projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: kill free_space pointer from inode structure
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
4b4e25f2 42#include "compat.h"
39279cc3
CM		 43#include "ctree.h"
44#include "disk-io.h"
45#include "transaction.h"
46#include "btrfs_inode.h"
47#include "ioctl.h"
48#include "print-tree.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 51#include "tree-log.h"
4a54c8c1 52#include "volumes.h"
c8b97818 53#include "compression.h"
b4ce94de 54#include "locking.h"
dc89e982 55#include "free-space-cache.h"
581bb050 56#include "inode-map.h"
39279cc3
CM		 57
58struct btrfs_iget_args {
59 u64 ino;
60 struct btrfs_root *root;
61};
62
6e1d5dcc
AD		 63static const struct inode_operations btrfs_dir_inode_operations;
64static const struct inode_operations btrfs_symlink_inode_operations;
65static const struct inode_operations btrfs_dir_ro_inode_operations;
66static const struct inode_operations btrfs_special_inode_operations;
67static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 68static const struct address_space_operations btrfs_aops;
69static const struct address_space_operations btrfs_symlink_aops;
828c0950 70static const struct file_operations btrfs_dir_file_operations;
d1310b2e 71static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 72
73static struct kmem_cache *btrfs_inode_cachep;
74struct kmem_cache *btrfs_trans_handle_cachep;
75struct kmem_cache *btrfs_transaction_cachep;
39279cc3 76struct kmem_cache *btrfs_path_cachep;
dc89e982 77struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 78
79#define S_SHIFT 12
80static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
88};
89
a41ad394
JB		 90static int btrfs_setsize(struct inode *inode, loff_t newsize);
91static int btrfs_truncate(struct inode *inode);
5fd02043 92static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 93static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
2115133f
CM		 97static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
7b128766 99
f34f57a3 100static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
0279b4cd
JO		 103{
104 int err;
105
f34f57a3 106 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 107 if (!err)
2a7dba39 108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 109 return err;
110}
111
c8b97818
CM		 112/*
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
116 */
d397712b 117static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
fe3f566c 120 int compress_type,
c8b97818
CM		 121 struct page **compressed_pages)
122{
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
127 char *kaddr;
128 unsigned long ptr;
129 struct btrfs_file_extent_item *ei;
130 int err = 0;
131 int ret;
132 size_t cur_size = size;
133 size_t datasize;
134 unsigned long offset;
c8b97818 135
fe3f566c 136 if (compressed_size && compressed_pages)
c8b97818 137 cur_size = compressed_size;
c8b97818 138
d397712b
CM		 139 path = btrfs_alloc_path();
140 if (!path)
c8b97818
CM		 141 return -ENOMEM;
142
b9473439 143 path->leave_spinning = 1;
c8b97818 144
33345d01 145 key.objectid = btrfs_ino(inode);
c8b97818
CM		 146 key.offset = start;
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 datasize);
c8b97818
CM		 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
7ac687d9 175 kaddr = kmap_atomic(cpage);
c8b97818 176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 177 kunmap_atomic(kaddr);
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);
7ac687d9 189 kaddr = kmap_atomic(page);
c8b97818
CM		 190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 192 kunmap_atomic(kaddr);
c8b97818
CM		 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 207 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 208 ret = btrfs_update_inode(trans, root, inode);
c2167754 209
79787eaa 210 return ret;
c8b97818
CM		 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);
79787eaa
JM		 252 if (ret)
253 return ret;
c8b97818
CM		 254
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
fe3f566c 259 compress_type, compressed_pages);
2adcac1a 260 if (ret && ret != -ENOSPC) {
79787eaa
JM		 261 btrfs_abort_transaction(trans, root, ret);
262 return ret;
2adcac1a
JB		 263 } else if (ret == -ENOSPC) {
264 return 1;
79787eaa 265 }
2adcac1a 266
0ca1f7ce 267 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 268 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 269 return 0;
270}
271
771ed689
CM		 272struct async_extent {
273 u64 start;
274 u64 ram_size;
275 u64 compressed_size;
276 struct page **pages;
277 unsigned long nr_pages;
261507a0 278 int compress_type;
771ed689
CM		 279 struct list_head list;
280};
281
282struct async_cow {
283 struct inode *inode;
284 struct btrfs_root *root;
285 struct page *locked_page;
286 u64 start;
287 u64 end;
288 struct list_head extents;
289 struct btrfs_work work;
290};
291
292static noinline int add_async_extent(struct async_cow *cow,
293 u64 start, u64 ram_size,
294 u64 compressed_size,
295 struct page **pages,
261507a0
LZ		 296 unsigned long nr_pages,
297 int compress_type)
771ed689
CM		 298{
299 struct async_extent *async_extent;
300
301 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 302 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 303 async_extent->start = start;
304 async_extent->ram_size = ram_size;
305 async_extent->compressed_size = compressed_size;
306 async_extent->pages = pages;
307 async_extent->nr_pages = nr_pages;
261507a0 308 async_extent->compress_type = compress_type;
771ed689
CM		 309 list_add_tail(&async_extent->list, &cow->extents);
310 return 0;
311}
312
d352ac68 313/*
771ed689
CM		 314 * we create compressed extents in two phases. The first
315 * phase compresses a range of pages that have already been
316 * locked (both pages and state bits are locked).
c8b97818 317 *
771ed689
CM		 318 * This is done inside an ordered work queue, and the compression
319 * is spread across many cpus. The actual IO submission is step
320 * two, and the ordered work queue takes care of making sure that
321 * happens in the same order things were put onto the queue by
322 * writepages and friends.
c8b97818 323 *
771ed689
CM		 324 * If this code finds it can't get good compression, it puts an
325 * entry onto the work queue to write the uncompressed bytes. This
326 * makes sure that both compressed inodes and uncompressed inodes
327 * are written in the same order that pdflush sent them down.
d352ac68 328 */
771ed689
CM		 329static noinline int compress_file_range(struct inode *inode,
330 struct page *locked_page,
331 u64 start, u64 end,
332 struct async_cow *async_cow,
333 int *num_added)
b888db2b
CM		 334{
335 struct btrfs_root *root = BTRFS_I(inode)->root;
336 struct btrfs_trans_handle *trans;
db94535d 337 u64 num_bytes;
db94535d 338 u64 blocksize = root->sectorsize;
c8b97818 339 u64 actual_end;
42dc7bab 340 u64 isize = i_size_read(inode);
e6dcd2dc 341 int ret = 0;
c8b97818
CM		 342 struct page **pages = NULL;
343 unsigned long nr_pages;
344 unsigned long nr_pages_ret = 0;
345 unsigned long total_compressed = 0;
346 unsigned long total_in = 0;
347 unsigned long max_compressed = 128 * 1024;
771ed689 348 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 349 int i;
350 int will_compress;
261507a0 351 int compress_type = root->fs_info->compress_type;
b888db2b 352
4cb13e5d
LB		 353 /* if this is a small write inside eof, kick off a defrag */
354 if ((end - start + 1) < 16 * 1024 &&
355 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 356 btrfs_add_inode_defrag(NULL, inode);
357
42dc7bab 358 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 359again:
360 will_compress = 0;
361 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
362 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 363
f03d9301
CM		 364 /*
365 * we don't want to send crud past the end of i_size through
366 * compression, that's just a waste of CPU time. So, if the
367 * end of the file is before the start of our current
368 * requested range of bytes, we bail out to the uncompressed
369 * cleanup code that can deal with all of this.
370 *
371 * It isn't really the fastest way to fix things, but this is a
372 * very uncommon corner.
373 */
374 if (actual_end <= start)
375 goto cleanup_and_bail_uncompressed;
376
c8b97818
CM		 377 total_compressed = actual_end - start;
378
379 /* we want to make sure that amount of ram required to uncompress
380 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 381 * of a compressed extent to 128k. This is a crucial number
382 * because it also controls how easily we can spread reads across
383 * cpus for decompression.
384 *
385 * We also want to make sure the amount of IO required to do
386 * a random read is reasonably small, so we limit the size of
387 * a compressed extent to 128k.
c8b97818
CM		 388 */
389 total_compressed = min(total_compressed, max_uncompressed);
db94535d 390 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 391 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 392 total_in = 0;
393 ret = 0;
db94535d 394
771ed689
CM		 395 /*
396 * we do compression for mount -o compress and when the
397 * inode has not been flagged as nocompress. This flag can
398 * change at any time if we discover bad compression ratios.
c8b97818 399 */
6cbff00f 400 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 401 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 402 (BTRFS_I(inode)->force_compress) ||
403 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 404 WARN_ON(pages);
cfbc246e 405 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 406 if (!pages) {
407 /* just bail out to the uncompressed code */
408 goto cont;
409 }
c8b97818 410
261507a0
LZ		 411 if (BTRFS_I(inode)->force_compress)
412 compress_type = BTRFS_I(inode)->force_compress;
413
414 ret = btrfs_compress_pages(compress_type,
415 inode->i_mapping, start,
416 total_compressed, pages,
417 nr_pages, &nr_pages_ret,
418 &total_in,
419 &total_compressed,
420 max_compressed);
c8b97818
CM		 421
422 if (!ret) {
423 unsigned long offset = total_compressed &
424 (PAGE_CACHE_SIZE - 1);
425 struct page *page = pages[nr_pages_ret - 1];
426 char *kaddr;
427
428 /* zero the tail end of the last page, we might be
429 * sending it down to disk
430 */
431 if (offset) {
7ac687d9 432 kaddr = kmap_atomic(page);
c8b97818
CM		 433 memset(kaddr + offset, 0,
434 PAGE_CACHE_SIZE - offset);
7ac687d9 435 kunmap_atomic(kaddr);
c8b97818
CM		 436 }
437 will_compress = 1;
438 }
439 }
560f7d75 440cont:
c8b97818 441 if (start == 0) {
7a7eaa40 442 trans = btrfs_join_transaction(root);
79787eaa
JM		 443 if (IS_ERR(trans)) {
444 ret = PTR_ERR(trans);
445 trans = NULL;
446 goto cleanup_and_out;
447 }
0ca1f7ce 448 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 449
c8b97818 450 /* lets try to make an inline extent */
771ed689 451 if (ret || total_in < (actual_end - start)) {
c8b97818 452 /* we didn't compress the entire range, try
771ed689 453 * to make an uncompressed inline extent.
c8b97818
CM		 454 */
455 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 456 start, end, 0, 0, NULL);
c8b97818 457 } else {
771ed689 458 /* try making a compressed inline extent */
c8b97818
CM		 459 ret = cow_file_range_inline(trans, root, inode,
460 start, end,
fe3f566c
LZ		 461 total_compressed,
462 compress_type, pages);
c8b97818 463 }
79787eaa 464 if (ret <= 0) {
771ed689 465 /*
79787eaa
JM		 466 * inline extent creation worked or returned error,
467 * we don't need to create any more async work items.
468 * Unlock and free up our temp pages.
771ed689 469 */
c8b97818 470 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 471 &BTRFS_I(inode)->io_tree,
472 start, end, NULL,
473 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 474 EXTENT_CLEAR_DELALLOC |
a791e35e 475 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 476
477 btrfs_end_transaction(trans, root);
c8b97818
CM		 478 goto free_pages_out;
479 }
c2167754 480 btrfs_end_transaction(trans, root);
c8b97818
CM		 481 }
482
483 if (will_compress) {
484 /*
485 * we aren't doing an inline extent round the compressed size
486 * up to a block size boundary so the allocator does sane
487 * things
488 */
489 total_compressed = (total_compressed + blocksize - 1) &
490 ~(blocksize - 1);
491
492 /*
493 * one last check to make sure the compression is really a
494 * win, compare the page count read with the blocks on disk
495 */
496 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
497 ~(PAGE_CACHE_SIZE - 1);
498 if (total_compressed >= total_in) {
499 will_compress = 0;
500 } else {
c8b97818
CM		 501 num_bytes = total_in;
502 }
503 }
504 if (!will_compress && pages) {
505 /*
506 * the compression code ran but failed to make things smaller,
507 * free any pages it allocated and our page pointer array
508 */
509 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 510 WARN_ON(pages[i]->mapping);
c8b97818
CM		 511 page_cache_release(pages[i]);
512 }
513 kfree(pages);
514 pages = NULL;
515 total_compressed = 0;
516 nr_pages_ret = 0;
517
518 /* flag the file so we don't compress in the future */
1e701a32
CM		 519 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
520 !(BTRFS_I(inode)->force_compress)) {
a555f810 521 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 522 }
c8b97818 523 }
771ed689
CM		 524 if (will_compress) {
525 *num_added += 1;
c8b97818 526
771ed689
CM		 527 /* the async work queues will take care of doing actual
528 * allocation on disk for these compressed pages,
529 * and will submit them to the elevator.
530 */
531 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 532 total_compressed, pages, nr_pages_ret,
533 compress_type);
179e29e4 534
24ae6365 535 if (start + num_bytes < end) {
771ed689
CM		 536 start += num_bytes;
537 pages = NULL;
538 cond_resched();
539 goto again;
540 }
541 } else {
f03d9301 542cleanup_and_bail_uncompressed:
771ed689
CM		 543 /*
544 * No compression, but we still need to write the pages in
545 * the file we've been given so far. redirty the locked
546 * page if it corresponds to our extent and set things up
547 * for the async work queue to run cow_file_range to do
548 * the normal delalloc dance
549 */
550 if (page_offset(locked_page) >= start &&
551 page_offset(locked_page) <= end) {
552 __set_page_dirty_nobuffers(locked_page);
553 /* unlocked later on in the async handlers */
554 }
261507a0
LZ		 555 add_async_extent(async_cow, start, end - start + 1,
556 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 557 *num_added += 1;
558 }
3b951516 559
771ed689 560out:
79787eaa 561 return ret;
771ed689
CM		 562
563free_pages_out:
564 for (i = 0; i < nr_pages_ret; i++) {
565 WARN_ON(pages[i]->mapping);
566 page_cache_release(pages[i]);
567 }
d397712b 568 kfree(pages);
771ed689
CM		 569
570 goto out;
79787eaa
JM		 571
572cleanup_and_out:
573 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
574 start, end, NULL,
575 EXTENT_CLEAR_UNLOCK_PAGE |
576 EXTENT_CLEAR_DIRTY |
577 EXTENT_CLEAR_DELALLOC |
578 EXTENT_SET_WRITEBACK |
579 EXTENT_END_WRITEBACK);
580 if (!trans || IS_ERR(trans))
581 btrfs_error(root->fs_info, ret, "Failed to join transaction");
582 else
583 btrfs_abort_transaction(trans, root, ret);
584 goto free_pages_out;
771ed689
CM		 585}
586
587/*
588 * phase two of compressed writeback. This is the ordered portion
589 * of the code, which only gets called in the order the work was
590 * queued. We walk all the async extents created by compress_file_range
591 * and send them down to the disk.
592 */
593static noinline int submit_compressed_extents(struct inode *inode,
594 struct async_cow *async_cow)
595{
596 struct async_extent *async_extent;
597 u64 alloc_hint = 0;
598 struct btrfs_trans_handle *trans;
599 struct btrfs_key ins;
600 struct extent_map *em;
601 struct btrfs_root *root = BTRFS_I(inode)->root;
602 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
603 struct extent_io_tree *io_tree;
f5a84ee3 604 int ret = 0;
771ed689
CM		 605
606 if (list_empty(&async_cow->extents))
607 return 0;
608
771ed689 609
d397712b 610 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 611 async_extent = list_entry(async_cow->extents.next,
612 struct async_extent, list);
613 list_del(&async_extent->list);
c8b97818 614
771ed689
CM		 615 io_tree = &BTRFS_I(inode)->io_tree;
616
f5a84ee3 617retry:
771ed689
CM		 618 /* did the compression code fall back to uncompressed IO? */
619 if (!async_extent->pages) {
620 int page_started = 0;
621 unsigned long nr_written = 0;
622
623 lock_extent(io_tree, async_extent->start,
2ac55d41 624 async_extent->start +
d0082371 625 async_extent->ram_size - 1);
771ed689
CM		 626
627 /* allocate blocks */
f5a84ee3
JB		 628 ret = cow_file_range(inode, async_cow->locked_page,
629 async_extent->start,
630 async_extent->start +
631 async_extent->ram_size - 1,
632 &page_started, &nr_written, 0);
771ed689 633
79787eaa
JM		 634 /* JDM XXX */
635
771ed689
CM		 636 /*
637 * if page_started, cow_file_range inserted an
638 * inline extent and took care of all the unlocking
639 * and IO for us. Otherwise, we need to submit
640 * all those pages down to the drive.
641 */
f5a84ee3 642 if (!page_started && !ret)
771ed689
CM		 643 extent_write_locked_range(io_tree,
644 inode, async_extent->start,
d397712b 645 async_extent->start +
771ed689
CM		 646 async_extent->ram_size - 1,
647 btrfs_get_extent,
648 WB_SYNC_ALL);
649 kfree(async_extent);
650 cond_resched();
651 continue;
652 }
653
654 lock_extent(io_tree, async_extent->start,
d0082371 655 async_extent->start + async_extent->ram_size - 1);
771ed689 656
7a7eaa40 657 trans = btrfs_join_transaction(root);
79787eaa
JM		 658 if (IS_ERR(trans)) {
659 ret = PTR_ERR(trans);
660 } else {
661 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
662 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 663 async_extent->compressed_size,
664 async_extent->compressed_size,
81c9ad23 665 0, alloc_hint, &ins, 1);
79787eaa
JM		 666 if (ret)
667 btrfs_abort_transaction(trans, root, ret);
668 btrfs_end_transaction(trans, root);
669 }
c2167754 670
f5a84ee3
JB		 671 if (ret) {
672 int i;
673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
676 }
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
680 unlock_extent(io_tree, async_extent->start,
681 async_extent->start +
d0082371 682 async_extent->ram_size - 1);
79787eaa
JM		 683 if (ret == -ENOSPC)
684 goto retry;
685 goto out_free; /* JDM: Requeue? */
f5a84ee3
JB		 686 }
687
c2167754
YZ		 688 /*
689 * here we're doing allocation and writeback of the
690 * compressed pages
691 */
692 btrfs_drop_extent_cache(inode, async_extent->start,
693 async_extent->start +
694 async_extent->ram_size - 1, 0);
695
172ddd60 696 em = alloc_extent_map();
79787eaa 697 BUG_ON(!em); /* -ENOMEM */
771ed689
CM		 698 em->start = async_extent->start;
699 em->len = async_extent->ram_size;
445a6944 700 em->orig_start = em->start;
c8b97818 701
771ed689
CM		 702 em->block_start = ins.objectid;
703 em->block_len = ins.offset;
704 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 705 em->compress_type = async_extent->compress_type;
771ed689
CM		 706 set_bit(EXTENT_FLAG_PINNED, &em->flags);
707 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
708
d397712b 709 while (1) {
890871be 710 write_lock(&em_tree->lock);
771ed689 711 ret = add_extent_mapping(em_tree, em);
890871be 712 write_unlock(&em_tree->lock);
771ed689
CM		 713 if (ret != -EEXIST) {
714 free_extent_map(em);
715 break;
716 }
717 btrfs_drop_extent_cache(inode, async_extent->start,
718 async_extent->start +
719 async_extent->ram_size - 1, 0);
720 }
721
261507a0
LZ		 722 ret = btrfs_add_ordered_extent_compress(inode,
723 async_extent->start,
724 ins.objectid,
725 async_extent->ram_size,
726 ins.offset,
727 BTRFS_ORDERED_COMPRESSED,
728 async_extent->compress_type);
79787eaa 729 BUG_ON(ret); /* -ENOMEM */
771ed689 730
771ed689
CM		 731 /*
732 * clear dirty, set writeback and unlock the pages.
733 */
734 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 735 &BTRFS_I(inode)->io_tree,
736 async_extent->start,
737 async_extent->start +
738 async_extent->ram_size - 1,
739 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
740 EXTENT_CLEAR_UNLOCK |
a3429ab7 741 EXTENT_CLEAR_DELALLOC |
a791e35e 742 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 743
744 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 745 async_extent->start,
746 async_extent->ram_size,
747 ins.objectid,
748 ins.offset, async_extent->pages,
749 async_extent->nr_pages);
771ed689 750
79787eaa 751 BUG_ON(ret); /* -ENOMEM */
771ed689
CM		 752 alloc_hint = ins.objectid + ins.offset;
753 kfree(async_extent);
754 cond_resched();
755 }
79787eaa
JM		 756 ret = 0;
757out:
758 return ret;
759out_free:
760 kfree(async_extent);
761 goto out;
771ed689
CM		 762}
763
4b46fce2
JB		 764static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
765 u64 num_bytes)
766{
767 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
768 struct extent_map *em;
769 u64 alloc_hint = 0;
770
771 read_lock(&em_tree->lock);
772 em = search_extent_mapping(em_tree, start, num_bytes);
773 if (em) {
774 /*
775 * if block start isn't an actual block number then find the
776 * first block in this inode and use that as a hint. If that
777 * block is also bogus then just don't worry about it.
778 */
779 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
780 free_extent_map(em);
781 em = search_extent_mapping(em_tree, 0, 0);
782 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
783 alloc_hint = em->block_start;
784 if (em)
785 free_extent_map(em);
786 } else {
787 alloc_hint = em->block_start;
788 free_extent_map(em);
789 }
790 }
791 read_unlock(&em_tree->lock);
792
793 return alloc_hint;
794}
795
771ed689
CM		 796/*
797 * when extent_io.c finds a delayed allocation range in the file,
798 * the call backs end up in this code. The basic idea is to
799 * allocate extents on disk for the range, and create ordered data structs
800 * in ram to track those extents.
801 *
802 * locked_page is the page that writepage had locked already. We use
803 * it to make sure we don't do extra locks or unlocks.
804 *
805 * *page_started is set to one if we unlock locked_page and do everything
806 * required to start IO on it. It may be clean and already done with
807 * IO when we return.
808 */
809static noinline int cow_file_range(struct inode *inode,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
813 int unlock)
814{
815 struct btrfs_root *root = BTRFS_I(inode)->root;
816 struct btrfs_trans_handle *trans;
817 u64 alloc_hint = 0;
818 u64 num_bytes;
819 unsigned long ram_size;
820 u64 disk_num_bytes;
821 u64 cur_alloc_size;
822 u64 blocksize = root->sectorsize;
771ed689
CM		 823 struct btrfs_key ins;
824 struct extent_map *em;
825 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826 int ret = 0;
827
83eea1f1 828 BUG_ON(btrfs_is_free_space_inode(inode));
7a7eaa40 829 trans = btrfs_join_transaction(root);
79787eaa
JM		 830 if (IS_ERR(trans)) {
831 extent_clear_unlock_delalloc(inode,
832 &BTRFS_I(inode)->io_tree,
beb42dd7 833 start, end, locked_page,
79787eaa
JM		 834 EXTENT_CLEAR_UNLOCK_PAGE |
835 EXTENT_CLEAR_UNLOCK |
836 EXTENT_CLEAR_DELALLOC |
837 EXTENT_CLEAR_DIRTY |
838 EXTENT_SET_WRITEBACK |
839 EXTENT_END_WRITEBACK);
840 return PTR_ERR(trans);
841 }
0ca1f7ce 842 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 843
771ed689
CM		 844 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
845 num_bytes = max(blocksize, num_bytes);
846 disk_num_bytes = num_bytes;
847 ret = 0;
848
4cb5300b 849 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 850 if (num_bytes < 64 * 1024 &&
851 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 852 btrfs_add_inode_defrag(trans, inode);
853
771ed689
CM		 854 if (start == 0) {
855 /* lets try to make an inline extent */
856 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 857 start, end, 0, 0, NULL);
771ed689
CM		 858 if (ret == 0) {
859 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 860 &BTRFS_I(inode)->io_tree,
861 start, end, NULL,
862 EXTENT_CLEAR_UNLOCK_PAGE |
863 EXTENT_CLEAR_UNLOCK |
864 EXTENT_CLEAR_DELALLOC |
865 EXTENT_CLEAR_DIRTY |
866 EXTENT_SET_WRITEBACK |
867 EXTENT_END_WRITEBACK);
c2167754 868
771ed689
CM		 869 *nr_written = *nr_written +
870 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
871 *page_started = 1;
771ed689 872 goto out;
79787eaa
JM		 873 } else if (ret < 0) {
874 btrfs_abort_transaction(trans, root, ret);
875 goto out_unlock;
771ed689
CM		 876 }
877 }
878
879 BUG_ON(disk_num_bytes >
6c41761f 880 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 881
4b46fce2 882 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 883 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
884
d397712b 885 while (disk_num_bytes > 0) {
a791e35e
CM		 886 unsigned long op;
887
287a0ab9 888 cur_alloc_size = disk_num_bytes;
e6dcd2dc 889 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 890 root->sectorsize, 0, alloc_hint,
81c9ad23 891 &ins, 1);
79787eaa
JM		 892 if (ret < 0) {
893 btrfs_abort_transaction(trans, root, ret);
894 goto out_unlock;
895 }
d397712b 896
172ddd60 897 em = alloc_extent_map();
79787eaa 898 BUG_ON(!em); /* -ENOMEM */
e6dcd2dc 899 em->start = start;
445a6944 900 em->orig_start = em->start;
771ed689
CM		 901 ram_size = ins.offset;
902 em->len = ins.offset;
c8b97818 903
e6dcd2dc 904 em->block_start = ins.objectid;
c8b97818 905 em->block_len = ins.offset;
e6dcd2dc 906 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 907 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 908
d397712b 909 while (1) {
890871be 910 write_lock(&em_tree->lock);
e6dcd2dc 911 ret = add_extent_mapping(em_tree, em);
890871be 912 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 913 if (ret != -EEXIST) {
914 free_extent_map(em);
915 break;
916 }
917 btrfs_drop_extent_cache(inode, start,
c8b97818 918 start + ram_size - 1, 0);
e6dcd2dc
CM		 919 }
920
98d20f67 921 cur_alloc_size = ins.offset;
e6dcd2dc 922 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 923 ram_size, cur_alloc_size, 0);
79787eaa 924 BUG_ON(ret); /* -ENOMEM */
c8b97818 925
17d217fe
YZ		 926 if (root->root_key.objectid ==
927 BTRFS_DATA_RELOC_TREE_OBJECTID) {
928 ret = btrfs_reloc_clone_csums(inode, start,
929 cur_alloc_size);
79787eaa
JM		 930 if (ret) {
931 btrfs_abort_transaction(trans, root, ret);
932 goto out_unlock;
933 }
17d217fe
YZ		 934 }
935
d397712b 936 if (disk_num_bytes < cur_alloc_size)
3b951516 937 break;
d397712b 938
c8b97818
CM		 939 /* we're not doing compressed IO, don't unlock the first
940 * page (which the caller expects to stay locked), don't
941 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 942 *
943 * Do set the Private2 bit so we know this page was properly
944 * setup for writepage
c8b97818 945 */
a791e35e
CM		 946 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
947 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
948 EXTENT_SET_PRIVATE2;
949
c8b97818
CM		 950 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
951 start, start + ram_size - 1,
a791e35e 952 locked_page, op);
c8b97818 953 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 954 num_bytes -= cur_alloc_size;
955 alloc_hint = ins.objectid + ins.offset;
956 start += cur_alloc_size;
b888db2b 957 }
771ed689 958 ret = 0;
79787eaa 959out:
b888db2b 960 btrfs_end_transaction(trans, root);
c8b97818 961
be20aa9d 962 return ret;
79787eaa
JM		 963out_unlock:
964 extent_clear_unlock_delalloc(inode,
965 &BTRFS_I(inode)->io_tree,
beb42dd7 966 start, end, locked_page,
79787eaa
JM		 967 EXTENT_CLEAR_UNLOCK_PAGE |
968 EXTENT_CLEAR_UNLOCK |
969 EXTENT_CLEAR_DELALLOC |
970 EXTENT_CLEAR_DIRTY |
971 EXTENT_SET_WRITEBACK |
972 EXTENT_END_WRITEBACK);
973
974 goto out;
771ed689 975}
c8b97818 976
771ed689
CM		 977/*
978 * work queue call back to started compression on a file and pages
979 */
980static noinline void async_cow_start(struct btrfs_work *work)
981{
982 struct async_cow *async_cow;
983 int num_added = 0;
984 async_cow = container_of(work, struct async_cow, work);
985
986 compress_file_range(async_cow->inode, async_cow->locked_page,
987 async_cow->start, async_cow->end, async_cow,
988 &num_added);
8180ef88 989 if (num_added == 0) {
cb77fcd8 990 btrfs_add_delayed_iput(async_cow->inode);
771ed689 991 async_cow->inode = NULL;
8180ef88 992 }
771ed689
CM		 993}
994
995/*
996 * work queue call back to submit previously compressed pages
997 */
998static noinline void async_cow_submit(struct btrfs_work *work)
999{
1000 struct async_cow *async_cow;
1001 struct btrfs_root *root;
1002 unsigned long nr_pages;
1003
1004 async_cow = container_of(work, struct async_cow, work);
1005
1006 root = async_cow->root;
1007 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1008 PAGE_CACHE_SHIFT;
1009
1010 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1011
1012 if (atomic_read(&root->fs_info->async_delalloc_pages) <
287082b0 1013 5 * 1024 * 1024 &&
771ed689
CM		 1014 waitqueue_active(&root->fs_info->async_submit_wait))
1015 wake_up(&root->fs_info->async_submit_wait);
1016
d397712b 1017 if (async_cow->inode)
771ed689 1018 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1019}
c8b97818 1020
771ed689
CM		 1021static noinline void async_cow_free(struct btrfs_work *work)
1022{
1023 struct async_cow *async_cow;
1024 async_cow = container_of(work, struct async_cow, work);
8180ef88 1025 if (async_cow->inode)
cb77fcd8 1026 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1027 kfree(async_cow);
1028}
1029
1030static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1031 u64 start, u64 end, int *page_started,
1032 unsigned long *nr_written)
1033{
1034 struct async_cow *async_cow;
1035 struct btrfs_root *root = BTRFS_I(inode)->root;
1036 unsigned long nr_pages;
1037 u64 cur_end;
287082b0 1038 int limit = 10 * 1024 * 1024;
771ed689 1039
a3429ab7
CM		 1040 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1041 1, 0, NULL, GFP_NOFS);
d397712b 1042 while (start < end) {
771ed689 1043 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1044 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1045 async_cow->inode = igrab(inode);
771ed689
CM		 1046 async_cow->root = root;
1047 async_cow->locked_page = locked_page;
1048 async_cow->start = start;
1049
6cbff00f 1050 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1051 cur_end = end;
1052 else
1053 cur_end = min(end, start + 512 * 1024 - 1);
1054
1055 async_cow->end = cur_end;
1056 INIT_LIST_HEAD(&async_cow->extents);
1057
1058 async_cow->work.func = async_cow_start;
1059 async_cow->work.ordered_func = async_cow_submit;
1060 async_cow->work.ordered_free = async_cow_free;
1061 async_cow->work.flags = 0;
1062
771ed689
CM		 1063 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1064 PAGE_CACHE_SHIFT;
1065 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1066
1067 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1068 &async_cow->work);
1069
1070 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1071 wait_event(root->fs_info->async_submit_wait,
1072 (atomic_read(&root->fs_info->async_delalloc_pages) <
1073 limit));
1074 }
1075
d397712b 1076 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1077 atomic_read(&root->fs_info->async_delalloc_pages)) {
1078 wait_event(root->fs_info->async_submit_wait,
1079 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1080 0));
1081 }
1082
1083 *nr_written += nr_pages;
1084 start = cur_end + 1;
1085 }
1086 *page_started = 1;
1087 return 0;
be20aa9d
CM		 1088}
1089
d397712b 1090static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1091 u64 bytenr, u64 num_bytes)
1092{
1093 int ret;
1094 struct btrfs_ordered_sum *sums;
1095 LIST_HEAD(list);
1096
07d400a6 1097 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1098 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1099 if (ret == 0 && list_empty(&list))
1100 return 0;
1101
1102 while (!list_empty(&list)) {
1103 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1104 list_del(&sums->list);
1105 kfree(sums);
1106 }
1107 return 1;
1108}
1109
d352ac68
CM		 1110/*
1111 * when nowcow writeback call back. This checks for snapshots or COW copies
1112 * of the extents that exist in the file, and COWs the file as required.
1113 *
1114 * If no cow copies or snapshots exist, we write directly to the existing
1115 * blocks on disk
1116 */
7f366cfe
CM		 1117static noinline int run_delalloc_nocow(struct inode *inode,
1118 struct page *locked_page,
771ed689
CM		 1119 u64 start, u64 end, int *page_started, int force,
1120 unsigned long *nr_written)
be20aa9d 1121{
be20aa9d 1122 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1123 struct btrfs_trans_handle *trans;
be20aa9d 1124 struct extent_buffer *leaf;
be20aa9d 1125 struct btrfs_path *path;
80ff3856 1126 struct btrfs_file_extent_item *fi;
be20aa9d 1127 struct btrfs_key found_key;
80ff3856
YZ		 1128 u64 cow_start;
1129 u64 cur_offset;
1130 u64 extent_end;
5d4f98a2 1131 u64 extent_offset;
80ff3856
YZ		 1132 u64 disk_bytenr;
1133 u64 num_bytes;
1134 int extent_type;
79787eaa 1135 int ret, err;
d899e052 1136 int type;
80ff3856
YZ		 1137 int nocow;
1138 int check_prev = 1;
82d5902d 1139 bool nolock;
33345d01 1140 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1141
1142 path = btrfs_alloc_path();
17ca04af
JB		 1143 if (!path) {
1144 extent_clear_unlock_delalloc(inode,
1145 &BTRFS_I(inode)->io_tree,
1146 start, end, locked_page,
1147 EXTENT_CLEAR_UNLOCK_PAGE |
1148 EXTENT_CLEAR_UNLOCK |
1149 EXTENT_CLEAR_DELALLOC |
1150 EXTENT_CLEAR_DIRTY |
1151 EXTENT_SET_WRITEBACK |
1152 EXTENT_END_WRITEBACK);
d8926bb3 1153 return -ENOMEM;
17ca04af 1154 }
82d5902d 1155
83eea1f1 1156 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1157
1158 if (nolock)
7a7eaa40 1159 trans = btrfs_join_transaction_nolock(root);
82d5902d 1160 else
7a7eaa40 1161 trans = btrfs_join_transaction(root);
ff5714cc 1162
79787eaa 1163 if (IS_ERR(trans)) {
17ca04af
JB		 1164 extent_clear_unlock_delalloc(inode,
1165 &BTRFS_I(inode)->io_tree,
1166 start, end, locked_page,
1167 EXTENT_CLEAR_UNLOCK_PAGE |
1168 EXTENT_CLEAR_UNLOCK |
1169 EXTENT_CLEAR_DELALLOC |
1170 EXTENT_CLEAR_DIRTY |
1171 EXTENT_SET_WRITEBACK |
1172 EXTENT_END_WRITEBACK);
79787eaa
JM		 1173 btrfs_free_path(path);
1174 return PTR_ERR(trans);
1175 }
1176
74b21075 1177 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1178
80ff3856
YZ		 1179 cow_start = (u64)-1;
1180 cur_offset = start;
1181 while (1) {
33345d01 1182 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1183 cur_offset, 0);
79787eaa
JM		 1184 if (ret < 0) {
1185 btrfs_abort_transaction(trans, root, ret);
1186 goto error;
1187 }
80ff3856
YZ		 1188 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1189 leaf = path->nodes[0];
1190 btrfs_item_key_to_cpu(leaf, &found_key,
1191 path->slots[0] - 1);
33345d01 1192 if (found_key.objectid == ino &&
80ff3856
YZ		 1193 found_key.type == BTRFS_EXTENT_DATA_KEY)
1194 path->slots[0]--;
1195 }
1196 check_prev = 0;
1197next_slot:
1198 leaf = path->nodes[0];
1199 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1200 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1201 if (ret < 0) {
1202 btrfs_abort_transaction(trans, root, ret);
1203 goto error;
1204 }
80ff3856
YZ		 1205 if (ret > 0)
1206 break;
1207 leaf = path->nodes[0];
1208 }
be20aa9d 1209
80ff3856
YZ		 1210 nocow = 0;
1211 disk_bytenr = 0;
17d217fe 1212 num_bytes = 0;
80ff3856
YZ		 1213 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1214
33345d01 1215 if (found_key.objectid > ino ||
80ff3856
YZ		 1216 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1217 found_key.offset > end)
1218 break;
1219
1220 if (found_key.offset > cur_offset) {
1221 extent_end = found_key.offset;
e9061e21 1222 extent_type = 0;
80ff3856
YZ		 1223 goto out_check;
1224 }
1225
1226 fi = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_file_extent_item);
1228 extent_type = btrfs_file_extent_type(leaf, fi);
1229
d899e052
YZ		 1230 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1231 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1232 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1233 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1234 extent_end = found_key.offset +
1235 btrfs_file_extent_num_bytes(leaf, fi);
1236 if (extent_end <= start) {
1237 path->slots[0]++;
1238 goto next_slot;
1239 }
17d217fe
YZ		 1240 if (disk_bytenr == 0)
1241 goto out_check;
80ff3856
YZ		 1242 if (btrfs_file_extent_compression(leaf, fi) ||
1243 btrfs_file_extent_encryption(leaf, fi) ||
1244 btrfs_file_extent_other_encoding(leaf, fi))
1245 goto out_check;
d899e052
YZ		 1246 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247 goto out_check;
d2fb3437 1248 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1249 goto out_check;
33345d01 1250 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1251 found_key.offset -
1252 extent_offset, disk_bytenr))
17d217fe 1253 goto out_check;
5d4f98a2 1254 disk_bytenr += extent_offset;
17d217fe
YZ		 1255 disk_bytenr += cur_offset - found_key.offset;
1256 num_bytes = min(end + 1, extent_end) - cur_offset;
1257 /*
1258 * force cow if csum exists in the range.
1259 * this ensure that csum for a given extent are
1260 * either valid or do not exist.
1261 */
1262 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263 goto out_check;
80ff3856
YZ		 1264 nocow = 1;
1265 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266 extent_end = found_key.offset +
1267 btrfs_file_extent_inline_len(leaf, fi);
1268 extent_end = ALIGN(extent_end, root->sectorsize);
1269 } else {
1270 BUG_ON(1);
1271 }
1272out_check:
1273 if (extent_end <= start) {
1274 path->slots[0]++;
1275 goto next_slot;
1276 }
1277 if (!nocow) {
1278 if (cow_start == (u64)-1)
1279 cow_start = cur_offset;
1280 cur_offset = extent_end;
1281 if (cur_offset > end)
1282 break;
1283 path->slots[0]++;
1284 goto next_slot;
7ea394f1
YZ		 1285 }
1286
b3b4aa74 1287 btrfs_release_path(path);
80ff3856
YZ		 1288 if (cow_start != (u64)-1) {
1289 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1290 found_key.offset - 1, page_started,
1291 nr_written, 1);
79787eaa
JM		 1292 if (ret) {
1293 btrfs_abort_transaction(trans, root, ret);
1294 goto error;
1295 }
80ff3856 1296 cow_start = (u64)-1;
7ea394f1 1297 }
80ff3856 1298
d899e052
YZ		 1299 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300 struct extent_map *em;
1301 struct extent_map_tree *em_tree;
1302 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1303 em = alloc_extent_map();
79787eaa 1304 BUG_ON(!em); /* -ENOMEM */
d899e052 1305 em->start = cur_offset;
445a6944 1306 em->orig_start = em->start;
d899e052
YZ		 1307 em->len = num_bytes;
1308 em->block_len = num_bytes;
1309 em->block_start = disk_bytenr;
1310 em->bdev = root->fs_info->fs_devices->latest_bdev;
1311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312 while (1) {
890871be 1313 write_lock(&em_tree->lock);
d899e052 1314 ret = add_extent_mapping(em_tree, em);
890871be 1315 write_unlock(&em_tree->lock);
d899e052
YZ		 1316 if (ret != -EEXIST) {
1317 free_extent_map(em);
1318 break;
1319 }
1320 btrfs_drop_extent_cache(inode, em->start,
1321 em->start + em->len - 1, 0);
1322 }
1323 type = BTRFS_ORDERED_PREALLOC;
1324 } else {
1325 type = BTRFS_ORDERED_NOCOW;
1326 }
80ff3856
YZ		 1327
1328 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1329 num_bytes, num_bytes, type);
79787eaa 1330 BUG_ON(ret); /* -ENOMEM */
771ed689 1331
efa56464
YZ		 1332 if (root->root_key.objectid ==
1333 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1334 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1335 num_bytes);
79787eaa
JM		 1336 if (ret) {
1337 btrfs_abort_transaction(trans, root, ret);
1338 goto error;
1339 }
efa56464
YZ		 1340 }
1341
d899e052 1342 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1343 cur_offset, cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1345 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1346 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1347 cur_offset = extent_end;
1348 if (cur_offset > end)
1349 break;
be20aa9d 1350 }
b3b4aa74 1351 btrfs_release_path(path);
80ff3856 1352
17ca04af 1353 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1354 cow_start = cur_offset;
17ca04af
JB		 1355 cur_offset = end;
1356 }
1357
80ff3856
YZ		 1358 if (cow_start != (u64)-1) {
1359 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1360 page_started, nr_written, 1);
79787eaa
JM		 1361 if (ret) {
1362 btrfs_abort_transaction(trans, root, ret);
1363 goto error;
1364 }
80ff3856
YZ		 1365 }
1366
79787eaa 1367error:
0cb59c99 1368 if (nolock) {
79787eaa 1369 err = btrfs_end_transaction_nolock(trans, root);
0cb59c99 1370 } else {
79787eaa 1371 err = btrfs_end_transaction(trans, root);
0cb59c99 1372 }
79787eaa
JM		 1373 if (!ret)
1374 ret = err;
1375
17ca04af
JB		 1376 if (ret && cur_offset < end)
1377 extent_clear_unlock_delalloc(inode,
1378 &BTRFS_I(inode)->io_tree,
1379 cur_offset, end, locked_page,
1380 EXTENT_CLEAR_UNLOCK_PAGE |
1381 EXTENT_CLEAR_UNLOCK |
1382 EXTENT_CLEAR_DELALLOC |
1383 EXTENT_CLEAR_DIRTY |
1384 EXTENT_SET_WRITEBACK |
1385 EXTENT_END_WRITEBACK);
1386
7ea394f1 1387 btrfs_free_path(path);
79787eaa 1388 return ret;
be20aa9d
CM		 1389}
1390
d352ac68
CM		 1391/*
1392 * extent_io.c call back to do delayed allocation processing
1393 */
c8b97818 1394static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1395 u64 start, u64 end, int *page_started,
1396 unsigned long *nr_written)
be20aa9d 1397{
be20aa9d 1398 int ret;
7f366cfe 1399 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1400
7ddf5a42 1401 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1402 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1403 page_started, 1, nr_written);
7ddf5a42 1404 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1405 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1406 page_started, 0, nr_written);
7ddf5a42
JB		 1407 } else if (!btrfs_test_opt(root, COMPRESS) &&
1408 !(BTRFS_I(inode)->force_compress) &&
1409 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1410 ret = cow_file_range(inode, locked_page, start, end,
1411 page_started, nr_written, 1);
7ddf5a42
JB		 1412 } else {
1413 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1414 &BTRFS_I(inode)->runtime_flags);
771ed689 1415 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1416 page_started, nr_written);
7ddf5a42 1417 }
b888db2b
CM		 1418 return ret;
1419}
1420
1bf85046
JM		 1421static void btrfs_split_extent_hook(struct inode *inode,
1422 struct extent_state *orig, u64 split)
9ed74f2d 1423{
0ca1f7ce 1424 /* not delalloc, ignore it */
9ed74f2d 1425 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1426 return;
9ed74f2d 1427
9e0baf60
JB		 1428 spin_lock(&BTRFS_I(inode)->lock);
1429 BTRFS_I(inode)->outstanding_extents++;
1430 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1431}
1432
1433/*
1434 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1435 * extents so we can keep track of new extents that are just merged onto old
1436 * extents, such as when we are doing sequential writes, so we can properly
1437 * account for the metadata space we'll need.
1438 */
1bf85046
JM		 1439static void btrfs_merge_extent_hook(struct inode *inode,
1440 struct extent_state *new,
1441 struct extent_state *other)
9ed74f2d 1442{
9ed74f2d
JB		 1443 /* not delalloc, ignore it */
1444 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1445 return;
9ed74f2d 1446
9e0baf60
JB		 1447 spin_lock(&BTRFS_I(inode)->lock);
1448 BTRFS_I(inode)->outstanding_extents--;
1449 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1450}
1451
d352ac68
CM		 1452/*
1453 * extent_io.c set_bit_hook, used to track delayed allocation
1454 * bytes in this file, and to maintain the list of inodes that
1455 * have pending delalloc work to be done.
1456 */
1bf85046
JM		 1457static void btrfs_set_bit_hook(struct inode *inode,
1458 struct extent_state *state, int *bits)
291d673e 1459{
9ed74f2d 1460
75eff68e
CM		 1461 /*
1462 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1463 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1464 * bit, which is only set or cleared with irqs on
1465 */
0ca1f7ce 1466 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1467 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1468 u64 len = state->end + 1 - state->start;
83eea1f1 1469 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1470
9e0baf60 1471 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1472 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1473 } else {
1474 spin_lock(&BTRFS_I(inode)->lock);
1475 BTRFS_I(inode)->outstanding_extents++;
1476 spin_unlock(&BTRFS_I(inode)->lock);
1477 }
287a0ab9 1478
75eff68e 1479 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1480 BTRFS_I(inode)->delalloc_bytes += len;
1481 root->fs_info->delalloc_bytes += len;
0cb59c99 1482 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
ea8c2819
CM		 1483 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1484 &root->fs_info->delalloc_inodes);
1485 }
75eff68e 1486 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1487 }
291d673e
CM		 1488}
1489
d352ac68
CM		 1490/*
1491 * extent_io.c clear_bit_hook, see set_bit_hook for why
1492 */
1bf85046
JM		 1493static void btrfs_clear_bit_hook(struct inode *inode,
1494 struct extent_state *state, int *bits)
291d673e 1495{
75eff68e
CM		 1496 /*
1497 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1498 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1499 * bit, which is only set or cleared with irqs on
1500 */
0ca1f7ce 1501 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1502 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1503 u64 len = state->end + 1 - state->start;
83eea1f1 1504 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1505
9e0baf60 1506 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1507 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1508 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1509 spin_lock(&BTRFS_I(inode)->lock);
1510 BTRFS_I(inode)->outstanding_extents--;
1511 spin_unlock(&BTRFS_I(inode)->lock);
1512 }
0ca1f7ce
YZ		 1513
1514 if (*bits & EXTENT_DO_ACCOUNTING)
1515 btrfs_delalloc_release_metadata(inode, len);
1516
0cb59c99
JB		 1517 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1518 && do_list)
0ca1f7ce 1519 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1520
75eff68e 1521 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1522 root->fs_info->delalloc_bytes -= len;
1523 BTRFS_I(inode)->delalloc_bytes -= len;
1524
0cb59c99 1525 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
ea8c2819
CM		 1526 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1528 }
75eff68e 1529 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1530 }
291d673e
CM		 1531}
1532
d352ac68
CM		 1533/*
1534 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1535 * we don't create bios that span stripes or chunks
1536 */
239b14b3 1537int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1538 size_t size, struct bio *bio,
1539 unsigned long bio_flags)
239b14b3
CM		 1540{
1541 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1542 struct btrfs_mapping_tree *map_tree;
a62b9401 1543 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1544 u64 length = 0;
1545 u64 map_length;
239b14b3
CM		 1546 int ret;
1547
771ed689
CM		 1548 if (bio_flags & EXTENT_BIO_COMPRESSED)
1549 return 0;
1550
f2d8d74d 1551 length = bio->bi_size;
239b14b3
CM		 1552 map_tree = &root->fs_info->mapping_tree;
1553 map_length = length;
cea9e445 1554 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1555 &map_length, NULL, 0);
3444a972
JM		 1556 /* Will always return 0 or 1 with map_multi == NULL */
1557 BUG_ON(ret < 0);
d397712b 1558 if (map_length < length + size)
239b14b3 1559 return 1;
3444a972 1560 return 0;
239b14b3
CM		 1561}
1562
d352ac68
CM		 1563/*
1564 * in order to insert checksums into the metadata in large chunks,
1565 * we wait until bio submission time. All the pages in the bio are
1566 * checksummed and sums are attached onto the ordered extent record.
1567 *
1568 * At IO completion time the cums attached on the ordered extent record
1569 * are inserted into the btree
1570 */
d397712b
CM		 1571static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1572 struct bio *bio, int mirror_num,
eaf25d93
CM		 1573 unsigned long bio_flags,
1574 u64 bio_offset)
065631f6 1575{
065631f6 1576 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1577 int ret = 0;
e015640f 1578
d20f7043 1579 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1580 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1581 return 0;
1582}
e015640f 1583
4a69a410
CM		 1584/*
1585 * in order to insert checksums into the metadata in large chunks,
1586 * we wait until bio submission time. All the pages in the bio are
1587 * checksummed and sums are attached onto the ordered extent record.
1588 *
1589 * At IO completion time the cums attached on the ordered extent record
1590 * are inserted into the btree
1591 */
b2950863 1592static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1593 int mirror_num, unsigned long bio_flags,
1594 u64 bio_offset)
4a69a410
CM		 1595{
1596 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1597 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1598}
1599
d352ac68 1600/*
cad321ad
CM		 1601 * extent_io.c submission hook. This does the right thing for csum calculation
1602 * on write, or reading the csums from the tree before a read
d352ac68 1603 */
b2950863 1604static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1605 int mirror_num, unsigned long bio_flags,
1606 u64 bio_offset)
44b8bd7e
CM		 1607{
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
1609 int ret = 0;
19b9bdb0 1610 int skip_sum;
0417341e 1611 int metadata = 0;
44b8bd7e 1612
6cbff00f 1613 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1614
83eea1f1 1615 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1616 metadata = 2;
1617
7b6d91da 1618 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1619 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1620 if (ret)
1621 return ret;
1622
d20f7043 1623 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1624 return btrfs_submit_compressed_read(inode, bio,
1625 mirror_num, bio_flags);
c2db1073
TI		 1626 } else if (!skip_sum) {
1627 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1628 if (ret)
1629 return ret;
1630 }
4d1b5fb4 1631 goto mapit;
19b9bdb0 1632 } else if (!skip_sum) {
17d217fe
YZ		 1633 /* csum items have already been cloned */
1634 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1635 goto mapit;
19b9bdb0
CM		 1636 /* we're doing a write, do the async checksumming */
1637 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1638 inode, rw, bio, mirror_num,
eaf25d93
CM		 1639 bio_flags, bio_offset,
1640 __btrfs_submit_bio_start,
4a69a410 1641 __btrfs_submit_bio_done);
19b9bdb0
CM		 1642 }
1643
0b86a832 1644mapit:
8b712842 1645 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1646}
6885f308 1647
d352ac68
CM		 1648/*
1649 * given a list of ordered sums record them in the inode. This happens
1650 * at IO completion time based on sums calculated at bio submission time.
1651 */
ba1da2f4 1652static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1653 struct inode *inode, u64 file_offset,
1654 struct list_head *list)
1655{
e6dcd2dc
CM		 1656 struct btrfs_ordered_sum *sum;
1657
c6e30871 1658 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1659 btrfs_csum_file_blocks(trans,
1660 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1661 }
1662 return 0;
1663}
1664
2ac55d41
JB		 1665int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1666 struct extent_state **cached_state)
ea8c2819 1667{
d397712b 1668 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1669 WARN_ON(1);
ea8c2819 1670 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1671 cached_state, GFP_NOFS);
ea8c2819
CM		 1672}
1673
d352ac68 1674/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1675struct btrfs_writepage_fixup {
1676 struct page *page;
1677 struct btrfs_work work;
1678};
1679
b2950863 1680static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1681{
1682 struct btrfs_writepage_fixup *fixup;
1683 struct btrfs_ordered_extent *ordered;
2ac55d41 1684 struct extent_state *cached_state = NULL;
247e743c
CM		 1685 struct page *page;
1686 struct inode *inode;
1687 u64 page_start;
1688 u64 page_end;
87826df0 1689 int ret;
247e743c
CM		 1690
1691 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1692 page = fixup->page;
4a096752 1693again:
247e743c
CM		 1694 lock_page(page);
1695 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1696 ClearPageChecked(page);
1697 goto out_page;
1698 }
1699
1700 inode = page->mapping->host;
1701 page_start = page_offset(page);
1702 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1703
2ac55d41 1704 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1705 &cached_state);
4a096752
CM		 1706
1707 /* already ordered? We're done */
8b62b72b 1708 if (PagePrivate2(page))
247e743c 1709 goto out;
4a096752
CM		 1710
1711 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1712 if (ordered) {
2ac55d41
JB		 1713 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1714 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1715 unlock_page(page);
1716 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1717 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1718 goto again;
1719 }
247e743c 1720
87826df0
JM		 1721 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1722 if (ret) {
1723 mapping_set_error(page->mapping, ret);
1724 end_extent_writepage(page, ret, page_start, page_end);
1725 ClearPageChecked(page);
1726 goto out;
1727 }
1728
2ac55d41 1729 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1730 ClearPageChecked(page);
87826df0 1731 set_page_dirty(page);
247e743c 1732out:
2ac55d41
JB		 1733 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1734 &cached_state, GFP_NOFS);
247e743c
CM		 1735out_page:
1736 unlock_page(page);
1737 page_cache_release(page);
b897abec 1738 kfree(fixup);
247e743c
CM		 1739}
1740
1741/*
1742 * There are a few paths in the higher layers of the kernel that directly
1743 * set the page dirty bit without asking the filesystem if it is a
1744 * good idea. This causes problems because we want to make sure COW
1745 * properly happens and the data=ordered rules are followed.
1746 *
c8b97818 1747 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1748 * hasn't been properly setup for IO. We kick off an async process
1749 * to fix it up. The async helper will wait for ordered extents, set
1750 * the delalloc bit and make it safe to write the page.
1751 */
b2950863 1752static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1753{
1754 struct inode *inode = page->mapping->host;
1755 struct btrfs_writepage_fixup *fixup;
1756 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1757
8b62b72b
CM		 1758 /* this page is properly in the ordered list */
1759 if (TestClearPagePrivate2(page))
247e743c
CM		 1760 return 0;
1761
1762 if (PageChecked(page))
1763 return -EAGAIN;
1764
1765 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1766 if (!fixup)
1767 return -EAGAIN;
f421950f 1768
247e743c
CM		 1769 SetPageChecked(page);
1770 page_cache_get(page);
1771 fixup->work.func = btrfs_writepage_fixup_worker;
1772 fixup->page = page;
1773 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1774 return -EBUSY;
247e743c
CM		 1775}
1776
d899e052
YZ		 1777static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1778 struct inode *inode, u64 file_pos,
1779 u64 disk_bytenr, u64 disk_num_bytes,
1780 u64 num_bytes, u64 ram_bytes,
1781 u8 compression, u8 encryption,
1782 u16 other_encoding, int extent_type)
1783{
1784 struct btrfs_root *root = BTRFS_I(inode)->root;
1785 struct btrfs_file_extent_item *fi;
1786 struct btrfs_path *path;
1787 struct extent_buffer *leaf;
1788 struct btrfs_key ins;
1789 u64 hint;
1790 int ret;
1791
1792 path = btrfs_alloc_path();
d8926bb3
MF		 1793 if (!path)
1794 return -ENOMEM;
d899e052 1795
b9473439 1796 path->leave_spinning = 1;
a1ed835e
CM		 1797
1798 /*
1799 * we may be replacing one extent in the tree with another.
1800 * The new extent is pinned in the extent map, and we don't want
1801 * to drop it from the cache until it is completely in the btree.
1802 *
1803 * So, tell btrfs_drop_extents to leave this extent in the cache.
1804 * the caller is expected to unpin it and allow it to be merged
1805 * with the others.
1806 */
920bbbfb
YZ		 1807 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1808 &hint, 0);
79787eaa
JM		 1809 if (ret)
1810 goto out;
d899e052 1811
33345d01 1812 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1813 ins.offset = file_pos;
1814 ins.type = BTRFS_EXTENT_DATA_KEY;
1815 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1816 if (ret)
1817 goto out;
d899e052
YZ		 1818 leaf = path->nodes[0];
1819 fi = btrfs_item_ptr(leaf, path->slots[0],
1820 struct btrfs_file_extent_item);
1821 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1822 btrfs_set_file_extent_type(leaf, fi, extent_type);
1823 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1824 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1825 btrfs_set_file_extent_offset(leaf, fi, 0);
1826 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1827 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1828 btrfs_set_file_extent_compression(leaf, fi, compression);
1829 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1830 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1831
1832 btrfs_unlock_up_safe(path, 1);
1833 btrfs_set_lock_blocking(leaf);
1834
d899e052
YZ		 1835 btrfs_mark_buffer_dirty(leaf);
1836
1837 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1838
1839 ins.objectid = disk_bytenr;
1840 ins.offset = disk_num_bytes;
1841 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1842 ret = btrfs_alloc_reserved_file_extent(trans, root,
1843 root->root_key.objectid,
33345d01 1844 btrfs_ino(inode), file_pos, &ins);
79787eaa 1845out:
d899e052 1846 btrfs_free_path(path);
b9473439 1847
79787eaa 1848 return ret;
d899e052
YZ		 1849}
1850
5d13a98f
CM		 1851/*
1852 * helper function for btrfs_finish_ordered_io, this
1853 * just reads in some of the csum leaves to prime them into ram
1854 * before we start the transaction. It limits the amount of btree
1855 * reads required while inside the transaction.
1856 */
d352ac68
CM		 1857/* as ordered data IO finishes, this gets called so we can finish
1858 * an ordered extent if the range of bytes in the file it covers are
1859 * fully written.
1860 */
5fd02043 1861static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 1862{
5fd02043 1863 struct inode *inode = ordered_extent->inode;
e6dcd2dc 1864 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1865 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 1866 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1867 struct extent_state *cached_state = NULL;
261507a0 1868 int compress_type = 0;
e6dcd2dc 1869 int ret;
82d5902d 1870 bool nolock;
e6dcd2dc 1871
83eea1f1 1872 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 1873
5fd02043
JB		 1874 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1875 ret = -EIO;
1876 goto out;
1877 }
1878
c2167754 1879 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 1880 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
c2167754
YZ		 1881 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1882 if (!ret) {
0cb59c99 1883 if (nolock)
7a7eaa40 1884 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1885 else
7a7eaa40 1886 trans = btrfs_join_transaction(root);
79787eaa
JM		 1887 if (IS_ERR(trans))
1888 return PTR_ERR(trans);
0ca1f7ce 1889 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2115133f 1890 ret = btrfs_update_inode_fallback(trans, root, inode);
79787eaa
JM		 1891 if (ret) /* -ENOMEM or corruption */
1892 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 1893 }
1894 goto out;
1895 }
e6dcd2dc 1896
2ac55d41
JB		 1897 lock_extent_bits(io_tree, ordered_extent->file_offset,
1898 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 1899 0, &cached_state);
e6dcd2dc 1900
0cb59c99 1901 if (nolock)
7a7eaa40 1902 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1903 else
7a7eaa40 1904 trans = btrfs_join_transaction(root);
79787eaa
JM		 1905 if (IS_ERR(trans)) {
1906 ret = PTR_ERR(trans);
1907 trans = NULL;
1908 goto out_unlock;
1909 }
0ca1f7ce 1910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1911
c8b97818 1912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1913 compress_type = ordered_extent->compress_type;
d899e052 1914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1915 BUG_ON(compress_type);
920bbbfb 1916 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1917 ordered_extent->file_offset,
1918 ordered_extent->file_offset +
1919 ordered_extent->len);
d899e052 1920 } else {
0af3d00b 1921 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1922 ret = insert_reserved_file_extent(trans, inode,
1923 ordered_extent->file_offset,
1924 ordered_extent->start,
1925 ordered_extent->disk_len,
1926 ordered_extent->len,
1927 ordered_extent->len,
261507a0 1928 compress_type, 0, 0,
d899e052 1929 BTRFS_FILE_EXTENT_REG);
a1ed835e
CM		 1930 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1931 ordered_extent->file_offset,
1932 ordered_extent->len);
d899e052 1933 }
5fd02043 1934
79787eaa
JM		 1935 if (ret < 0) {
1936 btrfs_abort_transaction(trans, root, ret);
5fd02043 1937 goto out_unlock;
79787eaa 1938 }
2ac55d41 1939
e6dcd2dc
CM		 1940 add_pending_csums(trans, inode, ordered_extent->file_offset,
1941 &ordered_extent->list);
1942
1ef30be1 1943 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1944 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2115133f 1945 ret = btrfs_update_inode_fallback(trans, root, inode);
79787eaa
JM		 1946 if (ret) { /* -ENOMEM or corruption */
1947 btrfs_abort_transaction(trans, root, ret);
5fd02043 1948 goto out_unlock;
79787eaa 1949 }
1ef30be1
JB		 1950 }
1951 ret = 0;
5fd02043
JB		 1952out_unlock:
1953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1954 ordered_extent->file_offset +
1955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 1956out:
5b0e95bf 1957 if (root != root->fs_info->tree_root)
0cb59c99 1958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
5b0e95bf
JB		 1959 if (trans) {
1960 if (nolock)
0cb59c99 1961 btrfs_end_transaction_nolock(trans, root);
5b0e95bf 1962 else
0cb59c99
JB		 1963 btrfs_end_transaction(trans, root);
1964 }
1965
5fd02043
JB		 1966 if (ret)
1967 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1968 ordered_extent->file_offset +
1969 ordered_extent->len - 1, NULL, GFP_NOFS);
1970
1971 /*
1972 * This needs to be dont to make sure anybody waiting knows we are done
1973 * upating everything for this ordered extent.
1974 */
1975 btrfs_remove_ordered_extent(inode, ordered_extent);
1976
e6dcd2dc
CM		 1977 /* once for us */
1978 btrfs_put_ordered_extent(ordered_extent);
1979 /* once for the tree */
1980 btrfs_put_ordered_extent(ordered_extent);
1981
5fd02043
JB		 1982 return ret;
1983}
1984
1985static void finish_ordered_fn(struct btrfs_work *work)
1986{
1987 struct btrfs_ordered_extent *ordered_extent;
1988 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1989 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 1990}
1991
b2950863 1992static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1993 struct extent_state *state, int uptodate)
1994{
5fd02043
JB		 1995 struct inode *inode = page->mapping->host;
1996 struct btrfs_root *root = BTRFS_I(inode)->root;
1997 struct btrfs_ordered_extent *ordered_extent = NULL;
1998 struct btrfs_workers *workers;
1999
1abe9b8a 2000 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2001
8b62b72b 2002 ClearPagePrivate2(page);
5fd02043
JB		 2003 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2004 end - start + 1, uptodate))
2005 return 0;
2006
2007 ordered_extent->work.func = finish_ordered_fn;
2008 ordered_extent->work.flags = 0;
2009
83eea1f1 2010 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2011 workers = &root->fs_info->endio_freespace_worker;
2012 else
2013 workers = &root->fs_info->endio_write_workers;
2014 btrfs_queue_worker(workers, &ordered_extent->work);
2015
2016 return 0;
211f90e6
CM		 2017}
2018
d352ac68
CM		 2019/*
2020 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2021 * if there's a match, we allow the bio to finish. If not, the code in
2022 * extent_io.c will try to find good copies for us.
d352ac68 2023 */
b2950863 2024static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2025 struct extent_state *state, int mirror)
07157aac 2026{
35ebb934 2027 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 2028 struct inode *inode = page->mapping->host;
d1310b2e 2029 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2030 char *kaddr;
aadfeb6e 2031 u64 private = ~(u32)0;
07157aac 2032 int ret;
ff79f819
CM		 2033 struct btrfs_root *root = BTRFS_I(inode)->root;
2034 u32 csum = ~(u32)0;
d1310b2e 2035
d20f7043
CM		 2036 if (PageChecked(page)) {
2037 ClearPageChecked(page);
2038 goto good;
2039 }
6cbff00f
CH		 2040
2041 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2042 goto good;
17d217fe
YZ		 2043
2044 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2045 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2046 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2047 GFP_NOFS);
b6cda9bc 2048 return 0;
17d217fe 2049 }
d20f7043 2050
c2e639f0 2051 if (state && state->start == start) {
70dec807
CM		 2052 private = state->private;
2053 ret = 0;
2054 } else {
2055 ret = get_state_private(io_tree, start, &private);
2056 }
7ac687d9 2057 kaddr = kmap_atomic(page);
d397712b 2058 if (ret)
07157aac 2059 goto zeroit;
d397712b 2060
ff79f819
CM		 2061 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2062 btrfs_csum_final(csum, (char *)&csum);
d397712b 2063 if (csum != private)
07157aac 2064 goto zeroit;
d397712b 2065
7ac687d9 2066 kunmap_atomic(kaddr);
d20f7043 2067good:
07157aac
CM		 2068 return 0;
2069
2070zeroit:
945d8962 2071 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 2072 "private %llu\n",
2073 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 2074 (unsigned long long)start, csum,
2075 (unsigned long long)private);
db94535d
CM		 2076 memset(kaddr + offset, 1, end - start + 1);
2077 flush_dcache_page(page);
7ac687d9 2078 kunmap_atomic(kaddr);
3b951516
CM		 2079 if (private == 0)
2080 return 0;
7e38326f 2081 return -EIO;
07157aac 2082}
b888db2b 2083
24bbcf04
YZ		 2084struct delayed_iput {
2085 struct list_head list;
2086 struct inode *inode;
2087};
2088
79787eaa
JM		 2089/* JDM: If this is fs-wide, why can't we add a pointer to
2090 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2091void btrfs_add_delayed_iput(struct inode *inode)
2092{
2093 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2094 struct delayed_iput *delayed;
2095
2096 if (atomic_add_unless(&inode->i_count, -1, 1))
2097 return;
2098
2099 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2100 delayed->inode = inode;
2101
2102 spin_lock(&fs_info->delayed_iput_lock);
2103 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2104 spin_unlock(&fs_info->delayed_iput_lock);
2105}
2106
2107void btrfs_run_delayed_iputs(struct btrfs_root *root)
2108{
2109 LIST_HEAD(list);
2110 struct btrfs_fs_info *fs_info = root->fs_info;
2111 struct delayed_iput *delayed;
2112 int empty;
2113
2114 spin_lock(&fs_info->delayed_iput_lock);
2115 empty = list_empty(&fs_info->delayed_iputs);
2116 spin_unlock(&fs_info->delayed_iput_lock);
2117 if (empty)
2118 return;
2119
2120 down_read(&root->fs_info->cleanup_work_sem);
2121 spin_lock(&fs_info->delayed_iput_lock);
2122 list_splice_init(&fs_info->delayed_iputs, &list);
2123 spin_unlock(&fs_info->delayed_iput_lock);
2124
2125 while (!list_empty(&list)) {
2126 delayed = list_entry(list.next, struct delayed_iput, list);
2127 list_del(&delayed->list);
2128 iput(delayed->inode);
2129 kfree(delayed);
2130 }
2131 up_read(&root->fs_info->cleanup_work_sem);
2132}
2133
d68fc57b
YZ		 2134enum btrfs_orphan_cleanup_state {
2135 ORPHAN_CLEANUP_STARTED = 1,
2136 ORPHAN_CLEANUP_DONE = 2,
2137};
2138
2139/*
42b2aa86 2140 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2141 * files in the subvolume, it removes orphan item and frees block_rsv
2142 * structure.
2143 */
2144void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2145 struct btrfs_root *root)
2146{
90290e19 2147 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2148 int ret;
2149
8a35d95f 2150 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2151 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2152 return;
2153
90290e19 2154 spin_lock(&root->orphan_lock);
8a35d95f 2155 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2156 spin_unlock(&root->orphan_lock);
2157 return;
2158 }
2159
2160 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2161 spin_unlock(&root->orphan_lock);
2162 return;
2163 }
2164
2165 block_rsv = root->orphan_block_rsv;
2166 root->orphan_block_rsv = NULL;
2167 spin_unlock(&root->orphan_lock);
2168
d68fc57b
YZ		 2169 if (root->orphan_item_inserted &&
2170 btrfs_root_refs(&root->root_item) > 0) {
2171 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172 root->root_key.objectid);
2173 BUG_ON(ret);
2174 root->orphan_item_inserted = 0;
2175 }
2176
90290e19
JB		 2177 if (block_rsv) {
2178 WARN_ON(block_rsv->size > 0);
2179 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2180 }
2181}
2182
7b128766
JB		 2183/*
2184 * This creates an orphan entry for the given inode in case something goes
2185 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2186 *
2187 * NOTE: caller of this function should reserve 5 units of metadata for
2188 * this function.
7b128766
JB		 2189 */
2190int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2191{
2192 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2193 struct btrfs_block_rsv *block_rsv = NULL;
2194 int reserve = 0;
2195 int insert = 0;
2196 int ret;
7b128766 2197
d68fc57b
YZ		 2198 if (!root->orphan_block_rsv) {
2199 block_rsv = btrfs_alloc_block_rsv(root);
b532402e
TI		 2200 if (!block_rsv)
2201 return -ENOMEM;
d68fc57b 2202 }
7b128766 2203
d68fc57b
YZ		 2204 spin_lock(&root->orphan_lock);
2205 if (!root->orphan_block_rsv) {
2206 root->orphan_block_rsv = block_rsv;
2207 } else if (block_rsv) {
2208 btrfs_free_block_rsv(root, block_rsv);
2209 block_rsv = NULL;
7b128766 2210 }
7b128766 2211
8a35d95f
JB		 2212 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2213 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 2214#if 0
2215 /*
2216 * For proper ENOSPC handling, we should do orphan
2217 * cleanup when mounting. But this introduces backward
2218 * compatibility issue.
2219 */
2220 if (!xchg(&root->orphan_item_inserted, 1))
2221 insert = 2;
2222 else
2223 insert = 1;
2224#endif
2225 insert = 1;
8a35d95f 2226 atomic_dec(&root->orphan_inodes);
7b128766
JB		 2227 }
2228
72ac3c0d
JB		 2229 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2230 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2231 reserve = 1;
d68fc57b 2232 spin_unlock(&root->orphan_lock);
7b128766 2233
d68fc57b
YZ		 2234 /* grab metadata reservation from transaction handle */
2235 if (reserve) {
2236 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 2237 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 2238 }
7b128766 2239
d68fc57b
YZ		 2240 /* insert an orphan item to track this unlinked/truncated file */
2241 if (insert >= 1) {
33345d01 2242 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2243 if (ret && ret != -EEXIST) {
8a35d95f
JB		 2244 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2245 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 2246 btrfs_abort_transaction(trans, root, ret);
2247 return ret;
2248 }
2249 ret = 0;
d68fc57b
YZ		 2250 }
2251
2252 /* insert an orphan item to track subvolume contains orphan files */
2253 if (insert >= 2) {
2254 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2255 root->root_key.objectid);
79787eaa
JM		 2256 if (ret && ret != -EEXIST) {
2257 btrfs_abort_transaction(trans, root, ret);
2258 return ret;
2259 }
d68fc57b
YZ		 2260 }
2261 return 0;
7b128766
JB		 2262}
2263
2264/*
2265 * We have done the truncate/delete so we can go ahead and remove the orphan
2266 * item for this particular inode.
2267 */
2268int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2269{
2270 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2271 int delete_item = 0;
2272 int release_rsv = 0;
7b128766
JB		 2273 int ret = 0;
2274
d68fc57b 2275 spin_lock(&root->orphan_lock);
8a35d95f
JB		 2276 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2277 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2278 delete_item = 1;
7b128766 2279
72ac3c0d
JB		 2280 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2281 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2282 release_rsv = 1;
d68fc57b 2283 spin_unlock(&root->orphan_lock);
7b128766 2284
d68fc57b 2285 if (trans && delete_item) {
33345d01 2286 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2287 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 2288 }
7b128766 2289
8a35d95f 2290 if (release_rsv) {
d68fc57b 2291 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 2292 atomic_dec(&root->orphan_inodes);
2293 }
7b128766 2294
d68fc57b 2295 return 0;
7b128766
JB		 2296}
2297
2298/*
2299 * this cleans up any orphans that may be left on the list from the last use
2300 * of this root.
2301 */
66b4ffd1 2302int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2303{
2304 struct btrfs_path *path;
2305 struct extent_buffer *leaf;
7b128766
JB		 2306 struct btrfs_key key, found_key;
2307 struct btrfs_trans_handle *trans;
2308 struct inode *inode;
8f6d7f4f 2309 u64 last_objectid = 0;
7b128766
JB		 2310 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2311
d68fc57b 2312 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2313 return 0;
c71bf099
YZ		 2314
2315 path = btrfs_alloc_path();
66b4ffd1
JB		 2316 if (!path) {
2317 ret = -ENOMEM;
2318 goto out;
2319 }
7b128766
JB		 2320 path->reada = -1;
2321
2322 key.objectid = BTRFS_ORPHAN_OBJECTID;
2323 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2324 key.offset = (u64)-1;
2325
7b128766
JB		 2326 while (1) {
2327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2328 if (ret < 0)
2329 goto out;
7b128766
JB		 2330
2331 /*
2332 * if ret == 0 means we found what we were searching for, which
25985edc 2333 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2334 * find the key and see if we have stuff that matches
2335 */
2336 if (ret > 0) {
66b4ffd1 2337 ret = 0;
7b128766
JB		 2338 if (path->slots[0] == 0)
2339 break;
2340 path->slots[0]--;
2341 }
2342
2343 /* pull out the item */
2344 leaf = path->nodes[0];
7b128766
JB		 2345 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2346
2347 /* make sure the item matches what we want */
2348 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2349 break;
2350 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2351 break;
2352
2353 /* release the path since we're done with it */
b3b4aa74 2354 btrfs_release_path(path);
7b128766
JB		 2355
2356 /*
2357 * this is where we are basically btrfs_lookup, without the
2358 * crossing root thing. we store the inode number in the
2359 * offset of the orphan item.
2360 */
8f6d7f4f
JB		 2361
2362 if (found_key.offset == last_objectid) {
2363 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2364 "stopping orphan cleanup\n");
2365 ret = -EINVAL;
2366 goto out;
2367 }
2368
2369 last_objectid = found_key.offset;
2370
5d4f98a2
YZ		 2371 found_key.objectid = found_key.offset;
2372 found_key.type = BTRFS_INODE_ITEM_KEY;
2373 found_key.offset = 0;
73f73415 2374 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2375 ret = PTR_RET(inode);
2376 if (ret && ret != -ESTALE)
66b4ffd1 2377 goto out;
7b128766 2378
f8e9e0b0
AJ		 2379 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2380 struct btrfs_root *dead_root;
2381 struct btrfs_fs_info *fs_info = root->fs_info;
2382 int is_dead_root = 0;
2383
2384 /*
2385 * this is an orphan in the tree root. Currently these
2386 * could come from 2 sources:
2387 * a) a snapshot deletion in progress
2388 * b) a free space cache inode
2389 * We need to distinguish those two, as the snapshot
2390 * orphan must not get deleted.
2391 * find_dead_roots already ran before us, so if this
2392 * is a snapshot deletion, we should find the root
2393 * in the dead_roots list
2394 */
2395 spin_lock(&fs_info->trans_lock);
2396 list_for_each_entry(dead_root, &fs_info->dead_roots,
2397 root_list) {
2398 if (dead_root->root_key.objectid ==
2399 found_key.objectid) {
2400 is_dead_root = 1;
2401 break;
2402 }
2403 }
2404 spin_unlock(&fs_info->trans_lock);
2405 if (is_dead_root) {
2406 /* prevent this orphan from being found again */
2407 key.offset = found_key.objectid - 1;
2408 continue;
2409 }
2410 }
7b128766 2411 /*
a8c9e576
JB		 2412 * Inode is already gone but the orphan item is still there,
2413 * kill the orphan item.
7b128766 2414 */
a8c9e576
JB		 2415 if (ret == -ESTALE) {
2416 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2417 if (IS_ERR(trans)) {
2418 ret = PTR_ERR(trans);
2419 goto out;
2420 }
8a35d95f
JB		 2421 printk(KERN_ERR "auto deleting %Lu\n",
2422 found_key.objectid);
a8c9e576
JB		 2423 ret = btrfs_del_orphan_item(trans, root,
2424 found_key.objectid);
79787eaa 2425 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 2426 btrfs_end_transaction(trans, root);
7b128766
JB		 2427 continue;
2428 }
2429
a8c9e576
JB		 2430 /*
2431 * add this inode to the orphan list so btrfs_orphan_del does
2432 * the proper thing when we hit it
2433 */
8a35d95f
JB		 2434 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2435 &BTRFS_I(inode)->runtime_flags);
a8c9e576 2436
7b128766
JB		 2437 /* if we have links, this was a truncate, lets do that */
2438 if (inode->i_nlink) {
a41ad394
JB		 2439 if (!S_ISREG(inode->i_mode)) {
2440 WARN_ON(1);
2441 iput(inode);
2442 continue;
2443 }
7b128766 2444 nr_truncate++;
66b4ffd1 2445 ret = btrfs_truncate(inode);
7b128766
JB		 2446 } else {
2447 nr_unlink++;
2448 }
2449
2450 /* this will do delete_inode and everything for us */
2451 iput(inode);
66b4ffd1
JB		 2452 if (ret)
2453 goto out;
7b128766 2454 }
3254c876
MX		 2455 /* release the path since we're done with it */
2456 btrfs_release_path(path);
2457
d68fc57b
YZ		 2458 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2459
2460 if (root->orphan_block_rsv)
2461 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2462 (u64)-1);
2463
2464 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2465 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2466 if (!IS_ERR(trans))
2467 btrfs_end_transaction(trans, root);
d68fc57b 2468 }
7b128766
JB		 2469
2470 if (nr_unlink)
2471 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2472 if (nr_truncate)
2473 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2474
2475out:
2476 if (ret)
2477 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2478 btrfs_free_path(path);
2479 return ret;
7b128766
JB		 2480}
2481
46a53cca
CM		 2482/*
2483 * very simple check to peek ahead in the leaf looking for xattrs. If we
2484 * don't find any xattrs, we know there can't be any acls.
2485 *
2486 * slot is the slot the inode is in, objectid is the objectid of the inode
2487 */
2488static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2489 int slot, u64 objectid)
2490{
2491 u32 nritems = btrfs_header_nritems(leaf);
2492 struct btrfs_key found_key;
2493 int scanned = 0;
2494
2495 slot++;
2496 while (slot < nritems) {
2497 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2498
2499 /* we found a different objectid, there must not be acls */
2500 if (found_key.objectid != objectid)
2501 return 0;
2502
2503 /* we found an xattr, assume we've got an acl */
2504 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2505 return 1;
2506
2507 /*
2508 * we found a key greater than an xattr key, there can't
2509 * be any acls later on
2510 */
2511 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2512 return 0;
2513
2514 slot++;
2515 scanned++;
2516
2517 /*
2518 * it goes inode, inode backrefs, xattrs, extents,
2519 * so if there are a ton of hard links to an inode there can
2520 * be a lot of backrefs. Don't waste time searching too hard,
2521 * this is just an optimization
2522 */
2523 if (scanned >= 8)
2524 break;
2525 }
2526 /* we hit the end of the leaf before we found an xattr or
2527 * something larger than an xattr. We have to assume the inode
2528 * has acls
2529 */
2530 return 1;
2531}
2532
d352ac68
CM		 2533/*
2534 * read an inode from the btree into the in-memory inode
2535 */
5d4f98a2 2536static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2537{
2538 struct btrfs_path *path;
5f39d397 2539 struct extent_buffer *leaf;
39279cc3 2540 struct btrfs_inode_item *inode_item;
0b86a832 2541 struct btrfs_timespec *tspec;
39279cc3
CM		 2542 struct btrfs_root *root = BTRFS_I(inode)->root;
2543 struct btrfs_key location;
46a53cca 2544 int maybe_acls;
618e21d5 2545 u32 rdev;
39279cc3 2546 int ret;
2f7e33d4
MX		 2547 bool filled = false;
2548
2549 ret = btrfs_fill_inode(inode, &rdev);
2550 if (!ret)
2551 filled = true;
39279cc3
CM		 2552
2553 path = btrfs_alloc_path();
1748f843
MF		 2554 if (!path)
2555 goto make_bad;
2556
d90c7321 2557 path->leave_spinning = 1;
39279cc3 2558 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2559
39279cc3 2560 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2561 if (ret)
39279cc3 2562 goto make_bad;
39279cc3 2563
5f39d397 2564 leaf = path->nodes[0];
2f7e33d4
MX		 2565
2566 if (filled)
2567 goto cache_acl;
2568
5f39d397
CM		 2569 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2570 struct btrfs_inode_item);
5f39d397 2571 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 2572 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
5f39d397
CM		 2573 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2574 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2575 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2576
2577 tspec = btrfs_inode_atime(inode_item);
2578 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2579 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2580
2581 tspec = btrfs_inode_mtime(inode_item);
2582 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584
2585 tspec = btrfs_inode_ctime(inode_item);
2586 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588
a76a3cd4 2589 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2590 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
0c4d2d95 2591 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2592 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2593 inode->i_rdev = 0;
5f39d397
CM		 2594 rdev = btrfs_inode_rdev(leaf, inode_item);
2595
aec7477b 2596 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2597 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2598cache_acl:
46a53cca
CM		 2599 /*
2600 * try to precache a NULL acl entry for files that don't have
2601 * any xattrs or acls
2602 */
33345d01
LZ		 2603 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2604 btrfs_ino(inode));
72c04902
AV		 2605 if (!maybe_acls)
2606 cache_no_acl(inode);
46a53cca 2607
39279cc3 2608 btrfs_free_path(path);
39279cc3 2609
39279cc3 2610 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2611 case S_IFREG:
2612 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2613 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2614 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2615 inode->i_fop = &btrfs_file_operations;
2616 inode->i_op = &btrfs_file_inode_operations;
2617 break;
2618 case S_IFDIR:
2619 inode->i_fop = &btrfs_dir_file_operations;
2620 if (root == root->fs_info->tree_root)
2621 inode->i_op = &btrfs_dir_ro_inode_operations;
2622 else
2623 inode->i_op = &btrfs_dir_inode_operations;
2624 break;
2625 case S_IFLNK:
2626 inode->i_op = &btrfs_symlink_inode_operations;
2627 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2628 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2629 break;
618e21d5 2630 default:
0279b4cd 2631 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2632 init_special_inode(inode, inode->i_mode, rdev);
2633 break;
39279cc3 2634 }
6cbff00f
CH		 2635
2636 btrfs_update_iflags(inode);
39279cc3
CM		 2637 return;
2638
2639make_bad:
39279cc3 2640 btrfs_free_path(path);
39279cc3
CM		 2641 make_bad_inode(inode);
2642}
2643
d352ac68
CM		 2644/*
2645 * given a leaf and an inode, copy the inode fields into the leaf
2646 */
e02119d5
CM		 2647static void fill_inode_item(struct btrfs_trans_handle *trans,
2648 struct extent_buffer *leaf,
5f39d397 2649 struct btrfs_inode_item *item,
39279cc3
CM		 2650 struct inode *inode)
2651{
5f39d397
CM		 2652 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2653 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2654 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2655 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2656 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2657
2658 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2659 inode->i_atime.tv_sec);
2660 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2661 inode->i_atime.tv_nsec);
2662
2663 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2664 inode->i_mtime.tv_sec);
2665 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2666 inode->i_mtime.tv_nsec);
2667
2668 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2669 inode->i_ctime.tv_sec);
2670 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2671 inode->i_ctime.tv_nsec);
2672
a76a3cd4 2673 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2674 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
0c4d2d95 2675 btrfs_set_inode_sequence(leaf, item, inode->i_version);
e02119d5 2676 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2677 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2678 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2679 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2680}
2681
d352ac68
CM		 2682/*
2683 * copy everything in the in-memory inode into the btree.
2684 */
2115133f 2685static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2686 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2687{
2688 struct btrfs_inode_item *inode_item;
2689 struct btrfs_path *path;
5f39d397 2690 struct extent_buffer *leaf;
39279cc3
CM		 2691 int ret;
2692
2693 path = btrfs_alloc_path();
16cdcec7
MX		 2694 if (!path)
2695 return -ENOMEM;
2696
b9473439 2697 path->leave_spinning = 1;
16cdcec7
MX		 2698 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2699 1);
39279cc3
CM		 2700 if (ret) {
2701 if (ret > 0)
2702 ret = -ENOENT;
2703 goto failed;
2704 }
2705
b4ce94de 2706 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2707 leaf = path->nodes[0];
2708 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2709 struct btrfs_inode_item);
39279cc3 2710
e02119d5 2711 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2712 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2713 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2714 ret = 0;
2715failed:
39279cc3
CM		 2716 btrfs_free_path(path);
2717 return ret;
2718}
2719
2115133f
CM		 2720/*
2721 * copy everything in the in-memory inode into the btree.
2722 */
2723noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root, struct inode *inode)
2725{
2726 int ret;
2727
2728 /*
2729 * If the inode is a free space inode, we can deadlock during commit
2730 * if we put it into the delayed code.
2731 *
2732 * The data relocation inode should also be directly updated
2733 * without delay
2734 */
83eea1f1 2735 if (!btrfs_is_free_space_inode(inode)
2115133f
CM		 2736 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737 ret = btrfs_delayed_update_inode(trans, root, inode);
2738 if (!ret)
2739 btrfs_set_inode_last_trans(trans, inode);
2740 return ret;
2741 }
2742
2743 return btrfs_update_inode_item(trans, root, inode);
2744}
2745
2746static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *root, struct inode *inode)
2748{
2749 int ret;
2750
2751 ret = btrfs_update_inode(trans, root, inode);
2752 if (ret == -ENOSPC)
2753 return btrfs_update_inode_item(trans, root, inode);
2754 return ret;
2755}
2756
d352ac68
CM		 2757/*
2758 * unlink helper that gets used here in inode.c and in the tree logging
2759 * recovery code. It remove a link in a directory with a given name, and
2760 * also drops the back refs in the inode to the directory
2761 */
92986796
AV		 2762static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2763 struct btrfs_root *root,
2764 struct inode *dir, struct inode *inode,
2765 const char *name, int name_len)
39279cc3
CM		 2766{
2767 struct btrfs_path *path;
39279cc3 2768 int ret = 0;
5f39d397 2769 struct extent_buffer *leaf;
39279cc3 2770 struct btrfs_dir_item *di;
5f39d397 2771 struct btrfs_key key;
aec7477b 2772 u64 index;
33345d01
LZ		 2773 u64 ino = btrfs_ino(inode);
2774 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2775
2776 path = btrfs_alloc_path();
54aa1f4d
CM		 2777 if (!path) {
2778 ret = -ENOMEM;
554233a6 2779 goto out;
54aa1f4d
CM		 2780 }
2781
b9473439 2782 path->leave_spinning = 1;
33345d01 2783 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2784 name, name_len, -1);
2785 if (IS_ERR(di)) {
2786 ret = PTR_ERR(di);
2787 goto err;
2788 }
2789 if (!di) {
2790 ret = -ENOENT;
2791 goto err;
2792 }
5f39d397
CM		 2793 leaf = path->nodes[0];
2794 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2795 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2796 if (ret)
2797 goto err;
b3b4aa74 2798 btrfs_release_path(path);
39279cc3 2799
33345d01
LZ		 2800 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2801 dir_ino, &index);
aec7477b 2802 if (ret) {
d397712b 2803 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2804 "inode %llu parent %llu\n", name_len, name,
2805 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 2806 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 2807 goto err;
2808 }
2809
16cdcec7 2810 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 2811 if (ret) {
2812 btrfs_abort_transaction(trans, root, ret);
39279cc3 2813 goto err;
79787eaa 2814 }
39279cc3 2815
e02119d5 2816 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2817 inode, dir_ino);
79787eaa
JM		 2818 if (ret != 0 && ret != -ENOENT) {
2819 btrfs_abort_transaction(trans, root, ret);
2820 goto err;
2821 }
e02119d5
CM		 2822
2823 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2824 dir, index);
6418c961
CM		 2825 if (ret == -ENOENT)
2826 ret = 0;
39279cc3
CM		 2827err:
2828 btrfs_free_path(path);
e02119d5
CM		 2829 if (ret)
2830 goto out;
2831
2832 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 2833 inode_inc_iversion(inode);
2834 inode_inc_iversion(dir);
e02119d5 2835 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 2836 ret = btrfs_update_inode(trans, root, dir);
e02119d5 2837out:
39279cc3
CM		 2838 return ret;
2839}
2840
92986796
AV		 2841int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2842 struct btrfs_root *root,
2843 struct inode *dir, struct inode *inode,
2844 const char *name, int name_len)
2845{
2846 int ret;
2847 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2848 if (!ret) {
2849 btrfs_drop_nlink(inode);
2850 ret = btrfs_update_inode(trans, root, inode);
2851 }
2852 return ret;
2853}
2854
2855
a22285a6
YZ		 2856/* helper to check if there is any shared block in the path */
2857static int check_path_shared(struct btrfs_root *root,
2858 struct btrfs_path *path)
39279cc3 2859{
a22285a6
YZ		 2860 struct extent_buffer *eb;
2861 int level;
0e4dcbef 2862 u64 refs = 1;
5df6a9f6 2863
a22285a6 2864 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2865 int ret;
2866
a22285a6
YZ		 2867 if (!path->nodes[level])
2868 break;
2869 eb = path->nodes[level];
2870 if (!btrfs_block_can_be_shared(root, eb))
2871 continue;
2872 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2873 &refs, NULL);
2874 if (refs > 1)
2875 return 1;
5df6a9f6 2876 }
dedefd72 2877 return 0;
39279cc3
CM		 2878}
2879
a22285a6
YZ		 2880/*
2881 * helper to start transaction for unlink and rmdir.
2882 *
2883 * unlink and rmdir are special in btrfs, they do not always free space.
2884 * so in enospc case, we should make sure they will free space before
2885 * allowing them to use the global metadata reservation.
2886 */
2887static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2888 struct dentry *dentry)
4df27c4d 2889{
39279cc3 2890 struct btrfs_trans_handle *trans;
a22285a6 2891 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2892 struct btrfs_path *path;
a22285a6 2893 struct btrfs_inode_ref *ref;
4df27c4d 2894 struct btrfs_dir_item *di;
7b128766 2895 struct inode *inode = dentry->d_inode;
4df27c4d 2896 u64 index;
a22285a6
YZ		 2897 int check_link = 1;
2898 int err = -ENOSPC;
4df27c4d 2899 int ret;
33345d01
LZ		 2900 u64 ino = btrfs_ino(inode);
2901 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2902
e70bea5f
JB		 2903 /*
2904 * 1 for the possible orphan item
2905 * 1 for the dir item
2906 * 1 for the dir index
2907 * 1 for the inode ref
2908 * 1 for the inode ref in the tree log
2909 * 2 for the dir entries in the log
2910 * 1 for the inode
2911 */
2912 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2913 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2914 return trans;
4df27c4d 2915
33345d01 2916 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2917 return ERR_PTR(-ENOSPC);
4df27c4d 2918
a22285a6
YZ		 2919 /* check if there is someone else holds reference */
2920 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2921 return ERR_PTR(-ENOSPC);
4df27c4d 2922
a22285a6
YZ		 2923 if (atomic_read(&inode->i_count) > 2)
2924 return ERR_PTR(-ENOSPC);
4df27c4d 2925
a22285a6
YZ		 2926 if (xchg(&root->fs_info->enospc_unlink, 1))
2927 return ERR_PTR(-ENOSPC);
2928
2929 path = btrfs_alloc_path();
2930 if (!path) {
2931 root->fs_info->enospc_unlink = 0;
2932 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2933 }
2934
3880a1b4
JB		 2935 /* 1 for the orphan item */
2936 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2937 if (IS_ERR(trans)) {
a22285a6
YZ		 2938 btrfs_free_path(path);
2939 root->fs_info->enospc_unlink = 0;
2940 return trans;
2941 }
4df27c4d 2942
a22285a6
YZ		 2943 path->skip_locking = 1;
2944 path->search_commit_root = 1;
4df27c4d 2945
a22285a6
YZ		 2946 ret = btrfs_lookup_inode(trans, root, path,
2947 &BTRFS_I(dir)->location, 0);
2948 if (ret < 0) {
2949 err = ret;
2950 goto out;
2951 }
2952 if (ret == 0) {
2953 if (check_path_shared(root, path))
2954 goto out;
2955 } else {
2956 check_link = 0;
5df6a9f6 2957 }
b3b4aa74 2958 btrfs_release_path(path);
a22285a6
YZ		 2959
2960 ret = btrfs_lookup_inode(trans, root, path,
2961 &BTRFS_I(inode)->location, 0);
2962 if (ret < 0) {
2963 err = ret;
2964 goto out;
2965 }
2966 if (ret == 0) {
2967 if (check_path_shared(root, path))
2968 goto out;
2969 } else {
2970 check_link = 0;
2971 }
b3b4aa74 2972 btrfs_release_path(path);
a22285a6
YZ		 2973
2974 if (ret == 0 && S_ISREG(inode->i_mode)) {
2975 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2976 ino, (u64)-1, 0);
a22285a6
YZ		 2977 if (ret < 0) {
2978 err = ret;
2979 goto out;
2980 }
79787eaa 2981 BUG_ON(ret == 0); /* Corruption */
a22285a6
YZ		 2982 if (check_path_shared(root, path))
2983 goto out;
b3b4aa74 2984 btrfs_release_path(path);
a22285a6
YZ		 2985 }
2986
2987 if (!check_link) {
2988 err = 0;
2989 goto out;
2990 }
2991
33345d01 2992 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2993 dentry->d_name.name, dentry->d_name.len, 0);
2994 if (IS_ERR(di)) {
2995 err = PTR_ERR(di);
2996 goto out;
2997 }
2998 if (di) {
2999 if (check_path_shared(root, path))
3000 goto out;
3001 } else {
3002 err = 0;
3003 goto out;
3004 }
b3b4aa74 3005 btrfs_release_path(path);
a22285a6
YZ		 3006
3007 ref = btrfs_lookup_inode_ref(trans, root, path,
3008 dentry->d_name.name, dentry->d_name.len,
33345d01 3009 ino, dir_ino, 0);
a22285a6
YZ		 3010 if (IS_ERR(ref)) {
3011 err = PTR_ERR(ref);
3012 goto out;
3013 }
79787eaa 3014 BUG_ON(!ref); /* Logic error */
a22285a6
YZ		 3015 if (check_path_shared(root, path))
3016 goto out;
3017 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 3018 btrfs_release_path(path);
a22285a6 3019
16cdcec7
MX		 3020 /*
3021 * This is a commit root search, if we can lookup inode item and other
3022 * relative items in the commit root, it means the transaction of
3023 * dir/file creation has been committed, and the dir index item that we
3024 * delay to insert has also been inserted into the commit root. So
3025 * we needn't worry about the delayed insertion of the dir index item
3026 * here.
3027 */
33345d01 3028 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 3029 dentry->d_name.name, dentry->d_name.len, 0);
3030 if (IS_ERR(di)) {
3031 err = PTR_ERR(di);
3032 goto out;
3033 }
3034 BUG_ON(ret == -ENOENT);
3035 if (check_path_shared(root, path))
3036 goto out;
3037
3038 err = 0;
3039out:
3040 btrfs_free_path(path);
3880a1b4
JB		 3041 /* Migrate the orphan reservation over */
3042 if (!err)
3043 err = btrfs_block_rsv_migrate(trans->block_rsv,
3044 &root->fs_info->global_block_rsv,
5a77d76c 3045 trans->bytes_reserved);
3880a1b4 3046
a22285a6
YZ		 3047 if (err) {
3048 btrfs_end_transaction(trans, root);
3049 root->fs_info->enospc_unlink = 0;
3050 return ERR_PTR(err);
3051 }
3052
3053 trans->block_rsv = &root->fs_info->global_block_rsv;
3054 return trans;
3055}
3056
3057static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3058 struct btrfs_root *root)
3059{
3060 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
5a77d76c
JB		 3061 btrfs_block_rsv_release(root, trans->block_rsv,
3062 trans->bytes_reserved);
3063 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 3064 BUG_ON(!root->fs_info->enospc_unlink);
3065 root->fs_info->enospc_unlink = 0;
3066 }
7ad85bb7 3067 btrfs_end_transaction(trans, root);
a22285a6
YZ		 3068}
3069
3070static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3071{
3072 struct btrfs_root *root = BTRFS_I(dir)->root;
3073 struct btrfs_trans_handle *trans;
3074 struct inode *inode = dentry->d_inode;
3075 int ret;
3076 unsigned long nr = 0;
3077
3078 trans = __unlink_start_trans(dir, dentry);
3079 if (IS_ERR(trans))
3080 return PTR_ERR(trans);
5f39d397 3081
12fcfd22
CM		 3082 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3083
e02119d5
CM		 3084 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3086 if (ret)
3087 goto out;
7b128766 3088
a22285a6 3089 if (inode->i_nlink == 0) {
7b128766 3090 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3091 if (ret)
3092 goto out;
a22285a6 3093 }
7b128766 3094
b532402e 3095out:
d3c2fdcf 3096 nr = trans->blocks_used;
a22285a6 3097 __unlink_end_trans(trans, root);
d3c2fdcf 3098 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3099 return ret;
3100}
3101
4df27c4d
YZ		 3102int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3103 struct btrfs_root *root,
3104 struct inode *dir, u64 objectid,
3105 const char *name, int name_len)
3106{
3107 struct btrfs_path *path;
3108 struct extent_buffer *leaf;
3109 struct btrfs_dir_item *di;
3110 struct btrfs_key key;
3111 u64 index;
3112 int ret;
33345d01 3113 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3114
3115 path = btrfs_alloc_path();
3116 if (!path)
3117 return -ENOMEM;
3118
33345d01 3119 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3120 name, name_len, -1);
79787eaa
JM		 3121 if (IS_ERR_OR_NULL(di)) {
3122 if (!di)
3123 ret = -ENOENT;
3124 else
3125 ret = PTR_ERR(di);
3126 goto out;
3127 }
4df27c4d
YZ		 3128
3129 leaf = path->nodes[0];
3130 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3131 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3132 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3133 if (ret) {
3134 btrfs_abort_transaction(trans, root, ret);
3135 goto out;
3136 }
b3b4aa74 3137 btrfs_release_path(path);
4df27c4d
YZ		 3138
3139 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3140 objectid, root->root_key.objectid,
33345d01 3141 dir_ino, &index, name, name_len);
4df27c4d 3142 if (ret < 0) {
79787eaa
JM		 3143 if (ret != -ENOENT) {
3144 btrfs_abort_transaction(trans, root, ret);
3145 goto out;
3146 }
33345d01 3147 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3148 name, name_len);
79787eaa
JM		 3149 if (IS_ERR_OR_NULL(di)) {
3150 if (!di)
3151 ret = -ENOENT;
3152 else
3153 ret = PTR_ERR(di);
3154 btrfs_abort_transaction(trans, root, ret);
3155 goto out;
3156 }
4df27c4d
YZ		 3157
3158 leaf = path->nodes[0];
3159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3160 btrfs_release_path(path);
4df27c4d
YZ		 3161 index = key.offset;
3162 }
945d8962 3163 btrfs_release_path(path);
4df27c4d 3164
16cdcec7 3165 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3166 if (ret) {
3167 btrfs_abort_transaction(trans, root, ret);
3168 goto out;
3169 }
4df27c4d
YZ		 3170
3171 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3172 inode_inc_iversion(dir);
4df27c4d
YZ		 3173 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3174 ret = btrfs_update_inode(trans, root, dir);
79787eaa
JM		 3175 if (ret)
3176 btrfs_abort_transaction(trans, root, ret);
3177out:
71d7aed0 3178 btrfs_free_path(path);
79787eaa 3179 return ret;
4df27c4d
YZ		 3180}
3181
39279cc3
CM		 3182static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3183{
3184 struct inode *inode = dentry->d_inode;
1832a6d5 3185 int err = 0;
39279cc3 3186 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3187 struct btrfs_trans_handle *trans;
1832a6d5 3188 unsigned long nr = 0;
39279cc3 3189
3394e160 3190 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 3191 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 3192 return -ENOTEMPTY;
3193
a22285a6
YZ		 3194 trans = __unlink_start_trans(dir, dentry);
3195 if (IS_ERR(trans))
5df6a9f6 3196 return PTR_ERR(trans);
5df6a9f6 3197
33345d01 3198 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3199 err = btrfs_unlink_subvol(trans, root, dir,
3200 BTRFS_I(inode)->location.objectid,
3201 dentry->d_name.name,
3202 dentry->d_name.len);
3203 goto out;
3204 }
3205
7b128766
JB		 3206 err = btrfs_orphan_add(trans, inode);
3207 if (err)
4df27c4d 3208 goto out;
7b128766 3209
39279cc3 3210 /* now the directory is empty */
e02119d5
CM		 3211 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3212 dentry->d_name.name, dentry->d_name.len);
d397712b 3213 if (!err)
dbe674a9 3214 btrfs_i_size_write(inode, 0);
4df27c4d 3215out:
d3c2fdcf 3216 nr = trans->blocks_used;
a22285a6 3217 __unlink_end_trans(trans, root);
d3c2fdcf 3218 btrfs_btree_balance_dirty(root, nr);
3954401f 3219
39279cc3
CM		 3220 return err;
3221}
3222
39279cc3
CM		 3223/*
3224 * this can truncate away extent items, csum items and directory items.
3225 * It starts at a high offset and removes keys until it can't find
d352ac68 3226 * any higher than new_size
39279cc3
CM		 3227 *
3228 * csum items that cross the new i_size are truncated to the new size
3229 * as well.
7b128766
JB		 3230 *
3231 * min_type is the minimum key type to truncate down to. If set to 0, this
3232 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3233 */
8082510e
YZ		 3234int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3235 struct btrfs_root *root,
3236 struct inode *inode,
3237 u64 new_size, u32 min_type)
39279cc3 3238{
39279cc3 3239 struct btrfs_path *path;
5f39d397 3240 struct extent_buffer *leaf;
39279cc3 3241 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3242 struct btrfs_key key;
3243 struct btrfs_key found_key;
39279cc3 3244 u64 extent_start = 0;
db94535d 3245 u64 extent_num_bytes = 0;
5d4f98a2 3246 u64 extent_offset = 0;
39279cc3 3247 u64 item_end = 0;
8082510e
YZ		 3248 u64 mask = root->sectorsize - 1;
3249 u32 found_type = (u8)-1;
39279cc3
CM		 3250 int found_extent;
3251 int del_item;
85e21bac
CM		 3252 int pending_del_nr = 0;
3253 int pending_del_slot = 0;
179e29e4 3254 int extent_type = -1;
8082510e
YZ		 3255 int ret;
3256 int err = 0;
33345d01 3257 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3258
3259 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3260
0eb0e19c
MF		 3261 path = btrfs_alloc_path();
3262 if (!path)
3263 return -ENOMEM;
3264 path->reada = -1;
3265
0af3d00b 3266 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 3267 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 3268
16cdcec7
MX		 3269 /*
3270 * This function is also used to drop the items in the log tree before
3271 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3272 * it is used to drop the loged items. So we shouldn't kill the delayed
3273 * items.
3274 */
3275 if (min_type == 0 && root == BTRFS_I(inode)->root)
3276 btrfs_kill_delayed_inode_items(inode);
3277
33345d01 3278 key.objectid = ino;
39279cc3 3279 key.offset = (u64)-1;
5f39d397
CM		 3280 key.type = (u8)-1;
3281
85e21bac 3282search_again:
b9473439 3283 path->leave_spinning = 1;
85e21bac 3284 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3285 if (ret < 0) {
3286 err = ret;
3287 goto out;
3288 }
d397712b 3289
85e21bac 3290 if (ret > 0) {
e02119d5
CM		 3291 /* there are no items in the tree for us to truncate, we're
3292 * done
3293 */
8082510e
YZ		 3294 if (path->slots[0] == 0)
3295 goto out;
85e21bac
CM		 3296 path->slots[0]--;
3297 }
3298
d397712b 3299 while (1) {
39279cc3 3300 fi = NULL;
5f39d397
CM		 3301 leaf = path->nodes[0];
3302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3303 found_type = btrfs_key_type(&found_key);
39279cc3 3304
33345d01 3305 if (found_key.objectid != ino)
39279cc3 3306 break;
5f39d397 3307
85e21bac 3308 if (found_type < min_type)
39279cc3
CM		 3309 break;
3310
5f39d397 3311 item_end = found_key.offset;
39279cc3 3312 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3313 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3314 struct btrfs_file_extent_item);
179e29e4
CM		 3315 extent_type = btrfs_file_extent_type(leaf, fi);
3316 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3317 item_end +=
db94535d 3318 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3319 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3320 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3321 fi);
39279cc3 3322 }
008630c1 3323 item_end--;
39279cc3 3324 }
8082510e
YZ		 3325 if (found_type > min_type) {
3326 del_item = 1;
3327 } else {
3328 if (item_end < new_size)
b888db2b 3329 break;
8082510e
YZ		 3330 if (found_key.offset >= new_size)
3331 del_item = 1;
3332 else
3333 del_item = 0;
39279cc3 3334 }
39279cc3 3335 found_extent = 0;
39279cc3 3336 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3337 if (found_type != BTRFS_EXTENT_DATA_KEY)
3338 goto delete;
3339
3340 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3341 u64 num_dec;
db94535d 3342 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3343 if (!del_item) {
db94535d
CM		 3344 u64 orig_num_bytes =
3345 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3346 extent_num_bytes = new_size -
5f39d397 3347 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3348 extent_num_bytes = extent_num_bytes &
3349 ~((u64)root->sectorsize - 1);
db94535d
CM		 3350 btrfs_set_file_extent_num_bytes(leaf, fi,
3351 extent_num_bytes);
3352 num_dec = (orig_num_bytes -
9069218d 3353 extent_num_bytes);
e02119d5 3354 if (root->ref_cows && extent_start != 0)
a76a3cd4 3355 inode_sub_bytes(inode, num_dec);
5f39d397 3356 btrfs_mark_buffer_dirty(leaf);
39279cc3 3357 } else {
db94535d
CM		 3358 extent_num_bytes =
3359 btrfs_file_extent_disk_num_bytes(leaf,
3360 fi);
5d4f98a2
YZ		 3361 extent_offset = found_key.offset -
3362 btrfs_file_extent_offset(leaf, fi);
3363
39279cc3 3364 /* FIXME blocksize != 4096 */
9069218d 3365 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3366 if (extent_start != 0) {
3367 found_extent = 1;
e02119d5 3368 if (root->ref_cows)
a76a3cd4 3369 inode_sub_bytes(inode, num_dec);
e02119d5 3370 }
39279cc3 3371 }
9069218d 3372 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3373 /*
3374 * we can't truncate inline items that have had
3375 * special encodings
3376 */
3377 if (!del_item &&
3378 btrfs_file_extent_compression(leaf, fi) == 0 &&
3379 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3380 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3381 u32 size = new_size - found_key.offset;
3382
3383 if (root->ref_cows) {
a76a3cd4
YZ		 3384 inode_sub_bytes(inode, item_end + 1 -
3385 new_size);
e02119d5
CM		 3386 }
3387 size =
3388 btrfs_file_extent_calc_inline_size(size);
143bede5
JM		 3389 btrfs_truncate_item(trans, root, path,
3390 size, 1);
e02119d5 3391 } else if (root->ref_cows) {
a76a3cd4
YZ		 3392 inode_sub_bytes(inode, item_end + 1 -
3393 found_key.offset);
9069218d 3394 }
39279cc3 3395 }
179e29e4 3396delete:
39279cc3 3397 if (del_item) {
85e21bac
CM		 3398 if (!pending_del_nr) {
3399 /* no pending yet, add ourselves */
3400 pending_del_slot = path->slots[0];
3401 pending_del_nr = 1;
3402 } else if (pending_del_nr &&
3403 path->slots[0] + 1 == pending_del_slot) {
3404 /* hop on the pending chunk */
3405 pending_del_nr++;
3406 pending_del_slot = path->slots[0];
3407 } else {
d397712b 3408 BUG();
85e21bac 3409 }
39279cc3
CM		 3410 } else {
3411 break;
3412 }
0af3d00b
JB		 3413 if (found_extent && (root->ref_cows ||
3414 root == root->fs_info->tree_root)) {
b9473439 3415 btrfs_set_path_blocking(path);
39279cc3 3416 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3417 extent_num_bytes, 0,
3418 btrfs_header_owner(leaf),
66d7e7f0 3419 ino, extent_offset, 0);
39279cc3
CM		 3420 BUG_ON(ret);
3421 }
85e21bac 3422
8082510e
YZ		 3423 if (found_type == BTRFS_INODE_ITEM_KEY)
3424 break;
3425
3426 if (path->slots[0] == 0 ||
3427 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3428 if (root->ref_cows &&
3429 BTRFS_I(inode)->location.objectid !=
3430 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3431 err = -EAGAIN;
3432 goto out;
3433 }
3434 if (pending_del_nr) {
3435 ret = btrfs_del_items(trans, root, path,
3436 pending_del_slot,
3437 pending_del_nr);
79787eaa
JM		 3438 if (ret) {
3439 btrfs_abort_transaction(trans,
3440 root, ret);
3441 goto error;
3442 }
8082510e
YZ		 3443 pending_del_nr = 0;
3444 }
b3b4aa74 3445 btrfs_release_path(path);
85e21bac 3446 goto search_again;
8082510e
YZ		 3447 } else {
3448 path->slots[0]--;
85e21bac 3449 }
39279cc3 3450 }
8082510e 3451out:
85e21bac
CM		 3452 if (pending_del_nr) {
3453 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3454 pending_del_nr);
79787eaa
JM		 3455 if (ret)
3456 btrfs_abort_transaction(trans, root, ret);
85e21bac 3457 }
79787eaa 3458error:
39279cc3 3459 btrfs_free_path(path);
8082510e 3460 return err;
39279cc3
CM		 3461}
3462
3463/*
3464 * taken from block_truncate_page, but does cow as it zeros out
3465 * any bytes left in the last page in the file.
3466 */
3467static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3468{
3469 struct inode *inode = mapping->host;
db94535d 3470 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3471 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3472 struct btrfs_ordered_extent *ordered;
2ac55d41 3473 struct extent_state *cached_state = NULL;
e6dcd2dc 3474 char *kaddr;
db94535d 3475 u32 blocksize = root->sectorsize;
39279cc3
CM		 3476 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3477 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3478 struct page *page;
3b16a4e3 3479 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3480 int ret = 0;
a52d9a80 3481 u64 page_start;
e6dcd2dc 3482 u64 page_end;
39279cc3
CM		 3483
3484 if ((offset & (blocksize - 1)) == 0)
3485 goto out;
0ca1f7ce 3486 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3487 if (ret)
3488 goto out;
39279cc3
CM		 3489
3490 ret = -ENOMEM;
211c17f5 3491again:
3b16a4e3 3492 page = find_or_create_page(mapping, index, mask);
5d5e103a 3493 if (!page) {
0ca1f7ce 3494 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3495 goto out;
5d5e103a 3496 }
e6dcd2dc
CM		 3497
3498 page_start = page_offset(page);
3499 page_end = page_start + PAGE_CACHE_SIZE - 1;
3500
39279cc3 3501 if (!PageUptodate(page)) {
9ebefb18 3502 ret = btrfs_readpage(NULL, page);
39279cc3 3503 lock_page(page);
211c17f5
CM		 3504 if (page->mapping != mapping) {
3505 unlock_page(page);
3506 page_cache_release(page);
3507 goto again;
3508 }
39279cc3
CM		 3509 if (!PageUptodate(page)) {
3510 ret = -EIO;
89642229 3511 goto out_unlock;
39279cc3
CM		 3512 }
3513 }
211c17f5 3514 wait_on_page_writeback(page);
e6dcd2dc 3515
d0082371 3516 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 3517 set_page_extent_mapped(page);
3518
3519 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3520 if (ordered) {
2ac55d41
JB		 3521 unlock_extent_cached(io_tree, page_start, page_end,
3522 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3523 unlock_page(page);
3524 page_cache_release(page);
eb84ae03 3525 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3526 btrfs_put_ordered_extent(ordered);
3527 goto again;
3528 }
3529
2ac55d41 3530 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3531 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3532 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3533
2ac55d41
JB		 3534 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3535 &cached_state);
9ed74f2d 3536 if (ret) {
2ac55d41
JB		 3537 unlock_extent_cached(io_tree, page_start, page_end,
3538 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3539 goto out_unlock;
3540 }
3541
e6dcd2dc
CM		 3542 ret = 0;
3543 if (offset != PAGE_CACHE_SIZE) {
3544 kaddr = kmap(page);
3545 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3546 flush_dcache_page(page);
3547 kunmap(page);
3548 }
247e743c 3549 ClearPageChecked(page);
e6dcd2dc 3550 set_page_dirty(page);
2ac55d41
JB		 3551 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3552 GFP_NOFS);
39279cc3 3553
89642229 3554out_unlock:
5d5e103a 3555 if (ret)
0ca1f7ce 3556 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3557 unlock_page(page);
3558 page_cache_release(page);
3559out:
3560 return ret;
3561}
3562
695a0d0d
JB		 3563/*
3564 * This function puts in dummy file extents for the area we're creating a hole
3565 * for. So if we are truncating this file to a larger size we need to insert
3566 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3567 * the range between oldsize and size
3568 */
a41ad394 3569int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3570{
9036c102
YZ		 3571 struct btrfs_trans_handle *trans;
3572 struct btrfs_root *root = BTRFS_I(inode)->root;
3573 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3574 struct extent_map *em = NULL;
2ac55d41 3575 struct extent_state *cached_state = NULL;
9036c102 3576 u64 mask = root->sectorsize - 1;
a41ad394 3577 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3578 u64 block_end = (size + mask) & ~mask;
3579 u64 last_byte;
3580 u64 cur_offset;
3581 u64 hole_size;
9ed74f2d 3582 int err = 0;
39279cc3 3583
9036c102
YZ		 3584 if (size <= hole_start)
3585 return 0;
3586
9036c102
YZ		 3587 while (1) {
3588 struct btrfs_ordered_extent *ordered;
3589 btrfs_wait_ordered_range(inode, hole_start,
3590 block_end - hole_start);
2ac55d41 3591 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 3592 &cached_state);
9036c102
YZ		 3593 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3594 if (!ordered)
3595 break;
2ac55d41
JB		 3596 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3597 &cached_state, GFP_NOFS);
9036c102
YZ		 3598 btrfs_put_ordered_extent(ordered);
3599 }
39279cc3 3600
9036c102
YZ		 3601 cur_offset = hole_start;
3602 while (1) {
3603 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3604 block_end - cur_offset, 0);
79787eaa
JM		 3605 if (IS_ERR(em)) {
3606 err = PTR_ERR(em);
3607 break;
3608 }
9036c102
YZ		 3609 last_byte = min(extent_map_end(em), block_end);
3610 last_byte = (last_byte + mask) & ~mask;
8082510e 3611 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3612 u64 hint_byte = 0;
9036c102 3613 hole_size = last_byte - cur_offset;
9ed74f2d 3614
3642320e 3615 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 3616 if (IS_ERR(trans)) {
3617 err = PTR_ERR(trans);
9ed74f2d 3618 break;
a22285a6 3619 }
8082510e
YZ		 3620
3621 err = btrfs_drop_extents(trans, inode, cur_offset,
3622 cur_offset + hole_size,
3623 &hint_byte, 1);
5b397377 3624 if (err) {
79787eaa 3625 btrfs_abort_transaction(trans, root, err);
5b397377 3626 btrfs_end_transaction(trans, root);
3893e33b 3627 break;
5b397377 3628 }
8082510e 3629
9036c102 3630 err = btrfs_insert_file_extent(trans, root,
33345d01 3631 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3632 0, hole_size, 0, hole_size,
3633 0, 0, 0);
5b397377 3634 if (err) {
79787eaa 3635 btrfs_abort_transaction(trans, root, err);
5b397377 3636 btrfs_end_transaction(trans, root);
3893e33b 3637 break;
5b397377 3638 }
8082510e 3639
9036c102
YZ		 3640 btrfs_drop_extent_cache(inode, hole_start,
3641 last_byte - 1, 0);
8082510e 3642
3642320e 3643 btrfs_update_inode(trans, root, inode);
8082510e 3644 btrfs_end_transaction(trans, root);
9036c102
YZ		 3645 }
3646 free_extent_map(em);
a22285a6 3647 em = NULL;
9036c102 3648 cur_offset = last_byte;
8082510e 3649 if (cur_offset >= block_end)
9036c102
YZ		 3650 break;
3651 }
1832a6d5 3652
a22285a6 3653 free_extent_map(em);
2ac55d41
JB		 3654 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3655 GFP_NOFS);
9036c102
YZ		 3656 return err;
3657}
39279cc3 3658
a41ad394 3659static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3660{
f4a2f4c5
MX		 3661 struct btrfs_root *root = BTRFS_I(inode)->root;
3662 struct btrfs_trans_handle *trans;
a41ad394 3663 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3664 int ret;
3665
a41ad394 3666 if (newsize == oldsize)
8082510e
YZ		 3667 return 0;
3668
a41ad394 3669 if (newsize > oldsize) {
a41ad394
JB		 3670 truncate_pagecache(inode, oldsize, newsize);
3671 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 3672 if (ret)
8082510e 3673 return ret;
8082510e 3674
f4a2f4c5
MX		 3675 trans = btrfs_start_transaction(root, 1);
3676 if (IS_ERR(trans))
3677 return PTR_ERR(trans);
3678
3679 i_size_write(inode, newsize);
3680 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3681 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 3682 btrfs_end_transaction(trans, root);
a41ad394 3683 } else {
8082510e 3684
a41ad394
JB		 3685 /*
3686 * We're truncating a file that used to have good data down to
3687 * zero. Make sure it gets into the ordered flush list so that
3688 * any new writes get down to disk quickly.
3689 */
3690 if (newsize == 0)
72ac3c0d
JB		 3691 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3692 &BTRFS_I(inode)->runtime_flags);
8082510e 3693
a41ad394
JB		 3694 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3695 truncate_setsize(inode, newsize);
3696 ret = btrfs_truncate(inode);
8082510e
YZ		 3697 }
3698
a41ad394 3699 return ret;
8082510e
YZ		 3700}
3701
9036c102
YZ		 3702static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3703{
3704 struct inode *inode = dentry->d_inode;
b83cc969 3705 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3706 int err;
39279cc3 3707
b83cc969
LZ		 3708 if (btrfs_root_readonly(root))
3709 return -EROFS;
3710
9036c102
YZ		 3711 err = inode_change_ok(inode, attr);
3712 if (err)
3713 return err;
2bf5a725 3714
5a3f23d5 3715 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3716 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3717 if (err)
3718 return err;
39279cc3 3719 }
9036c102 3720
1025774c
CH		 3721 if (attr->ia_valid) {
3722 setattr_copy(inode, attr);
0c4d2d95 3723 inode_inc_iversion(inode);
22c44fe6 3724 err = btrfs_dirty_inode(inode);
1025774c 3725
22c44fe6 3726 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 3727 err = btrfs_acl_chmod(inode);
3728 }
33268eaf 3729
39279cc3
CM		 3730 return err;
3731}
61295eb8 3732
bd555975 3733void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3734{
3735 struct btrfs_trans_handle *trans;
3736 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3737 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3738 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3739 unsigned long nr;
39279cc3
CM		 3740 int ret;
3741
1abe9b8a 3742 trace_btrfs_inode_evict(inode);
3743
39279cc3 3744 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3745 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 3746 btrfs_is_free_space_inode(inode)))
bd555975
AV		 3747 goto no_delete;
3748
39279cc3 3749 if (is_bad_inode(inode)) {
7b128766 3750 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3751 goto no_delete;
3752 }
bd555975 3753 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3754 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3755
c71bf099 3756 if (root->fs_info->log_root_recovering) {
6bf02314 3757 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 3758 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 3759 goto no_delete;
3760 }
3761
76dda93c
YZ		 3762 if (inode->i_nlink > 0) {
3763 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3764 goto no_delete;
3765 }
3766
4289a667
JB		 3767 rsv = btrfs_alloc_block_rsv(root);
3768 if (!rsv) {
3769 btrfs_orphan_del(NULL, inode);
3770 goto no_delete;
3771 }
4a338542 3772 rsv->size = min_size;
726c35fa 3773 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3774
dbe674a9 3775 btrfs_i_size_write(inode, 0);
5f39d397 3776
4289a667
JB		 3777 /*
3778 * This is a bit simpler than btrfs_truncate since
3779 *
3780 * 1) We've already reserved our space for our orphan item in the
3781 * unlink.
3782 * 2) We're going to delete the inode item, so we don't need to update
3783 * it at all.
3784 *
3785 * So we just need to reserve some slack space in case we add bytes when
3786 * doing the truncate.
3787 */
8082510e 3788 while (1) {
aa38a711 3789 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
726c35fa
JB		 3790
3791 /*
3792 * Try and steal from the global reserve since we will
3793 * likely not use this space anyway, we want to try as
3794 * hard as possible to get this to work.
3795 */
3796 if (ret)
3797 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 3798
d68fc57b 3799 if (ret) {
4289a667 3800 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3801 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3802 btrfs_orphan_del(NULL, inode);
3803 btrfs_free_block_rsv(root, rsv);
3804 goto no_delete;
d68fc57b 3805 }
7b128766 3806
4289a667
JB		 3807 trans = btrfs_start_transaction(root, 0);
3808 if (IS_ERR(trans)) {
3809 btrfs_orphan_del(NULL, inode);
3810 btrfs_free_block_rsv(root, rsv);
3811 goto no_delete;
d68fc57b 3812 }
7b128766 3813
4289a667
JB		 3814 trans->block_rsv = rsv;
3815
d68fc57b 3816 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3817 if (ret != -EAGAIN)
3818 break;
85e21bac 3819
8082510e
YZ		 3820 nr = trans->blocks_used;
3821 btrfs_end_transaction(trans, root);
3822 trans = NULL;
3823 btrfs_btree_balance_dirty(root, nr);
3824 }
5f39d397 3825
4289a667
JB		 3826 btrfs_free_block_rsv(root, rsv);
3827
8082510e 3828 if (ret == 0) {
4289a667 3829 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3830 ret = btrfs_orphan_del(trans, inode);
3831 BUG_ON(ret);
3832 }
54aa1f4d 3833
4289a667 3834 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3835 if (!(root == root->fs_info->tree_root ||
3836 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3837 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3838
d3c2fdcf 3839 nr = trans->blocks_used;
54aa1f4d 3840 btrfs_end_transaction(trans, root);
d3c2fdcf 3841 btrfs_btree_balance_dirty(root, nr);
39279cc3 3842no_delete:
dbd5768f 3843 clear_inode(inode);
8082510e 3844 return;
39279cc3
CM		 3845}
3846
3847/*
3848 * this returns the key found in the dir entry in the location pointer.
3849 * If no dir entries were found, location->objectid is 0.
3850 */
3851static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3852 struct btrfs_key *location)
3853{
3854 const char *name = dentry->d_name.name;
3855 int namelen = dentry->d_name.len;
3856 struct btrfs_dir_item *di;
3857 struct btrfs_path *path;
3858 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3859 int ret = 0;
39279cc3
CM		 3860
3861 path = btrfs_alloc_path();
d8926bb3
MF		 3862 if (!path)
3863 return -ENOMEM;
3954401f 3864
33345d01 3865 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3866 namelen, 0);
0d9f7f3e
Y		 3867 if (IS_ERR(di))
3868 ret = PTR_ERR(di);
d397712b 3869
c704005d 3870 if (IS_ERR_OR_NULL(di))
3954401f 3871 goto out_err;
d397712b 3872
5f39d397 3873 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3874out:
39279cc3
CM		 3875 btrfs_free_path(path);
3876 return ret;
3954401f
CM		 3877out_err:
3878 location->objectid = 0;
3879 goto out;
39279cc3
CM		 3880}
3881
3882/*
3883 * when we hit a tree root in a directory, the btrfs part of the inode
3884 * needs to be changed to reflect the root directory of the tree root. This
3885 * is kind of like crossing a mount point.
3886 */
3887static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3888 struct inode *dir,
3889 struct dentry *dentry,
3890 struct btrfs_key *location,
3891 struct btrfs_root **sub_root)
39279cc3 3892{
4df27c4d
YZ		 3893 struct btrfs_path *path;
3894 struct btrfs_root *new_root;
3895 struct btrfs_root_ref *ref;
3896 struct extent_buffer *leaf;
3897 int ret;
3898 int err = 0;
39279cc3 3899
4df27c4d
YZ		 3900 path = btrfs_alloc_path();
3901 if (!path) {
3902 err = -ENOMEM;
3903 goto out;
3904 }
39279cc3 3905
4df27c4d
YZ		 3906 err = -ENOENT;
3907 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3908 BTRFS_I(dir)->root->root_key.objectid,
3909 location->objectid);
3910 if (ret) {
3911 if (ret < 0)
3912 err = ret;
3913 goto out;
3914 }
39279cc3 3915
4df27c4d
YZ		 3916 leaf = path->nodes[0];
3917 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3918 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3919 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3920 goto out;
39279cc3 3921
4df27c4d
YZ		 3922 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3923 (unsigned long)(ref + 1),
3924 dentry->d_name.len);
3925 if (ret)
3926 goto out;
3927
b3b4aa74 3928 btrfs_release_path(path);
4df27c4d
YZ		 3929
3930 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3931 if (IS_ERR(new_root)) {
3932 err = PTR_ERR(new_root);
3933 goto out;
3934 }
3935
3936 if (btrfs_root_refs(&new_root->root_item) == 0) {
3937 err = -ENOENT;
3938 goto out;
3939 }
3940
3941 *sub_root = new_root;
3942 location->objectid = btrfs_root_dirid(&new_root->root_item);
3943 location->type = BTRFS_INODE_ITEM_KEY;
3944 location->offset = 0;
3945 err = 0;
3946out:
3947 btrfs_free_path(path);
3948 return err;
39279cc3
CM		 3949}
3950
5d4f98a2
YZ		 3951static void inode_tree_add(struct inode *inode)
3952{
3953 struct btrfs_root *root = BTRFS_I(inode)->root;
3954 struct btrfs_inode *entry;
03e860bd
FNP		 3955 struct rb_node **p;
3956 struct rb_node *parent;
33345d01 3957 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3958again:
3959 p = &root->inode_tree.rb_node;
3960 parent = NULL;
5d4f98a2 3961
1d3382cb 3962 if (inode_unhashed(inode))
76dda93c
YZ		 3963 return;
3964
5d4f98a2
YZ		 3965 spin_lock(&root->inode_lock);
3966 while (*p) {
3967 parent = *p;
3968 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3969
33345d01 3970 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3971 p = &parent->rb_left;
33345d01 3972 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3973 p = &parent->rb_right;
5d4f98a2
YZ		 3974 else {
3975 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3976 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3977 rb_erase(parent, &root->inode_tree);
3978 RB_CLEAR_NODE(parent);
3979 spin_unlock(&root->inode_lock);
3980 goto again;
5d4f98a2
YZ		 3981 }
3982 }
3983 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3984 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3985 spin_unlock(&root->inode_lock);
3986}
3987
3988static void inode_tree_del(struct inode *inode)
3989{
3990 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3991 int empty = 0;
5d4f98a2 3992
03e860bd 3993 spin_lock(&root->inode_lock);
5d4f98a2 3994 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3995 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3996 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3997 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3998 }
03e860bd 3999 spin_unlock(&root->inode_lock);
76dda93c 4000
0af3d00b
JB		 4001 /*
4002 * Free space cache has inodes in the tree root, but the tree root has a
4003 * root_refs of 0, so this could end up dropping the tree root as a
4004 * snapshot, so we need the extra !root->fs_info->tree_root check to
4005 * make sure we don't drop it.
4006 */
4007 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4008 root != root->fs_info->tree_root) {
76dda93c
YZ		 4009 synchronize_srcu(&root->fs_info->subvol_srcu);
4010 spin_lock(&root->inode_lock);
4011 empty = RB_EMPTY_ROOT(&root->inode_tree);
4012 spin_unlock(&root->inode_lock);
4013 if (empty)
4014 btrfs_add_dead_root(root);
4015 }
4016}
4017
143bede5 4018void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4019{
4020 struct rb_node *node;
4021 struct rb_node *prev;
4022 struct btrfs_inode *entry;
4023 struct inode *inode;
4024 u64 objectid = 0;
4025
4026 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4027
4028 spin_lock(&root->inode_lock);
4029again:
4030 node = root->inode_tree.rb_node;
4031 prev = NULL;
4032 while (node) {
4033 prev = node;
4034 entry = rb_entry(node, struct btrfs_inode, rb_node);
4035
33345d01 4036 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4037 node = node->rb_left;
33345d01 4038 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4039 node = node->rb_right;
4040 else
4041 break;
4042 }
4043 if (!node) {
4044 while (prev) {
4045 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4046 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4047 node = prev;
4048 break;
4049 }
4050 prev = rb_next(prev);
4051 }
4052 }
4053 while (node) {
4054 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4055 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4056 inode = igrab(&entry->vfs_inode);
4057 if (inode) {
4058 spin_unlock(&root->inode_lock);
4059 if (atomic_read(&inode->i_count) > 1)
4060 d_prune_aliases(inode);
4061 /*
45321ac5 4062 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4063 * the inode cache when its usage count
4064 * hits zero.
4065 */
4066 iput(inode);
4067 cond_resched();
4068 spin_lock(&root->inode_lock);
4069 goto again;
4070 }
4071
4072 if (cond_resched_lock(&root->inode_lock))
4073 goto again;
4074
4075 node = rb_next(node);
4076 }
4077 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4078}
4079
e02119d5
CM		 4080static int btrfs_init_locked_inode(struct inode *inode, void *p)
4081{
4082 struct btrfs_iget_args *args = p;
4083 inode->i_ino = args->ino;
e02119d5 4084 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4085 return 0;
4086}
4087
4088static int btrfs_find_actor(struct inode *inode, void *opaque)
4089{
4090 struct btrfs_iget_args *args = opaque;
33345d01 4091 return args->ino == btrfs_ino(inode) &&
d397712b 4092 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4093}
4094
5d4f98a2
YZ		 4095static struct inode *btrfs_iget_locked(struct super_block *s,
4096 u64 objectid,
4097 struct btrfs_root *root)
39279cc3
CM		 4098{
4099 struct inode *inode;
4100 struct btrfs_iget_args args;
4101 args.ino = objectid;
4102 args.root = root;
4103
4104 inode = iget5_locked(s, objectid, btrfs_find_actor,
4105 btrfs_init_locked_inode,
4106 (void *)&args);
4107 return inode;
4108}
4109
1a54ef8c
BR		 4110/* Get an inode object given its location and corresponding root.
4111 * Returns in *is_new if the inode was read from disk
4112 */
4113struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4114 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4115{
4116 struct inode *inode;
4117
4118 inode = btrfs_iget_locked(s, location->objectid, root);
4119 if (!inode)
5d4f98a2 4120 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4121
4122 if (inode->i_state & I_NEW) {
4123 BTRFS_I(inode)->root = root;
4124 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4125 btrfs_read_locked_inode(inode);
1748f843
MF		 4126 if (!is_bad_inode(inode)) {
4127 inode_tree_add(inode);
4128 unlock_new_inode(inode);
4129 if (new)
4130 *new = 1;
4131 } else {
e0b6d65b
ST		 4132 unlock_new_inode(inode);
4133 iput(inode);
4134 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4135 }
4136 }
4137
1a54ef8c
BR		 4138 return inode;
4139}
4140
4df27c4d
YZ		 4141static struct inode *new_simple_dir(struct super_block *s,
4142 struct btrfs_key *key,
4143 struct btrfs_root *root)
4144{
4145 struct inode *inode = new_inode(s);
4146
4147 if (!inode)
4148 return ERR_PTR(-ENOMEM);
4149
4df27c4d
YZ		 4150 BTRFS_I(inode)->root = root;
4151 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4152 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4153
4154 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4155 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4156 inode->i_fop = &simple_dir_operations;
4157 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4158 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4159
4160 return inode;
4161}
4162
3de4586c 4163struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4164{
d397712b 4165 struct inode *inode;
4df27c4d 4166 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4167 struct btrfs_root *sub_root = root;
4168 struct btrfs_key location;
76dda93c 4169 int index;
b4aff1f8 4170 int ret = 0;
39279cc3
CM		 4171
4172 if (dentry->d_name.len > BTRFS_NAME_LEN)
4173 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4174
b4aff1f8
JB		 4175 if (unlikely(d_need_lookup(dentry))) {
4176 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4177 kfree(dentry->d_fsdata);
4178 dentry->d_fsdata = NULL;
a66e7cc6
JB		 4179 /* This thing is hashed, drop it for now */
4180 d_drop(dentry);
b4aff1f8
JB		 4181 } else {
4182 ret = btrfs_inode_by_name(dir, dentry, &location);
4183 }
5f39d397 4184
39279cc3
CM		 4185 if (ret < 0)
4186 return ERR_PTR(ret);
5f39d397 4187
4df27c4d
YZ		 4188 if (location.objectid == 0)
4189 return NULL;
4190
4191 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4192 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4193 return inode;
4194 }
4195
4196 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4197
76dda93c 4198 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4199 ret = fixup_tree_root_location(root, dir, dentry,
4200 &location, &sub_root);
4201 if (ret < 0) {
4202 if (ret != -ENOENT)
4203 inode = ERR_PTR(ret);
4204 else
4205 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4206 } else {
73f73415 4207 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4208 }
76dda93c
YZ		 4209 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4210
34d19bad 4211 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4212 down_read(&root->fs_info->cleanup_work_sem);
4213 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4214 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4215 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 4216 if (ret)
4217 inode = ERR_PTR(ret);
c71bf099
YZ		 4218 }
4219
3de4586c
CM		 4220 return inode;
4221}
4222
fe15ce44 4223static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4224{
4225 struct btrfs_root *root;
848cce0d 4226 struct inode *inode = dentry->d_inode;
76dda93c 4227
848cce0d
LZ		 4228 if (!inode && !IS_ROOT(dentry))
4229 inode = dentry->d_parent->d_inode;
76dda93c 4230
848cce0d
LZ		 4231 if (inode) {
4232 root = BTRFS_I(inode)->root;
efefb143
YZ		 4233 if (btrfs_root_refs(&root->root_item) == 0)
4234 return 1;
848cce0d
LZ		 4235
4236 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4237 return 1;
efefb143 4238 }
76dda93c
YZ		 4239 return 0;
4240}
4241
b4aff1f8
JB		 4242static void btrfs_dentry_release(struct dentry *dentry)
4243{
4244 if (dentry->d_fsdata)
4245 kfree(dentry->d_fsdata);
4246}
4247
3de4586c
CM		 4248static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4249 struct nameidata *nd)
4250{
a66e7cc6
JB		 4251 struct dentry *ret;
4252
4253 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4254 if (unlikely(d_need_lookup(dentry))) {
4255 spin_lock(&dentry->d_lock);
4256 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4257 spin_unlock(&dentry->d_lock);
4258 }
4259 return ret;
39279cc3
CM		 4260}
4261
16cdcec7 4262unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4263 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4264};
4265
cbdf5a24
DW		 4266static int btrfs_real_readdir(struct file *filp, void *dirent,
4267 filldir_t filldir)
39279cc3 4268{
6da6abae 4269 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4270 struct btrfs_root *root = BTRFS_I(inode)->root;
4271 struct btrfs_item *item;
4272 struct btrfs_dir_item *di;
4273 struct btrfs_key key;
5f39d397 4274 struct btrfs_key found_key;
39279cc3 4275 struct btrfs_path *path;
16cdcec7
MX		 4276 struct list_head ins_list;
4277 struct list_head del_list;
39279cc3 4278 int ret;
5f39d397 4279 struct extent_buffer *leaf;
39279cc3 4280 int slot;
39279cc3
CM		 4281 unsigned char d_type;
4282 int over = 0;
4283 u32 di_cur;
4284 u32 di_total;
4285 u32 di_len;
4286 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4287 char tmp_name[32];
4288 char *name_ptr;
4289 int name_len;
16cdcec7 4290 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4291
4292 /* FIXME, use a real flag for deciding about the key type */
4293 if (root->fs_info->tree_root == root)
4294 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4295
3954401f
CM		 4296 /* special case for "." */
4297 if (filp->f_pos == 0) {
3765fefa
HS		 4298 over = filldir(dirent, ".", 1,
4299 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4300 if (over)
4301 return 0;
4302 filp->f_pos = 1;
4303 }
3954401f
CM		 4304 /* special case for .., just use the back ref */
4305 if (filp->f_pos == 1) {
5ecc7e5d 4306 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4307 over = filldir(dirent, "..", 2,
3765fefa 4308 filp->f_pos, pino, DT_DIR);
3954401f 4309 if (over)
49593bfa 4310 return 0;
3954401f
CM		 4311 filp->f_pos = 2;
4312 }
49593bfa 4313 path = btrfs_alloc_path();
16cdcec7
MX		 4314 if (!path)
4315 return -ENOMEM;
ff5714cc 4316
026fd317 4317 path->reada = 1;
49593bfa 4318
16cdcec7
MX		 4319 if (key_type == BTRFS_DIR_INDEX_KEY) {
4320 INIT_LIST_HEAD(&ins_list);
4321 INIT_LIST_HEAD(&del_list);
4322 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4323 }
4324
39279cc3
CM		 4325 btrfs_set_key_type(&key, key_type);
4326 key.offset = filp->f_pos;
33345d01 4327 key.objectid = btrfs_ino(inode);
5f39d397 4328
39279cc3
CM		 4329 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4330 if (ret < 0)
4331 goto err;
49593bfa
DW		 4332
4333 while (1) {
5f39d397 4334 leaf = path->nodes[0];
39279cc3 4335 slot = path->slots[0];
b9e03af0
LZ		 4336 if (slot >= btrfs_header_nritems(leaf)) {
4337 ret = btrfs_next_leaf(root, path);
4338 if (ret < 0)
4339 goto err;
4340 else if (ret > 0)
4341 break;
4342 continue;
39279cc3 4343 }
3de4586c 4344
5f39d397
CM		 4345 item = btrfs_item_nr(leaf, slot);
4346 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4347
4348 if (found_key.objectid != key.objectid)
39279cc3 4349 break;
5f39d397 4350 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4351 break;
5f39d397 4352 if (found_key.offset < filp->f_pos)
b9e03af0 4353 goto next;
16cdcec7
MX		 4354 if (key_type == BTRFS_DIR_INDEX_KEY &&
4355 btrfs_should_delete_dir_index(&del_list,
4356 found_key.offset))
4357 goto next;
5f39d397
CM		 4358
4359 filp->f_pos = found_key.offset;
16cdcec7 4360 is_curr = 1;
49593bfa 4361
39279cc3
CM		 4362 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4363 di_cur = 0;
5f39d397 4364 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4365
4366 while (di_cur < di_total) {
5f39d397
CM		 4367 struct btrfs_key location;
4368
22a94d44
JB		 4369 if (verify_dir_item(root, leaf, di))
4370 break;
4371
5f39d397 4372 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4373 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4374 name_ptr = tmp_name;
4375 } else {
4376 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4377 if (!name_ptr) {
4378 ret = -ENOMEM;
4379 goto err;
4380 }
5f39d397
CM		 4381 }
4382 read_extent_buffer(leaf, name_ptr,
4383 (unsigned long)(di + 1), name_len);
4384
4385 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4386 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4387
fede766f 4388
3de4586c 4389 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 4390 * skip it.
4391 *
4392 * In contrast to old kernels, we insert the snapshot's
4393 * dir item and dir index after it has been created, so
4394 * we won't find a reference to our own snapshot. We
4395 * still keep the following code for backward
4396 * compatibility.
3de4586c
CM		 4397 */
4398 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4399 location.objectid == root->root_key.objectid) {
4400 over = 0;
4401 goto skip;
4402 }
5f39d397 4403 over = filldir(dirent, name_ptr, name_len,
49593bfa 4404 found_key.offset, location.objectid,
39279cc3 4405 d_type);
5f39d397 4406
3de4586c 4407skip:
5f39d397
CM		 4408 if (name_ptr != tmp_name)
4409 kfree(name_ptr);
4410
39279cc3
CM		 4411 if (over)
4412 goto nopos;
5103e947 4413 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4414 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4415 di_cur += di_len;
4416 di = (struct btrfs_dir_item *)((char *)di + di_len);
4417 }
b9e03af0
LZ		 4418next:
4419 path->slots[0]++;
39279cc3 4420 }
49593bfa 4421
16cdcec7
MX		 4422 if (key_type == BTRFS_DIR_INDEX_KEY) {
4423 if (is_curr)
4424 filp->f_pos++;
4425 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4426 &ins_list);
4427 if (ret)
4428 goto nopos;
4429 }
4430
49593bfa 4431 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4432 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4433 /*
4434 * 32-bit glibc will use getdents64, but then strtol -
4435 * so the last number we can serve is this.
4436 */
4437 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4438 else
4439 filp->f_pos++;
39279cc3
CM		 4440nopos:
4441 ret = 0;
4442err:
16cdcec7
MX		 4443 if (key_type == BTRFS_DIR_INDEX_KEY)
4444 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4445 btrfs_free_path(path);
39279cc3
CM		 4446 return ret;
4447}
4448
a9185b41 4449int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4450{
4451 struct btrfs_root *root = BTRFS_I(inode)->root;
4452 struct btrfs_trans_handle *trans;
4453 int ret = 0;
0af3d00b 4454 bool nolock = false;
39279cc3 4455
72ac3c0d 4456 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 4457 return 0;
4458
83eea1f1 4459 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 4460 nolock = true;
0af3d00b 4461
a9185b41 4462 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4463 if (nolock)
7a7eaa40 4464 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4465 else
7a7eaa40 4466 trans = btrfs_join_transaction(root);
3612b495
TI		 4467 if (IS_ERR(trans))
4468 return PTR_ERR(trans);
0af3d00b
JB		 4469 if (nolock)
4470 ret = btrfs_end_transaction_nolock(trans, root);
4471 else
4472 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4473 }
4474 return ret;
4475}
4476
4477/*
54aa1f4d 4478 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4479 * inode changes. But, it is most likely to find the inode in cache.
4480 * FIXME, needs more benchmarking...there are no reasons other than performance
4481 * to keep or drop this code.
4482 */
22c44fe6 4483int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 4484{
4485 struct btrfs_root *root = BTRFS_I(inode)->root;
4486 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4487 int ret;
4488
72ac3c0d 4489 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 4490 return 0;
39279cc3 4491
7a7eaa40 4492 trans = btrfs_join_transaction(root);
22c44fe6
JB		 4493 if (IS_ERR(trans))
4494 return PTR_ERR(trans);
8929ecfa
YZ		 4495
4496 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4497 if (ret && ret == -ENOSPC) {
4498 /* whoops, lets try again with the full transaction */
4499 btrfs_end_transaction(trans, root);
4500 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 4501 if (IS_ERR(trans))
4502 return PTR_ERR(trans);
8929ecfa 4503
94b60442 4504 ret = btrfs_update_inode(trans, root, inode);
94b60442 4505 }
39279cc3 4506 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4507 if (BTRFS_I(inode)->delayed_node)
4508 btrfs_balance_delayed_items(root);
22c44fe6
JB		 4509
4510 return ret;
4511}
4512
4513/*
4514 * This is a copy of file_update_time. We need this so we can return error on
4515 * ENOSPC for updating the inode in the case of file write and mmap writes.
4516 */
e41f941a
JB		 4517static int btrfs_update_time(struct inode *inode, struct timespec *now,
4518 int flags)
22c44fe6 4519{
2bc55652
AB		 4520 struct btrfs_root *root = BTRFS_I(inode)->root;
4521
4522 if (btrfs_root_readonly(root))
4523 return -EROFS;
4524
e41f941a 4525 if (flags & S_VERSION)
22c44fe6 4526 inode_inc_iversion(inode);
e41f941a
JB		 4527 if (flags & S_CTIME)
4528 inode->i_ctime = *now;
4529 if (flags & S_MTIME)
4530 inode->i_mtime = *now;
4531 if (flags & S_ATIME)
4532 inode->i_atime = *now;
4533 return btrfs_dirty_inode(inode);
39279cc3
CM		 4534}
4535
d352ac68
CM		 4536/*
4537 * find the highest existing sequence number in a directory
4538 * and then set the in-memory index_cnt variable to reflect
4539 * free sequence numbers
4540 */
aec7477b
JB		 4541static int btrfs_set_inode_index_count(struct inode *inode)
4542{
4543 struct btrfs_root *root = BTRFS_I(inode)->root;
4544 struct btrfs_key key, found_key;
4545 struct btrfs_path *path;
4546 struct extent_buffer *leaf;
4547 int ret;
4548
33345d01 4549 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4550 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4551 key.offset = (u64)-1;
4552
4553 path = btrfs_alloc_path();
4554 if (!path)
4555 return -ENOMEM;
4556
4557 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4558 if (ret < 0)
4559 goto out;
4560 /* FIXME: we should be able to handle this */
4561 if (ret == 0)
4562 goto out;
4563 ret = 0;
4564
4565 /*
4566 * MAGIC NUMBER EXPLANATION:
4567 * since we search a directory based on f_pos we have to start at 2
4568 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4569 * else has to start at 2
4570 */
4571 if (path->slots[0] == 0) {
4572 BTRFS_I(inode)->index_cnt = 2;
4573 goto out;
4574 }
4575
4576 path->slots[0]--;
4577
4578 leaf = path->nodes[0];
4579 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4580
33345d01 4581 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4582 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4583 BTRFS_I(inode)->index_cnt = 2;
4584 goto out;
4585 }
4586
4587 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4588out:
4589 btrfs_free_path(path);
4590 return ret;
4591}
4592
d352ac68
CM		 4593/*
4594 * helper to find a free sequence number in a given directory. This current
4595 * code is very simple, later versions will do smarter things in the btree
4596 */
3de4586c 4597int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4598{
4599 int ret = 0;
4600
4601 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4602 ret = btrfs_inode_delayed_dir_index_count(dir);
4603 if (ret) {
4604 ret = btrfs_set_inode_index_count(dir);
4605 if (ret)
4606 return ret;
4607 }
aec7477b
JB		 4608 }
4609
00e4e6b3 4610 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4611 BTRFS_I(dir)->index_cnt++;
4612
4613 return ret;
4614}
4615
39279cc3
CM		 4616static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4617 struct btrfs_root *root,
aec7477b 4618 struct inode *dir,
9c58309d 4619 const char *name, int name_len,
175a4eb7
AV		 4620 u64 ref_objectid, u64 objectid,
4621 umode_t mode, u64 *index)
39279cc3
CM		 4622{
4623 struct inode *inode;
5f39d397 4624 struct btrfs_inode_item *inode_item;
39279cc3 4625 struct btrfs_key *location;
5f39d397 4626 struct btrfs_path *path;
9c58309d
CM		 4627 struct btrfs_inode_ref *ref;
4628 struct btrfs_key key[2];
4629 u32 sizes[2];
4630 unsigned long ptr;
39279cc3
CM		 4631 int ret;
4632 int owner;
4633
5f39d397 4634 path = btrfs_alloc_path();
d8926bb3
MF		 4635 if (!path)
4636 return ERR_PTR(-ENOMEM);
5f39d397 4637
39279cc3 4638 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4639 if (!inode) {
4640 btrfs_free_path(path);
39279cc3 4641 return ERR_PTR(-ENOMEM);
8fb27640 4642 }
39279cc3 4643
581bb050
LZ		 4644 /*
4645 * we have to initialize this early, so we can reclaim the inode
4646 * number if we fail afterwards in this function.
4647 */
4648 inode->i_ino = objectid;
4649
aec7477b 4650 if (dir) {
1abe9b8a 4651 trace_btrfs_inode_request(dir);
4652
3de4586c 4653 ret = btrfs_set_inode_index(dir, index);
09771430 4654 if (ret) {
8fb27640 4655 btrfs_free_path(path);
09771430 4656 iput(inode);
aec7477b 4657 return ERR_PTR(ret);
09771430 4658 }
aec7477b
JB		 4659 }
4660 /*
4661 * index_cnt is ignored for everything but a dir,
4662 * btrfs_get_inode_index_count has an explanation for the magic
4663 * number
4664 */
4665 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4666 BTRFS_I(inode)->root = root;
e02119d5 4667 BTRFS_I(inode)->generation = trans->transid;
76195853 4668 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 4669
569254b0 4670 if (S_ISDIR(mode))
39279cc3
CM		 4671 owner = 0;
4672 else
4673 owner = 1;
9c58309d
CM		 4674
4675 key[0].objectid = objectid;
4676 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4677 key[0].offset = 0;
4678
4679 key[1].objectid = objectid;
4680 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4681 key[1].offset = ref_objectid;
4682
4683 sizes[0] = sizeof(struct btrfs_inode_item);
4684 sizes[1] = name_len + sizeof(*ref);
4685
b9473439 4686 path->leave_spinning = 1;
9c58309d
CM		 4687 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4688 if (ret != 0)
5f39d397
CM		 4689 goto fail;
4690
ecc11fab 4691 inode_init_owner(inode, dir, mode);
a76a3cd4 4692 inode_set_bytes(inode, 0);
39279cc3 4693 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4694 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4695 struct btrfs_inode_item);
e02119d5 4696 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4697
4698 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4699 struct btrfs_inode_ref);
4700 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4701 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4702 ptr = (unsigned long)(ref + 1);
4703 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4704
5f39d397
CM		 4705 btrfs_mark_buffer_dirty(path->nodes[0]);
4706 btrfs_free_path(path);
4707
39279cc3
CM		 4708 location = &BTRFS_I(inode)->location;
4709 location->objectid = objectid;
39279cc3
CM		 4710 location->offset = 0;
4711 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4712
6cbff00f
CH		 4713 btrfs_inherit_iflags(inode, dir);
4714
569254b0 4715 if (S_ISREG(mode)) {
94272164
CM		 4716 if (btrfs_test_opt(root, NODATASUM))
4717 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4718 if (btrfs_test_opt(root, NODATACOW) ||
4719 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4720 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4721 }
4722
39279cc3 4723 insert_inode_hash(inode);
5d4f98a2 4724 inode_tree_add(inode);
1abe9b8a 4725
4726 trace_btrfs_inode_new(inode);
1973f0fa 4727 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4728
39279cc3 4729 return inode;
5f39d397 4730fail:
aec7477b
JB		 4731 if (dir)
4732 BTRFS_I(dir)->index_cnt--;
5f39d397 4733 btrfs_free_path(path);
09771430 4734 iput(inode);
5f39d397 4735 return ERR_PTR(ret);
39279cc3
CM		 4736}
4737
4738static inline u8 btrfs_inode_type(struct inode *inode)
4739{
4740 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4741}
4742
d352ac68
CM		 4743/*
4744 * utility function to add 'inode' into 'parent_inode' with
4745 * a give name and a given sequence number.
4746 * if 'add_backref' is true, also insert a backref from the
4747 * inode to the parent directory.
4748 */
e02119d5
CM		 4749int btrfs_add_link(struct btrfs_trans_handle *trans,
4750 struct inode *parent_inode, struct inode *inode,
4751 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4752{
4df27c4d 4753 int ret = 0;
39279cc3 4754 struct btrfs_key key;
e02119d5 4755 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4756 u64 ino = btrfs_ino(inode);
4757 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4758
33345d01 4759 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4760 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4761 } else {
33345d01 4762 key.objectid = ino;
4df27c4d
YZ		 4763 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4764 key.offset = 0;
4765 }
4766
33345d01 4767 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4768 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4769 key.objectid, root->root_key.objectid,
33345d01 4770 parent_ino, index, name, name_len);
4df27c4d 4771 } else if (add_backref) {
33345d01
LZ		 4772 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4773 parent_ino, index);
4df27c4d 4774 }
39279cc3 4775
79787eaa
JM		 4776 /* Nothing to clean up yet */
4777 if (ret)
4778 return ret;
4df27c4d 4779
79787eaa
JM		 4780 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4781 parent_inode, &key,
4782 btrfs_inode_type(inode), index);
4783 if (ret == -EEXIST)
4784 goto fail_dir_item;
4785 else if (ret) {
4786 btrfs_abort_transaction(trans, root, ret);
4787 return ret;
39279cc3 4788 }
79787eaa
JM		 4789
4790 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4791 name_len * 2);
0c4d2d95 4792 inode_inc_iversion(parent_inode);
79787eaa
JM		 4793 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4794 ret = btrfs_update_inode(trans, root, parent_inode);
4795 if (ret)
4796 btrfs_abort_transaction(trans, root, ret);
39279cc3 4797 return ret;
fe66a05a
CM		 4798
4799fail_dir_item:
4800 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4801 u64 local_index;
4802 int err;
4803 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4804 key.objectid, root->root_key.objectid,
4805 parent_ino, &local_index, name, name_len);
4806
4807 } else if (add_backref) {
4808 u64 local_index;
4809 int err;
4810
4811 err = btrfs_del_inode_ref(trans, root, name, name_len,
4812 ino, parent_ino, &local_index);
4813 }
4814 return ret;
39279cc3
CM		 4815}
4816
4817static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4818 struct inode *dir, struct dentry *dentry,
4819 struct inode *inode, int backref, u64 index)
39279cc3 4820{
a1b075d2
JB		 4821 int err = btrfs_add_link(trans, dir, inode,
4822 dentry->d_name.name, dentry->d_name.len,
4823 backref, index);
39279cc3
CM		 4824 if (err > 0)
4825 err = -EEXIST;
4826 return err;
4827}
4828
618e21d5 4829static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 4830 umode_t mode, dev_t rdev)
618e21d5
JB		 4831{
4832 struct btrfs_trans_handle *trans;
4833 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4834 struct inode *inode = NULL;
618e21d5
JB		 4835 int err;
4836 int drop_inode = 0;
4837 u64 objectid;
1832a6d5 4838 unsigned long nr = 0;
00e4e6b3 4839 u64 index = 0;
618e21d5
JB		 4840
4841 if (!new_valid_dev(rdev))
4842 return -EINVAL;
4843
9ed74f2d
JB		 4844 /*
4845 * 2 for inode item and ref
4846 * 2 for dir items
4847 * 1 for xattr if selinux is on
4848 */
a22285a6
YZ		 4849 trans = btrfs_start_transaction(root, 5);
4850 if (IS_ERR(trans))
4851 return PTR_ERR(trans);
1832a6d5 4852
581bb050
LZ		 4853 err = btrfs_find_free_ino(root, &objectid);
4854 if (err)
4855 goto out_unlock;
4856
aec7477b 4857 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4858 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4859 mode, &index);
7cf96da3
TI		 4860 if (IS_ERR(inode)) {
4861 err = PTR_ERR(inode);
618e21d5 4862 goto out_unlock;
7cf96da3 4863 }
618e21d5 4864
2a7dba39 4865 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4866 if (err) {
4867 drop_inode = 1;
4868 goto out_unlock;
4869 }
4870
ad19db71
CS		 4871 /*
4872 * If the active LSM wants to access the inode during
4873 * d_instantiate it needs these. Smack checks to see
4874 * if the filesystem supports xattrs by looking at the
4875 * ops vector.
4876 */
4877
4878 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 4879 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4880 if (err)
4881 drop_inode = 1;
4882 else {
618e21d5 4883 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4884 btrfs_update_inode(trans, root, inode);
08c422c2 4885 d_instantiate(dentry, inode);
618e21d5 4886 }
618e21d5 4887out_unlock:
d3c2fdcf 4888 nr = trans->blocks_used;
7ad85bb7 4889 btrfs_end_transaction(trans, root);
a22285a6 4890 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4891 if (drop_inode) {
4892 inode_dec_link_count(inode);
4893 iput(inode);
4894 }
618e21d5
JB		 4895 return err;
4896}
4897
39279cc3 4898static int btrfs_create(struct inode *dir, struct dentry *dentry,
4acdaf27 4899 umode_t mode, struct nameidata *nd)
39279cc3
CM		 4900{
4901 struct btrfs_trans_handle *trans;
4902 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4903 struct inode *inode = NULL;
39279cc3 4904 int drop_inode = 0;
a22285a6 4905 int err;
1832a6d5 4906 unsigned long nr = 0;
39279cc3 4907 u64 objectid;
00e4e6b3 4908 u64 index = 0;
39279cc3 4909
9ed74f2d
JB		 4910 /*
4911 * 2 for inode item and ref
4912 * 2 for dir items
4913 * 1 for xattr if selinux is on
4914 */
a22285a6
YZ		 4915 trans = btrfs_start_transaction(root, 5);
4916 if (IS_ERR(trans))
4917 return PTR_ERR(trans);
9ed74f2d 4918
581bb050
LZ		 4919 err = btrfs_find_free_ino(root, &objectid);
4920 if (err)
4921 goto out_unlock;
4922
aec7477b 4923 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4924 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4925 mode, &index);
7cf96da3
TI		 4926 if (IS_ERR(inode)) {
4927 err = PTR_ERR(inode);
39279cc3 4928 goto out_unlock;
7cf96da3 4929 }
39279cc3 4930
2a7dba39 4931 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4932 if (err) {
4933 drop_inode = 1;
4934 goto out_unlock;
4935 }
4936
ad19db71
CS		 4937 /*
4938 * If the active LSM wants to access the inode during
4939 * d_instantiate it needs these. Smack checks to see
4940 * if the filesystem supports xattrs by looking at the
4941 * ops vector.
4942 */
4943 inode->i_fop = &btrfs_file_operations;
4944 inode->i_op = &btrfs_file_inode_operations;
4945
a1b075d2 4946 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4947 if (err)
4948 drop_inode = 1;
4949 else {
4950 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4951 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 4952 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
08c422c2 4953 d_instantiate(dentry, inode);
39279cc3 4954 }
39279cc3 4955out_unlock:
d3c2fdcf 4956 nr = trans->blocks_used;
7ad85bb7 4957 btrfs_end_transaction(trans, root);
39279cc3
CM		 4958 if (drop_inode) {
4959 inode_dec_link_count(inode);
4960 iput(inode);
4961 }
d3c2fdcf 4962 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4963 return err;
4964}
4965
4966static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4967 struct dentry *dentry)
4968{
4969 struct btrfs_trans_handle *trans;
4970 struct btrfs_root *root = BTRFS_I(dir)->root;
4971 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4972 u64 index;
1832a6d5 4973 unsigned long nr = 0;
39279cc3
CM		 4974 int err;
4975 int drop_inode = 0;
4976
4a8be425
TH		 4977 /* do not allow sys_link's with other subvols of the same device */
4978 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4979 return -EXDEV;
4a8be425 4980
c055e99e
AV		 4981 if (inode->i_nlink == ~0U)
4982 return -EMLINK;
4a8be425 4983
3de4586c 4984 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4985 if (err)
4986 goto fail;
4987
a22285a6 4988 /*
7e6b6465 4989 * 2 items for inode and inode ref
a22285a6 4990 * 2 items for dir items
7e6b6465 4991 * 1 item for parent inode
a22285a6 4992 */
7e6b6465 4993 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4994 if (IS_ERR(trans)) {
4995 err = PTR_ERR(trans);
4996 goto fail;
4997 }
5f39d397 4998
3153495d 4999 btrfs_inc_nlink(inode);
0c4d2d95 5000 inode_inc_iversion(inode);
3153495d 5001 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5002 ihold(inode);
aec7477b 5003
a1b075d2 5004 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5005
a5719521 5006 if (err) {
54aa1f4d 5007 drop_inode = 1;
a5719521 5008 } else {
10d9f309 5009 struct dentry *parent = dentry->d_parent;
a5719521 5010 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5011 if (err)
5012 goto fail;
08c422c2 5013 d_instantiate(dentry, inode);
6a912213 5014 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5015 }
39279cc3 5016
d3c2fdcf 5017 nr = trans->blocks_used;
7ad85bb7 5018 btrfs_end_transaction(trans, root);
1832a6d5 5019fail:
39279cc3
CM		 5020 if (drop_inode) {
5021 inode_dec_link_count(inode);
5022 iput(inode);
5023 }
d3c2fdcf 5024 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5025 return err;
5026}
5027
18bb1db3 5028static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5029{
b9d86667 5030 struct inode *inode = NULL;
39279cc3
CM		 5031 struct btrfs_trans_handle *trans;
5032 struct btrfs_root *root = BTRFS_I(dir)->root;
5033 int err = 0;
5034 int drop_on_err = 0;
b9d86667 5035 u64 objectid = 0;
00e4e6b3 5036 u64 index = 0;
d3c2fdcf 5037 unsigned long nr = 1;
39279cc3 5038
9ed74f2d
JB		 5039 /*
5040 * 2 items for inode and ref
5041 * 2 items for dir items
5042 * 1 for xattr if selinux is on
5043 */
a22285a6
YZ		 5044 trans = btrfs_start_transaction(root, 5);
5045 if (IS_ERR(trans))
5046 return PTR_ERR(trans);
39279cc3 5047
581bb050
LZ		 5048 err = btrfs_find_free_ino(root, &objectid);
5049 if (err)
5050 goto out_fail;
5051
aec7477b 5052 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5053 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5054 S_IFDIR | mode, &index);
39279cc3
CM		 5055 if (IS_ERR(inode)) {
5056 err = PTR_ERR(inode);
5057 goto out_fail;
5058 }
5f39d397 5059
39279cc3 5060 drop_on_err = 1;
33268eaf 5061
2a7dba39 5062 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5063 if (err)
5064 goto out_fail;
5065
39279cc3
CM		 5066 inode->i_op = &btrfs_dir_inode_operations;
5067 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5068
dbe674a9 5069 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5070 err = btrfs_update_inode(trans, root, inode);
5071 if (err)
5072 goto out_fail;
5f39d397 5073
a1b075d2
JB		 5074 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5075 dentry->d_name.len, 0, index);
39279cc3
CM		 5076 if (err)
5077 goto out_fail;
5f39d397 5078
39279cc3
CM		 5079 d_instantiate(dentry, inode);
5080 drop_on_err = 0;
39279cc3
CM		 5081
5082out_fail:
d3c2fdcf 5083 nr = trans->blocks_used;
7ad85bb7 5084 btrfs_end_transaction(trans, root);
39279cc3
CM		 5085 if (drop_on_err)
5086 iput(inode);
d3c2fdcf 5087 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5088 return err;
5089}
5090
d352ac68
CM		 5091/* helper for btfs_get_extent. Given an existing extent in the tree,
5092 * and an extent that you want to insert, deal with overlap and insert
5093 * the new extent into the tree.
5094 */
3b951516
CM		 5095static int merge_extent_mapping(struct extent_map_tree *em_tree,
5096 struct extent_map *existing,
e6dcd2dc
CM		 5097 struct extent_map *em,
5098 u64 map_start, u64 map_len)
3b951516
CM		 5099{
5100 u64 start_diff;
3b951516 5101
e6dcd2dc
CM		 5102 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5103 start_diff = map_start - em->start;
5104 em->start = map_start;
5105 em->len = map_len;
c8b97818
CM		 5106 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5107 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5108 em->block_start += start_diff;
c8b97818
CM		 5109 em->block_len -= start_diff;
5110 }
e6dcd2dc 5111 return add_extent_mapping(em_tree, em);
3b951516
CM		 5112}
5113
c8b97818
CM		 5114static noinline int uncompress_inline(struct btrfs_path *path,
5115 struct inode *inode, struct page *page,
5116 size_t pg_offset, u64 extent_offset,
5117 struct btrfs_file_extent_item *item)
5118{
5119 int ret;
5120 struct extent_buffer *leaf = path->nodes[0];
5121 char *tmp;
5122 size_t max_size;
5123 unsigned long inline_size;
5124 unsigned long ptr;
261507a0 5125 int compress_type;
c8b97818
CM		 5126
5127 WARN_ON(pg_offset != 0);
261507a0 5128 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5129 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5130 inline_size = btrfs_file_extent_inline_item_len(leaf,
5131 btrfs_item_nr(leaf, path->slots[0]));
5132 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5133 if (!tmp)
5134 return -ENOMEM;
c8b97818
CM		 5135 ptr = btrfs_file_extent_inline_start(item);
5136
5137 read_extent_buffer(leaf, tmp, ptr, inline_size);
5138
5b050f04 5139 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5140 ret = btrfs_decompress(compress_type, tmp, page,
5141 extent_offset, inline_size, max_size);
c8b97818 5142 if (ret) {
7ac687d9 5143 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5144 unsigned long copy_size = min_t(u64,
5145 PAGE_CACHE_SIZE - pg_offset,
5146 max_size - extent_offset);
5147 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5148 kunmap_atomic(kaddr);
c8b97818
CM		 5149 }
5150 kfree(tmp);
5151 return 0;
5152}
5153
d352ac68
CM		 5154/*
5155 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5156 * the ugly parts come from merging extents from the disk with the in-ram
5157 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5158 * where the in-ram extents might be locked pending data=ordered completion.
5159 *
5160 * This also copies inline extents directly into the page.
5161 */
d397712b 5162
a52d9a80 5163struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5164 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5165 int create)
5166{
5167 int ret;
5168 int err = 0;
db94535d 5169 u64 bytenr;
a52d9a80
CM		 5170 u64 extent_start = 0;
5171 u64 extent_end = 0;
33345d01 5172 u64 objectid = btrfs_ino(inode);
a52d9a80 5173 u32 found_type;
f421950f 5174 struct btrfs_path *path = NULL;
a52d9a80
CM		 5175 struct btrfs_root *root = BTRFS_I(inode)->root;
5176 struct btrfs_file_extent_item *item;
5f39d397
CM		 5177 struct extent_buffer *leaf;
5178 struct btrfs_key found_key;
a52d9a80
CM		 5179 struct extent_map *em = NULL;
5180 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5181 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5182 struct btrfs_trans_handle *trans = NULL;
261507a0 5183 int compress_type;
a52d9a80 5184
a52d9a80 5185again:
890871be 5186 read_lock(&em_tree->lock);
d1310b2e 5187 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5188 if (em)
5189 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5190 read_unlock(&em_tree->lock);
d1310b2e 5191
a52d9a80 5192 if (em) {
e1c4b745
CM		 5193 if (em->start > start || em->start + em->len <= start)
5194 free_extent_map(em);
5195 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5196 free_extent_map(em);
5197 else
5198 goto out;
a52d9a80 5199 }
172ddd60 5200 em = alloc_extent_map();
a52d9a80 5201 if (!em) {
d1310b2e
CM		 5202 err = -ENOMEM;
5203 goto out;
a52d9a80 5204 }
e6dcd2dc 5205 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5206 em->start = EXTENT_MAP_HOLE;
445a6944 5207 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5208 em->len = (u64)-1;
c8b97818 5209 em->block_len = (u64)-1;
f421950f
CM		 5210
5211 if (!path) {
5212 path = btrfs_alloc_path();
026fd317
JB		 5213 if (!path) {
5214 err = -ENOMEM;
5215 goto out;
5216 }
5217 /*
5218 * Chances are we'll be called again, so go ahead and do
5219 * readahead
5220 */
5221 path->reada = 1;
f421950f
CM		 5222 }
5223
179e29e4
CM		 5224 ret = btrfs_lookup_file_extent(trans, root, path,
5225 objectid, start, trans != NULL);
a52d9a80
CM		 5226 if (ret < 0) {
5227 err = ret;
5228 goto out;
5229 }
5230
5231 if (ret != 0) {
5232 if (path->slots[0] == 0)
5233 goto not_found;
5234 path->slots[0]--;
5235 }
5236
5f39d397
CM		 5237 leaf = path->nodes[0];
5238 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5239 struct btrfs_file_extent_item);
a52d9a80 5240 /* are we inside the extent that was found? */
5f39d397
CM		 5241 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5242 found_type = btrfs_key_type(&found_key);
5243 if (found_key.objectid != objectid ||
a52d9a80
CM		 5244 found_type != BTRFS_EXTENT_DATA_KEY) {
5245 goto not_found;
5246 }
5247
5f39d397
CM		 5248 found_type = btrfs_file_extent_type(leaf, item);
5249 extent_start = found_key.offset;
261507a0 5250 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5251 if (found_type == BTRFS_FILE_EXTENT_REG ||
5252 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5253 extent_end = extent_start +
db94535d 5254 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5255 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5256 size_t size;
5257 size = btrfs_file_extent_inline_len(leaf, item);
5258 extent_end = (extent_start + size + root->sectorsize - 1) &
5259 ~((u64)root->sectorsize - 1);
5260 }
5261
5262 if (start >= extent_end) {
5263 path->slots[0]++;
5264 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5265 ret = btrfs_next_leaf(root, path);
5266 if (ret < 0) {
5267 err = ret;
5268 goto out;
a52d9a80 5269 }
9036c102
YZ		 5270 if (ret > 0)
5271 goto not_found;
5272 leaf = path->nodes[0];
a52d9a80 5273 }
9036c102
YZ		 5274 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5275 if (found_key.objectid != objectid ||
5276 found_key.type != BTRFS_EXTENT_DATA_KEY)
5277 goto not_found;
5278 if (start + len <= found_key.offset)
5279 goto not_found;
5280 em->start = start;
5281 em->len = found_key.offset - start;
5282 goto not_found_em;
5283 }
5284
d899e052
YZ		 5285 if (found_type == BTRFS_FILE_EXTENT_REG ||
5286 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5287 em->start = extent_start;
5288 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5289 em->orig_start = extent_start -
5290 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5291 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5292 if (bytenr == 0) {
5f39d397 5293 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5294 goto insert;
5295 }
261507a0 5296 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5297 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5298 em->compress_type = compress_type;
c8b97818
CM		 5299 em->block_start = bytenr;
5300 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5301 item);
5302 } else {
5303 bytenr += btrfs_file_extent_offset(leaf, item);
5304 em->block_start = bytenr;
5305 em->block_len = em->len;
d899e052
YZ		 5306 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5307 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5308 }
a52d9a80
CM		 5309 goto insert;
5310 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5311 unsigned long ptr;
a52d9a80 5312 char *map;
3326d1b0
CM		 5313 size_t size;
5314 size_t extent_offset;
5315 size_t copy_size;
a52d9a80 5316
689f9346 5317 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5318 if (!page || create) {
689f9346 5319 em->start = extent_start;
9036c102 5320 em->len = extent_end - extent_start;
689f9346
Y		 5321 goto out;
5322 }
5f39d397 5323
9036c102
YZ		 5324 size = btrfs_file_extent_inline_len(leaf, item);
5325 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5326 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5327 size - extent_offset);
3326d1b0 5328 em->start = extent_start + extent_offset;
70dec807
CM		 5329 em->len = (copy_size + root->sectorsize - 1) &
5330 ~((u64)root->sectorsize - 1);
ff5b7ee3 5331 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5332 if (compress_type) {
c8b97818 5333 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5334 em->compress_type = compress_type;
5335 }
689f9346 5336 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5337 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5338 if (btrfs_file_extent_compression(leaf, item) !=
5339 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5340 ret = uncompress_inline(path, inode, page,
5341 pg_offset,
5342 extent_offset, item);
79787eaa 5343 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 5344 } else {
5345 map = kmap(page);
5346 read_extent_buffer(leaf, map + pg_offset, ptr,
5347 copy_size);
93c82d57
CM		 5348 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5349 memset(map + pg_offset + copy_size, 0,
5350 PAGE_CACHE_SIZE - pg_offset -
5351 copy_size);
5352 }
c8b97818
CM		 5353 kunmap(page);
5354 }
179e29e4
CM		 5355 flush_dcache_page(page);
5356 } else if (create && PageUptodate(page)) {
6bf7e080 5357 BUG();
179e29e4
CM		 5358 if (!trans) {
5359 kunmap(page);
5360 free_extent_map(em);
5361 em = NULL;
ff5714cc 5362
b3b4aa74 5363 btrfs_release_path(path);
7a7eaa40 5364 trans = btrfs_join_transaction(root);
ff5714cc 5365
3612b495
TI		 5366 if (IS_ERR(trans))
5367 return ERR_CAST(trans);
179e29e4
CM		 5368 goto again;
5369 }
c8b97818 5370 map = kmap(page);
70dec807 5371 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5372 copy_size);
c8b97818 5373 kunmap(page);
179e29e4 5374 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5375 }
d1310b2e 5376 set_extent_uptodate(io_tree, em->start,
507903b8 5377 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5378 goto insert;
5379 } else {
d397712b 5380 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5381 WARN_ON(1);
5382 }
5383not_found:
5384 em->start = start;
d1310b2e 5385 em->len = len;
a52d9a80 5386not_found_em:
5f39d397 5387 em->block_start = EXTENT_MAP_HOLE;
9036c102 5388 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5389insert:
b3b4aa74 5390 btrfs_release_path(path);
d1310b2e 5391 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5392 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5393 "[%llu %llu]\n", (unsigned long long)em->start,
5394 (unsigned long long)em->len,
5395 (unsigned long long)start,
5396 (unsigned long long)len);
a52d9a80
CM		 5397 err = -EIO;
5398 goto out;
5399 }
d1310b2e
CM		 5400
5401 err = 0;
890871be 5402 write_lock(&em_tree->lock);
a52d9a80 5403 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5404 /* it is possible that someone inserted the extent into the tree
5405 * while we had the lock dropped. It is also possible that
5406 * an overlapping map exists in the tree
5407 */
a52d9a80 5408 if (ret == -EEXIST) {
3b951516 5409 struct extent_map *existing;
e6dcd2dc
CM		 5410
5411 ret = 0;
5412
3b951516 5413 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5414 if (existing && (existing->start > start ||
5415 existing->start + existing->len <= start)) {
5416 free_extent_map(existing);
5417 existing = NULL;
5418 }
3b951516
CM		 5419 if (!existing) {
5420 existing = lookup_extent_mapping(em_tree, em->start,
5421 em->len);
5422 if (existing) {
5423 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5424 em, start,
5425 root->sectorsize);
3b951516
CM		 5426 free_extent_map(existing);
5427 if (err) {
5428 free_extent_map(em);
5429 em = NULL;
5430 }
5431 } else {
5432 err = -EIO;
3b951516
CM		 5433 free_extent_map(em);
5434 em = NULL;
5435 }
5436 } else {
5437 free_extent_map(em);
5438 em = existing;
e6dcd2dc 5439 err = 0;
a52d9a80 5440 }
a52d9a80 5441 }
890871be 5442 write_unlock(&em_tree->lock);
a52d9a80 5443out:
1abe9b8a 5444
5445 trace_btrfs_get_extent(root, em);
5446
f421950f
CM		 5447 if (path)
5448 btrfs_free_path(path);
a52d9a80
CM		 5449 if (trans) {
5450 ret = btrfs_end_transaction(trans, root);
d397712b 5451 if (!err)
a52d9a80
CM		 5452 err = ret;
5453 }
a52d9a80
CM		 5454 if (err) {
5455 free_extent_map(em);
a52d9a80
CM		 5456 return ERR_PTR(err);
5457 }
79787eaa 5458 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 5459 return em;
5460}
5461
ec29ed5b
CM		 5462struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5463 size_t pg_offset, u64 start, u64 len,
5464 int create)
5465{
5466 struct extent_map *em;
5467 struct extent_map *hole_em = NULL;
5468 u64 range_start = start;
5469 u64 end;
5470 u64 found;
5471 u64 found_end;
5472 int err = 0;
5473
5474 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5475 if (IS_ERR(em))
5476 return em;
5477 if (em) {
5478 /*
5479 * if our em maps to a hole, there might
5480 * actually be delalloc bytes behind it
5481 */
5482 if (em->block_start != EXTENT_MAP_HOLE)
5483 return em;
5484 else
5485 hole_em = em;
5486 }
5487
5488 /* check to see if we've wrapped (len == -1 or similar) */
5489 end = start + len;
5490 if (end < start)
5491 end = (u64)-1;
5492 else
5493 end -= 1;
5494
5495 em = NULL;
5496
5497 /* ok, we didn't find anything, lets look for delalloc */
5498 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5499 end, len, EXTENT_DELALLOC, 1);
5500 found_end = range_start + found;
5501 if (found_end < range_start)
5502 found_end = (u64)-1;
5503
5504 /*
5505 * we didn't find anything useful, return
5506 * the original results from get_extent()
5507 */
5508 if (range_start > end || found_end <= start) {
5509 em = hole_em;
5510 hole_em = NULL;
5511 goto out;
5512 }
5513
5514 /* adjust the range_start to make sure it doesn't
5515 * go backwards from the start they passed in
5516 */
5517 range_start = max(start,range_start);
5518 found = found_end - range_start;
5519
5520 if (found > 0) {
5521 u64 hole_start = start;
5522 u64 hole_len = len;
5523
172ddd60 5524 em = alloc_extent_map();
ec29ed5b
CM		 5525 if (!em) {
5526 err = -ENOMEM;
5527 goto out;
5528 }
5529 /*
5530 * when btrfs_get_extent can't find anything it
5531 * returns one huge hole
5532 *
5533 * make sure what it found really fits our range, and
5534 * adjust to make sure it is based on the start from
5535 * the caller
5536 */
5537 if (hole_em) {
5538 u64 calc_end = extent_map_end(hole_em);
5539
5540 if (calc_end <= start || (hole_em->start > end)) {
5541 free_extent_map(hole_em);
5542 hole_em = NULL;
5543 } else {
5544 hole_start = max(hole_em->start, start);
5545 hole_len = calc_end - hole_start;
5546 }
5547 }
5548 em->bdev = NULL;
5549 if (hole_em && range_start > hole_start) {
5550 /* our hole starts before our delalloc, so we
5551 * have to return just the parts of the hole
5552 * that go until the delalloc starts
5553 */
5554 em->len = min(hole_len,
5555 range_start - hole_start);
5556 em->start = hole_start;
5557 em->orig_start = hole_start;
5558 /*
5559 * don't adjust block start at all,
5560 * it is fixed at EXTENT_MAP_HOLE
5561 */
5562 em->block_start = hole_em->block_start;
5563 em->block_len = hole_len;
5564 } else {
5565 em->start = range_start;
5566 em->len = found;
5567 em->orig_start = range_start;
5568 em->block_start = EXTENT_MAP_DELALLOC;
5569 em->block_len = found;
5570 }
5571 } else if (hole_em) {
5572 return hole_em;
5573 }
5574out:
5575
5576 free_extent_map(hole_em);
5577 if (err) {
5578 free_extent_map(em);
5579 return ERR_PTR(err);
5580 }
5581 return em;
5582}
5583
4b46fce2 5584static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5585 struct extent_map *em,
4b46fce2
JB		 5586 u64 start, u64 len)
5587{
5588 struct btrfs_root *root = BTRFS_I(inode)->root;
5589 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5590 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5591 struct btrfs_key ins;
5592 u64 alloc_hint;
5593 int ret;
16d299ac 5594 bool insert = false;
4b46fce2 5595
16d299ac
JB		 5596 /*
5597 * Ok if the extent map we looked up is a hole and is for the exact
5598 * range we want, there is no reason to allocate a new one, however if
5599 * it is not right then we need to free this one and drop the cache for
5600 * our range.
5601 */
5602 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5603 em->len != len) {
5604 free_extent_map(em);
5605 em = NULL;
5606 insert = true;
5607 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5608 }
4b46fce2 5609
7a7eaa40 5610 trans = btrfs_join_transaction(root);
3612b495
TI		 5611 if (IS_ERR(trans))
5612 return ERR_CAST(trans);
4b46fce2 5613
4cb5300b
CM		 5614 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5615 btrfs_add_inode_defrag(trans, inode);
5616
4b46fce2
JB		 5617 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5618
5619 alloc_hint = get_extent_allocation_hint(inode, start, len);
5620 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 5621 alloc_hint, &ins, 1);
4b46fce2
JB		 5622 if (ret) {
5623 em = ERR_PTR(ret);
5624 goto out;
5625 }
5626
4b46fce2 5627 if (!em) {
172ddd60 5628 em = alloc_extent_map();
16d299ac
JB		 5629 if (!em) {
5630 em = ERR_PTR(-ENOMEM);
5631 goto out;
5632 }
4b46fce2
JB		 5633 }
5634
5635 em->start = start;
5636 em->orig_start = em->start;
5637 em->len = ins.offset;
5638
5639 em->block_start = ins.objectid;
5640 em->block_len = ins.offset;
5641 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5642
5643 /*
5644 * We need to do this because if we're using the original em we searched
5645 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5646 */
5647 em->flags = 0;
4b46fce2
JB		 5648 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5649
16d299ac 5650 while (insert) {
4b46fce2
JB		 5651 write_lock(&em_tree->lock);
5652 ret = add_extent_mapping(em_tree, em);
5653 write_unlock(&em_tree->lock);
5654 if (ret != -EEXIST)
5655 break;
5656 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5657 }
5658
5659 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5660 ins.offset, ins.offset, 0);
5661 if (ret) {
5662 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5663 em = ERR_PTR(ret);
5664 }
5665out:
5666 btrfs_end_transaction(trans, root);
5667 return em;
5668}
5669
46bfbb5c
CM		 5670/*
5671 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5672 * block must be cow'd
5673 */
5674static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5675 struct inode *inode, u64 offset, u64 len)
5676{
5677 struct btrfs_path *path;
5678 int ret;
5679 struct extent_buffer *leaf;
5680 struct btrfs_root *root = BTRFS_I(inode)->root;
5681 struct btrfs_file_extent_item *fi;
5682 struct btrfs_key key;
5683 u64 disk_bytenr;
5684 u64 backref_offset;
5685 u64 extent_end;
5686 u64 num_bytes;
5687 int slot;
5688 int found_type;
5689
5690 path = btrfs_alloc_path();
5691 if (!path)
5692 return -ENOMEM;
5693
33345d01 5694 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5695 offset, 0);
5696 if (ret < 0)
5697 goto out;
5698
5699 slot = path->slots[0];
5700 if (ret == 1) {
5701 if (slot == 0) {
5702 /* can't find the item, must cow */
5703 ret = 0;
5704 goto out;
5705 }
5706 slot--;
5707 }
5708 ret = 0;
5709 leaf = path->nodes[0];
5710 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5711 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5712 key.type != BTRFS_EXTENT_DATA_KEY) {
5713 /* not our file or wrong item type, must cow */
5714 goto out;
5715 }
5716
5717 if (key.offset > offset) {
5718 /* Wrong offset, must cow */
5719 goto out;
5720 }
5721
5722 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5723 found_type = btrfs_file_extent_type(leaf, fi);
5724 if (found_type != BTRFS_FILE_EXTENT_REG &&
5725 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5726 /* not a regular extent, must cow */
5727 goto out;
5728 }
5729 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5730 backref_offset = btrfs_file_extent_offset(leaf, fi);
5731
5732 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5733 if (extent_end < offset + len) {
5734 /* extent doesn't include our full range, must cow */
5735 goto out;
5736 }
5737
5738 if (btrfs_extent_readonly(root, disk_bytenr))
5739 goto out;
5740
5741 /*
5742 * look for other files referencing this extent, if we
5743 * find any we must cow
5744 */
33345d01 5745 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5746 key.offset - backref_offset, disk_bytenr))
5747 goto out;
5748
5749 /*
5750 * adjust disk_bytenr and num_bytes to cover just the bytes
5751 * in this extent we are about to write. If there
5752 * are any csums in that range we have to cow in order
5753 * to keep the csums correct
5754 */
5755 disk_bytenr += backref_offset;
5756 disk_bytenr += offset - key.offset;
5757 num_bytes = min(offset + len, extent_end) - offset;
5758 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5759 goto out;
5760 /*
5761 * all of the above have passed, it is safe to overwrite this extent
5762 * without cow
5763 */
5764 ret = 1;
5765out:
5766 btrfs_free_path(path);
5767 return ret;
5768}
5769
4b46fce2
JB		 5770static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5771 struct buffer_head *bh_result, int create)
5772{
5773 struct extent_map *em;
5774 struct btrfs_root *root = BTRFS_I(inode)->root;
5775 u64 start = iblock << inode->i_blkbits;
5776 u64 len = bh_result->b_size;
46bfbb5c 5777 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5778
5779 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5780 if (IS_ERR(em))
5781 return PTR_ERR(em);
5782
5783 /*
5784 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5785 * io. INLINE is special, and we could probably kludge it in here, but
5786 * it's still buffered so for safety lets just fall back to the generic
5787 * buffered path.
5788 *
5789 * For COMPRESSED we _have_ to read the entire extent in so we can
5790 * decompress it, so there will be buffering required no matter what we
5791 * do, so go ahead and fallback to buffered.
5792 *
5793 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5794 * to buffered IO. Don't blame me, this is the price we pay for using
5795 * the generic code.
5796 */
5797 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5798 em->block_start == EXTENT_MAP_INLINE) {
5799 free_extent_map(em);
5800 return -ENOTBLK;
5801 }
5802
5803 /* Just a good old fashioned hole, return */
5804 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5805 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5806 free_extent_map(em);
5807 /* DIO will do one hole at a time, so just unlock a sector */
5808 unlock_extent(&BTRFS_I(inode)->io_tree, start,
d0082371 5809 start + root->sectorsize - 1);
4b46fce2
JB		 5810 return 0;
5811 }
5812
5813 /*
5814 * We don't allocate a new extent in the following cases
5815 *
5816 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5817 * existing extent.
5818 * 2) The extent is marked as PREALLOC. We're good to go here and can
5819 * just use the extent.
5820 *
5821 */
46bfbb5c
CM		 5822 if (!create) {
5823 len = em->len - (start - em->start);
4b46fce2 5824 goto map;
46bfbb5c 5825 }
4b46fce2
JB		 5826
5827 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5828 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5829 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5830 int type;
5831 int ret;
46bfbb5c 5832 u64 block_start;
4b46fce2
JB		 5833
5834 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5835 type = BTRFS_ORDERED_PREALLOC;
5836 else
5837 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5838 len = min(len, em->len - (start - em->start));
4b46fce2 5839 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5840
5841 /*
5842 * we're not going to log anything, but we do need
5843 * to make sure the current transaction stays open
5844 * while we look for nocow cross refs
5845 */
7a7eaa40 5846 trans = btrfs_join_transaction(root);
3612b495 5847 if (IS_ERR(trans))
46bfbb5c
CM		 5848 goto must_cow;
5849
5850 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5851 ret = btrfs_add_ordered_extent_dio(inode, start,
5852 block_start, len, len, type);
5853 btrfs_end_transaction(trans, root);
5854 if (ret) {
5855 free_extent_map(em);
5856 return ret;
5857 }
5858 goto unlock;
4b46fce2 5859 }
46bfbb5c 5860 btrfs_end_transaction(trans, root);
4b46fce2 5861 }
46bfbb5c
CM		 5862must_cow:
5863 /*
5864 * this will cow the extent, reset the len in case we changed
5865 * it above
5866 */
5867 len = bh_result->b_size;
16d299ac 5868 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5869 if (IS_ERR(em))
5870 return PTR_ERR(em);
5871 len = min(len, em->len - (start - em->start));
5872unlock:
4845e44f
CM		 5873 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5874 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5875 0, NULL, GFP_NOFS);
4b46fce2
JB		 5876map:
5877 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5878 inode->i_blkbits;
46bfbb5c 5879 bh_result->b_size = len;
4b46fce2
JB		 5880 bh_result->b_bdev = em->bdev;
5881 set_buffer_mapped(bh_result);
c3473e83
JB		 5882 if (create) {
5883 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5884 set_buffer_new(bh_result);
5885
5886 /*
5887 * Need to update the i_size under the extent lock so buffered
5888 * readers will get the updated i_size when we unlock.
5889 */
5890 if (start + len > i_size_read(inode))
5891 i_size_write(inode, start + len);
5892 }
4b46fce2
JB		 5893
5894 free_extent_map(em);
5895
5896 return 0;
5897}
5898
5899struct btrfs_dio_private {
5900 struct inode *inode;
5901 u64 logical_offset;
5902 u64 disk_bytenr;
5903 u64 bytes;
5904 u32 *csums;
5905 void *private;
e65e1535
MX		 5906
5907 /* number of bios pending for this dio */
5908 atomic_t pending_bios;
5909
5910 /* IO errors */
5911 int errors;
5912
5913 struct bio *orig_bio;
4b46fce2
JB		 5914};
5915
5916static void btrfs_endio_direct_read(struct bio *bio, int err)
5917{
e65e1535 5918 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5919 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5920 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5921 struct inode *inode = dip->inode;
5922 struct btrfs_root *root = BTRFS_I(inode)->root;
5923 u64 start;
5924 u32 *private = dip->csums;
5925
5926 start = dip->logical_offset;
5927 do {
5928 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5929 struct page *page = bvec->bv_page;
5930 char *kaddr;
5931 u32 csum = ~(u32)0;
5932 unsigned long flags;
5933
5934 local_irq_save(flags);
7ac687d9 5935 kaddr = kmap_atomic(page);
4b46fce2
JB		 5936 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5937 csum, bvec->bv_len);
5938 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 5939 kunmap_atomic(kaddr);
4b46fce2
JB		 5940 local_irq_restore(flags);
5941
5942 flush_dcache_page(bvec->bv_page);
5943 if (csum != *private) {
33345d01 5944 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5945 " %llu csum %u private %u\n",
33345d01
LZ		 5946 (unsigned long long)btrfs_ino(inode),
5947 (unsigned long long)start,
4b46fce2
JB		 5948 csum, *private);
5949 err = -EIO;
5950 }
5951 }
5952
5953 start += bvec->bv_len;
5954 private++;
5955 bvec++;
5956 } while (bvec <= bvec_end);
5957
5958 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 5959 dip->logical_offset + dip->bytes - 1);
4b46fce2
JB		 5960 bio->bi_private = dip->private;
5961
5962 kfree(dip->csums);
5963 kfree(dip);
c0da7aa1
JB		 5964
5965 /* If we had a csum failure make sure to clear the uptodate flag */
5966 if (err)
5967 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5968 dio_end_io(bio, err);
5969}
5970
5971static void btrfs_endio_direct_write(struct bio *bio, int err)
5972{
5973 struct btrfs_dio_private *dip = bio->bi_private;
5974 struct inode *inode = dip->inode;
5975 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 5976 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 5977 u64 ordered_offset = dip->logical_offset;
5978 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5979 int ret;
5980
5981 if (err)
5982 goto out_done;
163cf09c
CM		 5983again:
5984 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5985 &ordered_offset,
5fd02043 5986 ordered_bytes, !err);
4b46fce2 5987 if (!ret)
163cf09c 5988 goto out_test;
4b46fce2 5989
5fd02043
JB		 5990 ordered->work.func = finish_ordered_fn;
5991 ordered->work.flags = 0;
5992 btrfs_queue_worker(&root->fs_info->endio_write_workers,
5993 &ordered->work);
163cf09c
CM		 5994out_test:
5995 /*
5996 * our bio might span multiple ordered extents. If we haven't
5997 * completed the accounting for the whole dio, go back and try again
5998 */
5999 if (ordered_offset < dip->logical_offset + dip->bytes) {
6000 ordered_bytes = dip->logical_offset + dip->bytes -
6001 ordered_offset;
5fd02043 6002 ordered = NULL;
163cf09c
CM		 6003 goto again;
6004 }
4b46fce2
JB		 6005out_done:
6006 bio->bi_private = dip->private;
6007
4b46fce2 6008 kfree(dip);
c0da7aa1
JB		 6009
6010 /* If we had an error make sure to clear the uptodate flag */
6011 if (err)
6012 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 6013 dio_end_io(bio, err);
6014}
6015
eaf25d93
CM		 6016static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6017 struct bio *bio, int mirror_num,
6018 unsigned long bio_flags, u64 offset)
6019{
6020 int ret;
6021 struct btrfs_root *root = BTRFS_I(inode)->root;
6022 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6023 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6024 return 0;
6025}
6026
e65e1535
MX		 6027static void btrfs_end_dio_bio(struct bio *bio, int err)
6028{
6029 struct btrfs_dio_private *dip = bio->bi_private;
6030
6031 if (err) {
33345d01 6032 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6033 "sector %#Lx len %u err no %d\n",
33345d01 6034 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6035 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6036 dip->errors = 1;
6037
6038 /*
6039 * before atomic variable goto zero, we must make sure
6040 * dip->errors is perceived to be set.
6041 */
6042 smp_mb__before_atomic_dec();
6043 }
6044
6045 /* if there are more bios still pending for this dio, just exit */
6046 if (!atomic_dec_and_test(&dip->pending_bios))
6047 goto out;
6048
6049 if (dip->errors)
6050 bio_io_error(dip->orig_bio);
6051 else {
6052 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6053 bio_endio(dip->orig_bio, 0);
6054 }
6055out:
6056 bio_put(bio);
6057}
6058
6059static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6060 u64 first_sector, gfp_t gfp_flags)
6061{
6062 int nr_vecs = bio_get_nr_vecs(bdev);
6063 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6064}
6065
6066static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6067 int rw, u64 file_offset, int skip_sum,
1ae39938 6068 u32 *csums, int async_submit)
e65e1535
MX		 6069{
6070 int write = rw & REQ_WRITE;
6071 struct btrfs_root *root = BTRFS_I(inode)->root;
6072 int ret;
6073
6074 bio_get(bio);
5fd02043
JB		 6075
6076 if (!write) {
6077 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6078 if (ret)
6079 goto err;
6080 }
e65e1535 6081
1ae39938
JB		 6082 if (skip_sum)
6083 goto map;
6084
6085 if (write && async_submit) {
e65e1535
MX		 6086 ret = btrfs_wq_submit_bio(root->fs_info,
6087 inode, rw, bio, 0, 0,
6088 file_offset,
6089 __btrfs_submit_bio_start_direct_io,
6090 __btrfs_submit_bio_done);
6091 goto err;
1ae39938
JB		 6092 } else if (write) {
6093 /*
6094 * If we aren't doing async submit, calculate the csum of the
6095 * bio now.
6096 */
6097 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6098 if (ret)
6099 goto err;
c2db1073
TI		 6100 } else if (!skip_sum) {
6101 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 6102 file_offset, csums);
c2db1073
TI		 6103 if (ret)
6104 goto err;
6105 }
e65e1535 6106
1ae39938
JB		 6107map:
6108 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 6109err:
6110 bio_put(bio);
6111 return ret;
6112}
6113
6114static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6115 int skip_sum)
6116{
6117 struct inode *inode = dip->inode;
6118 struct btrfs_root *root = BTRFS_I(inode)->root;
6119 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6120 struct bio *bio;
6121 struct bio *orig_bio = dip->orig_bio;
6122 struct bio_vec *bvec = orig_bio->bi_io_vec;
6123 u64 start_sector = orig_bio->bi_sector;
6124 u64 file_offset = dip->logical_offset;
6125 u64 submit_len = 0;
6126 u64 map_length;
6127 int nr_pages = 0;
6128 u32 *csums = dip->csums;
6129 int ret = 0;
1ae39938 6130 int async_submit = 0;
98bc3149 6131 int write = rw & REQ_WRITE;
e65e1535 6132
e65e1535
MX		 6133 map_length = orig_bio->bi_size;
6134 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6135 &map_length, NULL, 0);
6136 if (ret) {
64728bbb 6137 bio_put(orig_bio);
e65e1535
MX		 6138 return -EIO;
6139 }
6140
02f57c7a
JB		 6141 if (map_length >= orig_bio->bi_size) {
6142 bio = orig_bio;
6143 goto submit;
6144 }
6145
1ae39938 6146 async_submit = 1;
02f57c7a
JB		 6147 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6148 if (!bio)
6149 return -ENOMEM;
6150 bio->bi_private = dip;
6151 bio->bi_end_io = btrfs_end_dio_bio;
6152 atomic_inc(&dip->pending_bios);
6153
e65e1535
MX		 6154 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6155 if (unlikely(map_length < submit_len + bvec->bv_len ||
6156 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6157 bvec->bv_offset) < bvec->bv_len)) {
6158 /*
6159 * inc the count before we submit the bio so
6160 * we know the end IO handler won't happen before
6161 * we inc the count. Otherwise, the dip might get freed
6162 * before we're done setting it up
6163 */
6164 atomic_inc(&dip->pending_bios);
6165 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6166 file_offset, skip_sum,
1ae39938 6167 csums, async_submit);
e65e1535
MX		 6168 if (ret) {
6169 bio_put(bio);
6170 atomic_dec(&dip->pending_bios);
6171 goto out_err;
6172 }
6173
98bc3149
JB		 6174 /* Write's use the ordered csums */
6175 if (!write && !skip_sum)
e65e1535
MX		 6176 csums = csums + nr_pages;
6177 start_sector += submit_len >> 9;
6178 file_offset += submit_len;
6179
6180 submit_len = 0;
6181 nr_pages = 0;
6182
6183 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6184 start_sector, GFP_NOFS);
6185 if (!bio)
6186 goto out_err;
6187 bio->bi_private = dip;
6188 bio->bi_end_io = btrfs_end_dio_bio;
6189
6190 map_length = orig_bio->bi_size;
6191 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6192 &map_length, NULL, 0);
6193 if (ret) {
6194 bio_put(bio);
6195 goto out_err;
6196 }
6197 } else {
6198 submit_len += bvec->bv_len;
6199 nr_pages ++;
6200 bvec++;
6201 }
6202 }
6203
02f57c7a 6204submit:
e65e1535 6205 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 6206 csums, async_submit);
e65e1535
MX		 6207 if (!ret)
6208 return 0;
6209
6210 bio_put(bio);
6211out_err:
6212 dip->errors = 1;
6213 /*
6214 * before atomic variable goto zero, we must
6215 * make sure dip->errors is perceived to be set.
6216 */
6217 smp_mb__before_atomic_dec();
6218 if (atomic_dec_and_test(&dip->pending_bios))
6219 bio_io_error(dip->orig_bio);
6220
6221 /* bio_end_io() will handle error, so we needn't return it */
6222 return 0;
6223}
6224
4b46fce2
JB		 6225static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6226 loff_t file_offset)
6227{
6228 struct btrfs_root *root = BTRFS_I(inode)->root;
6229 struct btrfs_dio_private *dip;
6230 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6231 int skip_sum;
7b6d91da 6232 int write = rw & REQ_WRITE;
4b46fce2
JB		 6233 int ret = 0;
6234
6235 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6236
6237 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6238 if (!dip) {
6239 ret = -ENOMEM;
6240 goto free_ordered;
6241 }
6242 dip->csums = NULL;
6243
98bc3149
JB		 6244 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6245 if (!write && !skip_sum) {
4b46fce2
JB		 6246 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6247 if (!dip->csums) {
b4966b77 6248 kfree(dip);
4b46fce2
JB		 6249 ret = -ENOMEM;
6250 goto free_ordered;
6251 }
6252 }
6253
6254 dip->private = bio->bi_private;
6255 dip->inode = inode;
6256 dip->logical_offset = file_offset;
6257
4b46fce2
JB		 6258 dip->bytes = 0;
6259 do {
6260 dip->bytes += bvec->bv_len;
6261 bvec++;
6262 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6263
46bfbb5c 6264 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6265 bio->bi_private = dip;
e65e1535
MX		 6266 dip->errors = 0;
6267 dip->orig_bio = bio;
6268 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6269
6270 if (write)
6271 bio->bi_end_io = btrfs_endio_direct_write;
6272 else
6273 bio->bi_end_io = btrfs_endio_direct_read;
6274
e65e1535
MX		 6275 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6276 if (!ret)
eaf25d93 6277 return;
4b46fce2
JB		 6278free_ordered:
6279 /*
6280 * If this is a write, we need to clean up the reserved space and kill
6281 * the ordered extent.
6282 */
6283 if (write) {
6284 struct btrfs_ordered_extent *ordered;
955256f2 6285 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6286 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6287 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6288 btrfs_free_reserved_extent(root, ordered->start,
6289 ordered->disk_len);
6290 btrfs_put_ordered_extent(ordered);
6291 btrfs_put_ordered_extent(ordered);
6292 }
6293 bio_endio(bio, ret);
6294}
6295
5a5f79b5
CM		 6296static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6297 const struct iovec *iov, loff_t offset,
6298 unsigned long nr_segs)
6299{
6300 int seg;
a1b75f7d 6301 int i;
5a5f79b5
CM		 6302 size_t size;
6303 unsigned long addr;
6304 unsigned blocksize_mask = root->sectorsize - 1;
6305 ssize_t retval = -EINVAL;
6306 loff_t end = offset;
6307
6308 if (offset & blocksize_mask)
6309 goto out;
6310
6311 /* Check the memory alignment. Blocks cannot straddle pages */
6312 for (seg = 0; seg < nr_segs; seg++) {
6313 addr = (unsigned long)iov[seg].iov_base;
6314 size = iov[seg].iov_len;
6315 end += size;
a1b75f7d 6316 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6317 goto out;
a1b75f7d
JB		 6318
6319 /* If this is a write we don't need to check anymore */
6320 if (rw & WRITE)
6321 continue;
6322
6323 /*
6324 * Check to make sure we don't have duplicate iov_base's in this
6325 * iovec, if so return EINVAL, otherwise we'll get csum errors
6326 * when reading back.
6327 */
6328 for (i = seg + 1; i < nr_segs; i++) {
6329 if (iov[seg].iov_base == iov[i].iov_base)
6330 goto out;
6331 }
5a5f79b5
CM		 6332 }
6333 retval = 0;
6334out:
6335 return retval;
6336}
16432985
CM		 6337static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6338 const struct iovec *iov, loff_t offset,
6339 unsigned long nr_segs)
6340{
4b46fce2
JB		 6341 struct file *file = iocb->ki_filp;
6342 struct inode *inode = file->f_mapping->host;
6343 struct btrfs_ordered_extent *ordered;
4845e44f 6344 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6345 u64 lockstart, lockend;
6346 ssize_t ret;
4845e44f
CM		 6347 int writing = rw & WRITE;
6348 int write_bits = 0;
3f7c579c 6349 size_t count = iov_length(iov, nr_segs);
4b46fce2 6350
5a5f79b5
CM		 6351 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6352 offset, nr_segs)) {
6353 return 0;
6354 }
6355
4b46fce2 6356 lockstart = offset;
3f7c579c
CM		 6357 lockend = offset + count - 1;
6358
6359 if (writing) {
6360 ret = btrfs_delalloc_reserve_space(inode, count);
6361 if (ret)
6362 goto out;
6363 }
4845e44f 6364
4b46fce2 6365 while (1) {
4845e44f 6366 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
d0082371 6367 0, &cached_state);
4b46fce2
JB		 6368 /*
6369 * We're concerned with the entire range that we're going to be
6370 * doing DIO to, so we need to make sure theres no ordered
6371 * extents in this range.
6372 */
6373 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6374 lockend - lockstart + 1);
c3473e83
JB		 6375
6376 /*
6377 * We need to make sure there are no buffered pages in this
6378 * range either, we could have raced between the invalidate in
6379 * generic_file_direct_write and locking the extent. The
6380 * invalidate needs to happen so that reads after a write do not
6381 * get stale data.
6382 */
6383 if (!ordered && (!writing ||
6384 !test_range_bit(&BTRFS_I(inode)->io_tree,
6385 lockstart, lockend, EXTENT_UPTODATE, 0,
6386 cached_state)))
4b46fce2 6387 break;
c3473e83 6388
4845e44f
CM		 6389 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6390 &cached_state, GFP_NOFS);
c3473e83
JB		 6391
6392 if (ordered) {
6393 btrfs_start_ordered_extent(inode, ordered, 1);
6394 btrfs_put_ordered_extent(ordered);
6395 } else {
6396 /* Screw you mmap */
6397 ret = filemap_write_and_wait_range(file->f_mapping,
6398 lockstart,
6399 lockend);
6400 if (ret)
6401 goto out;
6402
6403 /*
6404 * If we found a page that couldn't be invalidated just
6405 * fall back to buffered.
6406 */
6407 ret = invalidate_inode_pages2_range(file->f_mapping,
6408 lockstart >> PAGE_CACHE_SHIFT,
6409 lockend >> PAGE_CACHE_SHIFT);
6410 if (ret) {
6411 if (ret == -EBUSY)
6412 ret = 0;
6413 goto out;
6414 }
6415 }
6416
4b46fce2
JB		 6417 cond_resched();
6418 }
6419
4845e44f
CM		 6420 /*
6421 * we don't use btrfs_set_extent_delalloc because we don't want
6422 * the dirty or uptodate bits
6423 */
6424 if (writing) {
6425 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6426 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3fbe5c02 6427 EXTENT_DELALLOC, NULL, &cached_state,
4845e44f
CM		 6428 GFP_NOFS);
6429 if (ret) {
6430 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6431 lockend, EXTENT_LOCKED | write_bits,
6432 1, 0, &cached_state, GFP_NOFS);
6433 goto out;
6434 }
6435 }
6436
6437 free_extent_state(cached_state);
6438 cached_state = NULL;
6439
5a5f79b5
CM		 6440 ret = __blockdev_direct_IO(rw, iocb, inode,
6441 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6442 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6443 btrfs_submit_direct, 0);
4b46fce2
JB		 6444
6445 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6446 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6447 offset + iov_length(iov, nr_segs) - 1,
6448 EXTENT_LOCKED | write_bits, 1, 0,
6449 &cached_state, GFP_NOFS);
4b46fce2
JB		 6450 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6451 /*
6452 * We're falling back to buffered, unlock the section we didn't
6453 * do IO on.
6454 */
4845e44f
CM		 6455 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6456 offset + iov_length(iov, nr_segs) - 1,
6457 EXTENT_LOCKED | write_bits, 1, 0,
6458 &cached_state, GFP_NOFS);
4b46fce2 6459 }
4845e44f
CM		 6460out:
6461 free_extent_state(cached_state);
4b46fce2 6462 return ret;
16432985
CM		 6463}
6464
1506fcc8
YS		 6465static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6466 __u64 start, __u64 len)
6467{
ec29ed5b 6468 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6469}
6470
a52d9a80 6471int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6472{
d1310b2e
CM		 6473 struct extent_io_tree *tree;
6474 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6475 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6476}
1832a6d5 6477
a52d9a80 6478static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6479{
d1310b2e 6480 struct extent_io_tree *tree;
b888db2b
CM		 6481
6482
6483 if (current->flags & PF_MEMALLOC) {
6484 redirty_page_for_writepage(wbc, page);
6485 unlock_page(page);
6486 return 0;
6487 }
d1310b2e 6488 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6489 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6490}
6491
f421950f
CM		 6492int btrfs_writepages(struct address_space *mapping,
6493 struct writeback_control *wbc)
b293f02e 6494{
d1310b2e 6495 struct extent_io_tree *tree;
771ed689 6496
d1310b2e 6497 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6498 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6499}
6500
3ab2fb5a
CM		 6501static int
6502btrfs_readpages(struct file *file, struct address_space *mapping,
6503 struct list_head *pages, unsigned nr_pages)
6504{
d1310b2e
CM		 6505 struct extent_io_tree *tree;
6506 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6507 return extent_readpages(tree, mapping, pages, nr_pages,
6508 btrfs_get_extent);
6509}
e6dcd2dc 6510static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6511{
d1310b2e
CM		 6512 struct extent_io_tree *tree;
6513 struct extent_map_tree *map;
a52d9a80 6514 int ret;
8c2383c3 6515
d1310b2e
CM		 6516 tree = &BTRFS_I(page->mapping->host)->io_tree;
6517 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6518 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6519 if (ret == 1) {
6520 ClearPagePrivate(page);
6521 set_page_private(page, 0);
6522 page_cache_release(page);
39279cc3 6523 }
a52d9a80 6524 return ret;
39279cc3
CM		 6525}
6526
e6dcd2dc
CM		 6527static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6528{
98509cfc
CM		 6529 if (PageWriteback(page) || PageDirty(page))
6530 return 0;
b335b003 6531 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6532}
6533
a52d9a80 6534static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6535{
5fd02043 6536 struct inode *inode = page->mapping->host;
d1310b2e 6537 struct extent_io_tree *tree;
e6dcd2dc 6538 struct btrfs_ordered_extent *ordered;
2ac55d41 6539 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6540 u64 page_start = page_offset(page);
6541 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6542
8b62b72b
CM		 6543 /*
6544 * we have the page locked, so new writeback can't start,
6545 * and the dirty bit won't be cleared while we are here.
6546 *
6547 * Wait for IO on this page so that we can safely clear
6548 * the PagePrivate2 bit and do ordered accounting
6549 */
e6dcd2dc 6550 wait_on_page_writeback(page);
8b62b72b 6551
5fd02043 6552 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 6553 if (offset) {
6554 btrfs_releasepage(page, GFP_NOFS);
6555 return;
6556 }
d0082371 6557 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
5fd02043 6558 ordered = btrfs_lookup_ordered_extent(inode,
e6dcd2dc
CM		 6559 page_offset(page));
6560 if (ordered) {
eb84ae03
CM		 6561 /*
6562 * IO on this page will never be started, so we need
6563 * to account for any ordered extents now
6564 */
e6dcd2dc
CM		 6565 clear_extent_bit(tree, page_start, page_end,
6566 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6567 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6568 &cached_state, GFP_NOFS);
8b62b72b
CM		 6569 /*
6570 * whoever cleared the private bit is responsible
6571 * for the finish_ordered_io
6572 */
5fd02043
JB		 6573 if (TestClearPagePrivate2(page) &&
6574 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6575 PAGE_CACHE_SIZE, 1)) {
6576 btrfs_finish_ordered_io(ordered);
8b62b72b 6577 }
e6dcd2dc 6578 btrfs_put_ordered_extent(ordered);
2ac55d41 6579 cached_state = NULL;
d0082371 6580 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 6581 }
6582 clear_extent_bit(tree, page_start, page_end,
32c00aff 6583 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6584 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6585 __btrfs_releasepage(page, GFP_NOFS);
6586
4a096752 6587 ClearPageChecked(page);
9ad6b7bc 6588 if (PagePrivate(page)) {
9ad6b7bc
CM		 6589 ClearPagePrivate(page);
6590 set_page_private(page, 0);
6591 page_cache_release(page);
6592 }
39279cc3
CM		 6593}
6594
9ebefb18
CM		 6595/*
6596 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6597 * called from a page fault handler when a page is first dirtied. Hence we must
6598 * be careful to check for EOF conditions here. We set the page up correctly
6599 * for a written page which means we get ENOSPC checking when writing into
6600 * holes and correct delalloc and unwritten extent mapping on filesystems that
6601 * support these features.
6602 *
6603 * We are not allowed to take the i_mutex here so we have to play games to
6604 * protect against truncate races as the page could now be beyond EOF. Because
6605 * vmtruncate() writes the inode size before removing pages, once we have the
6606 * page lock we can determine safely if the page is beyond EOF. If it is not
6607 * beyond EOF, then the page is guaranteed safe against truncation until we
6608 * unlock the page.
6609 */
c2ec175c 6610int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6611{
c2ec175c 6612 struct page *page = vmf->page;
6da6abae 6613 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6614 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6615 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6616 struct btrfs_ordered_extent *ordered;
2ac55d41 6617 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6618 char *kaddr;
6619 unsigned long zero_start;
9ebefb18 6620 loff_t size;
1832a6d5 6621 int ret;
9998eb70 6622 int reserved = 0;
a52d9a80 6623 u64 page_start;
e6dcd2dc 6624 u64 page_end;
9ebefb18 6625
0ca1f7ce 6626 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 6627 if (!ret) {
e41f941a 6628 ret = file_update_time(vma->vm_file);
9998eb70
CM		 6629 reserved = 1;
6630 }
56a76f82
NP		 6631 if (ret) {
6632 if (ret == -ENOMEM)
6633 ret = VM_FAULT_OOM;
6634 else /* -ENOSPC, -EIO, etc */
6635 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 6636 if (reserved)
6637 goto out;
6638 goto out_noreserve;
56a76f82 6639 }
1832a6d5 6640
56a76f82 6641 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6642again:
9ebefb18 6643 lock_page(page);
9ebefb18 6644 size = i_size_read(inode);
e6dcd2dc
CM		 6645 page_start = page_offset(page);
6646 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6647
9ebefb18 6648 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6649 (page_start >= size)) {
9ebefb18
CM		 6650 /* page got truncated out from underneath us */
6651 goto out_unlock;
6652 }
e6dcd2dc
CM		 6653 wait_on_page_writeback(page);
6654
d0082371 6655 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 6656 set_page_extent_mapped(page);
6657
eb84ae03
CM		 6658 /*
6659 * we can't set the delalloc bits if there are pending ordered
6660 * extents. Drop our locks and wait for them to finish
6661 */
e6dcd2dc
CM		 6662 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6663 if (ordered) {
2ac55d41
JB		 6664 unlock_extent_cached(io_tree, page_start, page_end,
6665 &cached_state, GFP_NOFS);
e6dcd2dc 6666 unlock_page(page);
eb84ae03 6667 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6668 btrfs_put_ordered_extent(ordered);
6669 goto again;
6670 }
6671
fbf19087
JB		 6672 /*
6673 * XXX - page_mkwrite gets called every time the page is dirtied, even
6674 * if it was already dirty, so for space accounting reasons we need to
6675 * clear any delalloc bits for the range we are fixing to save. There
6676 * is probably a better way to do this, but for now keep consistent with
6677 * prepare_pages in the normal write path.
6678 */
2ac55d41 6679 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6680 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6681 0, 0, &cached_state, GFP_NOFS);
fbf19087 6682
2ac55d41
JB		 6683 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6684 &cached_state);
9ed74f2d 6685 if (ret) {
2ac55d41
JB		 6686 unlock_extent_cached(io_tree, page_start, page_end,
6687 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6688 ret = VM_FAULT_SIGBUS;
6689 goto out_unlock;
6690 }
e6dcd2dc 6691 ret = 0;
9ebefb18
CM		 6692
6693 /* page is wholly or partially inside EOF */
a52d9a80 6694 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6695 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6696 else
e6dcd2dc 6697 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6698
e6dcd2dc
CM		 6699 if (zero_start != PAGE_CACHE_SIZE) {
6700 kaddr = kmap(page);
6701 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6702 flush_dcache_page(page);
6703 kunmap(page);
6704 }
247e743c 6705 ClearPageChecked(page);
e6dcd2dc 6706 set_page_dirty(page);
50a9b214 6707 SetPageUptodate(page);
5a3f23d5 6708
257c62e1
CM		 6709 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6710 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6711
2ac55d41 6712 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6713
6714out_unlock:
50a9b214
CM		 6715 if (!ret)
6716 return VM_FAULT_LOCKED;
9ebefb18 6717 unlock_page(page);
1832a6d5 6718out:
ec39e180 6719 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 6720out_noreserve:
9ebefb18
CM		 6721 return ret;
6722}
6723
a41ad394 6724static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6725{
6726 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6727 struct btrfs_block_rsv *rsv;
39279cc3 6728 int ret;
3893e33b 6729 int err = 0;
39279cc3 6730 struct btrfs_trans_handle *trans;
d3c2fdcf 6731 unsigned long nr;
dbe674a9 6732 u64 mask = root->sectorsize - 1;
07127184 6733 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6734
5d5e103a
JB		 6735 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6736 if (ret)
a41ad394 6737 return ret;
8082510e 6738
4a096752 6739 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6740 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6741
fcb80c2a
JB		 6742 /*
6743 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6744 * 3 things going on here
6745 *
6746 * 1) We need to reserve space for our orphan item and the space to
6747 * delete our orphan item. Lord knows we don't want to have a dangling
6748 * orphan item because we didn't reserve space to remove it.
6749 *
6750 * 2) We need to reserve space to update our inode.
6751 *
6752 * 3) We need to have something to cache all the space that is going to
6753 * be free'd up by the truncate operation, but also have some slack
6754 * space reserved in case it uses space during the truncate (thank you
6755 * very much snapshotting).
6756 *
6757 * And we need these to all be seperate. The fact is we can use alot of
6758 * space doing the truncate, and we have no earthly idea how much space
6759 * we will use, so we need the truncate reservation to be seperate so it
6760 * doesn't end up using space reserved for updating the inode or
6761 * removing the orphan item. We also need to be able to stop the
6762 * transaction and start a new one, which means we need to be able to
6763 * update the inode several times, and we have no idea of knowing how
6764 * many times that will be, so we can't just reserve 1 item for the
6765 * entirety of the opration, so that has to be done seperately as well.
6766 * Then there is the orphan item, which does indeed need to be held on
6767 * to for the whole operation, and we need nobody to touch this reserved
6768 * space except the orphan code.
6769 *
6770 * So that leaves us with
6771 *
6772 * 1) root->orphan_block_rsv - for the orphan deletion.
6773 * 2) rsv - for the truncate reservation, which we will steal from the
6774 * transaction reservation.
6775 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6776 * updating the inode.
6777 */
6778 rsv = btrfs_alloc_block_rsv(root);
6779 if (!rsv)
6780 return -ENOMEM;
4a338542 6781 rsv->size = min_size;
f0cd846e 6782
907cbceb 6783 /*
07127184 6784 * 1 for the truncate slack space
907cbceb
JB		 6785 * 1 for the orphan item we're going to add
6786 * 1 for the orphan item deletion
6787 * 1 for updating the inode.
6788 */
fcb80c2a
JB		 6789 trans = btrfs_start_transaction(root, 4);
6790 if (IS_ERR(trans)) {
6791 err = PTR_ERR(trans);
6792 goto out;
6793 }
f0cd846e 6794
907cbceb
JB		 6795 /* Migrate the slack space for the truncate to our reserve */
6796 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6797 min_size);
fcb80c2a 6798 BUG_ON(ret);
f0cd846e
JB		 6799
6800 ret = btrfs_orphan_add(trans, inode);
6801 if (ret) {
6802 btrfs_end_transaction(trans, root);
fcb80c2a 6803 goto out;
f0cd846e
JB		 6804 }
6805
5a3f23d5
CM		 6806 /*
6807 * setattr is responsible for setting the ordered_data_close flag,
6808 * but that is only tested during the last file release. That
6809 * could happen well after the next commit, leaving a great big
6810 * window where new writes may get lost if someone chooses to write
6811 * to this file after truncating to zero
6812 *
6813 * The inode doesn't have any dirty data here, and so if we commit
6814 * this is a noop. If someone immediately starts writing to the inode
6815 * it is very likely we'll catch some of their writes in this
6816 * transaction, and the commit will find this file on the ordered
6817 * data list with good things to send down.
6818 *
6819 * This is a best effort solution, there is still a window where
6820 * using truncate to replace the contents of the file will
6821 * end up with a zero length file after a crash.
6822 */
72ac3c0d
JB		 6823 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6824 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 6825 btrfs_add_ordered_operation(trans, root, inode);
6826
8082510e 6827 while (1) {
36ba022a 6828 ret = btrfs_block_rsv_refill(root, rsv, min_size);
907cbceb
JB		 6829 if (ret) {
6830 /*
6831 * This can only happen with the original transaction we
6832 * started above, every other time we shouldn't have a
6833 * transaction started yet.
6834 */
6835 if (ret == -EAGAIN)
6836 goto end_trans;
6837 err = ret;
6838 break;
6839 }
6840
d68fc57b 6841 if (!trans) {
907cbceb
JB		 6842 /* Just need the 1 for updating the inode */
6843 trans = btrfs_start_transaction(root, 1);
fcb80c2a 6844 if (IS_ERR(trans)) {
7041ee97
JB		 6845 ret = err = PTR_ERR(trans);
6846 trans = NULL;
6847 break;
fcb80c2a 6848 }
d68fc57b
YZ		 6849 }
6850
907cbceb
JB		 6851 trans->block_rsv = rsv;
6852
8082510e
YZ		 6853 ret = btrfs_truncate_inode_items(trans, root, inode,
6854 inode->i_size,
6855 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6856 if (ret != -EAGAIN) {
6857 err = ret;
8082510e 6858 break;
3893e33b 6859 }
39279cc3 6860
fcb80c2a 6861 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6862 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6863 if (ret) {
6864 err = ret;
6865 break;
6866 }
907cbceb 6867end_trans:
8082510e
YZ		 6868 nr = trans->blocks_used;
6869 btrfs_end_transaction(trans, root);
d68fc57b 6870 trans = NULL;
8082510e 6871 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6872 }
6873
6874 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6875 trans->block_rsv = root->orphan_block_rsv;
8082510e 6876 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6877 if (ret)
6878 err = ret;
ded5db9d
JB		 6879 } else if (ret && inode->i_nlink > 0) {
6880 /*
6881 * Failed to do the truncate, remove us from the in memory
6882 * orphan list.
6883 */
6884 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6885 }
6886
917c16b2
CM		 6887 if (trans) {
6888 trans->block_rsv = &root->fs_info->trans_block_rsv;
6889 ret = btrfs_update_inode(trans, root, inode);
6890 if (ret && !err)
6891 err = ret;
7b128766 6892
917c16b2 6893 nr = trans->blocks_used;
7ad85bb7 6894 ret = btrfs_end_transaction(trans, root);
917c16b2
CM		 6895 btrfs_btree_balance_dirty(root, nr);
6896 }
fcb80c2a
JB		 6897
6898out:
6899 btrfs_free_block_rsv(root, rsv);
6900
3893e33b
JB		 6901 if (ret && !err)
6902 err = ret;
a41ad394 6903
3893e33b 6904 return err;
39279cc3
CM		 6905}
6906
d352ac68
CM		 6907/*
6908 * create a new subvolume directory/inode (helper for the ioctl).
6909 */
d2fb3437 6910int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6911 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6912{
39279cc3 6913 struct inode *inode;
76dda93c 6914 int err;
00e4e6b3 6915 u64 index = 0;
39279cc3 6916
12fc9d09
FA		 6917 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6918 new_dirid, new_dirid,
6919 S_IFDIR | (~current_umask() & S_IRWXUGO),
6920 &index);
54aa1f4d 6921 if (IS_ERR(inode))
f46b5a66 6922 return PTR_ERR(inode);
39279cc3
CM		 6923 inode->i_op = &btrfs_dir_inode_operations;
6924 inode->i_fop = &btrfs_dir_file_operations;
6925
bfe86848 6926 set_nlink(inode, 1);
dbe674a9 6927 btrfs_i_size_write(inode, 0);
3b96362c 6928
76dda93c 6929 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 6930
76dda93c 6931 iput(inode);
ce598979 6932 return err;
39279cc3
CM		 6933}
6934
39279cc3
CM		 6935struct inode *btrfs_alloc_inode(struct super_block *sb)
6936{
6937 struct btrfs_inode *ei;
2ead6ae7 6938 struct inode *inode;
39279cc3
CM		 6939
6940 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6941 if (!ei)
6942 return NULL;
2ead6ae7
YZ		 6943
6944 ei->root = NULL;
2ead6ae7 6945 ei->generation = 0;
15ee9bc7 6946 ei->last_trans = 0;
257c62e1 6947 ei->last_sub_trans = 0;
e02119d5 6948 ei->logged_trans = 0;
2ead6ae7 6949 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6950 ei->disk_i_size = 0;
6951 ei->flags = 0;
7709cde3 6952 ei->csum_bytes = 0;
2ead6ae7
YZ		 6953 ei->index_cnt = (u64)-1;
6954 ei->last_unlink_trans = 0;
6955
9e0baf60
JB		 6956 spin_lock_init(&ei->lock);
6957 ei->outstanding_extents = 0;
6958 ei->reserved_extents = 0;
2ead6ae7 6959
72ac3c0d 6960 ei->runtime_flags = 0;
261507a0 6961 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6962
16cdcec7
MX		 6963 ei->delayed_node = NULL;
6964
2ead6ae7 6965 inode = &ei->vfs_inode;
a8067e02 6966 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6967 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6968 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 6969 ei->io_tree.track_uptodate = 1;
6970 ei->io_failure_tree.track_uptodate = 1;
2ead6ae7 6971 mutex_init(&ei->log_mutex);
f248679e 6972 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 6973 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 6974 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6975 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6976 RB_CLEAR_NODE(&ei->rb_node);
6977
6978 return inode;
39279cc3
CM		 6979}
6980
fa0d7e3d
NP		 6981static void btrfs_i_callback(struct rcu_head *head)
6982{
6983 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 6984 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6985}
6986
39279cc3
CM		 6987void btrfs_destroy_inode(struct inode *inode)
6988{
e6dcd2dc 6989 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6990 struct btrfs_root *root = BTRFS_I(inode)->root;
6991
39279cc3
CM		 6992 WARN_ON(!list_empty(&inode->i_dentry));
6993 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6994 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6995 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6996 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6997 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6998
a6dbd429
JB		 6999 /*
7000 * This can happen where we create an inode, but somebody else also
7001 * created the same inode and we need to destroy the one we already
7002 * created.
7003 */
7004 if (!root)
7005 goto free;
7006
5a3f23d5
CM		 7007 /*
7008 * Make sure we're properly removed from the ordered operation
7009 * lists.
7010 */
7011 smp_mb();
7012 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7013 spin_lock(&root->fs_info->ordered_extent_lock);
7014 list_del_init(&BTRFS_I(inode)->ordered_operations);
7015 spin_unlock(&root->fs_info->ordered_extent_lock);
7016 }
7017
8a35d95f
JB		 7018 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7019 &BTRFS_I(inode)->runtime_flags)) {
33345d01
LZ		 7020 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7021 (unsigned long long)btrfs_ino(inode));
8a35d95f 7022 atomic_dec(&root->orphan_inodes);
7b128766 7023 }
7b128766 7024
d397712b 7025 while (1) {
e6dcd2dc
CM		 7026 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7027 if (!ordered)
7028 break;
7029 else {
d397712b
CM		 7030 printk(KERN_ERR "btrfs found ordered "
7031 "extent %llu %llu on inode cleanup\n",
7032 (unsigned long long)ordered->file_offset,
7033 (unsigned long long)ordered->len);
e6dcd2dc
CM		 7034 btrfs_remove_ordered_extent(inode, ordered);
7035 btrfs_put_ordered_extent(ordered);
7036 btrfs_put_ordered_extent(ordered);
7037 }
7038 }
5d4f98a2 7039 inode_tree_del(inode);
5b21f2ed 7040 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7041free:
16cdcec7 7042 btrfs_remove_delayed_node(inode);
fa0d7e3d 7043 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7044}
7045
45321ac5 7046int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7047{
7048 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7049
0af3d00b 7050 if (btrfs_root_refs(&root->root_item) == 0 &&
83eea1f1 7051 !btrfs_is_free_space_inode(inode))
45321ac5 7052 return 1;
76dda93c 7053 else
45321ac5 7054 return generic_drop_inode(inode);
76dda93c
YZ		 7055}
7056
0ee0fda0 7057static void init_once(void *foo)
39279cc3
CM		 7058{
7059 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7060
7061 inode_init_once(&ei->vfs_inode);
7062}
7063
7064void btrfs_destroy_cachep(void)
7065{
7066 if (btrfs_inode_cachep)
7067 kmem_cache_destroy(btrfs_inode_cachep);
7068 if (btrfs_trans_handle_cachep)
7069 kmem_cache_destroy(btrfs_trans_handle_cachep);
7070 if (btrfs_transaction_cachep)
7071 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7072 if (btrfs_path_cachep)
7073 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7074 if (btrfs_free_space_cachep)
7075 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 7076}
7077
7078int btrfs_init_cachep(void)
7079{
9601e3f6
CH		 7080 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7081 sizeof(struct btrfs_inode), 0,
7082 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7083 if (!btrfs_inode_cachep)
7084 goto fail;
9601e3f6
CH		 7085
7086 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7087 sizeof(struct btrfs_trans_handle), 0,
7088 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7089 if (!btrfs_trans_handle_cachep)
7090 goto fail;
9601e3f6
CH		 7091
7092 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7093 sizeof(struct btrfs_transaction), 0,
7094 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7095 if (!btrfs_transaction_cachep)
7096 goto fail;
9601e3f6
CH		 7097
7098 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7099 sizeof(struct btrfs_path), 0,
7100 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7101 if (!btrfs_path_cachep)
7102 goto fail;
9601e3f6 7103
dc89e982
JB		 7104 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7105 sizeof(struct btrfs_free_space), 0,
7106 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7107 if (!btrfs_free_space_cachep)
7108 goto fail;
7109
39279cc3
CM		 7110 return 0;
7111fail:
7112 btrfs_destroy_cachep();
7113 return -ENOMEM;
7114}
7115
7116static int btrfs_getattr(struct vfsmount *mnt,
7117 struct dentry *dentry, struct kstat *stat)
7118{
7119 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7120 u32 blocksize = inode->i_sb->s_blocksize;
7121
39279cc3 7122 generic_fillattr(inode, stat);
0ee5dc67 7123 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7124 stat->blksize = PAGE_CACHE_SIZE;
fadc0d8b
DS		 7125 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7126 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7127 return 0;
7128}
7129
75e7cb7f
LB		 7130/*
7131 * If a file is moved, it will inherit the cow and compression flags of the new
7132 * directory.
7133 */
7134static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7135{
7136 struct btrfs_inode *b_dir = BTRFS_I(dir);
7137 struct btrfs_inode *b_inode = BTRFS_I(inode);
7138
7139 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7140 b_inode->flags |= BTRFS_INODE_NODATACOW;
7141 else
7142 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7143
bc178237 7144 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
75e7cb7f 7145 b_inode->flags |= BTRFS_INODE_COMPRESS;
bc178237
LB		 7146 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7147 } else {
7148 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7149 BTRFS_INODE_NOCOMPRESS);
7150 }
75e7cb7f
LB		 7151}
7152
d397712b
CM		 7153static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7154 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7155{
7156 struct btrfs_trans_handle *trans;
7157 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7158 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7159 struct inode *new_inode = new_dentry->d_inode;
7160 struct inode *old_inode = old_dentry->d_inode;
7161 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7162 u64 index = 0;
4df27c4d 7163 u64 root_objectid;
39279cc3 7164 int ret;
33345d01 7165 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7166
33345d01 7167 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 7168 return -EPERM;
7169
4df27c4d 7170 /* we only allow rename subvolume link between subvolumes */
33345d01 7171 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 7172 return -EXDEV;
7173
33345d01
LZ		 7174 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7175 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 7176 return -ENOTEMPTY;
5f39d397 7177
4df27c4d
YZ		 7178 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7179 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7180 return -ENOTEMPTY;
5a3f23d5
CM		 7181 /*
7182 * we're using rename to replace one file with another.
7183 * and the replacement file is large. Start IO on it now so
7184 * we don't add too much work to the end of the transaction
7185 */
4baf8c92 7186 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 7187 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7188 filemap_flush(old_inode->i_mapping);
7189
76dda93c 7190 /* close the racy window with snapshot create/destroy ioctl */
33345d01 7191 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7192 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 7193 /*
7194 * We want to reserve the absolute worst case amount of items. So if
7195 * both inodes are subvols and we need to unlink them then that would
7196 * require 4 item modifications, but if they are both normal inodes it
7197 * would require 5 item modifications, so we'll assume their normal
7198 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7199 * should cover the worst case number of items we'll modify.
7200 */
7201 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 7202 if (IS_ERR(trans)) {
7203 ret = PTR_ERR(trans);
7204 goto out_notrans;
7205 }
76dda93c 7206
4df27c4d
YZ		 7207 if (dest != root)
7208 btrfs_record_root_in_trans(trans, dest);
5f39d397 7209
a5719521
YZ		 7210 ret = btrfs_set_inode_index(new_dir, &index);
7211 if (ret)
7212 goto out_fail;
5a3f23d5 7213
33345d01 7214 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7215 /* force full log commit if subvolume involved. */
7216 root->fs_info->last_trans_log_full_commit = trans->transid;
7217 } else {
a5719521
YZ		 7218 ret = btrfs_insert_inode_ref(trans, dest,
7219 new_dentry->d_name.name,
7220 new_dentry->d_name.len,
33345d01
LZ		 7221 old_ino,
7222 btrfs_ino(new_dir), index);
a5719521
YZ		 7223 if (ret)
7224 goto out_fail;
4df27c4d
YZ		 7225 /*
7226 * this is an ugly little race, but the rename is required
7227 * to make sure that if we crash, the inode is either at the
7228 * old name or the new one. pinning the log transaction lets
7229 * us make sure we don't allow a log commit to come in after
7230 * we unlink the name but before we add the new name back in.
7231 */
7232 btrfs_pin_log_trans(root);
7233 }
5a3f23d5
CM		 7234 /*
7235 * make sure the inode gets flushed if it is replacing
7236 * something.
7237 */
33345d01 7238 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7239 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7240
0c4d2d95
JB		 7241 inode_inc_iversion(old_dir);
7242 inode_inc_iversion(new_dir);
7243 inode_inc_iversion(old_inode);
39279cc3
CM		 7244 old_dir->i_ctime = old_dir->i_mtime = ctime;
7245 new_dir->i_ctime = new_dir->i_mtime = ctime;
7246 old_inode->i_ctime = ctime;
5f39d397 7247
12fcfd22
CM		 7248 if (old_dentry->d_parent != new_dentry->d_parent)
7249 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7250
33345d01 7251 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7252 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7253 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7254 old_dentry->d_name.name,
7255 old_dentry->d_name.len);
7256 } else {
92986796
AV		 7257 ret = __btrfs_unlink_inode(trans, root, old_dir,
7258 old_dentry->d_inode,
7259 old_dentry->d_name.name,
7260 old_dentry->d_name.len);
7261 if (!ret)
7262 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 7263 }
79787eaa
JM		 7264 if (ret) {
7265 btrfs_abort_transaction(trans, root, ret);
7266 goto out_fail;
7267 }
39279cc3
CM		 7268
7269 if (new_inode) {
0c4d2d95 7270 inode_inc_iversion(new_inode);
39279cc3 7271 new_inode->i_ctime = CURRENT_TIME;
33345d01 7272 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 7273 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7274 root_objectid = BTRFS_I(new_inode)->location.objectid;
7275 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7276 root_objectid,
7277 new_dentry->d_name.name,
7278 new_dentry->d_name.len);
7279 BUG_ON(new_inode->i_nlink == 0);
7280 } else {
7281 ret = btrfs_unlink_inode(trans, dest, new_dir,
7282 new_dentry->d_inode,
7283 new_dentry->d_name.name,
7284 new_dentry->d_name.len);
7285 }
79787eaa 7286 if (!ret && new_inode->i_nlink == 0) {
e02119d5 7287 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7288 BUG_ON(ret);
7b128766 7289 }
79787eaa
JM		 7290 if (ret) {
7291 btrfs_abort_transaction(trans, root, ret);
7292 goto out_fail;
7293 }
39279cc3 7294 }
aec7477b 7295
75e7cb7f
LB		 7296 fixup_inode_flags(new_dir, old_inode);
7297
4df27c4d
YZ		 7298 ret = btrfs_add_link(trans, new_dir, old_inode,
7299 new_dentry->d_name.name,
a5719521 7300 new_dentry->d_name.len, 0, index);
79787eaa
JM		 7301 if (ret) {
7302 btrfs_abort_transaction(trans, root, ret);
7303 goto out_fail;
7304 }
39279cc3 7305
33345d01 7306 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7307 struct dentry *parent = new_dentry->d_parent;
6a912213 7308 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7309 btrfs_end_log_trans(root);
7310 }
39279cc3 7311out_fail:
7ad85bb7 7312 btrfs_end_transaction(trans, root);
b44c59a8 7313out_notrans:
33345d01 7314 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7315 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7316
39279cc3
CM		 7317 return ret;
7318}
7319
d352ac68
CM		 7320/*
7321 * some fairly slow code that needs optimization. This walks the list
7322 * of all the inodes with pending delalloc and forces them to disk.
7323 */
24bbcf04 7324int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7325{
7326 struct list_head *head = &root->fs_info->delalloc_inodes;
7327 struct btrfs_inode *binode;
5b21f2ed 7328 struct inode *inode;
ea8c2819 7329
c146afad
YZ		 7330 if (root->fs_info->sb->s_flags & MS_RDONLY)
7331 return -EROFS;
7332
75eff68e 7333 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7334 while (!list_empty(head)) {
ea8c2819
CM		 7335 binode = list_entry(head->next, struct btrfs_inode,
7336 delalloc_inodes);
5b21f2ed
ZY		 7337 inode = igrab(&binode->vfs_inode);
7338 if (!inode)
7339 list_del_init(&binode->delalloc_inodes);
75eff68e 7340 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7341 if (inode) {
8c8bee1d 7342 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7343 if (delay_iput)
7344 btrfs_add_delayed_iput(inode);
7345 else
7346 iput(inode);
5b21f2ed
ZY		 7347 }
7348 cond_resched();
75eff68e 7349 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7350 }
75eff68e 7351 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7352
7353 /* the filemap_flush will queue IO into the worker threads, but
7354 * we have to make sure the IO is actually started and that
7355 * ordered extents get created before we return
7356 */
7357 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7358 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7359 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7360 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7361 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7362 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7363 }
7364 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7365 return 0;
7366}
7367
39279cc3
CM		 7368static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7369 const char *symname)
7370{
7371 struct btrfs_trans_handle *trans;
7372 struct btrfs_root *root = BTRFS_I(dir)->root;
7373 struct btrfs_path *path;
7374 struct btrfs_key key;
1832a6d5 7375 struct inode *inode = NULL;
39279cc3
CM		 7376 int err;
7377 int drop_inode = 0;
7378 u64 objectid;
00e4e6b3 7379 u64 index = 0 ;
39279cc3
CM		 7380 int name_len;
7381 int datasize;
5f39d397 7382 unsigned long ptr;
39279cc3 7383 struct btrfs_file_extent_item *ei;
5f39d397 7384 struct extent_buffer *leaf;
1832a6d5 7385 unsigned long nr = 0;
39279cc3
CM		 7386
7387 name_len = strlen(symname) + 1;
7388 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7389 return -ENAMETOOLONG;
1832a6d5 7390
9ed74f2d
JB		 7391 /*
7392 * 2 items for inode item and ref
7393 * 2 items for dir items
7394 * 1 item for xattr if selinux is on
7395 */
a22285a6
YZ		 7396 trans = btrfs_start_transaction(root, 5);
7397 if (IS_ERR(trans))
7398 return PTR_ERR(trans);
1832a6d5 7399
581bb050
LZ		 7400 err = btrfs_find_free_ino(root, &objectid);
7401 if (err)
7402 goto out_unlock;
7403
aec7477b 7404 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7405 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7406 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7407 if (IS_ERR(inode)) {
7408 err = PTR_ERR(inode);
39279cc3 7409 goto out_unlock;
7cf96da3 7410 }
39279cc3 7411
2a7dba39 7412 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7413 if (err) {
7414 drop_inode = 1;
7415 goto out_unlock;
7416 }
7417
ad19db71
CS		 7418 /*
7419 * If the active LSM wants to access the inode during
7420 * d_instantiate it needs these. Smack checks to see
7421 * if the filesystem supports xattrs by looking at the
7422 * ops vector.
7423 */
7424 inode->i_fop = &btrfs_file_operations;
7425 inode->i_op = &btrfs_file_inode_operations;
7426
a1b075d2 7427 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7428 if (err)
7429 drop_inode = 1;
7430 else {
7431 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7432 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 7433 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7434 }
39279cc3
CM		 7435 if (drop_inode)
7436 goto out_unlock;
7437
7438 path = btrfs_alloc_path();
d8926bb3
MF		 7439 if (!path) {
7440 err = -ENOMEM;
7441 drop_inode = 1;
7442 goto out_unlock;
7443 }
33345d01 7444 key.objectid = btrfs_ino(inode);
39279cc3 7445 key.offset = 0;
39279cc3
CM		 7446 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7447 datasize = btrfs_file_extent_calc_inline_size(name_len);
7448 err = btrfs_insert_empty_item(trans, root, path, &key,
7449 datasize);
54aa1f4d
CM		 7450 if (err) {
7451 drop_inode = 1;
b0839166 7452 btrfs_free_path(path);
54aa1f4d
CM		 7453 goto out_unlock;
7454 }
5f39d397
CM		 7455 leaf = path->nodes[0];
7456 ei = btrfs_item_ptr(leaf, path->slots[0],
7457 struct btrfs_file_extent_item);
7458 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7459 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7460 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7461 btrfs_set_file_extent_encryption(leaf, ei, 0);
7462 btrfs_set_file_extent_compression(leaf, ei, 0);
7463 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7464 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7465
39279cc3 7466 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7467 write_extent_buffer(leaf, symname, ptr, name_len);
7468 btrfs_mark_buffer_dirty(leaf);
39279cc3 7469 btrfs_free_path(path);
5f39d397 7470
39279cc3
CM		 7471 inode->i_op = &btrfs_symlink_inode_operations;
7472 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7473 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7474 inode_set_bytes(inode, name_len);
dbe674a9 7475 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7476 err = btrfs_update_inode(trans, root, inode);
7477 if (err)
7478 drop_inode = 1;
39279cc3
CM		 7479
7480out_unlock:
08c422c2
AV		 7481 if (!err)
7482 d_instantiate(dentry, inode);
d3c2fdcf 7483 nr = trans->blocks_used;
7ad85bb7 7484 btrfs_end_transaction(trans, root);
39279cc3
CM		 7485 if (drop_inode) {
7486 inode_dec_link_count(inode);
7487 iput(inode);
7488 }
d3c2fdcf 7489 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7490 return err;
7491}
16432985 7492
0af3d00b
JB		 7493static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7494 u64 start, u64 num_bytes, u64 min_size,
7495 loff_t actual_len, u64 *alloc_hint,
7496 struct btrfs_trans_handle *trans)
d899e052 7497{
d899e052
YZ		 7498 struct btrfs_root *root = BTRFS_I(inode)->root;
7499 struct btrfs_key ins;
d899e052 7500 u64 cur_offset = start;
55a61d1d 7501 u64 i_size;
d899e052 7502 int ret = 0;
0af3d00b 7503 bool own_trans = true;
d899e052 7504
0af3d00b
JB		 7505 if (trans)
7506 own_trans = false;
d899e052 7507 while (num_bytes > 0) {
0af3d00b
JB		 7508 if (own_trans) {
7509 trans = btrfs_start_transaction(root, 3);
7510 if (IS_ERR(trans)) {
7511 ret = PTR_ERR(trans);
7512 break;
7513 }
5a303d5d
YZ		 7514 }
7515
efa56464 7516 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
81c9ad23 7517 0, *alloc_hint, &ins, 1);
5a303d5d 7518 if (ret) {
0af3d00b
JB		 7519 if (own_trans)
7520 btrfs_end_transaction(trans, root);
a22285a6 7521 break;
d899e052 7522 }
5a303d5d 7523
d899e052
YZ		 7524 ret = insert_reserved_file_extent(trans, inode,
7525 cur_offset, ins.objectid,
7526 ins.offset, ins.offset,
920bbbfb 7527 ins.offset, 0, 0, 0,
d899e052 7528 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 7529 if (ret) {
7530 btrfs_abort_transaction(trans, root, ret);
7531 if (own_trans)
7532 btrfs_end_transaction(trans, root);
7533 break;
7534 }
a1ed835e
CM		 7535 btrfs_drop_extent_cache(inode, cur_offset,
7536 cur_offset + ins.offset -1, 0);
5a303d5d 7537
d899e052
YZ		 7538 num_bytes -= ins.offset;
7539 cur_offset += ins.offset;
efa56464 7540 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7541
0c4d2d95 7542 inode_inc_iversion(inode);
d899e052 7543 inode->i_ctime = CURRENT_TIME;
6cbff00f 7544 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7545 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7546 (actual_len > inode->i_size) &&
7547 (cur_offset > inode->i_size)) {
d1ea6a61 7548 if (cur_offset > actual_len)
55a61d1d 7549 i_size = actual_len;
d1ea6a61 7550 else
55a61d1d
JB		 7551 i_size = cur_offset;
7552 i_size_write(inode, i_size);
7553 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7554 }
7555
d899e052 7556 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 7557
7558 if (ret) {
7559 btrfs_abort_transaction(trans, root, ret);
7560 if (own_trans)
7561 btrfs_end_transaction(trans, root);
7562 break;
7563 }
d899e052 7564
0af3d00b
JB		 7565 if (own_trans)
7566 btrfs_end_transaction(trans, root);
5a303d5d 7567 }
d899e052
YZ		 7568 return ret;
7569}
7570
0af3d00b
JB		 7571int btrfs_prealloc_file_range(struct inode *inode, int mode,
7572 u64 start, u64 num_bytes, u64 min_size,
7573 loff_t actual_len, u64 *alloc_hint)
7574{
7575 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7576 min_size, actual_len, alloc_hint,
7577 NULL);
7578}
7579
7580int btrfs_prealloc_file_range_trans(struct inode *inode,
7581 struct btrfs_trans_handle *trans, int mode,
7582 u64 start, u64 num_bytes, u64 min_size,
7583 loff_t actual_len, u64 *alloc_hint)
7584{
7585 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7586 min_size, actual_len, alloc_hint, trans);
7587}
7588
e6dcd2dc
CM		 7589static int btrfs_set_page_dirty(struct page *page)
7590{
e6dcd2dc
CM		 7591 return __set_page_dirty_nobuffers(page);
7592}
7593
10556cb2 7594static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7595{
b83cc969 7596 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7597 umode_t mode = inode->i_mode;
b83cc969 7598
cb6db4e5
JM		 7599 if (mask & MAY_WRITE &&
7600 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7601 if (btrfs_root_readonly(root))
7602 return -EROFS;
7603 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7604 return -EACCES;
7605 }
2830ba7f 7606 return generic_permission(inode, mask);
fdebe2bd 7607}
39279cc3 7608
6e1d5dcc 7609static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7610 .getattr = btrfs_getattr,
39279cc3
CM		 7611 .lookup = btrfs_lookup,
7612 .create = btrfs_create,
7613 .unlink = btrfs_unlink,
7614 .link = btrfs_link,
7615 .mkdir = btrfs_mkdir,
7616 .rmdir = btrfs_rmdir,
7617 .rename = btrfs_rename,
7618 .symlink = btrfs_symlink,
7619 .setattr = btrfs_setattr,
618e21d5 7620 .mknod = btrfs_mknod,
95819c05
CH		 7621 .setxattr = btrfs_setxattr,
7622 .getxattr = btrfs_getxattr,
5103e947 7623 .listxattr = btrfs_listxattr,
95819c05 7624 .removexattr = btrfs_removexattr,
fdebe2bd 7625 .permission = btrfs_permission,
4e34e719 7626 .get_acl = btrfs_get_acl,
39279cc3 7627};
6e1d5dcc 7628static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7629 .lookup = btrfs_lookup,
fdebe2bd 7630 .permission = btrfs_permission,
4e34e719 7631 .get_acl = btrfs_get_acl,
39279cc3 7632};
76dda93c 7633
828c0950 7634static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7635 .llseek = generic_file_llseek,
7636 .read = generic_read_dir,
cbdf5a24 7637 .readdir = btrfs_real_readdir,
34287aa3 7638 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7639#ifdef CONFIG_COMPAT
34287aa3 7640 .compat_ioctl = btrfs_ioctl,
39279cc3 7641#endif
6bf13c0c 7642 .release = btrfs_release_file,
e02119d5 7643 .fsync = btrfs_sync_file,
39279cc3
CM		 7644};
7645
d1310b2e 7646static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7647 .fill_delalloc = run_delalloc_range,
065631f6 7648 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7649 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7650 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7651 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7652 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 7653 .set_bit_hook = btrfs_set_bit_hook,
7654 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7655 .merge_extent_hook = btrfs_merge_extent_hook,
7656 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7657};
7658
35054394
CM		 7659/*
7660 * btrfs doesn't support the bmap operation because swapfiles
7661 * use bmap to make a mapping of extents in the file. They assume
7662 * these extents won't change over the life of the file and they
7663 * use the bmap result to do IO directly to the drive.
7664 *
7665 * the btrfs bmap call would return logical addresses that aren't
7666 * suitable for IO and they also will change frequently as COW
7667 * operations happen. So, swapfile + btrfs == corruption.
7668 *
7669 * For now we're avoiding this by dropping bmap.
7670 */
7f09410b 7671static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7672 .readpage = btrfs_readpage,
7673 .writepage = btrfs_writepage,
b293f02e 7674 .writepages = btrfs_writepages,
3ab2fb5a 7675 .readpages = btrfs_readpages,
16432985 7676 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7677 .invalidatepage = btrfs_invalidatepage,
7678 .releasepage = btrfs_releasepage,
e6dcd2dc 7679 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7680 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7681};
7682
7f09410b 7683static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7684 .readpage = btrfs_readpage,
7685 .writepage = btrfs_writepage,
2bf5a725
CM		 7686 .invalidatepage = btrfs_invalidatepage,
7687 .releasepage = btrfs_releasepage,
39279cc3
CM		 7688};
7689
6e1d5dcc 7690static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7691 .getattr = btrfs_getattr,
7692 .setattr = btrfs_setattr,
95819c05
CH		 7693 .setxattr = btrfs_setxattr,
7694 .getxattr = btrfs_getxattr,
5103e947 7695 .listxattr = btrfs_listxattr,
95819c05 7696 .removexattr = btrfs_removexattr,
fdebe2bd 7697 .permission = btrfs_permission,
1506fcc8 7698 .fiemap = btrfs_fiemap,
4e34e719 7699 .get_acl = btrfs_get_acl,
e41f941a 7700 .update_time = btrfs_update_time,
39279cc3 7701};
6e1d5dcc 7702static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7703 .getattr = btrfs_getattr,
7704 .setattr = btrfs_setattr,
fdebe2bd 7705 .permission = btrfs_permission,
95819c05
CH		 7706 .setxattr = btrfs_setxattr,
7707 .getxattr = btrfs_getxattr,
33268eaf 7708 .listxattr = btrfs_listxattr,
95819c05 7709 .removexattr = btrfs_removexattr,
4e34e719 7710 .get_acl = btrfs_get_acl,
e41f941a 7711 .update_time = btrfs_update_time,
618e21d5 7712};
6e1d5dcc 7713static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7714 .readlink = generic_readlink,
7715 .follow_link = page_follow_link_light,
7716 .put_link = page_put_link,
f209561a 7717 .getattr = btrfs_getattr,
22c44fe6 7718 .setattr = btrfs_setattr,
fdebe2bd 7719 .permission = btrfs_permission,
0279b4cd
JO		 7720 .setxattr = btrfs_setxattr,
7721 .getxattr = btrfs_getxattr,
7722 .listxattr = btrfs_listxattr,
7723 .removexattr = btrfs_removexattr,
4e34e719 7724 .get_acl = btrfs_get_acl,
e41f941a 7725 .update_time = btrfs_update_time,
39279cc3 7726};
76dda93c 7727
82d339d9 7728const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7729 .d_delete = btrfs_dentry_delete,
b4aff1f8 7730 .d_release = btrfs_dentry_release,
76dda93c 7731};
Linux 6.x block layer and io_uring tree(s)