end;
5353 u64 num_bytes;
5354 int slot;
5355 int found_type;
5356
5357 path = btrfs_alloc_path();
5358 if (!path)
5359 return -ENOMEM;
5360
33345d01 5361 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5362 offset, 0);
5363 if (ret < 0)
5364 goto out;
5365
5366 slot = path->slots[0];
5367 if (ret == 1) {
5368 if (slot == 0) {
5369 /* can't find the item, must cow */
5370 ret = 0;
5371 goto out;
5372 }
5373 slot--;
5374 }
5375 ret = 0;
5376 leaf = path->nodes[0];
5377 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5378 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5379 key.type != BTRFS_EXTENT_DATA_KEY) {
5380 /* not our file or wrong item type, must cow */
5381 goto out;
5382 }
5383
5384 if (key.offset > offset) {
5385 /* Wrong offset, must cow */
5386 goto out;
5387 }
5388
5389 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5390 found_type = btrfs_file_extent_type(leaf, fi);
5391 if (found_type != BTRFS_FILE_EXTENT_REG &&
5392 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5393 /* not a regular extent, must cow */
5394 goto out;
5395 }
5396 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5397 backref_offset = btrfs_file_extent_offset(leaf, fi);
5398
5399 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5400 if (extent_end < offset + len) {
5401 /* extent doesn't include our full range, must cow */
5402 goto out;
5403 }
5404
5405 if (btrfs_extent_readonly(root, disk_bytenr))
5406 goto out;
5407
5408 /*
5409 * look for other files referencing this extent, if we
5410 * find any we must cow
5411 */
33345d01 5412 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5413 key.offset - backref_offset, disk_bytenr))
5414 goto out;
5415
5416 /*
5417 * adjust disk_bytenr and num_bytes to cover just the bytes
5418 * in this extent we are about to write. If there
5419 * are any csums in that range we have to cow in order
5420 * to keep the csums correct
5421 */
5422 disk_bytenr += backref_offset;
5423 disk_bytenr += offset - key.offset;
5424 num_bytes = min(offset + len, extent_end) - offset;
5425 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5426 goto out;
5427 /*
5428 * all of the above have passed, it is safe to overwrite this extent
5429 * without cow
5430 */
5431 ret = 1;
5432out:
5433 btrfs_free_path(path);
5434 return ret;
5435}
5436
4b46fce2
JB		 5437static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5438 struct buffer_head *bh_result, int create)
5439{
5440 struct extent_map *em;
5441 struct btrfs_root *root = BTRFS_I(inode)->root;
5442 u64 start = iblock << inode->i_blkbits;
5443 u64 len = bh_result->b_size;
46bfbb5c 5444 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5445
5446 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5447 if (IS_ERR(em))
5448 return PTR_ERR(em);
5449
5450 /*
5451 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5452 * io. INLINE is special, and we could probably kludge it in here, but
5453 * it's still buffered so for safety lets just fall back to the generic
5454 * buffered path.
5455 *
5456 * For COMPRESSED we _have_ to read the entire extent in so we can
5457 * decompress it, so there will be buffering required no matter what we
5458 * do, so go ahead and fallback to buffered.
5459 *
5460 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5461 * to buffered IO. Don't blame me, this is the price we pay for using
5462 * the generic code.
5463 */
5464 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5465 em->block_start == EXTENT_MAP_INLINE) {
5466 free_extent_map(em);
5467 return -ENOTBLK;
5468 }
5469
5470 /* Just a good old fashioned hole, return */
5471 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5472 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5473 free_extent_map(em);
5474 /* DIO will do one hole at a time, so just unlock a sector */
5475 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5476 start + root->sectorsize - 1, GFP_NOFS);
5477 return 0;
5478 }
5479
5480 /*
5481 * We don't allocate a new extent in the following cases
5482 *
5483 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5484 * existing extent.
5485 * 2) The extent is marked as PREALLOC. We're good to go here and can
5486 * just use the extent.
5487 *
5488 */
46bfbb5c
CM		 5489 if (!create) {
5490 len = em->len - (start - em->start);
4b46fce2 5491 goto map;
46bfbb5c 5492 }
4b46fce2
JB		 5493
5494 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5495 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5496 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5497 int type;
5498 int ret;
46bfbb5c 5499 u64 block_start;
4b46fce2
JB		 5500
5501 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5502 type = BTRFS_ORDERED_PREALLOC;
5503 else
5504 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5505 len = min(len, em->len - (start - em->start));
4b46fce2 5506 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5507
5508 /*
5509 * we're not going to log anything, but we do need
5510 * to make sure the current transaction stays open
5511 * while we look for nocow cross refs
5512 */
7a7eaa40 5513 trans = btrfs_join_transaction(root);
3612b495 5514 if (IS_ERR(trans))
46bfbb5c
CM		 5515 goto must_cow;
5516
5517 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5518 ret = btrfs_add_ordered_extent_dio(inode, start,
5519 block_start, len, len, type);
5520 btrfs_end_transaction(trans, root);
5521 if (ret) {
5522 free_extent_map(em);
5523 return ret;
5524 }
5525 goto unlock;
4b46fce2 5526 }
46bfbb5c 5527 btrfs_end_transaction(trans, root);
4b46fce2 5528 }
46bfbb5c
CM		 5529must_cow:
5530 /*
5531 * this will cow the extent, reset the len in case we changed
5532 * it above
5533 */
5534 len = bh_result->b_size;
16d299ac 5535 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5536 if (IS_ERR(em))
5537 return PTR_ERR(em);
5538 len = min(len, em->len - (start - em->start));
5539unlock:
4845e44f
CM		 5540 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5541 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5542 0, NULL, GFP_NOFS);
4b46fce2
JB		 5543map:
5544 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5545 inode->i_blkbits;
46bfbb5c 5546 bh_result->b_size = len;
4b46fce2
JB		 5547 bh_result->b_bdev = em->bdev;
5548 set_buffer_mapped(bh_result);
5549 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5550 set_buffer_new(bh_result);
5551
5552 free_extent_map(em);
5553
5554 return 0;
5555}
5556
5557struct btrfs_dio_private {
5558 struct inode *inode;
5559 u64 logical_offset;
5560 u64 disk_bytenr;
5561 u64 bytes;
5562 u32 *csums;
5563 void *private;
e65e1535
MX		 5564
5565 /* number of bios pending for this dio */
5566 atomic_t pending_bios;
5567
5568 /* IO errors */
5569 int errors;
5570
5571 struct bio *orig_bio;
4b46fce2
JB		 5572};
5573
5574static void btrfs_endio_direct_read(struct bio *bio, int err)
5575{
e65e1535 5576 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5577 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5578 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5579 struct inode *inode = dip->inode;
5580 struct btrfs_root *root = BTRFS_I(inode)->root;
5581 u64 start;
5582 u32 *private = dip->csums;
5583
5584 start = dip->logical_offset;
5585 do {
5586 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5587 struct page *page = bvec->bv_page;
5588 char *kaddr;
5589 u32 csum = ~(u32)0;
5590 unsigned long flags;
5591
5592 local_irq_save(flags);
5593 kaddr = kmap_atomic(page, KM_IRQ0);
5594 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5595 csum, bvec->bv_len);
5596 btrfs_csum_final(csum, (char *)&csum);
5597 kunmap_atomic(kaddr, KM_IRQ0);
5598 local_irq_restore(flags);
5599
5600 flush_dcache_page(bvec->bv_page);
5601 if (csum != *private) {
33345d01 5602 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5603 " %llu csum %u private %u\n",
33345d01
LZ		 5604 (unsigned long long)btrfs_ino(inode),
5605 (unsigned long long)start,
4b46fce2
JB		 5606 csum, *private);
5607 err = -EIO;
5608 }
5609 }
5610
5611 start += bvec->bv_len;
5612 private++;
5613 bvec++;
5614 } while (bvec <= bvec_end);
5615
5616 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5617 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5618 bio->bi_private = dip->private;
5619
5620 kfree(dip->csums);
5621 kfree(dip);
c0da7aa1
JB		 5622
5623 /* If we had a csum failure make sure to clear the uptodate flag */
5624 if (err)
5625 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5626 dio_end_io(bio, err);
5627}
5628
5629static void btrfs_endio_direct_write(struct bio *bio, int err)
5630{
5631 struct btrfs_dio_private *dip = bio->bi_private;
5632 struct inode *inode = dip->inode;
5633 struct btrfs_root *root = BTRFS_I(inode)->root;
5634 struct btrfs_trans_handle *trans;
5635 struct btrfs_ordered_extent *ordered = NULL;
5636 struct extent_state *cached_state = NULL;
163cf09c
CM		 5637 u64 ordered_offset = dip->logical_offset;
5638 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5639 int ret;
5640
5641 if (err)
5642 goto out_done;
163cf09c
CM		 5643again:
5644 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5645 &ordered_offset,
5646 ordered_bytes);
4b46fce2 5647 if (!ret)
163cf09c 5648 goto out_test;
4b46fce2
JB		 5649
5650 BUG_ON(!ordered);
5651
7a7eaa40 5652 trans = btrfs_join_transaction(root);
3612b495 5653 if (IS_ERR(trans)) {
4b46fce2
JB		 5654 err = -ENOMEM;
5655 goto out;
5656 }
5657 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5658
5659 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5660 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5661 if (!ret)
2115133f 5662 err = btrfs_update_inode_fallback(trans, root, inode);
4b46fce2
JB		 5663 goto out;
5664 }
5665
5666 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5667 ordered->file_offset + ordered->len - 1, 0,
5668 &cached_state, GFP_NOFS);
5669
5670 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5671 ret = btrfs_mark_extent_written(trans, inode,
5672 ordered->file_offset,
5673 ordered->file_offset +
5674 ordered->len);
5675 if (ret) {
5676 err = ret;
5677 goto out_unlock;
5678 }
5679 } else {
5680 ret = insert_reserved_file_extent(trans, inode,
5681 ordered->file_offset,
5682 ordered->start,
5683 ordered->disk_len,
5684 ordered->len,
5685 ordered->len,
5686 0, 0, 0,
5687 BTRFS_FILE_EXTENT_REG);
5688 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5689 ordered->file_offset, ordered->len);
5690 if (ret) {
5691 err = ret;
5692 WARN_ON(1);
5693 goto out_unlock;
5694 }
5695 }
5696
5697 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
1ef30be1 5698 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
a39f7521 5699 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
2115133f 5700 btrfs_update_inode_fallback(trans, root, inode);
1ef30be1 5701 ret = 0;
4b46fce2
JB		 5702out_unlock:
5703 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5704 ordered->file_offset + ordered->len - 1,
5705 &cached_state, GFP_NOFS);
5706out:
5707 btrfs_delalloc_release_metadata(inode, ordered->len);
5708 btrfs_end_transaction(trans, root);
163cf09c 5709 ordered_offset = ordered->file_offset + ordered->len;
4b46fce2
JB		 5710 btrfs_put_ordered_extent(ordered);
5711 btrfs_put_ordered_extent(ordered);
163cf09c
CM		 5712
5713out_test:
5714 /*
5715 * our bio might span multiple ordered extents. If we haven't
5716 * completed the accounting for the whole dio, go back and try again
5717 */
5718 if (ordered_offset < dip->logical_offset + dip->bytes) {
5719 ordered_bytes = dip->logical_offset + dip->bytes -
5720 ordered_offset;
5721 goto again;
5722 }
4b46fce2
JB		 5723out_done:
5724 bio->bi_private = dip->private;
5725
5726 kfree(dip->csums);
5727 kfree(dip);
c0da7aa1
JB		 5728
5729 /* If we had an error make sure to clear the uptodate flag */
5730 if (err)
5731 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5732 dio_end_io(bio, err);
5733}
5734
eaf25d93
CM		 5735static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5736 struct bio *bio, int mirror_num,
5737 unsigned long bio_flags, u64 offset)
5738{
5739 int ret;
5740 struct btrfs_root *root = BTRFS_I(inode)->root;
5741 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5742 BUG_ON(ret);
5743 return 0;
5744}
5745
e65e1535
MX		 5746static void btrfs_end_dio_bio(struct bio *bio, int err)
5747{
5748 struct btrfs_dio_private *dip = bio->bi_private;
5749
5750 if (err) {
33345d01 5751 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 5752 "sector %#Lx len %u err no %d\n",
33345d01 5753 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 5754 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 5755 dip->errors = 1;
5756
5757 /*
5758 * before atomic variable goto zero, we must make sure
5759 * dip->errors is perceived to be set.
5760 */
5761 smp_mb__before_atomic_dec();
5762 }
5763
5764 /* if there are more bios still pending for this dio, just exit */
5765 if (!atomic_dec_and_test(&dip->pending_bios))
5766 goto out;
5767
5768 if (dip->errors)
5769 bio_io_error(dip->orig_bio);
5770 else {
5771 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5772 bio_endio(dip->orig_bio, 0);
5773 }
5774out:
5775 bio_put(bio);
5776}
5777
5778static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5779 u64 first_sector, gfp_t gfp_flags)
5780{
5781 int nr_vecs = bio_get_nr_vecs(bdev);
5782 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5783}
5784
5785static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5786 int rw, u64 file_offset, int skip_sum,
1ae39938 5787 u32 *csums, int async_submit)
e65e1535
MX		 5788{
5789 int write = rw & REQ_WRITE;
5790 struct btrfs_root *root = BTRFS_I(inode)->root;
5791 int ret;
5792
5793 bio_get(bio);
5794 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5795 if (ret)
5796 goto err;
5797
1ae39938
JB		 5798 if (skip_sum)
5799 goto map;
5800
5801 if (write && async_submit) {
e65e1535
MX		 5802 ret = btrfs_wq_submit_bio(root->fs_info,
5803 inode, rw, bio, 0, 0,
5804 file_offset,
5805 __btrfs_submit_bio_start_direct_io,
5806 __btrfs_submit_bio_done);
5807 goto err;
1ae39938
JB		 5808 } else if (write) {
5809 /*
5810 * If we aren't doing async submit, calculate the csum of the
5811 * bio now.
5812 */
5813 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5814 if (ret)
5815 goto err;
c2db1073
TI		 5816 } else if (!skip_sum) {
5817 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 5818 file_offset, csums);
c2db1073
TI		 5819 if (ret)
5820 goto err;
5821 }
e65e1535 5822
1ae39938
JB		 5823map:
5824 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 5825err:
5826 bio_put(bio);
5827 return ret;
5828}
5829
5830static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5831 int skip_sum)
5832{
5833 struct inode *inode = dip->inode;
5834 struct btrfs_root *root = BTRFS_I(inode)->root;
5835 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5836 struct bio *bio;
5837 struct bio *orig_bio = dip->orig_bio;
5838 struct bio_vec *bvec = orig_bio->bi_io_vec;
5839 u64 start_sector = orig_bio->bi_sector;
5840 u64 file_offset = dip->logical_offset;
5841 u64 submit_len = 0;
5842 u64 map_length;
5843 int nr_pages = 0;
5844 u32 *csums = dip->csums;
5845 int ret = 0;
1ae39938 5846 int async_submit = 0;
98bc3149 5847 int write = rw & REQ_WRITE;
e65e1535 5848
e65e1535
MX		 5849 map_length = orig_bio->bi_size;
5850 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5851 &map_length, NULL, 0);
5852 if (ret) {
64728bbb 5853 bio_put(orig_bio);
e65e1535
MX		 5854 return -EIO;
5855 }
5856
02f57c7a
JB		 5857 if (map_length >= orig_bio->bi_size) {
5858 bio = orig_bio;
5859 goto submit;
5860 }
5861
1ae39938 5862 async_submit = 1;
02f57c7a
JB		 5863 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5864 if (!bio)
5865 return -ENOMEM;
5866 bio->bi_private = dip;
5867 bio->bi_end_io = btrfs_end_dio_bio;
5868 atomic_inc(&dip->pending_bios);
5869
e65e1535
MX		 5870 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5871 if (unlikely(map_length < submit_len + bvec->bv_len ||
5872 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5873 bvec->bv_offset) < bvec->bv_len)) {
5874 /*
5875 * inc the count before we submit the bio so
5876 * we know the end IO handler won't happen before
5877 * we inc the count. Otherwise, the dip might get freed
5878 * before we're done setting it up
5879 */
5880 atomic_inc(&dip->pending_bios);
5881 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5882 file_offset, skip_sum,
1ae39938 5883 csums, async_submit);
e65e1535
MX		 5884 if (ret) {
5885 bio_put(bio);
5886 atomic_dec(&dip->pending_bios);
5887 goto out_err;
5888 }
5889
98bc3149
JB		 5890 /* Write's use the ordered csums */
5891 if (!write && !skip_sum)
e65e1535
MX		 5892 csums = csums + nr_pages;
5893 start_sector += submit_len >> 9;
5894 file_offset += submit_len;
5895
5896 submit_len = 0;
5897 nr_pages = 0;
5898
5899 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5900 start_sector, GFP_NOFS);
5901 if (!bio)
5902 goto out_err;
5903 bio->bi_private = dip;
5904 bio->bi_end_io = btrfs_end_dio_bio;
5905
5906 map_length = orig_bio->bi_size;
5907 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5908 &map_length, NULL, 0);
5909 if (ret) {
5910 bio_put(bio);
5911 goto out_err;
5912 }
5913 } else {
5914 submit_len += bvec->bv_len;
5915 nr_pages ++;
5916 bvec++;
5917 }
5918 }
5919
02f57c7a 5920submit:
e65e1535 5921 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 5922 csums, async_submit);
e65e1535
MX		 5923 if (!ret)
5924 return 0;
5925
5926 bio_put(bio);
5927out_err:
5928 dip->errors = 1;
5929 /*
5930 * before atomic variable goto zero, we must
5931 * make sure dip->errors is perceived to be set.
5932 */
5933 smp_mb__before_atomic_dec();
5934 if (atomic_dec_and_test(&dip->pending_bios))
5935 bio_io_error(dip->orig_bio);
5936
5937 /* bio_end_io() will handle error, so we needn't return it */
5938 return 0;
5939}
5940
4b46fce2
JB		 5941static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5942 loff_t file_offset)
5943{
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 struct btrfs_dio_private *dip;
5946 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 5947 int skip_sum;
7b6d91da 5948 int write = rw & REQ_WRITE;
4b46fce2
JB		 5949 int ret = 0;
5950
5951 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5952
5953 dip = kmalloc(sizeof(*dip), GFP_NOFS);
5954 if (!dip) {
5955 ret = -ENOMEM;
5956 goto free_ordered;
5957 }
5958 dip->csums = NULL;
5959
98bc3149
JB		 5960 /* Write's use the ordered csum stuff, so we don't need dip->csums */
5961 if (!write && !skip_sum) {
4b46fce2
JB		 5962 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5963 if (!dip->csums) {
b4966b77 5964 kfree(dip);
4b46fce2
JB		 5965 ret = -ENOMEM;
5966 goto free_ordered;
5967 }
5968 }
5969
5970 dip->private = bio->bi_private;
5971 dip->inode = inode;
5972 dip->logical_offset = file_offset;
5973
4b46fce2
JB		 5974 dip->bytes = 0;
5975 do {
5976 dip->bytes += bvec->bv_len;
5977 bvec++;
5978 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5979
46bfbb5c 5980 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 5981 bio->bi_private = dip;
e65e1535
MX		 5982 dip->errors = 0;
5983 dip->orig_bio = bio;
5984 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 5985
5986 if (write)
5987 bio->bi_end_io = btrfs_endio_direct_write;
5988 else
5989 bio->bi_end_io = btrfs_endio_direct_read;
5990
e65e1535
MX		 5991 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5992 if (!ret)
eaf25d93 5993 return;
4b46fce2
JB		 5994free_ordered:
5995 /*
5996 * If this is a write, we need to clean up the reserved space and kill
5997 * the ordered extent.
5998 */
5999 if (write) {
6000 struct btrfs_ordered_extent *ordered;
955256f2 6001 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6002 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6003 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6004 btrfs_free_reserved_extent(root, ordered->start,
6005 ordered->disk_len);
6006 btrfs_put_ordered_extent(ordered);
6007 btrfs_put_ordered_extent(ordered);
6008 }
6009 bio_endio(bio, ret);
6010}
6011
5a5f79b5
CM		 6012static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6013 const struct iovec *iov, loff_t offset,
6014 unsigned long nr_segs)
6015{
6016 int seg;
a1b75f7d 6017 int i;
5a5f79b5
CM		 6018 size_t size;
6019 unsigned long addr;
6020 unsigned blocksize_mask = root->sectorsize - 1;
6021 ssize_t retval = -EINVAL;
6022 loff_t end = offset;
6023
6024 if (offset & blocksize_mask)
6025 goto out;
6026
6027 /* Check the memory alignment. Blocks cannot straddle pages */
6028 for (seg = 0; seg < nr_segs; seg++) {
6029 addr = (unsigned long)iov[seg].iov_base;
6030 size = iov[seg].iov_len;
6031 end += size;
a1b75f7d 6032 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6033 goto out;
a1b75f7d
JB		 6034
6035 /* If this is a write we don't need to check anymore */
6036 if (rw & WRITE)
6037 continue;
6038
6039 /*
6040 * Check to make sure we don't have duplicate iov_base's in this
6041 * iovec, if so return EINVAL, otherwise we'll get csum errors
6042 * when reading back.
6043 */
6044 for (i = seg + 1; i < nr_segs; i++) {
6045 if (iov[seg].iov_base == iov[i].iov_base)
6046 goto out;
6047 }
5a5f79b5
CM		 6048 }
6049 retval = 0;
6050out:
6051 return retval;
6052}
16432985
CM		 6053static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6054 const struct iovec *iov, loff_t offset,
6055 unsigned long nr_segs)
6056{
4b46fce2
JB		 6057 struct file *file = iocb->ki_filp;
6058 struct inode *inode = file->f_mapping->host;
6059 struct btrfs_ordered_extent *ordered;
4845e44f 6060 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6061 u64 lockstart, lockend;
6062 ssize_t ret;
4845e44f
CM		 6063 int writing = rw & WRITE;
6064 int write_bits = 0;
3f7c579c 6065 size_t count = iov_length(iov, nr_segs);
4b46fce2 6066
5a5f79b5
CM		 6067 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6068 offset, nr_segs)) {
6069 return 0;
6070 }
6071
4b46fce2 6072 lockstart = offset;
3f7c579c
CM		 6073 lockend = offset + count - 1;
6074
6075 if (writing) {
6076 ret = btrfs_delalloc_reserve_space(inode, count);
6077 if (ret)
6078 goto out;
6079 }
4845e44f 6080
4b46fce2 6081 while (1) {
4845e44f
CM		 6082 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6083 0, &cached_state, GFP_NOFS);
4b46fce2
JB		 6084 /*
6085 * We're concerned with the entire range that we're going to be
6086 * doing DIO to, so we need to make sure theres no ordered
6087 * extents in this range.
6088 */
6089 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6090 lockend - lockstart + 1);
6091 if (!ordered)
6092 break;
4845e44f
CM		 6093 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6094 &cached_state, GFP_NOFS);
4b46fce2
JB		 6095 btrfs_start_ordered_extent(inode, ordered, 1);
6096 btrfs_put_ordered_extent(ordered);
6097 cond_resched();
6098 }
6099
4845e44f
CM		 6100 /*
6101 * we don't use btrfs_set_extent_delalloc because we don't want
6102 * the dirty or uptodate bits
6103 */
6104 if (writing) {
6105 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6106 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6107 EXTENT_DELALLOC, 0, NULL, &cached_state,
6108 GFP_NOFS);
6109 if (ret) {
6110 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6111 lockend, EXTENT_LOCKED | write_bits,
6112 1, 0, &cached_state, GFP_NOFS);
6113 goto out;
6114 }
6115 }
6116
6117 free_extent_state(cached_state);
6118 cached_state = NULL;
6119
5a5f79b5
CM		 6120 ret = __blockdev_direct_IO(rw, iocb, inode,
6121 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6122 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6123 btrfs_submit_direct, 0);
4b46fce2
JB		 6124
6125 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6126 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6127 offset + iov_length(iov, nr_segs) - 1,
6128 EXTENT_LOCKED | write_bits, 1, 0,
6129 &cached_state, GFP_NOFS);
4b46fce2
JB		 6130 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6131 /*
6132 * We're falling back to buffered, unlock the section we didn't
6133 * do IO on.
6134 */
4845e44f
CM		 6135 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6136 offset + iov_length(iov, nr_segs) - 1,
6137 EXTENT_LOCKED | write_bits, 1, 0,
6138 &cached_state, GFP_NOFS);
4b46fce2 6139 }
4845e44f
CM		 6140out:
6141 free_extent_state(cached_state);
4b46fce2 6142 return ret;
16432985
CM		 6143}
6144
1506fcc8
YS		 6145static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6146 __u64 start, __u64 len)
6147{
ec29ed5b 6148 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6149}
6150
a52d9a80 6151int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6152{
d1310b2e
CM		 6153 struct extent_io_tree *tree;
6154 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6155 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6156}
1832a6d5 6157
a52d9a80 6158static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6159{
d1310b2e 6160 struct extent_io_tree *tree;
b888db2b
CM		 6161
6162
6163 if (current->flags & PF_MEMALLOC) {
6164 redirty_page_for_writepage(wbc, page);
6165 unlock_page(page);
6166 return 0;
6167 }
d1310b2e 6168 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6169 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6170}
6171
f421950f
CM		 6172int btrfs_writepages(struct address_space *mapping,
6173 struct writeback_control *wbc)
b293f02e 6174{
d1310b2e 6175 struct extent_io_tree *tree;
771ed689 6176
d1310b2e 6177 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6178 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6179}
6180
3ab2fb5a
CM		 6181static int
6182btrfs_readpages(struct file *file, struct address_space *mapping,
6183 struct list_head *pages, unsigned nr_pages)
6184{
d1310b2e
CM		 6185 struct extent_io_tree *tree;
6186 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6187 return extent_readpages(tree, mapping, pages, nr_pages,
6188 btrfs_get_extent);
6189}
e6dcd2dc 6190static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6191{
d1310b2e
CM		 6192 struct extent_io_tree *tree;
6193 struct extent_map_tree *map;
a52d9a80 6194 int ret;
8c2383c3 6195
d1310b2e
CM		 6196 tree = &BTRFS_I(page->mapping->host)->io_tree;
6197 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6198 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6199 if (ret == 1) {
6200 ClearPagePrivate(page);
6201 set_page_private(page, 0);
6202 page_cache_release(page);
39279cc3 6203 }
a52d9a80 6204 return ret;
39279cc3
CM		 6205}
6206
e6dcd2dc
CM		 6207static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6208{
98509cfc
CM		 6209 if (PageWriteback(page) || PageDirty(page))
6210 return 0;
b335b003 6211 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6212}
6213
a52d9a80 6214static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6215{
d1310b2e 6216 struct extent_io_tree *tree;
e6dcd2dc 6217 struct btrfs_ordered_extent *ordered;
2ac55d41 6218 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6219 u64 page_start = page_offset(page);
6220 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6221
8b62b72b
CM		 6222
6223 /*
6224 * we have the page locked, so new writeback can't start,
6225 * and the dirty bit won't be cleared while we are here.
6226 *
6227 * Wait for IO on this page so that we can safely clear
6228 * the PagePrivate2 bit and do ordered accounting
6229 */
e6dcd2dc 6230 wait_on_page_writeback(page);
8b62b72b 6231
d1310b2e 6232 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 6233 if (offset) {
6234 btrfs_releasepage(page, GFP_NOFS);
6235 return;
6236 }
2ac55d41
JB		 6237 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6238 GFP_NOFS);
e6dcd2dc
CM		 6239 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6240 page_offset(page));
6241 if (ordered) {
eb84ae03
CM		 6242 /*
6243 * IO on this page will never be started, so we need
6244 * to account for any ordered extents now
6245 */
e6dcd2dc
CM		 6246 clear_extent_bit(tree, page_start, page_end,
6247 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6248 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6249 &cached_state, GFP_NOFS);
8b62b72b
CM		 6250 /*
6251 * whoever cleared the private bit is responsible
6252 * for the finish_ordered_io
6253 */
6254 if (TestClearPagePrivate2(page)) {
6255 btrfs_finish_ordered_io(page->mapping->host,
6256 page_start, page_end);
6257 }
e6dcd2dc 6258 btrfs_put_ordered_extent(ordered);
2ac55d41
JB		 6259 cached_state = NULL;
6260 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6261 GFP_NOFS);
e6dcd2dc
CM		 6262 }
6263 clear_extent_bit(tree, page_start, page_end,
32c00aff 6264 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6265 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6266 __btrfs_releasepage(page, GFP_NOFS);
6267
4a096752 6268 ClearPageChecked(page);
9ad6b7bc 6269 if (PagePrivate(page)) {
9ad6b7bc
CM		 6270 ClearPagePrivate(page);
6271 set_page_private(page, 0);
6272 page_cache_release(page);
6273 }
39279cc3
CM		 6274}
6275
9ebefb18
CM		 6276/*
6277 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6278 * called from a page fault handler when a page is first dirtied. Hence we must
6279 * be careful to check for EOF conditions here. We set the page up correctly
6280 * for a written page which means we get ENOSPC checking when writing into
6281 * holes and correct delalloc and unwritten extent mapping on filesystems that
6282 * support these features.
6283 *
6284 * We are not allowed to take the i_mutex here so we have to play games to
6285 * protect against truncate races as the page could now be beyond EOF. Because
6286 * vmtruncate() writes the inode size before removing pages, once we have the
6287 * page lock we can determine safely if the page is beyond EOF. If it is not
6288 * beyond EOF, then the page is guaranteed safe against truncation until we
6289 * unlock the page.
6290 */
c2ec175c 6291int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6292{
c2ec175c 6293 struct page *page = vmf->page;
6da6abae 6294 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6295 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6296 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6297 struct btrfs_ordered_extent *ordered;
2ac55d41 6298 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6299 char *kaddr;
6300 unsigned long zero_start;
9ebefb18 6301 loff_t size;
1832a6d5 6302 int ret;
a52d9a80 6303 u64 page_start;
e6dcd2dc 6304 u64 page_end;
9ebefb18 6305
0ca1f7ce 6306 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
56a76f82
NP		 6307 if (ret) {
6308 if (ret == -ENOMEM)
6309 ret = VM_FAULT_OOM;
6310 else /* -ENOSPC, -EIO, etc */
6311 ret = VM_FAULT_SIGBUS;
1832a6d5 6312 goto out;
56a76f82 6313 }
1832a6d5 6314
56a76f82 6315 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6316again:
9ebefb18 6317 lock_page(page);
9ebefb18 6318 size = i_size_read(inode);
e6dcd2dc
CM		 6319 page_start = page_offset(page);
6320 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6321
9ebefb18 6322 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6323 (page_start >= size)) {
9ebefb18
CM		 6324 /* page got truncated out from underneath us */
6325 goto out_unlock;
6326 }
e6dcd2dc
CM		 6327 wait_on_page_writeback(page);
6328
2ac55d41
JB		 6329 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6330 GFP_NOFS);
e6dcd2dc
CM		 6331 set_page_extent_mapped(page);
6332
eb84ae03
CM		 6333 /*
6334 * we can't set the delalloc bits if there are pending ordered
6335 * extents. Drop our locks and wait for them to finish
6336 */
e6dcd2dc
CM		 6337 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6338 if (ordered) {
2ac55d41
JB		 6339 unlock_extent_cached(io_tree, page_start, page_end,
6340 &cached_state, GFP_NOFS);
e6dcd2dc 6341 unlock_page(page);
eb84ae03 6342 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6343 btrfs_put_ordered_extent(ordered);
6344 goto again;
6345 }
6346
fbf19087
JB		 6347 /*
6348 * XXX - page_mkwrite gets called every time the page is dirtied, even
6349 * if it was already dirty, so for space accounting reasons we need to
6350 * clear any delalloc bits for the range we are fixing to save. There
6351 * is probably a better way to do this, but for now keep consistent with
6352 * prepare_pages in the normal write path.
6353 */
2ac55d41 6354 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6355 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6356 0, 0, &cached_state, GFP_NOFS);
fbf19087 6357
2ac55d41
JB		 6358 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6359 &cached_state);
9ed74f2d 6360 if (ret) {
2ac55d41
JB		 6361 unlock_extent_cached(io_tree, page_start, page_end,
6362 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6363 ret = VM_FAULT_SIGBUS;
6364 goto out_unlock;
6365 }
e6dcd2dc 6366 ret = 0;
9ebefb18
CM		 6367
6368 /* page is wholly or partially inside EOF */
a52d9a80 6369 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6370 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6371 else
e6dcd2dc 6372 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6373
e6dcd2dc
CM		 6374 if (zero_start != PAGE_CACHE_SIZE) {
6375 kaddr = kmap(page);
6376 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6377 flush_dcache_page(page);
6378 kunmap(page);
6379 }
247e743c 6380 ClearPageChecked(page);
e6dcd2dc 6381 set_page_dirty(page);
50a9b214 6382 SetPageUptodate(page);
5a3f23d5 6383
257c62e1
CM		 6384 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6385 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6386
2ac55d41 6387 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6388
6389out_unlock:
50a9b214
CM		 6390 if (!ret)
6391 return VM_FAULT_LOCKED;
9ebefb18 6392 unlock_page(page);
0ca1f7ce 6393 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
1832a6d5 6394out:
9ebefb18
CM		 6395 return ret;
6396}
6397
a41ad394 6398static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6399{
6400 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6401 struct btrfs_block_rsv *rsv;
39279cc3 6402 int ret;
3893e33b 6403 int err = 0;
39279cc3 6404 struct btrfs_trans_handle *trans;
d3c2fdcf 6405 unsigned long nr;
dbe674a9 6406 u64 mask = root->sectorsize - 1;
07127184 6407 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6408
5d5e103a
JB		 6409 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6410 if (ret)
a41ad394 6411 return ret;
8082510e 6412
4a096752 6413 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6414 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6415
fcb80c2a
JB		 6416 /*
6417 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6418 * 3 things going on here
6419 *
6420 * 1) We need to reserve space for our orphan item and the space to
6421 * delete our orphan item. Lord knows we don't want to have a dangling
6422 * orphan item because we didn't reserve space to remove it.
6423 *
6424 * 2) We need to reserve space to update our inode.
6425 *
6426 * 3) We need to have something to cache all the space that is going to
6427 * be free'd up by the truncate operation, but also have some slack
6428 * space reserved in case it uses space during the truncate (thank you
6429 * very much snapshotting).
6430 *
6431 * And we need these to all be seperate. The fact is we can use alot of
6432 * space doing the truncate, and we have no earthly idea how much space
6433 * we will use, so we need the truncate reservation to be seperate so it
6434 * doesn't end up using space reserved for updating the inode or
6435 * removing the orphan item. We also need to be able to stop the
6436 * transaction and start a new one, which means we need to be able to
6437 * update the inode several times, and we have no idea of knowing how
6438 * many times that will be, so we can't just reserve 1 item for the
6439 * entirety of the opration, so that has to be done seperately as well.
6440 * Then there is the orphan item, which does indeed need to be held on
6441 * to for the whole operation, and we need nobody to touch this reserved
6442 * space except the orphan code.
6443 *
6444 * So that leaves us with
6445 *
6446 * 1) root->orphan_block_rsv - for the orphan deletion.
6447 * 2) rsv - for the truncate reservation, which we will steal from the
6448 * transaction reservation.
6449 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6450 * updating the inode.
6451 */
6452 rsv = btrfs_alloc_block_rsv(root);
6453 if (!rsv)
6454 return -ENOMEM;
4a338542 6455 rsv->size = min_size;
f0cd846e 6456
907cbceb 6457 /*
07127184 6458 * 1 for the truncate slack space
907cbceb
JB		 6459 * 1 for the orphan item we're going to add
6460 * 1 for the orphan item deletion
6461 * 1 for updating the inode.
6462 */
07127184 6463 trans = btrfs_start_transaction(root, 4);
fcb80c2a
JB		 6464 if (IS_ERR(trans)) {
6465 err = PTR_ERR(trans);
6466 goto out;
6467 }
f0cd846e 6468
907cbceb
JB		 6469 /* Migrate the slack space for the truncate to our reserve */
6470 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6471 min_size);
fcb80c2a 6472 BUG_ON(ret);
f0cd846e
JB		 6473
6474 ret = btrfs_orphan_add(trans, inode);
6475 if (ret) {
6476 btrfs_end_transaction(trans, root);
fcb80c2a 6477 goto out;
f0cd846e
JB		 6478 }
6479
5a3f23d5
CM		 6480 /*
6481 * setattr is responsible for setting the ordered_data_close flag,
6482 * but that is only tested during the last file release. That
6483 * could happen well after the next commit, leaving a great big
6484 * window where new writes may get lost if someone chooses to write
6485 * to this file after truncating to zero
6486 *
6487 * The inode doesn't have any dirty data here, and so if we commit
6488 * this is a noop. If someone immediately starts writing to the inode
6489 * it is very likely we'll catch some of their writes in this
6490 * transaction, and the commit will find this file on the ordered
6491 * data list with good things to send down.
6492 *
6493 * This is a best effort solution, there is still a window where
6494 * using truncate to replace the contents of the file will
6495 * end up with a zero length file after a crash.
6496 */
6497 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6498 btrfs_add_ordered_operation(trans, root, inode);
6499
8082510e 6500 while (1) {
36ba022a 6501 ret = btrfs_block_rsv_refill(root, rsv, min_size);
907cbceb
JB		 6502 if (ret) {
6503 /*
6504 * This can only happen with the original transaction we
6505 * started above, every other time we shouldn't have a
6506 * transaction started yet.
6507 */
6508 if (ret == -EAGAIN)
6509 goto end_trans;
6510 err = ret;
6511 break;
6512 }
6513
d68fc57b 6514 if (!trans) {
907cbceb
JB		 6515 /* Just need the 1 for updating the inode */
6516 trans = btrfs_start_transaction(root, 1);
fcb80c2a
JB		 6517 if (IS_ERR(trans)) {
6518 err = PTR_ERR(trans);
6519 goto out;
6520 }
d68fc57b
YZ		 6521 }
6522
907cbceb
JB		 6523 trans->block_rsv = rsv;
6524
8082510e
YZ		 6525 ret = btrfs_truncate_inode_items(trans, root, inode,
6526 inode->i_size,
6527 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6528 if (ret != -EAGAIN) {
6529 err = ret;
8082510e 6530 break;
3893e33b 6531 }
39279cc3 6532
fcb80c2a 6533 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6534 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6535 if (ret) {
6536 err = ret;
6537 break;
6538 }
907cbceb 6539end_trans:
8082510e
YZ		 6540 nr = trans->blocks_used;
6541 btrfs_end_transaction(trans, root);
d68fc57b 6542 trans = NULL;
8082510e 6543 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6544 }
6545
6546 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6547 trans->block_rsv = root->orphan_block_rsv;
8082510e 6548 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6549 if (ret)
6550 err = ret;
ded5db9d
JB		 6551 } else if (ret && inode->i_nlink > 0) {
6552 /*
6553 * Failed to do the truncate, remove us from the in memory
6554 * orphan list.
6555 */
6556 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6557 }
6558
917c16b2
CM		 6559 if (trans) {
6560 trans->block_rsv = &root->fs_info->trans_block_rsv;
6561 ret = btrfs_update_inode(trans, root, inode);
6562 if (ret && !err)
6563 err = ret;
7b128766 6564
917c16b2
CM		 6565 nr = trans->blocks_used;
6566 ret = btrfs_end_transaction_throttle(trans, root);
6567 btrfs_btree_balance_dirty(root, nr);
6568 }
fcb80c2a
JB		 6569
6570out:
6571 btrfs_free_block_rsv(root, rsv);
6572
3893e33b
JB		 6573 if (ret && !err)
6574 err = ret;
a41ad394 6575
3893e33b 6576 return err;
39279cc3
CM		 6577}
6578
d352ac68
CM		 6579/*
6580 * create a new subvolume directory/inode (helper for the ioctl).
6581 */
d2fb3437 6582int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6583 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6584{
39279cc3 6585 struct inode *inode;
76dda93c 6586 int err;
00e4e6b3 6587 u64 index = 0;
39279cc3 6588
aec7477b 6589 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d82a6f1d 6590 new_dirid, S_IFDIR | 0700, &index);
54aa1f4d 6591 if (IS_ERR(inode))
f46b5a66 6592 return PTR_ERR(inode);
39279cc3
CM		 6593 inode->i_op = &btrfs_dir_inode_operations;
6594 inode->i_fop = &btrfs_dir_file_operations;
6595
39279cc3 6596 inode->i_nlink = 1;
dbe674a9 6597 btrfs_i_size_write(inode, 0);
3b96362c 6598
76dda93c
YZ		 6599 err = btrfs_update_inode(trans, new_root, inode);
6600 BUG_ON(err);
cb8e7090 6601
76dda93c 6602 iput(inode);
cb8e7090 6603 return 0;
39279cc3
CM		 6604}
6605
39279cc3
CM		 6606struct inode *btrfs_alloc_inode(struct super_block *sb)
6607{
6608 struct btrfs_inode *ei;
2ead6ae7 6609 struct inode *inode;
39279cc3
CM		 6610
6611 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6612 if (!ei)
6613 return NULL;
2ead6ae7
YZ		 6614
6615 ei->root = NULL;
6616 ei->space_info = NULL;
6617 ei->generation = 0;
6618 ei->sequence = 0;
15ee9bc7 6619 ei->last_trans = 0;
257c62e1 6620 ei->last_sub_trans = 0;
e02119d5 6621 ei->logged_trans = 0;
2ead6ae7 6622 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6623 ei->disk_i_size = 0;
6624 ei->flags = 0;
7709cde3 6625 ei->csum_bytes = 0;
2ead6ae7
YZ		 6626 ei->index_cnt = (u64)-1;
6627 ei->last_unlink_trans = 0;
6628
9e0baf60
JB		 6629 spin_lock_init(&ei->lock);
6630 ei->outstanding_extents = 0;
6631 ei->reserved_extents = 0;
2ead6ae7
YZ		 6632
6633 ei->ordered_data_close = 0;
d68fc57b 6634 ei->orphan_meta_reserved = 0;
2ead6ae7 6635 ei->dummy_inode = 0;
4cb5300b 6636 ei->in_defrag = 0;
7fd2ae21 6637 ei->delalloc_meta_reserved = 0;
261507a0 6638 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6639
16cdcec7
MX		 6640 ei->delayed_node = NULL;
6641
2ead6ae7 6642 inode = &ei->vfs_inode;
a8067e02 6643 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6644 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6645 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
2ead6ae7 6646 mutex_init(&ei->log_mutex);
e6dcd2dc 6647 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7b128766 6648 INIT_LIST_HEAD(&ei->i_orphan);
2ead6ae7 6649 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6650 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6651 RB_CLEAR_NODE(&ei->rb_node);
6652
6653 return inode;
39279cc3
CM		 6654}
6655
fa0d7e3d
NP		 6656static void btrfs_i_callback(struct rcu_head *head)
6657{
6658 struct inode *inode = container_of(head, struct inode, i_rcu);
6659 INIT_LIST_HEAD(&inode->i_dentry);
6660 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6661}
6662
39279cc3
CM		 6663void btrfs_destroy_inode(struct inode *inode)
6664{
e6dcd2dc 6665 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6666 struct btrfs_root *root = BTRFS_I(inode)->root;
6667
39279cc3
CM		 6668 WARN_ON(!list_empty(&inode->i_dentry));
6669 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6670 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6671 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6672 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6673 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6674
a6dbd429
JB		 6675 /*
6676 * This can happen where we create an inode, but somebody else also
6677 * created the same inode and we need to destroy the one we already
6678 * created.
6679 */
6680 if (!root)
6681 goto free;
6682
5a3f23d5
CM		 6683 /*
6684 * Make sure we're properly removed from the ordered operation
6685 * lists.
6686 */
6687 smp_mb();
6688 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6689 spin_lock(&root->fs_info->ordered_extent_lock);
6690 list_del_init(&BTRFS_I(inode)->ordered_operations);
6691 spin_unlock(&root->fs_info->ordered_extent_lock);
6692 }
6693
d68fc57b 6694 spin_lock(&root->orphan_lock);
7b128766 6695 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
33345d01
LZ		 6696 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6697 (unsigned long long)btrfs_ino(inode));
8082510e 6698 list_del_init(&BTRFS_I(inode)->i_orphan);
7b128766 6699 }
d68fc57b 6700 spin_unlock(&root->orphan_lock);
7b128766 6701
d397712b 6702 while (1) {
e6dcd2dc
CM		 6703 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6704 if (!ordered)
6705 break;
6706 else {
d397712b
CM		 6707 printk(KERN_ERR "btrfs found ordered "
6708 "extent %llu %llu on inode cleanup\n",
6709 (unsigned long long)ordered->file_offset,
6710 (unsigned long long)ordered->len);
e6dcd2dc
CM		 6711 btrfs_remove_ordered_extent(inode, ordered);
6712 btrfs_put_ordered_extent(ordered);
6713 btrfs_put_ordered_extent(ordered);
6714 }
6715 }
5d4f98a2 6716 inode_tree_del(inode);
5b21f2ed 6717 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 6718free:
16cdcec7 6719 btrfs_remove_delayed_node(inode);
fa0d7e3d 6720 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 6721}
6722
45321ac5 6723int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 6724{
6725 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 6726
0af3d00b 6727 if (btrfs_root_refs(&root->root_item) == 0 &&
2cf8572d 6728 !btrfs_is_free_space_inode(root, inode))
45321ac5 6729 return 1;
76dda93c 6730 else
45321ac5 6731 return generic_drop_inode(inode);
76dda93c
YZ		 6732}
6733
0ee0fda0 6734static void init_once(void *foo)
39279cc3
CM		 6735{
6736 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6737
6738 inode_init_once(&ei->vfs_inode);
6739}
6740
6741void btrfs_destroy_cachep(void)
6742{
6743 if (btrfs_inode_cachep)
6744 kmem_cache_destroy(btrfs_inode_cachep);
6745 if (btrfs_trans_handle_cachep)
6746 kmem_cache_destroy(btrfs_trans_handle_cachep);
6747 if (btrfs_transaction_cachep)
6748 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 6749 if (btrfs_path_cachep)
6750 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 6751 if (btrfs_free_space_cachep)
6752 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 6753}
6754
6755int btrfs_init_cachep(void)
6756{
9601e3f6
CH		 6757 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6758 sizeof(struct btrfs_inode), 0,
6759 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 6760 if (!btrfs_inode_cachep)
6761 goto fail;
9601e3f6
CH		 6762
6763 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6764 sizeof(struct btrfs_trans_handle), 0,
6765 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6766 if (!btrfs_trans_handle_cachep)
6767 goto fail;
9601e3f6
CH		 6768
6769 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6770 sizeof(struct btrfs_transaction), 0,
6771 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6772 if (!btrfs_transaction_cachep)
6773 goto fail;
9601e3f6
CH		 6774
6775 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6776 sizeof(struct btrfs_path), 0,
6777 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6778 if (!btrfs_path_cachep)
6779 goto fail;
9601e3f6 6780
dc89e982
JB		 6781 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6782 sizeof(struct btrfs_free_space), 0,
6783 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6784 if (!btrfs_free_space_cachep)
6785 goto fail;
6786
39279cc3
CM		 6787 return 0;
6788fail:
6789 btrfs_destroy_cachep();
6790 return -ENOMEM;
6791}
6792
6793static int btrfs_getattr(struct vfsmount *mnt,
6794 struct dentry *dentry, struct kstat *stat)
6795{
6796 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 6797 u32 blocksize = inode->i_sb->s_blocksize;
6798
39279cc3 6799 generic_fillattr(inode, stat);
0ee5dc67 6800 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 6801 stat->blksize = PAGE_CACHE_SIZE;
fadc0d8b
DS		 6802 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6803 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 6804 return 0;
6805}
6806
75e7cb7f
LB		 6807/*
6808 * If a file is moved, it will inherit the cow and compression flags of the new
6809 * directory.
6810 */
6811static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6812{
6813 struct btrfs_inode *b_dir = BTRFS_I(dir);
6814 struct btrfs_inode *b_inode = BTRFS_I(inode);
6815
6816 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6817 b_inode->flags |= BTRFS_INODE_NODATACOW;
6818 else
6819 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6820
6821 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6822 b_inode->flags |= BTRFS_INODE_COMPRESS;
6823 else
6824 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6825}
6826
d397712b
CM		 6827static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6828 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 6829{
6830 struct btrfs_trans_handle *trans;
6831 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 6832 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 6833 struct inode *new_inode = new_dentry->d_inode;
6834 struct inode *old_inode = old_dentry->d_inode;
6835 struct timespec ctime = CURRENT_TIME;
00e4e6b3 6836 u64 index = 0;
4df27c4d 6837 u64 root_objectid;
39279cc3 6838 int ret;
33345d01 6839 u64 old_ino = btrfs_ino(old_inode);
39279cc3 6840
33345d01 6841 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 6842 return -EPERM;
6843
4df27c4d 6844 /* we only allow rename subvolume link between subvolumes */
33345d01 6845 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 6846 return -EXDEV;
6847
33345d01
LZ		 6848 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6849 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 6850 return -ENOTEMPTY;
5f39d397 6851
4df27c4d
YZ		 6852 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6853 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6854 return -ENOTEMPTY;
5a3f23d5
CM		 6855 /*
6856 * we're using rename to replace one file with another.
6857 * and the replacement file is large. Start IO on it now so
6858 * we don't add too much work to the end of the transaction
6859 */
4baf8c92 6860 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 6861 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6862 filemap_flush(old_inode->i_mapping);
6863
76dda93c 6864 /* close the racy window with snapshot create/destroy ioctl */
33345d01 6865 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 6866 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 6867 /*
6868 * We want to reserve the absolute worst case amount of items. So if
6869 * both inodes are subvols and we need to unlink them then that would
6870 * require 4 item modifications, but if they are both normal inodes it
6871 * would require 5 item modifications, so we'll assume their normal
6872 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6873 * should cover the worst case number of items we'll modify.
6874 */
6875 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 6876 if (IS_ERR(trans)) {
6877 ret = PTR_ERR(trans);
6878 goto out_notrans;
6879 }
76dda93c 6880
4df27c4d
YZ		 6881 if (dest != root)
6882 btrfs_record_root_in_trans(trans, dest);
5f39d397 6883
a5719521
YZ		 6884 ret = btrfs_set_inode_index(new_dir, &index);
6885 if (ret)
6886 goto out_fail;
5a3f23d5 6887
33345d01 6888 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6889 /* force full log commit if subvolume involved. */
6890 root->fs_info->last_trans_log_full_commit = trans->transid;
6891 } else {
a5719521
YZ		 6892 ret = btrfs_insert_inode_ref(trans, dest,
6893 new_dentry->d_name.name,
6894 new_dentry->d_name.len,
33345d01
LZ		 6895 old_ino,
6896 btrfs_ino(new_dir), index);
a5719521
YZ		 6897 if (ret)
6898 goto out_fail;
4df27c4d
YZ		 6899 /*
6900 * this is an ugly little race, but the rename is required
6901 * to make sure that if we crash, the inode is either at the
6902 * old name or the new one. pinning the log transaction lets
6903 * us make sure we don't allow a log commit to come in after
6904 * we unlink the name but before we add the new name back in.
6905 */
6906 btrfs_pin_log_trans(root);
6907 }
5a3f23d5
CM		 6908 /*
6909 * make sure the inode gets flushed if it is replacing
6910 * something.
6911 */
33345d01 6912 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 6913 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 6914
39279cc3
CM		 6915 old_dir->i_ctime = old_dir->i_mtime = ctime;
6916 new_dir->i_ctime = new_dir->i_mtime = ctime;
6917 old_inode->i_ctime = ctime;
5f39d397 6918
12fcfd22
CM		 6919 if (old_dentry->d_parent != new_dentry->d_parent)
6920 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6921
33345d01 6922 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6923 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6924 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6925 old_dentry->d_name.name,
6926 old_dentry->d_name.len);
6927 } else {
92986796
AV		 6928 ret = __btrfs_unlink_inode(trans, root, old_dir,
6929 old_dentry->d_inode,
6930 old_dentry->d_name.name,
6931 old_dentry->d_name.len);
6932 if (!ret)
6933 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d
YZ		 6934 }
6935 BUG_ON(ret);
39279cc3
CM		 6936
6937 if (new_inode) {
6938 new_inode->i_ctime = CURRENT_TIME;
33345d01 6939 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 6940 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6941 root_objectid = BTRFS_I(new_inode)->location.objectid;
6942 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6943 root_objectid,
6944 new_dentry->d_name.name,
6945 new_dentry->d_name.len);
6946 BUG_ON(new_inode->i_nlink == 0);
6947 } else {
6948 ret = btrfs_unlink_inode(trans, dest, new_dir,
6949 new_dentry->d_inode,
6950 new_dentry->d_name.name,
6951 new_dentry->d_name.len);
6952 }
6953 BUG_ON(ret);
7b128766 6954 if (new_inode->i_nlink == 0) {
e02119d5 6955 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 6956 BUG_ON(ret);
7b128766 6957 }
39279cc3 6958 }
aec7477b 6959
75e7cb7f
LB		 6960 fixup_inode_flags(new_dir, old_inode);
6961
4df27c4d
YZ		 6962 ret = btrfs_add_link(trans, new_dir, old_inode,
6963 new_dentry->d_name.name,
a5719521 6964 new_dentry->d_name.len, 0, index);
4df27c4d 6965 BUG_ON(ret);
39279cc3 6966
33345d01 6967 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 6968 struct dentry *parent = new_dentry->d_parent;
6a912213 6969 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 6970 btrfs_end_log_trans(root);
6971 }
39279cc3 6972out_fail:
ab78c84d 6973 btrfs_end_transaction_throttle(trans, root);
b44c59a8 6974out_notrans:
33345d01 6975 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 6976 up_read(&root->fs_info->subvol_sem);
9ed74f2d 6977
39279cc3
CM		 6978 return ret;
6979}
6980
d352ac68
CM		 6981/*
6982 * some fairly slow code that needs optimization. This walks the list
6983 * of all the inodes with pending delalloc and forces them to disk.
6984 */
24bbcf04 6985int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 6986{
6987 struct list_head *head = &root->fs_info->delalloc_inodes;
6988 struct btrfs_inode *binode;
5b21f2ed 6989 struct inode *inode;
ea8c2819 6990
c146afad
YZ		 6991 if (root->fs_info->sb->s_flags & MS_RDONLY)
6992 return -EROFS;
6993
75eff68e 6994 spin_lock(&root->fs_info->delalloc_lock);
d397712b 6995 while (!list_empty(head)) {
ea8c2819
CM		 6996 binode = list_entry(head->next, struct btrfs_inode,
6997 delalloc_inodes);
5b21f2ed
ZY		 6998 inode = igrab(&binode->vfs_inode);
6999 if (!inode)
7000 list_del_init(&binode->delalloc_inodes);
75eff68e 7001 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7002 if (inode) {
8c8bee1d 7003 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7004 if (delay_iput)
7005 btrfs_add_delayed_iput(inode);
7006 else
7007 iput(inode);
5b21f2ed
ZY		 7008 }
7009 cond_resched();
75eff68e 7010 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7011 }
75eff68e 7012 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7013
7014 /* the filemap_flush will queue IO into the worker threads, but
7015 * we have to make sure the IO is actually started and that
7016 * ordered extents get created before we return
7017 */
7018 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7019 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7020 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7021 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7022 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7023 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7024 }
7025 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7026 return 0;
7027}
7028
39279cc3
CM		 7029static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7030 const char *symname)
7031{
7032 struct btrfs_trans_handle *trans;
7033 struct btrfs_root *root = BTRFS_I(dir)->root;
7034 struct btrfs_path *path;
7035 struct btrfs_key key;
1832a6d5 7036 struct inode *inode = NULL;
39279cc3
CM		 7037 int err;
7038 int drop_inode = 0;
7039 u64 objectid;
00e4e6b3 7040 u64 index = 0 ;
39279cc3
CM		 7041 int name_len;
7042 int datasize;
5f39d397 7043 unsigned long ptr;
39279cc3 7044 struct btrfs_file_extent_item *ei;
5f39d397 7045 struct extent_buffer *leaf;
1832a6d5 7046 unsigned long nr = 0;
39279cc3
CM		 7047
7048 name_len = strlen(symname) + 1;
7049 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7050 return -ENAMETOOLONG;
1832a6d5 7051
9ed74f2d
JB		 7052 /*
7053 * 2 items for inode item and ref
7054 * 2 items for dir items
7055 * 1 item for xattr if selinux is on
7056 */
a22285a6
YZ		 7057 trans = btrfs_start_transaction(root, 5);
7058 if (IS_ERR(trans))
7059 return PTR_ERR(trans);
1832a6d5 7060
581bb050
LZ		 7061 err = btrfs_find_free_ino(root, &objectid);
7062 if (err)
7063 goto out_unlock;
7064
aec7477b 7065 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7066 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7067 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7068 if (IS_ERR(inode)) {
7069 err = PTR_ERR(inode);
39279cc3 7070 goto out_unlock;
7cf96da3 7071 }
39279cc3 7072
2a7dba39 7073 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7074 if (err) {
7075 drop_inode = 1;
7076 goto out_unlock;
7077 }
7078
a1b075d2 7079 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7080 if (err)
7081 drop_inode = 1;
7082 else {
7083 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7084 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 7085 inode->i_fop = &btrfs_file_operations;
7086 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 7087 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7088 }
39279cc3
CM		 7089 if (drop_inode)
7090 goto out_unlock;
7091
7092 path = btrfs_alloc_path();
d8926bb3
MF		 7093 if (!path) {
7094 err = -ENOMEM;
7095 drop_inode = 1;
7096 goto out_unlock;
7097 }
33345d01 7098 key.objectid = btrfs_ino(inode);
39279cc3 7099 key.offset = 0;
39279cc3
CM		 7100 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7101 datasize = btrfs_file_extent_calc_inline_size(name_len);
7102 err = btrfs_insert_empty_item(trans, root, path, &key,
7103 datasize);
54aa1f4d
CM		 7104 if (err) {
7105 drop_inode = 1;
b0839166 7106 btrfs_free_path(path);
54aa1f4d
CM		 7107 goto out_unlock;
7108 }
5f39d397
CM		 7109 leaf = path->nodes[0];
7110 ei = btrfs_item_ptr(leaf, path->slots[0],
7111 struct btrfs_file_extent_item);
7112 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7113 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7114 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7115 btrfs_set_file_extent_encryption(leaf, ei, 0);
7116 btrfs_set_file_extent_compression(leaf, ei, 0);
7117 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7118 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7119
39279cc3 7120 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7121 write_extent_buffer(leaf, symname, ptr, name_len);
7122 btrfs_mark_buffer_dirty(leaf);
39279cc3 7123 btrfs_free_path(path);
5f39d397 7124
39279cc3
CM		 7125 inode->i_op = &btrfs_symlink_inode_operations;
7126 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7127 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7128 inode_set_bytes(inode, name_len);
dbe674a9 7129 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7130 err = btrfs_update_inode(trans, root, inode);
7131 if (err)
7132 drop_inode = 1;
39279cc3
CM		 7133
7134out_unlock:
d3c2fdcf 7135 nr = trans->blocks_used;
ab78c84d 7136 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 7137 if (drop_inode) {
7138 inode_dec_link_count(inode);
7139 iput(inode);
7140 }
d3c2fdcf 7141 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7142 return err;
7143}
16432985 7144
0af3d00b
JB		 7145static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7146 u64 start, u64 num_bytes, u64 min_size,
7147 loff_t actual_len, u64 *alloc_hint,
7148 struct btrfs_trans_handle *trans)
d899e052 7149{
d899e052
YZ		 7150 struct btrfs_root *root = BTRFS_I(inode)->root;
7151 struct btrfs_key ins;
d899e052 7152 u64 cur_offset = start;
55a61d1d 7153 u64 i_size;
d899e052 7154 int ret = 0;
0af3d00b 7155 bool own_trans = true;
d899e052 7156
0af3d00b
JB		 7157 if (trans)
7158 own_trans = false;
d899e052 7159 while (num_bytes > 0) {
0af3d00b
JB		 7160 if (own_trans) {
7161 trans = btrfs_start_transaction(root, 3);
7162 if (IS_ERR(trans)) {
7163 ret = PTR_ERR(trans);
7164 break;
7165 }
5a303d5d
YZ		 7166 }
7167
efa56464
YZ		 7168 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7169 0, *alloc_hint, (u64)-1, &ins, 1);
5a303d5d 7170 if (ret) {
0af3d00b
JB		 7171 if (own_trans)
7172 btrfs_end_transaction(trans, root);
a22285a6 7173 break;
d899e052 7174 }
5a303d5d 7175
d899e052
YZ		 7176 ret = insert_reserved_file_extent(trans, inode,
7177 cur_offset, ins.objectid,
7178 ins.offset, ins.offset,
920bbbfb 7179 ins.offset, 0, 0, 0,
d899e052
YZ		 7180 BTRFS_FILE_EXTENT_PREALLOC);
7181 BUG_ON(ret);
a1ed835e
CM		 7182 btrfs_drop_extent_cache(inode, cur_offset,
7183 cur_offset + ins.offset -1, 0);
5a303d5d 7184
d899e052
YZ		 7185 num_bytes -= ins.offset;
7186 cur_offset += ins.offset;
efa56464 7187 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7188
d899e052 7189 inode->i_ctime = CURRENT_TIME;
6cbff00f 7190 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7191 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7192 (actual_len > inode->i_size) &&
7193 (cur_offset > inode->i_size)) {
d1ea6a61 7194 if (cur_offset > actual_len)
55a61d1d 7195 i_size = actual_len;
d1ea6a61 7196 else
55a61d1d
JB		 7197 i_size = cur_offset;
7198 i_size_write(inode, i_size);
7199 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7200 }
7201
d899e052
YZ		 7202 ret = btrfs_update_inode(trans, root, inode);
7203 BUG_ON(ret);
d899e052 7204
0af3d00b
JB		 7205 if (own_trans)
7206 btrfs_end_transaction(trans, root);
5a303d5d 7207 }
d899e052
YZ		 7208 return ret;
7209}
7210
0af3d00b
JB		 7211int btrfs_prealloc_file_range(struct inode *inode, int mode,
7212 u64 start, u64 num_bytes, u64 min_size,
7213 loff_t actual_len, u64 *alloc_hint)
7214{
7215 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7216 min_size, actual_len, alloc_hint,
7217 NULL);
7218}
7219
7220int btrfs_prealloc_file_range_trans(struct inode *inode,
7221 struct btrfs_trans_handle *trans, int mode,
7222 u64 start, u64 num_bytes, u64 min_size,
7223 loff_t actual_len, u64 *alloc_hint)
7224{
7225 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7226 min_size, actual_len, alloc_hint, trans);
7227}
7228
e6dcd2dc
CM		 7229static int btrfs_set_page_dirty(struct page *page)
7230{
e6dcd2dc
CM		 7231 return __set_page_dirty_nobuffers(page);
7232}
7233
10556cb2 7234static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7235{
b83cc969 7236 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7237 umode_t mode = inode->i_mode;
b83cc969 7238
cb6db4e5
JM		 7239 if (mask & MAY_WRITE &&
7240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7241 if (btrfs_root_readonly(root))
7242 return -EROFS;
7243 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7244 return -EACCES;
7245 }
2830ba7f 7246 return generic_permission(inode, mask);
fdebe2bd 7247}
39279cc3 7248
6e1d5dcc 7249static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7250 .getattr = btrfs_getattr,
39279cc3
CM		 7251 .lookup = btrfs_lookup,
7252 .create = btrfs_create,
7253 .unlink = btrfs_unlink,
7254 .link = btrfs_link,
7255 .mkdir = btrfs_mkdir,
7256 .rmdir = btrfs_rmdir,
7257 .rename = btrfs_rename,
7258 .symlink = btrfs_symlink,
7259 .setattr = btrfs_setattr,
618e21d5 7260 .mknod = btrfs_mknod,
95819c05
CH		 7261 .setxattr = btrfs_setxattr,
7262 .getxattr = btrfs_getxattr,
5103e947 7263 .listxattr = btrfs_listxattr,
95819c05 7264 .removexattr = btrfs_removexattr,
fdebe2bd 7265 .permission = btrfs_permission,
4e34e719 7266 .get_acl = btrfs_get_acl,
39279cc3 7267};
6e1d5dcc 7268static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7269 .lookup = btrfs_lookup,
fdebe2bd 7270 .permission = btrfs_permission,
4e34e719 7271 .get_acl = btrfs_get_acl,
39279cc3 7272};
76dda93c 7273
828c0950 7274static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7275 .llseek = generic_file_llseek,
7276 .read = generic_read_dir,
cbdf5a24 7277 .readdir = btrfs_real_readdir,
34287aa3 7278 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7279#ifdef CONFIG_COMPAT
34287aa3 7280 .compat_ioctl = btrfs_ioctl,
39279cc3 7281#endif
6bf13c0c 7282 .release = btrfs_release_file,
e02119d5 7283 .fsync = btrfs_sync_file,
39279cc3
CM		 7284};
7285
d1310b2e 7286static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7287 .fill_delalloc = run_delalloc_range,
065631f6 7288 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7289 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7290 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7291 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7292 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 7293 .set_bit_hook = btrfs_set_bit_hook,
7294 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7295 .merge_extent_hook = btrfs_merge_extent_hook,
7296 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7297};
7298
35054394
CM		 7299/*
7300 * btrfs doesn't support the bmap operation because swapfiles
7301 * use bmap to make a mapping of extents in the file. They assume
7302 * these extents won't change over the life of the file and they
7303 * use the bmap result to do IO directly to the drive.
7304 *
7305 * the btrfs bmap call would return logical addresses that aren't
7306 * suitable for IO and they also will change frequently as COW
7307 * operations happen. So, swapfile + btrfs == corruption.
7308 *
7309 * For now we're avoiding this by dropping bmap.
7310 */
7f09410b 7311static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7312 .readpage = btrfs_readpage,
7313 .writepage = btrfs_writepage,
b293f02e 7314 .writepages = btrfs_writepages,
3ab2fb5a 7315 .readpages = btrfs_readpages,
16432985 7316 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7317 .invalidatepage = btrfs_invalidatepage,
7318 .releasepage = btrfs_releasepage,
e6dcd2dc 7319 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7320 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7321};
7322
7f09410b 7323static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7324 .readpage = btrfs_readpage,
7325 .writepage = btrfs_writepage,
2bf5a725
CM		 7326 .invalidatepage = btrfs_invalidatepage,
7327 .releasepage = btrfs_releasepage,
39279cc3
CM		 7328};
7329
6e1d5dcc 7330static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7331 .getattr = btrfs_getattr,
7332 .setattr = btrfs_setattr,
95819c05
CH		 7333 .setxattr = btrfs_setxattr,
7334 .getxattr = btrfs_getxattr,
5103e947 7335 .listxattr = btrfs_listxattr,
95819c05 7336 .removexattr = btrfs_removexattr,
fdebe2bd 7337 .permission = btrfs_permission,
1506fcc8 7338 .fiemap = btrfs_fiemap,
4e34e719 7339 .get_acl = btrfs_get_acl,
39279cc3 7340};
6e1d5dcc 7341static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7342 .getattr = btrfs_getattr,
7343 .setattr = btrfs_setattr,
fdebe2bd 7344 .permission = btrfs_permission,
95819c05
CH		 7345 .setxattr = btrfs_setxattr,
7346 .getxattr = btrfs_getxattr,
33268eaf 7347 .listxattr = btrfs_listxattr,
95819c05 7348 .removexattr = btrfs_removexattr,
4e34e719 7349 .get_acl = btrfs_get_acl,
618e21d5 7350};
6e1d5dcc 7351static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7352 .readlink = generic_readlink,
7353 .follow_link = page_follow_link_light,
7354 .put_link = page_put_link,
f209561a 7355 .getattr = btrfs_getattr,
fdebe2bd 7356 .permission = btrfs_permission,
0279b4cd
JO		 7357 .setxattr = btrfs_setxattr,
7358 .getxattr = btrfs_getxattr,
7359 .listxattr = btrfs_listxattr,
7360 .removexattr = btrfs_removexattr,
4e34e719 7361 .get_acl = btrfs_get_acl,
39279cc3 7362};
76dda93c 7363
82d339d9 7364const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7365 .d_delete = btrfs_dentry_delete,
b4aff1f8 7366 .d_release = btrfs_dentry_release,
76dda93c 7367};
Linux 6.x block layer and io_uring tree(s)