projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix accessing a freed tree root
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
53b381b3 43#include <linux/blkdev.h>
4b4e25f2 44#include "compat.h"
39279cc3
CM		 45#include "ctree.h"
46#include "disk-io.h"
47#include "transaction.h"
48#include "btrfs_inode.h"
39279cc3 49#include "print-tree.h"
e6dcd2dc 50#include "ordered-data.h"
95819c05 51#include "xattr.h"
e02119d5 52#include "tree-log.h"
4a54c8c1 53#include "volumes.h"
c8b97818 54#include "compression.h"
b4ce94de 55#include "locking.h"
dc89e982 56#include "free-space-cache.h"
581bb050 57#include "inode-map.h"
38c227d8 58#include "backref.h"
39279cc3
CM		 59
60struct btrfs_iget_args {
61 u64 ino;
62 struct btrfs_root *root;
63};
64
6e1d5dcc
AD		 65static const struct inode_operations btrfs_dir_inode_operations;
66static const struct inode_operations btrfs_symlink_inode_operations;
67static const struct inode_operations btrfs_dir_ro_inode_operations;
68static const struct inode_operations btrfs_special_inode_operations;
69static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 70static const struct address_space_operations btrfs_aops;
71static const struct address_space_operations btrfs_symlink_aops;
828c0950 72static const struct file_operations btrfs_dir_file_operations;
d1310b2e 73static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 74
75static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 76static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 77struct kmem_cache *btrfs_trans_handle_cachep;
78struct kmem_cache *btrfs_transaction_cachep;
39279cc3 79struct kmem_cache *btrfs_path_cachep;
dc89e982 80struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 81
82#define S_SHIFT 12
83static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
84 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
85 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
86 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
87 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
88 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
89 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
90 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91};
92
3972f260 93static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 94static int btrfs_truncate(struct inode *inode);
5fd02043 95static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 96static noinline int cow_file_range(struct inode *inode,
97 struct page *locked_page,
98 u64 start, u64 end, int *page_started,
99 unsigned long *nr_written, int unlock);
70c8a91c
JB		 100static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
101 u64 len, u64 orig_start,
102 u64 block_start, u64 block_len,
cc95bef6
JB		 103 u64 orig_block_len, u64 ram_bytes,
104 int type);
7b128766 105
48a3b636
ES		 106static int btrfs_dirty_inode(struct inode *inode);
107
f34f57a3 108static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 109 struct inode *inode, struct inode *dir,
110 const struct qstr *qstr)
0279b4cd
JO		 111{
112 int err;
113
f34f57a3 114 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 115 if (!err)
2a7dba39 116 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 117 return err;
118}
119
c8b97818
CM		 120/*
121 * this does all the hard work for inserting an inline extent into
122 * the btree. The caller should have done a btrfs_drop_extents so that
123 * no overlapping inline items exist in the btree
124 */
d397712b 125static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 126 struct btrfs_root *root, struct inode *inode,
127 u64 start, size_t size, size_t compressed_size,
fe3f566c 128 int compress_type,
c8b97818
CM		 129 struct page **compressed_pages)
130{
131 struct btrfs_key key;
132 struct btrfs_path *path;
133 struct extent_buffer *leaf;
134 struct page *page = NULL;
135 char *kaddr;
136 unsigned long ptr;
137 struct btrfs_file_extent_item *ei;
138 int err = 0;
139 int ret;
140 size_t cur_size = size;
141 size_t datasize;
142 unsigned long offset;
c8b97818 143
fe3f566c 144 if (compressed_size && compressed_pages)
c8b97818 145 cur_size = compressed_size;
c8b97818 146
d397712b
CM		 147 path = btrfs_alloc_path();
148 if (!path)
c8b97818
CM		 149 return -ENOMEM;
150
b9473439 151 path->leave_spinning = 1;
c8b97818 152
33345d01 153 key.objectid = btrfs_ino(inode);
c8b97818
CM		 154 key.offset = start;
155 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 156 datasize = btrfs_file_extent_calc_inline_size(cur_size);
157
158 inode_add_bytes(inode, size);
159 ret = btrfs_insert_empty_item(trans, root, path, &key,
160 datasize);
c8b97818
CM		 161 if (ret) {
162 err = ret;
c8b97818
CM		 163 goto fail;
164 }
165 leaf = path->nodes[0];
166 ei = btrfs_item_ptr(leaf, path->slots[0],
167 struct btrfs_file_extent_item);
168 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
169 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
170 btrfs_set_file_extent_encryption(leaf, ei, 0);
171 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
172 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
173 ptr = btrfs_file_extent_inline_start(ei);
174
261507a0 175 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 176 struct page *cpage;
177 int i = 0;
d397712b 178 while (compressed_size > 0) {
c8b97818 179 cpage = compressed_pages[i];
5b050f04 180 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 181 PAGE_CACHE_SIZE);
182
7ac687d9 183 kaddr = kmap_atomic(cpage);
c8b97818 184 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 185 kunmap_atomic(kaddr);
c8b97818
CM		 186
187 i++;
188 ptr += cur_size;
189 compressed_size -= cur_size;
190 }
191 btrfs_set_file_extent_compression(leaf, ei,
261507a0 192 compress_type);
c8b97818
CM		 193 } else {
194 page = find_get_page(inode->i_mapping,
195 start >> PAGE_CACHE_SHIFT);
196 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 197 kaddr = kmap_atomic(page);
c8b97818
CM		 198 offset = start & (PAGE_CACHE_SIZE - 1);
199 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 200 kunmap_atomic(kaddr);
c8b97818
CM		 201 page_cache_release(page);
202 }
203 btrfs_mark_buffer_dirty(leaf);
204 btrfs_free_path(path);
205
c2167754
YZ		 206 /*
207 * we're an inline extent, so nobody can
208 * extend the file past i_size without locking
209 * a page we already have locked.
210 *
211 * We must do any isize and inode updates
212 * before we unlock the pages. Otherwise we
213 * could end up racing with unlink.
214 */
c8b97818 215 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 216 ret = btrfs_update_inode(trans, root, inode);
c2167754 217
79787eaa 218 return ret;
c8b97818
CM		 219fail:
220 btrfs_free_path(path);
221 return err;
222}
223
224
225/*
226 * conditionally insert an inline extent into the file. This
227 * does the checks required to make sure the data is small enough
228 * to fit as an inline extent.
229 */
7f366cfe 230static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 231 struct btrfs_root *root,
232 struct inode *inode, u64 start, u64 end,
fe3f566c 233 size_t compressed_size, int compress_type,
c8b97818
CM		 234 struct page **compressed_pages)
235{
236 u64 isize = i_size_read(inode);
237 u64 actual_end = min(end + 1, isize);
238 u64 inline_len = actual_end - start;
fda2832f 239 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 240 u64 data_len = inline_len;
241 int ret;
242
243 if (compressed_size)
244 data_len = compressed_size;
245
246 if (start > 0 ||
70b99e69 247 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 248 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (!compressed_size &&
250 (actual_end & (root->sectorsize - 1)) == 0) ||
251 end + 1 < isize ||
252 data_len > root->fs_info->max_inline) {
253 return 1;
254 }
255
2671485d 256 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM		 257 if (ret)
258 return ret;
c8b97818
CM		 259
260 if (isize > actual_end)
261 inline_len = min_t(u64, isize, actual_end);
262 ret = insert_inline_extent(trans, root, inode, start,
263 inline_len, compressed_size,
fe3f566c 264 compress_type, compressed_pages);
2adcac1a 265 if (ret && ret != -ENOSPC) {
79787eaa
JM		 266 btrfs_abort_transaction(trans, root, ret);
267 return ret;
2adcac1a
JB		 268 } else if (ret == -ENOSPC) {
269 return 1;
79787eaa 270 }
2adcac1a 271
bdc20e67 272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 273 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 274 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 275 return 0;
276}
277
771ed689
CM		 278struct async_extent {
279 u64 start;
280 u64 ram_size;
281 u64 compressed_size;
282 struct page **pages;
283 unsigned long nr_pages;
261507a0 284 int compress_type;
771ed689
CM		 285 struct list_head list;
286};
287
288struct async_cow {
289 struct inode *inode;
290 struct btrfs_root *root;
291 struct page *locked_page;
292 u64 start;
293 u64 end;
294 struct list_head extents;
295 struct btrfs_work work;
296};
297
298static noinline int add_async_extent(struct async_cow *cow,
299 u64 start, u64 ram_size,
300 u64 compressed_size,
301 struct page **pages,
261507a0
LZ		 302 unsigned long nr_pages,
303 int compress_type)
771ed689
CM		 304{
305 struct async_extent *async_extent;
306
307 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 308 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 309 async_extent->start = start;
310 async_extent->ram_size = ram_size;
311 async_extent->compressed_size = compressed_size;
312 async_extent->pages = pages;
313 async_extent->nr_pages = nr_pages;
261507a0 314 async_extent->compress_type = compress_type;
771ed689
CM		 315 list_add_tail(&async_extent->list, &cow->extents);
316 return 0;
317}
318
d352ac68 319/*
771ed689
CM		 320 * we create compressed extents in two phases. The first
321 * phase compresses a range of pages that have already been
322 * locked (both pages and state bits are locked).
c8b97818 323 *
771ed689
CM		 324 * This is done inside an ordered work queue, and the compression
325 * is spread across many cpus. The actual IO submission is step
326 * two, and the ordered work queue takes care of making sure that
327 * happens in the same order things were put onto the queue by
328 * writepages and friends.
c8b97818 329 *
771ed689
CM		 330 * If this code finds it can't get good compression, it puts an
331 * entry onto the work queue to write the uncompressed bytes. This
332 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 333 * are written in the same order that the flusher thread sent them
334 * down.
d352ac68 335 */
771ed689
CM		 336static noinline int compress_file_range(struct inode *inode,
337 struct page *locked_page,
338 u64 start, u64 end,
339 struct async_cow *async_cow,
340 int *num_added)
b888db2b
CM		 341{
342 struct btrfs_root *root = BTRFS_I(inode)->root;
343 struct btrfs_trans_handle *trans;
db94535d 344 u64 num_bytes;
db94535d 345 u64 blocksize = root->sectorsize;
c8b97818 346 u64 actual_end;
42dc7bab 347 u64 isize = i_size_read(inode);
e6dcd2dc 348 int ret = 0;
c8b97818
CM		 349 struct page **pages = NULL;
350 unsigned long nr_pages;
351 unsigned long nr_pages_ret = 0;
352 unsigned long total_compressed = 0;
353 unsigned long total_in = 0;
354 unsigned long max_compressed = 128 * 1024;
771ed689 355 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 356 int i;
357 int will_compress;
261507a0 358 int compress_type = root->fs_info->compress_type;
4adaa611 359 int redirty = 0;
b888db2b 360
4cb13e5d
LB		 361 /* if this is a small write inside eof, kick off a defrag */
362 if ((end - start + 1) < 16 * 1024 &&
363 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 364 btrfs_add_inode_defrag(NULL, inode);
365
42dc7bab 366 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 367again:
368 will_compress = 0;
369 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
370 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 371
f03d9301
CM		 372 /*
373 * we don't want to send crud past the end of i_size through
374 * compression, that's just a waste of CPU time. So, if the
375 * end of the file is before the start of our current
376 * requested range of bytes, we bail out to the uncompressed
377 * cleanup code that can deal with all of this.
378 *
379 * It isn't really the fastest way to fix things, but this is a
380 * very uncommon corner.
381 */
382 if (actual_end <= start)
383 goto cleanup_and_bail_uncompressed;
384
c8b97818
CM		 385 total_compressed = actual_end - start;
386
387 /* we want to make sure that amount of ram required to uncompress
388 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 389 * of a compressed extent to 128k. This is a crucial number
390 * because it also controls how easily we can spread reads across
391 * cpus for decompression.
392 *
393 * We also want to make sure the amount of IO required to do
394 * a random read is reasonably small, so we limit the size of
395 * a compressed extent to 128k.
c8b97818
CM		 396 */
397 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 398 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 399 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 400 total_in = 0;
401 ret = 0;
db94535d 402
771ed689
CM		 403 /*
404 * we do compression for mount -o compress and when the
405 * inode has not been flagged as nocompress. This flag can
406 * change at any time if we discover bad compression ratios.
c8b97818 407 */
6cbff00f 408 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 409 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 410 (BTRFS_I(inode)->force_compress) ||
411 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 412 WARN_ON(pages);
cfbc246e 413 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 414 if (!pages) {
415 /* just bail out to the uncompressed code */
416 goto cont;
417 }
c8b97818 418
261507a0
LZ		 419 if (BTRFS_I(inode)->force_compress)
420 compress_type = BTRFS_I(inode)->force_compress;
421
4adaa611
CM		 422 /*
423 * we need to call clear_page_dirty_for_io on each
424 * page in the range. Otherwise applications with the file
425 * mmap'd can wander in and change the page contents while
426 * we are compressing them.
427 *
428 * If the compression fails for any reason, we set the pages
429 * dirty again later on.
430 */
431 extent_range_clear_dirty_for_io(inode, start, end);
432 redirty = 1;
261507a0
LZ		 433 ret = btrfs_compress_pages(compress_type,
434 inode->i_mapping, start,
435 total_compressed, pages,
436 nr_pages, &nr_pages_ret,
437 &total_in,
438 &total_compressed,
439 max_compressed);
c8b97818
CM		 440
441 if (!ret) {
442 unsigned long offset = total_compressed &
443 (PAGE_CACHE_SIZE - 1);
444 struct page *page = pages[nr_pages_ret - 1];
445 char *kaddr;
446
447 /* zero the tail end of the last page, we might be
448 * sending it down to disk
449 */
450 if (offset) {
7ac687d9 451 kaddr = kmap_atomic(page);
c8b97818
CM		 452 memset(kaddr + offset, 0,
453 PAGE_CACHE_SIZE - offset);
7ac687d9 454 kunmap_atomic(kaddr);
c8b97818
CM		 455 }
456 will_compress = 1;
457 }
458 }
560f7d75 459cont:
c8b97818 460 if (start == 0) {
7a7eaa40 461 trans = btrfs_join_transaction(root);
79787eaa
JM		 462 if (IS_ERR(trans)) {
463 ret = PTR_ERR(trans);
464 trans = NULL;
465 goto cleanup_and_out;
466 }
0ca1f7ce 467 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 468
c8b97818 469 /* lets try to make an inline extent */
771ed689 470 if (ret || total_in < (actual_end - start)) {
c8b97818 471 /* we didn't compress the entire range, try
771ed689 472 * to make an uncompressed inline extent.
c8b97818
CM		 473 */
474 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 475 start, end, 0, 0, NULL);
c8b97818 476 } else {
771ed689 477 /* try making a compressed inline extent */
c8b97818
CM		 478 ret = cow_file_range_inline(trans, root, inode,
479 start, end,
fe3f566c
LZ		 480 total_compressed,
481 compress_type, pages);
c8b97818 482 }
79787eaa 483 if (ret <= 0) {
771ed689 484 /*
79787eaa
JM		 485 * inline extent creation worked or returned error,
486 * we don't need to create any more async work items.
487 * Unlock and free up our temp pages.
771ed689 488 */
c8b97818 489 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 490 &BTRFS_I(inode)->io_tree,
491 start, end, NULL,
492 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 493 EXTENT_CLEAR_DELALLOC |
a791e35e 494 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 495
496 btrfs_end_transaction(trans, root);
c8b97818
CM		 497 goto free_pages_out;
498 }
c2167754 499 btrfs_end_transaction(trans, root);
c8b97818
CM		 500 }
501
502 if (will_compress) {
503 /*
504 * we aren't doing an inline extent round the compressed size
505 * up to a block size boundary so the allocator does sane
506 * things
507 */
fda2832f 508 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 509
510 /*
511 * one last check to make sure the compression is really a
512 * win, compare the page count read with the blocks on disk
513 */
fda2832f 514 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 515 if (total_compressed >= total_in) {
516 will_compress = 0;
517 } else {
c8b97818
CM		 518 num_bytes = total_in;
519 }
520 }
521 if (!will_compress && pages) {
522 /*
523 * the compression code ran but failed to make things smaller,
524 * free any pages it allocated and our page pointer array
525 */
526 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 527 WARN_ON(pages[i]->mapping);
c8b97818
CM		 528 page_cache_release(pages[i]);
529 }
530 kfree(pages);
531 pages = NULL;
532 total_compressed = 0;
533 nr_pages_ret = 0;
534
535 /* flag the file so we don't compress in the future */
1e701a32
CM		 536 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
537 !(BTRFS_I(inode)->force_compress)) {
a555f810 538 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 539 }
c8b97818 540 }
771ed689
CM		 541 if (will_compress) {
542 *num_added += 1;
c8b97818 543
771ed689
CM		 544 /* the async work queues will take care of doing actual
545 * allocation on disk for these compressed pages,
546 * and will submit them to the elevator.
547 */
548 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 549 total_compressed, pages, nr_pages_ret,
550 compress_type);
179e29e4 551
24ae6365 552 if (start + num_bytes < end) {
771ed689
CM		 553 start += num_bytes;
554 pages = NULL;
555 cond_resched();
556 goto again;
557 }
558 } else {
f03d9301 559cleanup_and_bail_uncompressed:
771ed689
CM		 560 /*
561 * No compression, but we still need to write the pages in
562 * the file we've been given so far. redirty the locked
563 * page if it corresponds to our extent and set things up
564 * for the async work queue to run cow_file_range to do
565 * the normal delalloc dance
566 */
567 if (page_offset(locked_page) >= start &&
568 page_offset(locked_page) <= end) {
569 __set_page_dirty_nobuffers(locked_page);
570 /* unlocked later on in the async handlers */
571 }
4adaa611
CM		 572 if (redirty)
573 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 574 add_async_extent(async_cow, start, end - start + 1,
575 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 576 *num_added += 1;
577 }
3b951516 578
771ed689 579out:
79787eaa 580 return ret;
771ed689
CM		 581
582free_pages_out:
583 for (i = 0; i < nr_pages_ret; i++) {
584 WARN_ON(pages[i]->mapping);
585 page_cache_release(pages[i]);
586 }
d397712b 587 kfree(pages);
771ed689
CM		 588
589 goto out;
79787eaa
JM		 590
591cleanup_and_out:
592 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
593 start, end, NULL,
594 EXTENT_CLEAR_UNLOCK_PAGE |
595 EXTENT_CLEAR_DIRTY |
596 EXTENT_CLEAR_DELALLOC |
597 EXTENT_SET_WRITEBACK |
598 EXTENT_END_WRITEBACK);
599 if (!trans || IS_ERR(trans))
600 btrfs_error(root->fs_info, ret, "Failed to join transaction");
601 else
602 btrfs_abort_transaction(trans, root, ret);
603 goto free_pages_out;
771ed689
CM		 604}
605
606/*
607 * phase two of compressed writeback. This is the ordered portion
608 * of the code, which only gets called in the order the work was
609 * queued. We walk all the async extents created by compress_file_range
610 * and send them down to the disk.
611 */
612static noinline int submit_compressed_extents(struct inode *inode,
613 struct async_cow *async_cow)
614{
615 struct async_extent *async_extent;
616 u64 alloc_hint = 0;
617 struct btrfs_trans_handle *trans;
618 struct btrfs_key ins;
619 struct extent_map *em;
620 struct btrfs_root *root = BTRFS_I(inode)->root;
621 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
622 struct extent_io_tree *io_tree;
f5a84ee3 623 int ret = 0;
771ed689
CM		 624
625 if (list_empty(&async_cow->extents))
626 return 0;
627
3e04e7f1 628again:
d397712b 629 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 630 async_extent = list_entry(async_cow->extents.next,
631 struct async_extent, list);
632 list_del(&async_extent->list);
c8b97818 633
771ed689
CM		 634 io_tree = &BTRFS_I(inode)->io_tree;
635
f5a84ee3 636retry:
771ed689
CM		 637 /* did the compression code fall back to uncompressed IO? */
638 if (!async_extent->pages) {
639 int page_started = 0;
640 unsigned long nr_written = 0;
641
642 lock_extent(io_tree, async_extent->start,
2ac55d41 643 async_extent->start +
d0082371 644 async_extent->ram_size - 1);
771ed689
CM		 645
646 /* allocate blocks */
f5a84ee3
JB		 647 ret = cow_file_range(inode, async_cow->locked_page,
648 async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 &page_started, &nr_written, 0);
771ed689 652
79787eaa
JM		 653 /* JDM XXX */
654
771ed689
CM		 655 /*
656 * if page_started, cow_file_range inserted an
657 * inline extent and took care of all the unlocking
658 * and IO for us. Otherwise, we need to submit
659 * all those pages down to the drive.
660 */
f5a84ee3 661 if (!page_started && !ret)
771ed689
CM		 662 extent_write_locked_range(io_tree,
663 inode, async_extent->start,
d397712b 664 async_extent->start +
771ed689
CM		 665 async_extent->ram_size - 1,
666 btrfs_get_extent,
667 WB_SYNC_ALL);
3e04e7f1
JB		 668 else if (ret)
669 unlock_page(async_cow->locked_page);
771ed689
CM		 670 kfree(async_extent);
671 cond_resched();
672 continue;
673 }
674
675 lock_extent(io_tree, async_extent->start,
d0082371 676 async_extent->start + async_extent->ram_size - 1);
771ed689 677
7a7eaa40 678 trans = btrfs_join_transaction(root);
79787eaa
JM		 679 if (IS_ERR(trans)) {
680 ret = PTR_ERR(trans);
681 } else {
682 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
683 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 684 async_extent->compressed_size,
685 async_extent->compressed_size,
81c9ad23 686 0, alloc_hint, &ins, 1);
962197ba 687 if (ret && ret != -ENOSPC)
79787eaa
JM		 688 btrfs_abort_transaction(trans, root, ret);
689 btrfs_end_transaction(trans, root);
690 }
c2167754 691
f5a84ee3
JB		 692 if (ret) {
693 int i;
3e04e7f1 694
f5a84ee3
JB		 695 for (i = 0; i < async_extent->nr_pages; i++) {
696 WARN_ON(async_extent->pages[i]->mapping);
697 page_cache_release(async_extent->pages[i]);
698 }
699 kfree(async_extent->pages);
700 async_extent->nr_pages = 0;
701 async_extent->pages = NULL;
3e04e7f1 702
79787eaa
JM		 703 if (ret == -ENOSPC)
704 goto retry;
3e04e7f1 705 goto out_free;
f5a84ee3
JB		 706 }
707
c2167754
YZ		 708 /*
709 * here we're doing allocation and writeback of the
710 * compressed pages
711 */
712 btrfs_drop_extent_cache(inode, async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1, 0);
715
172ddd60 716 em = alloc_extent_map();
b9aa55be
LB		 717 if (!em) {
718 ret = -ENOMEM;
3e04e7f1 719 goto out_free_reserve;
b9aa55be 720 }
771ed689
CM		 721 em->start = async_extent->start;
722 em->len = async_extent->ram_size;
445a6944 723 em->orig_start = em->start;
2ab28f32
JB		 724 em->mod_start = em->start;
725 em->mod_len = em->len;
c8b97818 726
771ed689
CM		 727 em->block_start = ins.objectid;
728 em->block_len = ins.offset;
b4939680 729 em->orig_block_len = ins.offset;
cc95bef6 730 em->ram_bytes = async_extent->ram_size;
771ed689 731 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 732 em->compress_type = async_extent->compress_type;
771ed689
CM		 733 set_bit(EXTENT_FLAG_PINNED, &em->flags);
734 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 735 em->generation = -1;
771ed689 736
d397712b 737 while (1) {
890871be 738 write_lock(&em_tree->lock);
09a2a8f9 739 ret = add_extent_mapping(em_tree, em, 1);
890871be 740 write_unlock(&em_tree->lock);
771ed689
CM		 741 if (ret != -EEXIST) {
742 free_extent_map(em);
743 break;
744 }
745 btrfs_drop_extent_cache(inode, async_extent->start,
746 async_extent->start +
747 async_extent->ram_size - 1, 0);
748 }
749
3e04e7f1
JB		 750 if (ret)
751 goto out_free_reserve;
752
261507a0
LZ		 753 ret = btrfs_add_ordered_extent_compress(inode,
754 async_extent->start,
755 ins.objectid,
756 async_extent->ram_size,
757 ins.offset,
758 BTRFS_ORDERED_COMPRESSED,
759 async_extent->compress_type);
3e04e7f1
JB		 760 if (ret)
761 goto out_free_reserve;
771ed689 762
771ed689
CM		 763 /*
764 * clear dirty, set writeback and unlock the pages.
765 */
766 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 767 &BTRFS_I(inode)->io_tree,
768 async_extent->start,
769 async_extent->start +
770 async_extent->ram_size - 1,
771 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
772 EXTENT_CLEAR_UNLOCK |
a3429ab7 773 EXTENT_CLEAR_DELALLOC |
a791e35e 774 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 775
776 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 777 async_extent->start,
778 async_extent->ram_size,
779 ins.objectid,
780 ins.offset, async_extent->pages,
781 async_extent->nr_pages);
771ed689
CM		 782 alloc_hint = ins.objectid + ins.offset;
783 kfree(async_extent);
3e04e7f1
JB		 784 if (ret)
785 goto out;
771ed689
CM		 786 cond_resched();
787 }
79787eaa
JM		 788 ret = 0;
789out:
790 return ret;
3e04e7f1
JB		 791out_free_reserve:
792 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 793out_free:
3e04e7f1
JB		 794 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
795 async_extent->start,
796 async_extent->start +
797 async_extent->ram_size - 1,
798 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
801 EXTENT_CLEAR_DIRTY |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
79787eaa 804 kfree(async_extent);
3e04e7f1 805 goto again;
771ed689
CM		 806}
807
4b46fce2
JB		 808static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
809 u64 num_bytes)
810{
811 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
812 struct extent_map *em;
813 u64 alloc_hint = 0;
814
815 read_lock(&em_tree->lock);
816 em = search_extent_mapping(em_tree, start, num_bytes);
817 if (em) {
818 /*
819 * if block start isn't an actual block number then find the
820 * first block in this inode and use that as a hint. If that
821 * block is also bogus then just don't worry about it.
822 */
823 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
824 free_extent_map(em);
825 em = search_extent_mapping(em_tree, 0, 0);
826 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
827 alloc_hint = em->block_start;
828 if (em)
829 free_extent_map(em);
830 } else {
831 alloc_hint = em->block_start;
832 free_extent_map(em);
833 }
834 }
835 read_unlock(&em_tree->lock);
836
837 return alloc_hint;
838}
839
771ed689
CM		 840/*
841 * when extent_io.c finds a delayed allocation range in the file,
842 * the call backs end up in this code. The basic idea is to
843 * allocate extents on disk for the range, and create ordered data structs
844 * in ram to track those extents.
845 *
846 * locked_page is the page that writepage had locked already. We use
847 * it to make sure we don't do extra locks or unlocks.
848 *
849 * *page_started is set to one if we unlock locked_page and do everything
850 * required to start IO on it. It may be clean and already done with
851 * IO when we return.
852 */
b7d5b0a8
MX		 853static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
854 struct inode *inode,
855 struct btrfs_root *root,
856 struct page *locked_page,
857 u64 start, u64 end, int *page_started,
858 unsigned long *nr_written,
859 int unlock)
771ed689 860{
771ed689
CM		 861 u64 alloc_hint = 0;
862 u64 num_bytes;
863 unsigned long ram_size;
864 u64 disk_num_bytes;
865 u64 cur_alloc_size;
866 u64 blocksize = root->sectorsize;
771ed689
CM		 867 struct btrfs_key ins;
868 struct extent_map *em;
869 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
870 int ret = 0;
871
83eea1f1 872 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 873
fda2832f 874 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 875 num_bytes = max(blocksize, num_bytes);
876 disk_num_bytes = num_bytes;
771ed689 877
4cb5300b 878 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 879 if (num_bytes < 64 * 1024 &&
880 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 881 btrfs_add_inode_defrag(trans, inode);
882
771ed689
CM		 883 if (start == 0) {
884 /* lets try to make an inline extent */
885 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 886 start, end, 0, 0, NULL);
771ed689
CM		 887 if (ret == 0) {
888 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 889 &BTRFS_I(inode)->io_tree,
890 start, end, NULL,
891 EXTENT_CLEAR_UNLOCK_PAGE |
892 EXTENT_CLEAR_UNLOCK |
893 EXTENT_CLEAR_DELALLOC |
894 EXTENT_CLEAR_DIRTY |
895 EXTENT_SET_WRITEBACK |
896 EXTENT_END_WRITEBACK);
c2167754 897
771ed689
CM		 898 *nr_written = *nr_written +
899 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
900 *page_started = 1;
771ed689 901 goto out;
79787eaa
JM		 902 } else if (ret < 0) {
903 btrfs_abort_transaction(trans, root, ret);
904 goto out_unlock;
771ed689
CM		 905 }
906 }
907
908 BUG_ON(disk_num_bytes >
6c41761f 909 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 910
4b46fce2 911 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 912 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
913
d397712b 914 while (disk_num_bytes > 0) {
a791e35e
CM		 915 unsigned long op;
916
287a0ab9 917 cur_alloc_size = disk_num_bytes;
e6dcd2dc 918 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 919 root->sectorsize, 0, alloc_hint,
81c9ad23 920 &ins, 1);
79787eaa
JM		 921 if (ret < 0) {
922 btrfs_abort_transaction(trans, root, ret);
923 goto out_unlock;
924 }
d397712b 925
172ddd60 926 em = alloc_extent_map();
b9aa55be
LB		 927 if (!em) {
928 ret = -ENOMEM;
ace68bac 929 goto out_reserve;
b9aa55be 930 }
e6dcd2dc 931 em->start = start;
445a6944 932 em->orig_start = em->start;
771ed689
CM		 933 ram_size = ins.offset;
934 em->len = ins.offset;
2ab28f32
JB		 935 em->mod_start = em->start;
936 em->mod_len = em->len;
c8b97818 937
e6dcd2dc 938 em->block_start = ins.objectid;
c8b97818 939 em->block_len = ins.offset;
b4939680 940 em->orig_block_len = ins.offset;
cc95bef6 941 em->ram_bytes = ram_size;
e6dcd2dc 942 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 943 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 944 em->generation = -1;
c8b97818 945
d397712b 946 while (1) {
890871be 947 write_lock(&em_tree->lock);
09a2a8f9 948 ret = add_extent_mapping(em_tree, em, 1);
890871be 949 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 950 if (ret != -EEXIST) {
951 free_extent_map(em);
952 break;
953 }
954 btrfs_drop_extent_cache(inode, start,
c8b97818 955 start + ram_size - 1, 0);
e6dcd2dc 956 }
ace68bac
LB		 957 if (ret)
958 goto out_reserve;
e6dcd2dc 959
98d20f67 960 cur_alloc_size = ins.offset;
e6dcd2dc 961 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 962 ram_size, cur_alloc_size, 0);
ace68bac
LB		 963 if (ret)
964 goto out_reserve;
c8b97818 965
17d217fe
YZ		 966 if (root->root_key.objectid ==
967 BTRFS_DATA_RELOC_TREE_OBJECTID) {
968 ret = btrfs_reloc_clone_csums(inode, start,
969 cur_alloc_size);
79787eaa
JM		 970 if (ret) {
971 btrfs_abort_transaction(trans, root, ret);
ace68bac 972 goto out_reserve;
79787eaa 973 }
17d217fe
YZ		 974 }
975
d397712b 976 if (disk_num_bytes < cur_alloc_size)
3b951516 977 break;
d397712b 978
c8b97818
CM		 979 /* we're not doing compressed IO, don't unlock the first
980 * page (which the caller expects to stay locked), don't
981 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 982 *
983 * Do set the Private2 bit so we know this page was properly
984 * setup for writepage
c8b97818 985 */
a791e35e
CM		 986 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
987 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
988 EXTENT_SET_PRIVATE2;
989
c8b97818
CM		 990 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
991 start, start + ram_size - 1,
a791e35e 992 locked_page, op);
c8b97818 993 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 994 num_bytes -= cur_alloc_size;
995 alloc_hint = ins.objectid + ins.offset;
996 start += cur_alloc_size;
b888db2b 997 }
79787eaa 998out:
be20aa9d 999 return ret;
b7d5b0a8 1000
ace68bac
LB		 1001out_reserve:
1002 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa
JM		 1003out_unlock:
1004 extent_clear_unlock_delalloc(inode,
1005 &BTRFS_I(inode)->io_tree,
beb42dd7 1006 start, end, locked_page,
79787eaa
JM		 1007 EXTENT_CLEAR_UNLOCK_PAGE |
1008 EXTENT_CLEAR_UNLOCK |
1009 EXTENT_CLEAR_DELALLOC |
1010 EXTENT_CLEAR_DIRTY |
1011 EXTENT_SET_WRITEBACK |
1012 EXTENT_END_WRITEBACK);
1013
1014 goto out;
771ed689 1015}
c8b97818 1016
b7d5b0a8
MX		 1017static noinline int cow_file_range(struct inode *inode,
1018 struct page *locked_page,
1019 u64 start, u64 end, int *page_started,
1020 unsigned long *nr_written,
1021 int unlock)
1022{
1023 struct btrfs_trans_handle *trans;
1024 struct btrfs_root *root = BTRFS_I(inode)->root;
1025 int ret;
1026
1027 trans = btrfs_join_transaction(root);
1028 if (IS_ERR(trans)) {
1029 extent_clear_unlock_delalloc(inode,
1030 &BTRFS_I(inode)->io_tree,
1031 start, end, locked_page,
1032 EXTENT_CLEAR_UNLOCK_PAGE |
1033 EXTENT_CLEAR_UNLOCK |
1034 EXTENT_CLEAR_DELALLOC |
1035 EXTENT_CLEAR_DIRTY |
1036 EXTENT_SET_WRITEBACK |
1037 EXTENT_END_WRITEBACK);
1038 return PTR_ERR(trans);
1039 }
1040 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1041
1042 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043 page_started, nr_written, unlock);
1044
1045 btrfs_end_transaction(trans, root);
1046
1047 return ret;
1048}
1049
771ed689
CM		 1050/*
1051 * work queue call back to started compression on a file and pages
1052 */
1053static noinline void async_cow_start(struct btrfs_work *work)
1054{
1055 struct async_cow *async_cow;
1056 int num_added = 0;
1057 async_cow = container_of(work, struct async_cow, work);
1058
1059 compress_file_range(async_cow->inode, async_cow->locked_page,
1060 async_cow->start, async_cow->end, async_cow,
1061 &num_added);
8180ef88 1062 if (num_added == 0) {
cb77fcd8 1063 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1064 async_cow->inode = NULL;
8180ef88 1065 }
771ed689
CM		 1066}
1067
1068/*
1069 * work queue call back to submit previously compressed pages
1070 */
1071static noinline void async_cow_submit(struct btrfs_work *work)
1072{
1073 struct async_cow *async_cow;
1074 struct btrfs_root *root;
1075 unsigned long nr_pages;
1076
1077 async_cow = container_of(work, struct async_cow, work);
1078
1079 root = async_cow->root;
1080 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1081 PAGE_CACHE_SHIFT;
1082
66657b31 1083 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1084 5 * 1024 * 1024 &&
771ed689
CM		 1085 waitqueue_active(&root->fs_info->async_submit_wait))
1086 wake_up(&root->fs_info->async_submit_wait);
1087
d397712b 1088 if (async_cow->inode)
771ed689 1089 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1090}
c8b97818 1091
771ed689
CM		 1092static noinline void async_cow_free(struct btrfs_work *work)
1093{
1094 struct async_cow *async_cow;
1095 async_cow = container_of(work, struct async_cow, work);
8180ef88 1096 if (async_cow->inode)
cb77fcd8 1097 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1098 kfree(async_cow);
1099}
1100
1101static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102 u64 start, u64 end, int *page_started,
1103 unsigned long *nr_written)
1104{
1105 struct async_cow *async_cow;
1106 struct btrfs_root *root = BTRFS_I(inode)->root;
1107 unsigned long nr_pages;
1108 u64 cur_end;
287082b0 1109 int limit = 10 * 1024 * 1024;
771ed689 1110
a3429ab7
CM		 1111 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112 1, 0, NULL, GFP_NOFS);
d397712b 1113 while (start < end) {
771ed689 1114 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1115 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1116 async_cow->inode = igrab(inode);
771ed689
CM		 1117 async_cow->root = root;
1118 async_cow->locked_page = locked_page;
1119 async_cow->start = start;
1120
6cbff00f 1121 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1122 cur_end = end;
1123 else
1124 cur_end = min(end, start + 512 * 1024 - 1);
1125
1126 async_cow->end = cur_end;
1127 INIT_LIST_HEAD(&async_cow->extents);
1128
1129 async_cow->work.func = async_cow_start;
1130 async_cow->work.ordered_func = async_cow_submit;
1131 async_cow->work.ordered_free = async_cow_free;
1132 async_cow->work.flags = 0;
1133
771ed689
CM		 1134 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1135 PAGE_CACHE_SHIFT;
1136 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1137
1138 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1139 &async_cow->work);
1140
1141 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142 wait_event(root->fs_info->async_submit_wait,
1143 (atomic_read(&root->fs_info->async_delalloc_pages) <
1144 limit));
1145 }
1146
d397712b 1147 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1148 atomic_read(&root->fs_info->async_delalloc_pages)) {
1149 wait_event(root->fs_info->async_submit_wait,
1150 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1151 0));
1152 }
1153
1154 *nr_written += nr_pages;
1155 start = cur_end + 1;
1156 }
1157 *page_started = 1;
1158 return 0;
be20aa9d
CM		 1159}
1160
d397712b 1161static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1162 u64 bytenr, u64 num_bytes)
1163{
1164 int ret;
1165 struct btrfs_ordered_sum *sums;
1166 LIST_HEAD(list);
1167
07d400a6 1168 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1169 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1170 if (ret == 0 && list_empty(&list))
1171 return 0;
1172
1173 while (!list_empty(&list)) {
1174 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175 list_del(&sums->list);
1176 kfree(sums);
1177 }
1178 return 1;
1179}
1180
d352ac68
CM		 1181/*
1182 * when nowcow writeback call back. This checks for snapshots or COW copies
1183 * of the extents that exist in the file, and COWs the file as required.
1184 *
1185 * If no cow copies or snapshots exist, we write directly to the existing
1186 * blocks on disk
1187 */
7f366cfe
CM		 1188static noinline int run_delalloc_nocow(struct inode *inode,
1189 struct page *locked_page,
771ed689
CM		 1190 u64 start, u64 end, int *page_started, int force,
1191 unsigned long *nr_written)
be20aa9d 1192{
be20aa9d 1193 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1194 struct btrfs_trans_handle *trans;
be20aa9d 1195 struct extent_buffer *leaf;
be20aa9d 1196 struct btrfs_path *path;
80ff3856 1197 struct btrfs_file_extent_item *fi;
be20aa9d 1198 struct btrfs_key found_key;
80ff3856
YZ		 1199 u64 cow_start;
1200 u64 cur_offset;
1201 u64 extent_end;
5d4f98a2 1202 u64 extent_offset;
80ff3856
YZ		 1203 u64 disk_bytenr;
1204 u64 num_bytes;
b4939680 1205 u64 disk_num_bytes;
cc95bef6 1206 u64 ram_bytes;
80ff3856 1207 int extent_type;
79787eaa 1208 int ret, err;
d899e052 1209 int type;
80ff3856
YZ		 1210 int nocow;
1211 int check_prev = 1;
82d5902d 1212 bool nolock;
33345d01 1213 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1214
1215 path = btrfs_alloc_path();
17ca04af
JB		 1216 if (!path) {
1217 extent_clear_unlock_delalloc(inode,
1218 &BTRFS_I(inode)->io_tree,
1219 start, end, locked_page,
1220 EXTENT_CLEAR_UNLOCK_PAGE |
1221 EXTENT_CLEAR_UNLOCK |
1222 EXTENT_CLEAR_DELALLOC |
1223 EXTENT_CLEAR_DIRTY |
1224 EXTENT_SET_WRITEBACK |
1225 EXTENT_END_WRITEBACK);
d8926bb3 1226 return -ENOMEM;
17ca04af 1227 }
82d5902d 1228
83eea1f1 1229 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1230
1231 if (nolock)
7a7eaa40 1232 trans = btrfs_join_transaction_nolock(root);
82d5902d 1233 else
7a7eaa40 1234 trans = btrfs_join_transaction(root);
ff5714cc 1235
79787eaa 1236 if (IS_ERR(trans)) {
17ca04af
JB		 1237 extent_clear_unlock_delalloc(inode,
1238 &BTRFS_I(inode)->io_tree,
1239 start, end, locked_page,
1240 EXTENT_CLEAR_UNLOCK_PAGE |
1241 EXTENT_CLEAR_UNLOCK |
1242 EXTENT_CLEAR_DELALLOC |
1243 EXTENT_CLEAR_DIRTY |
1244 EXTENT_SET_WRITEBACK |
1245 EXTENT_END_WRITEBACK);
79787eaa
JM		 1246 btrfs_free_path(path);
1247 return PTR_ERR(trans);
1248 }
1249
74b21075 1250 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1251
80ff3856
YZ		 1252 cow_start = (u64)-1;
1253 cur_offset = start;
1254 while (1) {
33345d01 1255 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1256 cur_offset, 0);
79787eaa
JM		 1257 if (ret < 0) {
1258 btrfs_abort_transaction(trans, root, ret);
1259 goto error;
1260 }
80ff3856
YZ		 1261 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &found_key,
1264 path->slots[0] - 1);
33345d01 1265 if (found_key.objectid == ino &&
80ff3856
YZ		 1266 found_key.type == BTRFS_EXTENT_DATA_KEY)
1267 path->slots[0]--;
1268 }
1269 check_prev = 0;
1270next_slot:
1271 leaf = path->nodes[0];
1272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1274 if (ret < 0) {
1275 btrfs_abort_transaction(trans, root, ret);
1276 goto error;
1277 }
80ff3856
YZ		 1278 if (ret > 0)
1279 break;
1280 leaf = path->nodes[0];
1281 }
be20aa9d 1282
80ff3856
YZ		 1283 nocow = 0;
1284 disk_bytenr = 0;
17d217fe 1285 num_bytes = 0;
80ff3856
YZ		 1286 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1287
33345d01 1288 if (found_key.objectid > ino ||
80ff3856
YZ		 1289 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290 found_key.offset > end)
1291 break;
1292
1293 if (found_key.offset > cur_offset) {
1294 extent_end = found_key.offset;
e9061e21 1295 extent_type = 0;
80ff3856
YZ		 1296 goto out_check;
1297 }
1298
1299 fi = btrfs_item_ptr(leaf, path->slots[0],
1300 struct btrfs_file_extent_item);
1301 extent_type = btrfs_file_extent_type(leaf, fi);
1302
cc95bef6 1303 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1304 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1306 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1307 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1308 extent_end = found_key.offset +
1309 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1310 disk_num_bytes =
1311 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1312 if (extent_end <= start) {
1313 path->slots[0]++;
1314 goto next_slot;
1315 }
17d217fe
YZ		 1316 if (disk_bytenr == 0)
1317 goto out_check;
80ff3856
YZ		 1318 if (btrfs_file_extent_compression(leaf, fi) ||
1319 btrfs_file_extent_encryption(leaf, fi) ||
1320 btrfs_file_extent_other_encoding(leaf, fi))
1321 goto out_check;
d899e052
YZ		 1322 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1323 goto out_check;
d2fb3437 1324 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1325 goto out_check;
33345d01 1326 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1327 found_key.offset -
1328 extent_offset, disk_bytenr))
17d217fe 1329 goto out_check;
5d4f98a2 1330 disk_bytenr += extent_offset;
17d217fe
YZ		 1331 disk_bytenr += cur_offset - found_key.offset;
1332 num_bytes = min(end + 1, extent_end) - cur_offset;
1333 /*
1334 * force cow if csum exists in the range.
1335 * this ensure that csum for a given extent are
1336 * either valid or do not exist.
1337 */
1338 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1339 goto out_check;
80ff3856
YZ		 1340 nocow = 1;
1341 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342 extent_end = found_key.offset +
1343 btrfs_file_extent_inline_len(leaf, fi);
1344 extent_end = ALIGN(extent_end, root->sectorsize);
1345 } else {
1346 BUG_ON(1);
1347 }
1348out_check:
1349 if (extent_end <= start) {
1350 path->slots[0]++;
1351 goto next_slot;
1352 }
1353 if (!nocow) {
1354 if (cow_start == (u64)-1)
1355 cow_start = cur_offset;
1356 cur_offset = extent_end;
1357 if (cur_offset > end)
1358 break;
1359 path->slots[0]++;
1360 goto next_slot;
7ea394f1
YZ		 1361 }
1362
b3b4aa74 1363 btrfs_release_path(path);
80ff3856 1364 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1365 ret = __cow_file_range(trans, inode, root, locked_page,
1366 cow_start, found_key.offset - 1,
1367 page_started, nr_written, 1);
79787eaa
JM		 1368 if (ret) {
1369 btrfs_abort_transaction(trans, root, ret);
1370 goto error;
1371 }
80ff3856 1372 cow_start = (u64)-1;
7ea394f1 1373 }
80ff3856 1374
d899e052
YZ		 1375 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376 struct extent_map *em;
1377 struct extent_map_tree *em_tree;
1378 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1379 em = alloc_extent_map();
79787eaa 1380 BUG_ON(!em); /* -ENOMEM */
d899e052 1381 em->start = cur_offset;
70c8a91c 1382 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1383 em->len = num_bytes;
1384 em->block_len = num_bytes;
1385 em->block_start = disk_bytenr;
b4939680 1386 em->orig_block_len = disk_num_bytes;
cc95bef6 1387 em->ram_bytes = ram_bytes;
d899e052 1388 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1389 em->mod_start = em->start;
1390 em->mod_len = em->len;
d899e052 1391 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1392 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1393 em->generation = -1;
d899e052 1394 while (1) {
890871be 1395 write_lock(&em_tree->lock);
09a2a8f9 1396 ret = add_extent_mapping(em_tree, em, 1);
890871be 1397 write_unlock(&em_tree->lock);
d899e052
YZ		 1398 if (ret != -EEXIST) {
1399 free_extent_map(em);
1400 break;
1401 }
1402 btrfs_drop_extent_cache(inode, em->start,
1403 em->start + em->len - 1, 0);
1404 }
1405 type = BTRFS_ORDERED_PREALLOC;
1406 } else {
1407 type = BTRFS_ORDERED_NOCOW;
1408 }
80ff3856
YZ		 1409
1410 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1411 num_bytes, num_bytes, type);
79787eaa 1412 BUG_ON(ret); /* -ENOMEM */
771ed689 1413
efa56464
YZ		 1414 if (root->root_key.objectid ==
1415 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1417 num_bytes);
79787eaa
JM		 1418 if (ret) {
1419 btrfs_abort_transaction(trans, root, ret);
1420 goto error;
1421 }
efa56464
YZ		 1422 }
1423
d899e052 1424 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1425 cur_offset, cur_offset + num_bytes - 1,
1426 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1429 cur_offset = extent_end;
1430 if (cur_offset > end)
1431 break;
be20aa9d 1432 }
b3b4aa74 1433 btrfs_release_path(path);
80ff3856 1434
17ca04af 1435 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1436 cow_start = cur_offset;
17ca04af
JB		 1437 cur_offset = end;
1438 }
1439
80ff3856 1440 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1441 ret = __cow_file_range(trans, inode, root, locked_page,
1442 cow_start, end,
1443 page_started, nr_written, 1);
79787eaa
JM		 1444 if (ret) {
1445 btrfs_abort_transaction(trans, root, ret);
1446 goto error;
1447 }
80ff3856
YZ		 1448 }
1449
79787eaa 1450error:
a698d075 1451 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1452 if (!ret)
1453 ret = err;
1454
17ca04af
JB		 1455 if (ret && cur_offset < end)
1456 extent_clear_unlock_delalloc(inode,
1457 &BTRFS_I(inode)->io_tree,
1458 cur_offset, end, locked_page,
1459 EXTENT_CLEAR_UNLOCK_PAGE |
1460 EXTENT_CLEAR_UNLOCK |
1461 EXTENT_CLEAR_DELALLOC |
1462 EXTENT_CLEAR_DIRTY |
1463 EXTENT_SET_WRITEBACK |
1464 EXTENT_END_WRITEBACK);
1465
7ea394f1 1466 btrfs_free_path(path);
79787eaa 1467 return ret;
be20aa9d
CM		 1468}
1469
d352ac68
CM		 1470/*
1471 * extent_io.c call back to do delayed allocation processing
1472 */
c8b97818 1473static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1474 u64 start, u64 end, int *page_started,
1475 unsigned long *nr_written)
be20aa9d 1476{
be20aa9d 1477 int ret;
7f366cfe 1478 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1479
7ddf5a42 1480 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1481 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1482 page_started, 1, nr_written);
7ddf5a42 1483 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1484 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1485 page_started, 0, nr_written);
7ddf5a42
JB		 1486 } else if (!btrfs_test_opt(root, COMPRESS) &&
1487 !(BTRFS_I(inode)->force_compress) &&
1488 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1489 ret = cow_file_range(inode, locked_page, start, end,
1490 page_started, nr_written, 1);
7ddf5a42
JB		 1491 } else {
1492 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493 &BTRFS_I(inode)->runtime_flags);
771ed689 1494 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1495 page_started, nr_written);
7ddf5a42 1496 }
b888db2b
CM		 1497 return ret;
1498}
1499
1bf85046
JM		 1500static void btrfs_split_extent_hook(struct inode *inode,
1501 struct extent_state *orig, u64 split)
9ed74f2d 1502{
0ca1f7ce 1503 /* not delalloc, ignore it */
9ed74f2d 1504 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1505 return;
9ed74f2d 1506
9e0baf60
JB		 1507 spin_lock(&BTRFS_I(inode)->lock);
1508 BTRFS_I(inode)->outstanding_extents++;
1509 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1510}
1511
1512/*
1513 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514 * extents so we can keep track of new extents that are just merged onto old
1515 * extents, such as when we are doing sequential writes, so we can properly
1516 * account for the metadata space we'll need.
1517 */
1bf85046
JM		 1518static void btrfs_merge_extent_hook(struct inode *inode,
1519 struct extent_state *new,
1520 struct extent_state *other)
9ed74f2d 1521{
9ed74f2d
JB		 1522 /* not delalloc, ignore it */
1523 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1524 return;
9ed74f2d 1525
9e0baf60
JB		 1526 spin_lock(&BTRFS_I(inode)->lock);
1527 BTRFS_I(inode)->outstanding_extents--;
1528 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1529}
1530
d352ac68
CM		 1531/*
1532 * extent_io.c set_bit_hook, used to track delayed allocation
1533 * bytes in this file, and to maintain the list of inodes that
1534 * have pending delalloc work to be done.
1535 */
1bf85046 1536static void btrfs_set_bit_hook(struct inode *inode,
41074888 1537 struct extent_state *state, unsigned long *bits)
291d673e 1538{
9ed74f2d 1539
75eff68e
CM		 1540 /*
1541 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1542 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1543 * bit, which is only set or cleared with irqs on
1544 */
0ca1f7ce 1545 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1546 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1547 u64 len = state->end + 1 - state->start;
83eea1f1 1548 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1549
9e0baf60 1550 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1551 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1552 } else {
1553 spin_lock(&BTRFS_I(inode)->lock);
1554 BTRFS_I(inode)->outstanding_extents++;
1555 spin_unlock(&BTRFS_I(inode)->lock);
1556 }
287a0ab9 1557
963d678b
MX		 1558 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1559 root->fs_info->delalloc_batch);
df0af1a5 1560 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1561 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5
MX		 1562 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1563 &BTRFS_I(inode)->runtime_flags)) {
1564 spin_lock(&root->fs_info->delalloc_lock);
1565 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567 &root->fs_info->delalloc_inodes);
1568 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569 &BTRFS_I(inode)->runtime_flags);
1570 }
1571 spin_unlock(&root->fs_info->delalloc_lock);
ea8c2819 1572 }
df0af1a5 1573 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1574 }
291d673e
CM		 1575}
1576
d352ac68
CM		 1577/*
1578 * extent_io.c clear_bit_hook, see set_bit_hook for why
1579 */
1bf85046 1580static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1581 struct extent_state *state,
1582 unsigned long *bits)
291d673e 1583{
75eff68e
CM		 1584 /*
1585 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1586 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1587 * bit, which is only set or cleared with irqs on
1588 */
0ca1f7ce 1589 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1590 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1591 u64 len = state->end + 1 - state->start;
83eea1f1 1592 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1593
9e0baf60 1594 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1595 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1596 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1597 spin_lock(&BTRFS_I(inode)->lock);
1598 BTRFS_I(inode)->outstanding_extents--;
1599 spin_unlock(&BTRFS_I(inode)->lock);
1600 }
0ca1f7ce
YZ		 1601
1602 if (*bits & EXTENT_DO_ACCOUNTING)
1603 btrfs_delalloc_release_metadata(inode, len);
1604
0cb59c99
JB		 1605 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1606 && do_list)
0ca1f7ce 1607 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1608
963d678b
MX		 1609 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1610 root->fs_info->delalloc_batch);
df0af1a5 1611 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1612 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1613 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5
MX		 1614 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1615 &BTRFS_I(inode)->runtime_flags)) {
1616 spin_lock(&root->fs_info->delalloc_lock);
1617 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1618 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1619 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1620 &BTRFS_I(inode)->runtime_flags);
1621 }
1622 spin_unlock(&root->fs_info->delalloc_lock);
ea8c2819 1623 }
df0af1a5 1624 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1625 }
291d673e
CM		 1626}
1627
d352ac68
CM		 1628/*
1629 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1630 * we don't create bios that span stripes or chunks
1631 */
64a16701 1632int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1633 size_t size, struct bio *bio,
1634 unsigned long bio_flags)
239b14b3
CM		 1635{
1636 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1637 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1638 u64 length = 0;
1639 u64 map_length;
239b14b3
CM		 1640 int ret;
1641
771ed689
CM		 1642 if (bio_flags & EXTENT_BIO_COMPRESSED)
1643 return 0;
1644
f2d8d74d 1645 length = bio->bi_size;
239b14b3 1646 map_length = length;
64a16701 1647 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1648 &map_length, NULL, 0);
3ec706c8 1649 /* Will always return 0 with map_multi == NULL */
3444a972 1650 BUG_ON(ret < 0);
d397712b 1651 if (map_length < length + size)
239b14b3 1652 return 1;
3444a972 1653 return 0;
239b14b3
CM		 1654}
1655
d352ac68
CM		 1656/*
1657 * in order to insert checksums into the metadata in large chunks,
1658 * we wait until bio submission time. All the pages in the bio are
1659 * checksummed and sums are attached onto the ordered extent record.
1660 *
1661 * At IO completion time the cums attached on the ordered extent record
1662 * are inserted into the btree
1663 */
d397712b
CM		 1664static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1665 struct bio *bio, int mirror_num,
eaf25d93
CM		 1666 unsigned long bio_flags,
1667 u64 bio_offset)
065631f6 1668{
065631f6 1669 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1670 int ret = 0;
e015640f 1671
d20f7043 1672 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1673 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1674 return 0;
1675}
e015640f 1676
4a69a410
CM		 1677/*
1678 * in order to insert checksums into the metadata in large chunks,
1679 * we wait until bio submission time. All the pages in the bio are
1680 * checksummed and sums are attached onto the ordered extent record.
1681 *
1682 * At IO completion time the cums attached on the ordered extent record
1683 * are inserted into the btree
1684 */
b2950863 1685static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1686 int mirror_num, unsigned long bio_flags,
1687 u64 bio_offset)
4a69a410
CM		 1688{
1689 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1690 int ret;
1691
1692 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1693 if (ret)
1694 bio_endio(bio, ret);
1695 return ret;
44b8bd7e
CM		 1696}
1697
d352ac68 1698/*
cad321ad
CM		 1699 * extent_io.c submission hook. This does the right thing for csum calculation
1700 * on write, or reading the csums from the tree before a read
d352ac68 1701 */
b2950863 1702static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1703 int mirror_num, unsigned long bio_flags,
1704 u64 bio_offset)
44b8bd7e
CM		 1705{
1706 struct btrfs_root *root = BTRFS_I(inode)->root;
1707 int ret = 0;
19b9bdb0 1708 int skip_sum;
0417341e 1709 int metadata = 0;
b812ce28 1710 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1711
6cbff00f 1712 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1713
83eea1f1 1714 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1715 metadata = 2;
1716
7b6d91da 1717 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1718 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1719 if (ret)
61891923 1720 goto out;
5fd02043 1721
d20f7043 1722 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1723 ret = btrfs_submit_compressed_read(inode, bio,
1724 mirror_num,
1725 bio_flags);
1726 goto out;
c2db1073
TI		 1727 } else if (!skip_sum) {
1728 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1729 if (ret)
61891923 1730 goto out;
c2db1073 1731 }
4d1b5fb4 1732 goto mapit;
b812ce28 1733 } else if (async && !skip_sum) {
17d217fe
YZ		 1734 /* csum items have already been cloned */
1735 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1736 goto mapit;
19b9bdb0 1737 /* we're doing a write, do the async checksumming */
61891923 1738 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1739 inode, rw, bio, mirror_num,
eaf25d93
CM		 1740 bio_flags, bio_offset,
1741 __btrfs_submit_bio_start,
4a69a410 1742 __btrfs_submit_bio_done);
61891923 1743 goto out;
b812ce28
JB		 1744 } else if (!skip_sum) {
1745 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1746 if (ret)
1747 goto out;
19b9bdb0
CM		 1748 }
1749
0b86a832 1750mapit:
61891923
SB		 1751 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1752
1753out:
1754 if (ret < 0)
1755 bio_endio(bio, ret);
1756 return ret;
065631f6 1757}
6885f308 1758
d352ac68
CM		 1759/*
1760 * given a list of ordered sums record them in the inode. This happens
1761 * at IO completion time based on sums calculated at bio submission time.
1762 */
ba1da2f4 1763static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1764 struct inode *inode, u64 file_offset,
1765 struct list_head *list)
1766{
e6dcd2dc
CM		 1767 struct btrfs_ordered_sum *sum;
1768
c6e30871 1769 list_for_each_entry(sum, list, list) {
39847c4d 1770 trans->adding_csums = 1;
d20f7043
CM		 1771 btrfs_csum_file_blocks(trans,
1772 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1773 trans->adding_csums = 0;
e6dcd2dc
CM		 1774 }
1775 return 0;
1776}
1777
2ac55d41
JB		 1778int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1779 struct extent_state **cached_state)
ea8c2819 1780{
6c1500f2 1781 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1782 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1783 cached_state, GFP_NOFS);
ea8c2819
CM		 1784}
1785
d352ac68 1786/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1787struct btrfs_writepage_fixup {
1788 struct page *page;
1789 struct btrfs_work work;
1790};
1791
b2950863 1792static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1793{
1794 struct btrfs_writepage_fixup *fixup;
1795 struct btrfs_ordered_extent *ordered;
2ac55d41 1796 struct extent_state *cached_state = NULL;
247e743c
CM		 1797 struct page *page;
1798 struct inode *inode;
1799 u64 page_start;
1800 u64 page_end;
87826df0 1801 int ret;
247e743c
CM		 1802
1803 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1804 page = fixup->page;
4a096752 1805again:
247e743c
CM		 1806 lock_page(page);
1807 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1808 ClearPageChecked(page);
1809 goto out_page;
1810 }
1811
1812 inode = page->mapping->host;
1813 page_start = page_offset(page);
1814 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1815
2ac55d41 1816 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1817 &cached_state);
4a096752
CM		 1818
1819 /* already ordered? We're done */
8b62b72b 1820 if (PagePrivate2(page))
247e743c 1821 goto out;
4a096752
CM		 1822
1823 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1824 if (ordered) {
2ac55d41
JB		 1825 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1826 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1827 unlock_page(page);
1828 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1829 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1830 goto again;
1831 }
247e743c 1832
87826df0
JM		 1833 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1834 if (ret) {
1835 mapping_set_error(page->mapping, ret);
1836 end_extent_writepage(page, ret, page_start, page_end);
1837 ClearPageChecked(page);
1838 goto out;
1839 }
1840
2ac55d41 1841 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1842 ClearPageChecked(page);
87826df0 1843 set_page_dirty(page);
247e743c 1844out:
2ac55d41
JB		 1845 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1846 &cached_state, GFP_NOFS);
247e743c
CM		 1847out_page:
1848 unlock_page(page);
1849 page_cache_release(page);
b897abec 1850 kfree(fixup);
247e743c
CM		 1851}
1852
1853/*
1854 * There are a few paths in the higher layers of the kernel that directly
1855 * set the page dirty bit without asking the filesystem if it is a
1856 * good idea. This causes problems because we want to make sure COW
1857 * properly happens and the data=ordered rules are followed.
1858 *
c8b97818 1859 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1860 * hasn't been properly setup for IO. We kick off an async process
1861 * to fix it up. The async helper will wait for ordered extents, set
1862 * the delalloc bit and make it safe to write the page.
1863 */
b2950863 1864static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1865{
1866 struct inode *inode = page->mapping->host;
1867 struct btrfs_writepage_fixup *fixup;
1868 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1869
8b62b72b
CM		 1870 /* this page is properly in the ordered list */
1871 if (TestClearPagePrivate2(page))
247e743c
CM		 1872 return 0;
1873
1874 if (PageChecked(page))
1875 return -EAGAIN;
1876
1877 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1878 if (!fixup)
1879 return -EAGAIN;
f421950f 1880
247e743c
CM		 1881 SetPageChecked(page);
1882 page_cache_get(page);
1883 fixup->work.func = btrfs_writepage_fixup_worker;
1884 fixup->page = page;
1885 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1886 return -EBUSY;
247e743c
CM		 1887}
1888
d899e052
YZ		 1889static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1890 struct inode *inode, u64 file_pos,
1891 u64 disk_bytenr, u64 disk_num_bytes,
1892 u64 num_bytes, u64 ram_bytes,
1893 u8 compression, u8 encryption,
1894 u16 other_encoding, int extent_type)
1895{
1896 struct btrfs_root *root = BTRFS_I(inode)->root;
1897 struct btrfs_file_extent_item *fi;
1898 struct btrfs_path *path;
1899 struct extent_buffer *leaf;
1900 struct btrfs_key ins;
d899e052
YZ		 1901 int ret;
1902
1903 path = btrfs_alloc_path();
d8926bb3
MF		 1904 if (!path)
1905 return -ENOMEM;
d899e052 1906
b9473439 1907 path->leave_spinning = 1;
a1ed835e
CM		 1908
1909 /*
1910 * we may be replacing one extent in the tree with another.
1911 * The new extent is pinned in the extent map, and we don't want
1912 * to drop it from the cache until it is completely in the btree.
1913 *
1914 * So, tell btrfs_drop_extents to leave this extent in the cache.
1915 * the caller is expected to unpin it and allow it to be merged
1916 * with the others.
1917 */
5dc562c5 1918 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1919 file_pos + num_bytes, 0);
79787eaa
JM		 1920 if (ret)
1921 goto out;
d899e052 1922
33345d01 1923 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1924 ins.offset = file_pos;
1925 ins.type = BTRFS_EXTENT_DATA_KEY;
1926 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1927 if (ret)
1928 goto out;
d899e052
YZ		 1929 leaf = path->nodes[0];
1930 fi = btrfs_item_ptr(leaf, path->slots[0],
1931 struct btrfs_file_extent_item);
1932 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1933 btrfs_set_file_extent_type(leaf, fi, extent_type);
1934 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1935 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1936 btrfs_set_file_extent_offset(leaf, fi, 0);
1937 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1938 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1939 btrfs_set_file_extent_compression(leaf, fi, compression);
1940 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1941 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1942
d899e052 1943 btrfs_mark_buffer_dirty(leaf);
ce195332 1944 btrfs_release_path(path);
d899e052
YZ		 1945
1946 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1947
1948 ins.objectid = disk_bytenr;
1949 ins.offset = disk_num_bytes;
1950 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1951 ret = btrfs_alloc_reserved_file_extent(trans, root,
1952 root->root_key.objectid,
33345d01 1953 btrfs_ino(inode), file_pos, &ins);
79787eaa 1954out:
d899e052 1955 btrfs_free_path(path);
b9473439 1956
79787eaa 1957 return ret;
d899e052
YZ		 1958}
1959
38c227d8
LB		 1960/* snapshot-aware defrag */
1961struct sa_defrag_extent_backref {
1962 struct rb_node node;
1963 struct old_sa_defrag_extent *old;
1964 u64 root_id;
1965 u64 inum;
1966 u64 file_pos;
1967 u64 extent_offset;
1968 u64 num_bytes;
1969 u64 generation;
1970};
1971
1972struct old_sa_defrag_extent {
1973 struct list_head list;
1974 struct new_sa_defrag_extent *new;
1975
1976 u64 extent_offset;
1977 u64 bytenr;
1978 u64 offset;
1979 u64 len;
1980 int count;
1981};
1982
1983struct new_sa_defrag_extent {
1984 struct rb_root root;
1985 struct list_head head;
1986 struct btrfs_path *path;
1987 struct inode *inode;
1988 u64 file_pos;
1989 u64 len;
1990 u64 bytenr;
1991 u64 disk_len;
1992 u8 compress_type;
1993};
1994
1995static int backref_comp(struct sa_defrag_extent_backref *b1,
1996 struct sa_defrag_extent_backref *b2)
1997{
1998 if (b1->root_id < b2->root_id)
1999 return -1;
2000 else if (b1->root_id > b2->root_id)
2001 return 1;
2002
2003 if (b1->inum < b2->inum)
2004 return -1;
2005 else if (b1->inum > b2->inum)
2006 return 1;
2007
2008 if (b1->file_pos < b2->file_pos)
2009 return -1;
2010 else if (b1->file_pos > b2->file_pos)
2011 return 1;
2012
2013 /*
2014 * [------------------------------] ===> (a range of space)
2015 * |<--->| |<---->| =============> (fs/file tree A)
2016 * |<---------------------------->| ===> (fs/file tree B)
2017 *
2018 * A range of space can refer to two file extents in one tree while
2019 * refer to only one file extent in another tree.
2020 *
2021 * So we may process a disk offset more than one time(two extents in A)
2022 * and locate at the same extent(one extent in B), then insert two same
2023 * backrefs(both refer to the extent in B).
2024 */
2025 return 0;
2026}
2027
2028static void backref_insert(struct rb_root *root,
2029 struct sa_defrag_extent_backref *backref)
2030{
2031 struct rb_node **p = &root->rb_node;
2032 struct rb_node *parent = NULL;
2033 struct sa_defrag_extent_backref *entry;
2034 int ret;
2035
2036 while (*p) {
2037 parent = *p;
2038 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2039
2040 ret = backref_comp(backref, entry);
2041 if (ret < 0)
2042 p = &(*p)->rb_left;
2043 else
2044 p = &(*p)->rb_right;
2045 }
2046
2047 rb_link_node(&backref->node, parent, p);
2048 rb_insert_color(&backref->node, root);
2049}
2050
2051/*
2052 * Note the backref might has changed, and in this case we just return 0.
2053 */
2054static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2055 void *ctx)
2056{
2057 struct btrfs_file_extent_item *extent;
2058 struct btrfs_fs_info *fs_info;
2059 struct old_sa_defrag_extent *old = ctx;
2060 struct new_sa_defrag_extent *new = old->new;
2061 struct btrfs_path *path = new->path;
2062 struct btrfs_key key;
2063 struct btrfs_root *root;
2064 struct sa_defrag_extent_backref *backref;
2065 struct extent_buffer *leaf;
2066 struct inode *inode = new->inode;
2067 int slot;
2068 int ret;
2069 u64 extent_offset;
2070 u64 num_bytes;
2071
2072 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2073 inum == btrfs_ino(inode))
2074 return 0;
2075
2076 key.objectid = root_id;
2077 key.type = BTRFS_ROOT_ITEM_KEY;
2078 key.offset = (u64)-1;
2079
2080 fs_info = BTRFS_I(inode)->root->fs_info;
2081 root = btrfs_read_fs_root_no_name(fs_info, &key);
2082 if (IS_ERR(root)) {
2083 if (PTR_ERR(root) == -ENOENT)
2084 return 0;
2085 WARN_ON(1);
2086 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2087 inum, offset, root_id);
2088 return PTR_ERR(root);
2089 }
2090
2091 key.objectid = inum;
2092 key.type = BTRFS_EXTENT_DATA_KEY;
2093 if (offset > (u64)-1 << 32)
2094 key.offset = 0;
2095 else
2096 key.offset = offset;
2097
2098 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2099 if (ret < 0) {
2100 WARN_ON(1);
2101 return ret;
2102 }
2103
2104 while (1) {
2105 cond_resched();
2106
2107 leaf = path->nodes[0];
2108 slot = path->slots[0];
2109
2110 if (slot >= btrfs_header_nritems(leaf)) {
2111 ret = btrfs_next_leaf(root, path);
2112 if (ret < 0) {
2113 goto out;
2114 } else if (ret > 0) {
2115 ret = 0;
2116 goto out;
2117 }
2118 continue;
2119 }
2120
2121 path->slots[0]++;
2122
2123 btrfs_item_key_to_cpu(leaf, &key, slot);
2124
2125 if (key.objectid > inum)
2126 goto out;
2127
2128 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2129 continue;
2130
2131 extent = btrfs_item_ptr(leaf, slot,
2132 struct btrfs_file_extent_item);
2133
2134 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2135 continue;
2136
2137 extent_offset = btrfs_file_extent_offset(leaf, extent);
2138 if (key.offset - extent_offset != offset)
2139 continue;
2140
2141 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2142 if (extent_offset >= old->extent_offset + old->offset +
2143 old->len || extent_offset + num_bytes <=
2144 old->extent_offset + old->offset)
2145 continue;
2146
2147 break;
2148 }
2149
2150 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2151 if (!backref) {
2152 ret = -ENOENT;
2153 goto out;
2154 }
2155
2156 backref->root_id = root_id;
2157 backref->inum = inum;
2158 backref->file_pos = offset + extent_offset;
2159 backref->num_bytes = num_bytes;
2160 backref->extent_offset = extent_offset;
2161 backref->generation = btrfs_file_extent_generation(leaf, extent);
2162 backref->old = old;
2163 backref_insert(&new->root, backref);
2164 old->count++;
2165out:
2166 btrfs_release_path(path);
2167 WARN_ON(ret);
2168 return ret;
2169}
2170
2171static noinline bool record_extent_backrefs(struct btrfs_path *path,
2172 struct new_sa_defrag_extent *new)
2173{
2174 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2175 struct old_sa_defrag_extent *old, *tmp;
2176 int ret;
2177
2178 new->path = path;
2179
2180 list_for_each_entry_safe(old, tmp, &new->head, list) {
2181 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2182 path, record_one_backref,
2183 old);
2184 BUG_ON(ret < 0 && ret != -ENOENT);
2185
2186 /* no backref to be processed for this extent */
2187 if (!old->count) {
2188 list_del(&old->list);
2189 kfree(old);
2190 }
2191 }
2192
2193 if (list_empty(&new->head))
2194 return false;
2195
2196 return true;
2197}
2198
2199static int relink_is_mergable(struct extent_buffer *leaf,
2200 struct btrfs_file_extent_item *fi,
2201 u64 disk_bytenr)
2202{
2203 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2204 return 0;
2205
2206 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2207 return 0;
2208
2209 if (btrfs_file_extent_compression(leaf, fi) ||
2210 btrfs_file_extent_encryption(leaf, fi) ||
2211 btrfs_file_extent_other_encoding(leaf, fi))
2212 return 0;
2213
2214 return 1;
2215}
2216
2217/*
2218 * Note the backref might has changed, and in this case we just return 0.
2219 */
2220static noinline int relink_extent_backref(struct btrfs_path *path,
2221 struct sa_defrag_extent_backref *prev,
2222 struct sa_defrag_extent_backref *backref)
2223{
2224 struct btrfs_file_extent_item *extent;
2225 struct btrfs_file_extent_item *item;
2226 struct btrfs_ordered_extent *ordered;
2227 struct btrfs_trans_handle *trans;
2228 struct btrfs_fs_info *fs_info;
2229 struct btrfs_root *root;
2230 struct btrfs_key key;
2231 struct extent_buffer *leaf;
2232 struct old_sa_defrag_extent *old = backref->old;
2233 struct new_sa_defrag_extent *new = old->new;
2234 struct inode *src_inode = new->inode;
2235 struct inode *inode;
2236 struct extent_state *cached = NULL;
2237 int ret = 0;
2238 u64 start;
2239 u64 len;
2240 u64 lock_start;
2241 u64 lock_end;
2242 bool merge = false;
2243 int index;
2244
2245 if (prev && prev->root_id == backref->root_id &&
2246 prev->inum == backref->inum &&
2247 prev->file_pos + prev->num_bytes == backref->file_pos)
2248 merge = true;
2249
2250 /* step 1: get root */
2251 key.objectid = backref->root_id;
2252 key.type = BTRFS_ROOT_ITEM_KEY;
2253 key.offset = (u64)-1;
2254
2255 fs_info = BTRFS_I(src_inode)->root->fs_info;
2256 index = srcu_read_lock(&fs_info->subvol_srcu);
2257
2258 root = btrfs_read_fs_root_no_name(fs_info, &key);
2259 if (IS_ERR(root)) {
2260 srcu_read_unlock(&fs_info->subvol_srcu, index);
2261 if (PTR_ERR(root) == -ENOENT)
2262 return 0;
2263 return PTR_ERR(root);
2264 }
2265 if (btrfs_root_refs(&root->root_item) == 0) {
2266 srcu_read_unlock(&fs_info->subvol_srcu, index);
2267 /* parse ENOENT to 0 */
2268 return 0;
2269 }
2270
2271 /* step 2: get inode */
2272 key.objectid = backref->inum;
2273 key.type = BTRFS_INODE_ITEM_KEY;
2274 key.offset = 0;
2275
2276 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2277 if (IS_ERR(inode)) {
2278 srcu_read_unlock(&fs_info->subvol_srcu, index);
2279 return 0;
2280 }
2281
2282 srcu_read_unlock(&fs_info->subvol_srcu, index);
2283
2284 /* step 3: relink backref */
2285 lock_start = backref->file_pos;
2286 lock_end = backref->file_pos + backref->num_bytes - 1;
2287 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2288 0, &cached);
2289
2290 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2291 if (ordered) {
2292 btrfs_put_ordered_extent(ordered);
2293 goto out_unlock;
2294 }
2295
2296 trans = btrfs_join_transaction(root);
2297 if (IS_ERR(trans)) {
2298 ret = PTR_ERR(trans);
2299 goto out_unlock;
2300 }
2301
2302 key.objectid = backref->inum;
2303 key.type = BTRFS_EXTENT_DATA_KEY;
2304 key.offset = backref->file_pos;
2305
2306 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2307 if (ret < 0) {
2308 goto out_free_path;
2309 } else if (ret > 0) {
2310 ret = 0;
2311 goto out_free_path;
2312 }
2313
2314 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2315 struct btrfs_file_extent_item);
2316
2317 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2318 backref->generation)
2319 goto out_free_path;
2320
2321 btrfs_release_path(path);
2322
2323 start = backref->file_pos;
2324 if (backref->extent_offset < old->extent_offset + old->offset)
2325 start += old->extent_offset + old->offset -
2326 backref->extent_offset;
2327
2328 len = min(backref->extent_offset + backref->num_bytes,
2329 old->extent_offset + old->offset + old->len);
2330 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2331
2332 ret = btrfs_drop_extents(trans, root, inode, start,
2333 start + len, 1);
2334 if (ret)
2335 goto out_free_path;
2336again:
2337 key.objectid = btrfs_ino(inode);
2338 key.type = BTRFS_EXTENT_DATA_KEY;
2339 key.offset = start;
2340
a09a0a70 2341 path->leave_spinning = 1;
38c227d8
LB		 2342 if (merge) {
2343 struct btrfs_file_extent_item *fi;
2344 u64 extent_len;
2345 struct btrfs_key found_key;
2346
2347 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2348 if (ret < 0)
2349 goto out_free_path;
2350
2351 path->slots[0]--;
2352 leaf = path->nodes[0];
2353 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2354
2355 fi = btrfs_item_ptr(leaf, path->slots[0],
2356 struct btrfs_file_extent_item);
2357 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2358
2359 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2360 extent_len + found_key.offset == start) {
2361 btrfs_set_file_extent_num_bytes(leaf, fi,
2362 extent_len + len);
2363 btrfs_mark_buffer_dirty(leaf);
2364 inode_add_bytes(inode, len);
2365
2366 ret = 1;
2367 goto out_free_path;
2368 } else {
2369 merge = false;
2370 btrfs_release_path(path);
2371 goto again;
2372 }
2373 }
2374
2375 ret = btrfs_insert_empty_item(trans, root, path, &key,
2376 sizeof(*extent));
2377 if (ret) {
2378 btrfs_abort_transaction(trans, root, ret);
2379 goto out_free_path;
2380 }
2381
2382 leaf = path->nodes[0];
2383 item = btrfs_item_ptr(leaf, path->slots[0],
2384 struct btrfs_file_extent_item);
2385 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2386 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2387 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2388 btrfs_set_file_extent_num_bytes(leaf, item, len);
2389 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2390 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2391 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2392 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2393 btrfs_set_file_extent_encryption(leaf, item, 0);
2394 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2395
2396 btrfs_mark_buffer_dirty(leaf);
2397 inode_add_bytes(inode, len);
a09a0a70 2398 btrfs_release_path(path);
38c227d8
LB		 2399
2400 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2401 new->disk_len, 0,
2402 backref->root_id, backref->inum,
2403 new->file_pos, 0); /* start - extent_offset */
2404 if (ret) {
2405 btrfs_abort_transaction(trans, root, ret);
2406 goto out_free_path;
2407 }
2408
2409 ret = 1;
2410out_free_path:
2411 btrfs_release_path(path);
a09a0a70 2412 path->leave_spinning = 0;
38c227d8
LB		 2413 btrfs_end_transaction(trans, root);
2414out_unlock:
2415 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2416 &cached, GFP_NOFS);
2417 iput(inode);
2418 return ret;
2419}
2420
2421static void relink_file_extents(struct new_sa_defrag_extent *new)
2422{
2423 struct btrfs_path *path;
2424 struct old_sa_defrag_extent *old, *tmp;
2425 struct sa_defrag_extent_backref *backref;
2426 struct sa_defrag_extent_backref *prev = NULL;
2427 struct inode *inode;
2428 struct btrfs_root *root;
2429 struct rb_node *node;
2430 int ret;
2431
2432 inode = new->inode;
2433 root = BTRFS_I(inode)->root;
2434
2435 path = btrfs_alloc_path();
2436 if (!path)
2437 return;
2438
2439 if (!record_extent_backrefs(path, new)) {
2440 btrfs_free_path(path);
2441 goto out;
2442 }
2443 btrfs_release_path(path);
2444
2445 while (1) {
2446 node = rb_first(&new->root);
2447 if (!node)
2448 break;
2449 rb_erase(node, &new->root);
2450
2451 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2452
2453 ret = relink_extent_backref(path, prev, backref);
2454 WARN_ON(ret < 0);
2455
2456 kfree(prev);
2457
2458 if (ret == 1)
2459 prev = backref;
2460 else
2461 prev = NULL;
2462 cond_resched();
2463 }
2464 kfree(prev);
2465
2466 btrfs_free_path(path);
2467
2468 list_for_each_entry_safe(old, tmp, &new->head, list) {
2469 list_del(&old->list);
2470 kfree(old);
2471 }
2472out:
2473 atomic_dec(&root->fs_info->defrag_running);
2474 wake_up(&root->fs_info->transaction_wait);
2475
2476 kfree(new);
2477}
2478
2479static struct new_sa_defrag_extent *
2480record_old_file_extents(struct inode *inode,
2481 struct btrfs_ordered_extent *ordered)
2482{
2483 struct btrfs_root *root = BTRFS_I(inode)->root;
2484 struct btrfs_path *path;
2485 struct btrfs_key key;
2486 struct old_sa_defrag_extent *old, *tmp;
2487 struct new_sa_defrag_extent *new;
2488 int ret;
2489
2490 new = kmalloc(sizeof(*new), GFP_NOFS);
2491 if (!new)
2492 return NULL;
2493
2494 new->inode = inode;
2495 new->file_pos = ordered->file_offset;
2496 new->len = ordered->len;
2497 new->bytenr = ordered->start;
2498 new->disk_len = ordered->disk_len;
2499 new->compress_type = ordered->compress_type;
2500 new->root = RB_ROOT;
2501 INIT_LIST_HEAD(&new->head);
2502
2503 path = btrfs_alloc_path();
2504 if (!path)
2505 goto out_kfree;
2506
2507 key.objectid = btrfs_ino(inode);
2508 key.type = BTRFS_EXTENT_DATA_KEY;
2509 key.offset = new->file_pos;
2510
2511 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512 if (ret < 0)
2513 goto out_free_path;
2514 if (ret > 0 && path->slots[0] > 0)
2515 path->slots[0]--;
2516
2517 /* find out all the old extents for the file range */
2518 while (1) {
2519 struct btrfs_file_extent_item *extent;
2520 struct extent_buffer *l;
2521 int slot;
2522 u64 num_bytes;
2523 u64 offset;
2524 u64 end;
2525 u64 disk_bytenr;
2526 u64 extent_offset;
2527
2528 l = path->nodes[0];
2529 slot = path->slots[0];
2530
2531 if (slot >= btrfs_header_nritems(l)) {
2532 ret = btrfs_next_leaf(root, path);
2533 if (ret < 0)
2534 goto out_free_list;
2535 else if (ret > 0)
2536 break;
2537 continue;
2538 }
2539
2540 btrfs_item_key_to_cpu(l, &key, slot);
2541
2542 if (key.objectid != btrfs_ino(inode))
2543 break;
2544 if (key.type != BTRFS_EXTENT_DATA_KEY)
2545 break;
2546 if (key.offset >= new->file_pos + new->len)
2547 break;
2548
2549 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2550
2551 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2552 if (key.offset + num_bytes < new->file_pos)
2553 goto next;
2554
2555 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2556 if (!disk_bytenr)
2557 goto next;
2558
2559 extent_offset = btrfs_file_extent_offset(l, extent);
2560
2561 old = kmalloc(sizeof(*old), GFP_NOFS);
2562 if (!old)
2563 goto out_free_list;
2564
2565 offset = max(new->file_pos, key.offset);
2566 end = min(new->file_pos + new->len, key.offset + num_bytes);
2567
2568 old->bytenr = disk_bytenr;
2569 old->extent_offset = extent_offset;
2570 old->offset = offset - key.offset;
2571 old->len = end - offset;
2572 old->new = new;
2573 old->count = 0;
2574 list_add_tail(&old->list, &new->head);
2575next:
2576 path->slots[0]++;
2577 cond_resched();
2578 }
2579
2580 btrfs_free_path(path);
2581 atomic_inc(&root->fs_info->defrag_running);
2582
2583 return new;
2584
2585out_free_list:
2586 list_for_each_entry_safe(old, tmp, &new->head, list) {
2587 list_del(&old->list);
2588 kfree(old);
2589 }
2590out_free_path:
2591 btrfs_free_path(path);
2592out_kfree:
2593 kfree(new);
2594 return NULL;
2595}
2596
5d13a98f
CM		 2597/*
2598 * helper function for btrfs_finish_ordered_io, this
2599 * just reads in some of the csum leaves to prime them into ram
2600 * before we start the transaction. It limits the amount of btree
2601 * reads required while inside the transaction.
2602 */
d352ac68
CM		 2603/* as ordered data IO finishes, this gets called so we can finish
2604 * an ordered extent if the range of bytes in the file it covers are
2605 * fully written.
2606 */
5fd02043 2607static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2608{
5fd02043 2609 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2610 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2611 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2612 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2613 struct extent_state *cached_state = NULL;
38c227d8 2614 struct new_sa_defrag_extent *new = NULL;
261507a0 2615 int compress_type = 0;
e6dcd2dc 2616 int ret;
82d5902d 2617 bool nolock;
e6dcd2dc 2618
83eea1f1 2619 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2620
5fd02043
JB		 2621 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2622 ret = -EIO;
2623 goto out;
2624 }
2625
c2167754 2626 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2627 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2628 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2629 if (nolock)
2630 trans = btrfs_join_transaction_nolock(root);
2631 else
2632 trans = btrfs_join_transaction(root);
2633 if (IS_ERR(trans)) {
2634 ret = PTR_ERR(trans);
2635 trans = NULL;
2636 goto out;
c2167754 2637 }
6c760c07
JB		 2638 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2639 ret = btrfs_update_inode_fallback(trans, root, inode);
2640 if (ret) /* -ENOMEM or corruption */
2641 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2642 goto out;
2643 }
e6dcd2dc 2644
2ac55d41
JB		 2645 lock_extent_bits(io_tree, ordered_extent->file_offset,
2646 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2647 0, &cached_state);
e6dcd2dc 2648
38c227d8
LB		 2649 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2650 ordered_extent->file_offset + ordered_extent->len - 1,
2651 EXTENT_DEFRAG, 1, cached_state);
2652 if (ret) {
2653 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2654 if (last_snapshot >= BTRFS_I(inode)->generation)
2655 /* the inode is shared */
2656 new = record_old_file_extents(inode, ordered_extent);
2657
2658 clear_extent_bit(io_tree, ordered_extent->file_offset,
2659 ordered_extent->file_offset + ordered_extent->len - 1,
2660 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2661 }
2662
0cb59c99 2663 if (nolock)
7a7eaa40 2664 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2665 else
7a7eaa40 2666 trans = btrfs_join_transaction(root);
79787eaa
JM		 2667 if (IS_ERR(trans)) {
2668 ret = PTR_ERR(trans);
2669 trans = NULL;
2670 goto out_unlock;
2671 }
0ca1f7ce 2672 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2673
c8b97818 2674 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2675 compress_type = ordered_extent->compress_type;
d899e052 2676 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2677 BUG_ON(compress_type);
920bbbfb 2678 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2679 ordered_extent->file_offset,
2680 ordered_extent->file_offset +
2681 ordered_extent->len);
d899e052 2682 } else {
0af3d00b 2683 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2684 ret = insert_reserved_file_extent(trans, inode,
2685 ordered_extent->file_offset,
2686 ordered_extent->start,
2687 ordered_extent->disk_len,
2688 ordered_extent->len,
2689 ordered_extent->len,
261507a0 2690 compress_type, 0, 0,
d899e052 2691 BTRFS_FILE_EXTENT_REG);
d899e052 2692 }
5dc562c5
JB		 2693 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2694 ordered_extent->file_offset, ordered_extent->len,
2695 trans->transid);
79787eaa
JM		 2696 if (ret < 0) {
2697 btrfs_abort_transaction(trans, root, ret);
5fd02043 2698 goto out_unlock;
79787eaa 2699 }
2ac55d41 2700
e6dcd2dc
CM		 2701 add_pending_csums(trans, inode, ordered_extent->file_offset,
2702 &ordered_extent->list);
2703
6c760c07
JB		 2704 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2705 ret = btrfs_update_inode_fallback(trans, root, inode);
2706 if (ret) { /* -ENOMEM or corruption */
2707 btrfs_abort_transaction(trans, root, ret);
2708 goto out_unlock;
1ef30be1
JB		 2709 }
2710 ret = 0;
5fd02043
JB		 2711out_unlock:
2712 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2713 ordered_extent->file_offset +
2714 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2715out:
5b0e95bf 2716 if (root != root->fs_info->tree_root)
0cb59c99 2717 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2718 if (trans)
2719 btrfs_end_transaction(trans, root);
0cb59c99 2720
0bec9ef5 2721 if (ret) {
5fd02043
JB		 2722 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2723 ordered_extent->file_offset +
2724 ordered_extent->len - 1, NULL, GFP_NOFS);
2725
0bec9ef5
JB		 2726 /*
2727 * If the ordered extent had an IOERR or something else went
2728 * wrong we need to return the space for this ordered extent
2729 * back to the allocator.
2730 */
2731 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2732 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2733 btrfs_free_reserved_extent(root, ordered_extent->start,
2734 ordered_extent->disk_len);
2735 }
2736
2737
5fd02043 2738 /*
8bad3c02
LB		 2739 * This needs to be done to make sure anybody waiting knows we are done
2740 * updating everything for this ordered extent.
5fd02043
JB		 2741 */
2742 btrfs_remove_ordered_extent(inode, ordered_extent);
2743
38c227d8
LB		 2744 /* for snapshot-aware defrag */
2745 if (new)
2746 relink_file_extents(new);
2747
e6dcd2dc
CM		 2748 /* once for us */
2749 btrfs_put_ordered_extent(ordered_extent);
2750 /* once for the tree */
2751 btrfs_put_ordered_extent(ordered_extent);
2752
5fd02043
JB		 2753 return ret;
2754}
2755
2756static void finish_ordered_fn(struct btrfs_work *work)
2757{
2758 struct btrfs_ordered_extent *ordered_extent;
2759 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2760 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2761}
2762
b2950863 2763static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2764 struct extent_state *state, int uptodate)
2765{
5fd02043
JB		 2766 struct inode *inode = page->mapping->host;
2767 struct btrfs_root *root = BTRFS_I(inode)->root;
2768 struct btrfs_ordered_extent *ordered_extent = NULL;
2769 struct btrfs_workers *workers;
2770
1abe9b8a 2771 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2772
8b62b72b 2773 ClearPagePrivate2(page);
5fd02043
JB		 2774 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2775 end - start + 1, uptodate))
2776 return 0;
2777
2778 ordered_extent->work.func = finish_ordered_fn;
2779 ordered_extent->work.flags = 0;
2780
83eea1f1 2781 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2782 workers = &root->fs_info->endio_freespace_worker;
2783 else
2784 workers = &root->fs_info->endio_write_workers;
2785 btrfs_queue_worker(workers, &ordered_extent->work);
2786
2787 return 0;
211f90e6
CM		 2788}
2789
d352ac68
CM		 2790/*
2791 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2792 * if there's a match, we allow the bio to finish. If not, the code in
2793 * extent_io.c will try to find good copies for us.
d352ac68 2794 */
b2950863 2795static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2796 struct extent_state *state, int mirror)
07157aac 2797{
4eee4fa4 2798 size_t offset = start - page_offset(page);
07157aac 2799 struct inode *inode = page->mapping->host;
d1310b2e 2800 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2801 char *kaddr;
aadfeb6e 2802 u64 private = ~(u32)0;
07157aac 2803 int ret;
ff79f819
CM		 2804 struct btrfs_root *root = BTRFS_I(inode)->root;
2805 u32 csum = ~(u32)0;
c2cf52eb
SK		 2806 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2807 DEFAULT_RATELIMIT_BURST);
d1310b2e 2808
d20f7043
CM		 2809 if (PageChecked(page)) {
2810 ClearPageChecked(page);
2811 goto good;
2812 }
6cbff00f
CH		 2813
2814 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2815 goto good;
17d217fe
YZ		 2816
2817 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2818 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2819 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2820 GFP_NOFS);
b6cda9bc 2821 return 0;
17d217fe 2822 }
d20f7043 2823
c2e639f0 2824 if (state && state->start == start) {
70dec807
CM		 2825 private = state->private;
2826 ret = 0;
2827 } else {
2828 ret = get_state_private(io_tree, start, &private);
2829 }
7ac687d9 2830 kaddr = kmap_atomic(page);
d397712b 2831 if (ret)
07157aac 2832 goto zeroit;
d397712b 2833
b0496686 2834 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2835 btrfs_csum_final(csum, (char *)&csum);
d397712b 2836 if (csum != private)
07157aac 2837 goto zeroit;
d397712b 2838
7ac687d9 2839 kunmap_atomic(kaddr);
d20f7043 2840good:
07157aac
CM		 2841 return 0;
2842
2843zeroit:
c2cf52eb
SK		 2844 if (__ratelimit(&_rs))
2845 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2846 (unsigned long long)btrfs_ino(page->mapping->host),
2847 (unsigned long long)start, csum,
2848 (unsigned long long)private);
db94535d
CM		 2849 memset(kaddr + offset, 1, end - start + 1);
2850 flush_dcache_page(page);
7ac687d9 2851 kunmap_atomic(kaddr);
3b951516
CM		 2852 if (private == 0)
2853 return 0;
7e38326f 2854 return -EIO;
07157aac 2855}
b888db2b 2856
24bbcf04
YZ		 2857struct delayed_iput {
2858 struct list_head list;
2859 struct inode *inode;
2860};
2861
79787eaa
JM		 2862/* JDM: If this is fs-wide, why can't we add a pointer to
2863 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2864void btrfs_add_delayed_iput(struct inode *inode)
2865{
2866 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2867 struct delayed_iput *delayed;
2868
2869 if (atomic_add_unless(&inode->i_count, -1, 1))
2870 return;
2871
2872 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2873 delayed->inode = inode;
2874
2875 spin_lock(&fs_info->delayed_iput_lock);
2876 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2877 spin_unlock(&fs_info->delayed_iput_lock);
2878}
2879
2880void btrfs_run_delayed_iputs(struct btrfs_root *root)
2881{
2882 LIST_HEAD(list);
2883 struct btrfs_fs_info *fs_info = root->fs_info;
2884 struct delayed_iput *delayed;
2885 int empty;
2886
2887 spin_lock(&fs_info->delayed_iput_lock);
2888 empty = list_empty(&fs_info->delayed_iputs);
2889 spin_unlock(&fs_info->delayed_iput_lock);
2890 if (empty)
2891 return;
2892
24bbcf04
YZ		 2893 spin_lock(&fs_info->delayed_iput_lock);
2894 list_splice_init(&fs_info->delayed_iputs, &list);
2895 spin_unlock(&fs_info->delayed_iput_lock);
2896
2897 while (!list_empty(&list)) {
2898 delayed = list_entry(list.next, struct delayed_iput, list);
2899 list_del(&delayed->list);
2900 iput(delayed->inode);
2901 kfree(delayed);
2902 }
24bbcf04
YZ		 2903}
2904
d68fc57b 2905/*
42b2aa86 2906 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2907 * files in the subvolume, it removes orphan item and frees block_rsv
2908 * structure.
2909 */
2910void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2911 struct btrfs_root *root)
2912{
90290e19 2913 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2914 int ret;
2915
8a35d95f 2916 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2917 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2918 return;
2919
90290e19 2920 spin_lock(&root->orphan_lock);
8a35d95f 2921 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2922 spin_unlock(&root->orphan_lock);
2923 return;
2924 }
2925
2926 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2927 spin_unlock(&root->orphan_lock);
2928 return;
2929 }
2930
2931 block_rsv = root->orphan_block_rsv;
2932 root->orphan_block_rsv = NULL;
2933 spin_unlock(&root->orphan_lock);
2934
d68fc57b
YZ		 2935 if (root->orphan_item_inserted &&
2936 btrfs_root_refs(&root->root_item) > 0) {
2937 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2938 root->root_key.objectid);
2939 BUG_ON(ret);
2940 root->orphan_item_inserted = 0;
2941 }
2942
90290e19
JB		 2943 if (block_rsv) {
2944 WARN_ON(block_rsv->size > 0);
2945 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2946 }
2947}
2948
7b128766
JB		 2949/*
2950 * This creates an orphan entry for the given inode in case something goes
2951 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2952 *
2953 * NOTE: caller of this function should reserve 5 units of metadata for
2954 * this function.
7b128766
JB		 2955 */
2956int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2957{
2958 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2959 struct btrfs_block_rsv *block_rsv = NULL;
2960 int reserve = 0;
2961 int insert = 0;
2962 int ret;
7b128766 2963
d68fc57b 2964 if (!root->orphan_block_rsv) {
66d8f3dd 2965 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2966 if (!block_rsv)
2967 return -ENOMEM;
d68fc57b 2968 }
7b128766 2969
d68fc57b
YZ		 2970 spin_lock(&root->orphan_lock);
2971 if (!root->orphan_block_rsv) {
2972 root->orphan_block_rsv = block_rsv;
2973 } else if (block_rsv) {
2974 btrfs_free_block_rsv(root, block_rsv);
2975 block_rsv = NULL;
7b128766 2976 }
7b128766 2977
8a35d95f
JB		 2978 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2979 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 2980#if 0
2981 /*
2982 * For proper ENOSPC handling, we should do orphan
2983 * cleanup when mounting. But this introduces backward
2984 * compatibility issue.
2985 */
2986 if (!xchg(&root->orphan_item_inserted, 1))
2987 insert = 2;
2988 else
2989 insert = 1;
2990#endif
2991 insert = 1;
321f0e70 2992 atomic_inc(&root->orphan_inodes);
7b128766
JB		 2993 }
2994
72ac3c0d
JB		 2995 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2996 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2997 reserve = 1;
d68fc57b 2998 spin_unlock(&root->orphan_lock);
7b128766 2999
d68fc57b
YZ		 3000 /* grab metadata reservation from transaction handle */
3001 if (reserve) {
3002 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3003 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3004 }
7b128766 3005
d68fc57b
YZ		 3006 /* insert an orphan item to track this unlinked/truncated file */
3007 if (insert >= 1) {
33345d01 3008 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3009 if (ret && ret != -EEXIST) {
8a35d95f
JB		 3010 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3011 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 3012 btrfs_abort_transaction(trans, root, ret);
3013 return ret;
3014 }
3015 ret = 0;
d68fc57b
YZ		 3016 }
3017
3018 /* insert an orphan item to track subvolume contains orphan files */
3019 if (insert >= 2) {
3020 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3021 root->root_key.objectid);
79787eaa
JM		 3022 if (ret && ret != -EEXIST) {
3023 btrfs_abort_transaction(trans, root, ret);
3024 return ret;
3025 }
d68fc57b
YZ		 3026 }
3027 return 0;
7b128766
JB		 3028}
3029
3030/*
3031 * We have done the truncate/delete so we can go ahead and remove the orphan
3032 * item for this particular inode.
3033 */
48a3b636
ES		 3034static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3035 struct inode *inode)
7b128766
JB		 3036{
3037 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3038 int delete_item = 0;
3039 int release_rsv = 0;
7b128766
JB		 3040 int ret = 0;
3041
d68fc57b 3042 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3043 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3044 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3045 delete_item = 1;
7b128766 3046
72ac3c0d
JB		 3047 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3048 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3049 release_rsv = 1;
d68fc57b 3050 spin_unlock(&root->orphan_lock);
7b128766 3051
d68fc57b 3052 if (trans && delete_item) {
33345d01 3053 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3054 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 3055 }
7b128766 3056
8a35d95f 3057 if (release_rsv) {
d68fc57b 3058 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 3059 atomic_dec(&root->orphan_inodes);
3060 }
7b128766 3061
d68fc57b 3062 return 0;
7b128766
JB		 3063}
3064
3065/*
3066 * this cleans up any orphans that may be left on the list from the last use
3067 * of this root.
3068 */
66b4ffd1 3069int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3070{
3071 struct btrfs_path *path;
3072 struct extent_buffer *leaf;
7b128766
JB		 3073 struct btrfs_key key, found_key;
3074 struct btrfs_trans_handle *trans;
3075 struct inode *inode;
8f6d7f4f 3076 u64 last_objectid = 0;
7b128766
JB		 3077 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3078
d68fc57b 3079 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3080 return 0;
c71bf099
YZ		 3081
3082 path = btrfs_alloc_path();
66b4ffd1
JB		 3083 if (!path) {
3084 ret = -ENOMEM;
3085 goto out;
3086 }
7b128766
JB		 3087 path->reada = -1;
3088
3089 key.objectid = BTRFS_ORPHAN_OBJECTID;
3090 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3091 key.offset = (u64)-1;
3092
7b128766
JB		 3093 while (1) {
3094 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3095 if (ret < 0)
3096 goto out;
7b128766
JB		 3097
3098 /*
3099 * if ret == 0 means we found what we were searching for, which
25985edc 3100 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3101 * find the key and see if we have stuff that matches
3102 */
3103 if (ret > 0) {
66b4ffd1 3104 ret = 0;
7b128766
JB		 3105 if (path->slots[0] == 0)
3106 break;
3107 path->slots[0]--;
3108 }
3109
3110 /* pull out the item */
3111 leaf = path->nodes[0];
7b128766
JB		 3112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3113
3114 /* make sure the item matches what we want */
3115 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3116 break;
3117 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3118 break;
3119
3120 /* release the path since we're done with it */
b3b4aa74 3121 btrfs_release_path(path);
7b128766
JB		 3122
3123 /*
3124 * this is where we are basically btrfs_lookup, without the
3125 * crossing root thing. we store the inode number in the
3126 * offset of the orphan item.
3127 */
8f6d7f4f
JB		 3128
3129 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3130 btrfs_err(root->fs_info,
3131 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3132 ret = -EINVAL;
3133 goto out;
3134 }
3135
3136 last_objectid = found_key.offset;
3137
5d4f98a2
YZ		 3138 found_key.objectid = found_key.offset;
3139 found_key.type = BTRFS_INODE_ITEM_KEY;
3140 found_key.offset = 0;
73f73415 3141 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 3142 ret = PTR_RET(inode);
3143 if (ret && ret != -ESTALE)
66b4ffd1 3144 goto out;
7b128766 3145
f8e9e0b0
AJ		 3146 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3147 struct btrfs_root *dead_root;
3148 struct btrfs_fs_info *fs_info = root->fs_info;
3149 int is_dead_root = 0;
3150
3151 /*
3152 * this is an orphan in the tree root. Currently these
3153 * could come from 2 sources:
3154 * a) a snapshot deletion in progress
3155 * b) a free space cache inode
3156 * We need to distinguish those two, as the snapshot
3157 * orphan must not get deleted.
3158 * find_dead_roots already ran before us, so if this
3159 * is a snapshot deletion, we should find the root
3160 * in the dead_roots list
3161 */
3162 spin_lock(&fs_info->trans_lock);
3163 list_for_each_entry(dead_root, &fs_info->dead_roots,
3164 root_list) {
3165 if (dead_root->root_key.objectid ==
3166 found_key.objectid) {
3167 is_dead_root = 1;
3168 break;
3169 }
3170 }
3171 spin_unlock(&fs_info->trans_lock);
3172 if (is_dead_root) {
3173 /* prevent this orphan from being found again */
3174 key.offset = found_key.objectid - 1;
3175 continue;
3176 }
3177 }
7b128766 3178 /*
a8c9e576
JB		 3179 * Inode is already gone but the orphan item is still there,
3180 * kill the orphan item.
7b128766 3181 */
a8c9e576
JB		 3182 if (ret == -ESTALE) {
3183 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3184 if (IS_ERR(trans)) {
3185 ret = PTR_ERR(trans);
3186 goto out;
3187 }
c2cf52eb
SK		 3188 btrfs_debug(root->fs_info, "auto deleting %Lu",
3189 found_key.objectid);
a8c9e576
JB		 3190 ret = btrfs_del_orphan_item(trans, root,
3191 found_key.objectid);
79787eaa 3192 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 3193 btrfs_end_transaction(trans, root);
7b128766
JB		 3194 continue;
3195 }
3196
a8c9e576
JB		 3197 /*
3198 * add this inode to the orphan list so btrfs_orphan_del does
3199 * the proper thing when we hit it
3200 */
8a35d95f
JB		 3201 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3202 &BTRFS_I(inode)->runtime_flags);
925396ec 3203 atomic_inc(&root->orphan_inodes);
a8c9e576 3204
7b128766
JB		 3205 /* if we have links, this was a truncate, lets do that */
3206 if (inode->i_nlink) {
a41ad394
JB		 3207 if (!S_ISREG(inode->i_mode)) {
3208 WARN_ON(1);
3209 iput(inode);
3210 continue;
3211 }
7b128766 3212 nr_truncate++;
f3fe820c
JB		 3213
3214 /* 1 for the orphan item deletion. */
3215 trans = btrfs_start_transaction(root, 1);
3216 if (IS_ERR(trans)) {
3217 ret = PTR_ERR(trans);
3218 goto out;
3219 }
3220 ret = btrfs_orphan_add(trans, inode);
3221 btrfs_end_transaction(trans, root);
3222 if (ret)
3223 goto out;
3224
66b4ffd1 3225 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3226 if (ret)
3227 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3228 } else {
3229 nr_unlink++;
3230 }
3231
3232 /* this will do delete_inode and everything for us */
3233 iput(inode);
66b4ffd1
JB		 3234 if (ret)
3235 goto out;
7b128766 3236 }
3254c876
MX		 3237 /* release the path since we're done with it */
3238 btrfs_release_path(path);
3239
d68fc57b
YZ		 3240 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3241
3242 if (root->orphan_block_rsv)
3243 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3244 (u64)-1);
3245
3246 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3247 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3248 if (!IS_ERR(trans))
3249 btrfs_end_transaction(trans, root);
d68fc57b 3250 }
7b128766
JB		 3251
3252 if (nr_unlink)
4884b476 3253 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3254 if (nr_truncate)
4884b476 3255 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3256
3257out:
3258 if (ret)
c2cf52eb
SK		 3259 btrfs_crit(root->fs_info,
3260 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3261 btrfs_free_path(path);
3262 return ret;
7b128766
JB		 3263}
3264
46a53cca
CM		 3265/*
3266 * very simple check to peek ahead in the leaf looking for xattrs. If we
3267 * don't find any xattrs, we know there can't be any acls.
3268 *
3269 * slot is the slot the inode is in, objectid is the objectid of the inode
3270 */
3271static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3272 int slot, u64 objectid)
3273{
3274 u32 nritems = btrfs_header_nritems(leaf);
3275 struct btrfs_key found_key;
3276 int scanned = 0;
3277
3278 slot++;
3279 while (slot < nritems) {
3280 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3281
3282 /* we found a different objectid, there must not be acls */
3283 if (found_key.objectid != objectid)
3284 return 0;
3285
3286 /* we found an xattr, assume we've got an acl */
3287 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3288 return 1;
3289
3290 /*
3291 * we found a key greater than an xattr key, there can't
3292 * be any acls later on
3293 */
3294 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3295 return 0;
3296
3297 slot++;
3298 scanned++;
3299
3300 /*
3301 * it goes inode, inode backrefs, xattrs, extents,
3302 * so if there are a ton of hard links to an inode there can
3303 * be a lot of backrefs. Don't waste time searching too hard,
3304 * this is just an optimization
3305 */
3306 if (scanned >= 8)
3307 break;
3308 }
3309 /* we hit the end of the leaf before we found an xattr or
3310 * something larger than an xattr. We have to assume the inode
3311 * has acls
3312 */
3313 return 1;
3314}
3315
d352ac68
CM		 3316/*
3317 * read an inode from the btree into the in-memory inode
3318 */
5d4f98a2 3319static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3320{
3321 struct btrfs_path *path;
5f39d397 3322 struct extent_buffer *leaf;
39279cc3 3323 struct btrfs_inode_item *inode_item;
0b86a832 3324 struct btrfs_timespec *tspec;
39279cc3
CM		 3325 struct btrfs_root *root = BTRFS_I(inode)->root;
3326 struct btrfs_key location;
46a53cca 3327 int maybe_acls;
618e21d5 3328 u32 rdev;
39279cc3 3329 int ret;
2f7e33d4
MX		 3330 bool filled = false;
3331
3332 ret = btrfs_fill_inode(inode, &rdev);
3333 if (!ret)
3334 filled = true;
39279cc3
CM		 3335
3336 path = btrfs_alloc_path();
1748f843
MF		 3337 if (!path)
3338 goto make_bad;
3339
d90c7321 3340 path->leave_spinning = 1;
39279cc3 3341 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3342
39279cc3 3343 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3344 if (ret)
39279cc3 3345 goto make_bad;
39279cc3 3346
5f39d397 3347 leaf = path->nodes[0];
2f7e33d4
MX		 3348
3349 if (filled)
3350 goto cache_acl;
3351
5f39d397
CM		 3352 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3353 struct btrfs_inode_item);
5f39d397 3354 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3355 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3356 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3357 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3358 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3359
3360 tspec = btrfs_inode_atime(inode_item);
3361 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3362 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3363
3364 tspec = btrfs_inode_mtime(inode_item);
3365 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3366 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3367
3368 tspec = btrfs_inode_ctime(inode_item);
3369 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3370 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3371
a76a3cd4 3372 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3373 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3374 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3375
3376 /*
3377 * If we were modified in the current generation and evicted from memory
3378 * and then re-read we need to do a full sync since we don't have any
3379 * idea about which extents were modified before we were evicted from
3380 * cache.
3381 */
3382 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3383 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3384 &BTRFS_I(inode)->runtime_flags);
3385
0c4d2d95 3386 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3387 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3388 inode->i_rdev = 0;
5f39d397
CM		 3389 rdev = btrfs_inode_rdev(leaf, inode_item);
3390
aec7477b 3391 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3392 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3393cache_acl:
46a53cca
CM		 3394 /*
3395 * try to precache a NULL acl entry for files that don't have
3396 * any xattrs or acls
3397 */
33345d01
LZ		 3398 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3399 btrfs_ino(inode));
72c04902
AV		 3400 if (!maybe_acls)
3401 cache_no_acl(inode);
46a53cca 3402
39279cc3 3403 btrfs_free_path(path);
39279cc3 3404
39279cc3 3405 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3406 case S_IFREG:
3407 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3408 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3409 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3410 inode->i_fop = &btrfs_file_operations;
3411 inode->i_op = &btrfs_file_inode_operations;
3412 break;
3413 case S_IFDIR:
3414 inode->i_fop = &btrfs_dir_file_operations;
3415 if (root == root->fs_info->tree_root)
3416 inode->i_op = &btrfs_dir_ro_inode_operations;
3417 else
3418 inode->i_op = &btrfs_dir_inode_operations;
3419 break;
3420 case S_IFLNK:
3421 inode->i_op = &btrfs_symlink_inode_operations;
3422 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3423 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3424 break;
618e21d5 3425 default:
0279b4cd 3426 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3427 init_special_inode(inode, inode->i_mode, rdev);
3428 break;
39279cc3 3429 }
6cbff00f
CH		 3430
3431 btrfs_update_iflags(inode);
39279cc3
CM		 3432 return;
3433
3434make_bad:
39279cc3 3435 btrfs_free_path(path);
39279cc3
CM		 3436 make_bad_inode(inode);
3437}
3438
d352ac68
CM		 3439/*
3440 * given a leaf and an inode, copy the inode fields into the leaf
3441 */
e02119d5
CM		 3442static void fill_inode_item(struct btrfs_trans_handle *trans,
3443 struct extent_buffer *leaf,
5f39d397 3444 struct btrfs_inode_item *item,
39279cc3
CM		 3445 struct inode *inode)
3446{
51fab693
LB		 3447 struct btrfs_map_token token;
3448
3449 btrfs_init_map_token(&token);
5f39d397 3450
51fab693
LB		 3451 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3452 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3453 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3454 &token);
3455 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3456 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3457
51fab693
LB		 3458 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3459 inode->i_atime.tv_sec, &token);
3460 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3461 inode->i_atime.tv_nsec, &token);
5f39d397 3462
51fab693
LB		 3463 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3464 inode->i_mtime.tv_sec, &token);
3465 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3466 inode->i_mtime.tv_nsec, &token);
5f39d397 3467
51fab693
LB		 3468 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3469 inode->i_ctime.tv_sec, &token);
3470 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3471 inode->i_ctime.tv_nsec, &token);
5f39d397 3472
51fab693
LB		 3473 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3474 &token);
3475 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3476 &token);
3477 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3478 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3479 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3480 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3481 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3482}
3483
d352ac68
CM		 3484/*
3485 * copy everything in the in-memory inode into the btree.
3486 */
2115133f 3487static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3488 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3489{
3490 struct btrfs_inode_item *inode_item;
3491 struct btrfs_path *path;
5f39d397 3492 struct extent_buffer *leaf;
39279cc3
CM		 3493 int ret;
3494
3495 path = btrfs_alloc_path();
16cdcec7
MX		 3496 if (!path)
3497 return -ENOMEM;
3498
b9473439 3499 path->leave_spinning = 1;
16cdcec7
MX		 3500 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3501 1);
39279cc3
CM		 3502 if (ret) {
3503 if (ret > 0)
3504 ret = -ENOENT;
3505 goto failed;
3506 }
3507
b4ce94de 3508 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3509 leaf = path->nodes[0];
3510 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3511 struct btrfs_inode_item);
39279cc3 3512
e02119d5 3513 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3514 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3515 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3516 ret = 0;
3517failed:
39279cc3
CM		 3518 btrfs_free_path(path);
3519 return ret;
3520}
3521
2115133f
CM		 3522/*
3523 * copy everything in the in-memory inode into the btree.
3524 */
3525noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3526 struct btrfs_root *root, struct inode *inode)
3527{
3528 int ret;
3529
3530 /*
3531 * If the inode is a free space inode, we can deadlock during commit
3532 * if we put it into the delayed code.
3533 *
3534 * The data relocation inode should also be directly updated
3535 * without delay
3536 */
83eea1f1 3537 if (!btrfs_is_free_space_inode(inode)
2115133f 3538 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3539 btrfs_update_root_times(trans, root);
3540
2115133f
CM		 3541 ret = btrfs_delayed_update_inode(trans, root, inode);
3542 if (!ret)
3543 btrfs_set_inode_last_trans(trans, inode);
3544 return ret;
3545 }
3546
3547 return btrfs_update_inode_item(trans, root, inode);
3548}
3549
be6aef60
JB		 3550noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3551 struct btrfs_root *root,
3552 struct inode *inode)
2115133f
CM		 3553{
3554 int ret;
3555
3556 ret = btrfs_update_inode(trans, root, inode);
3557 if (ret == -ENOSPC)
3558 return btrfs_update_inode_item(trans, root, inode);
3559 return ret;
3560}
3561
d352ac68
CM		 3562/*
3563 * unlink helper that gets used here in inode.c and in the tree logging
3564 * recovery code. It remove a link in a directory with a given name, and
3565 * also drops the back refs in the inode to the directory
3566 */
92986796
AV		 3567static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3568 struct btrfs_root *root,
3569 struct inode *dir, struct inode *inode,
3570 const char *name, int name_len)
39279cc3
CM		 3571{
3572 struct btrfs_path *path;
39279cc3 3573 int ret = 0;
5f39d397 3574 struct extent_buffer *leaf;
39279cc3 3575 struct btrfs_dir_item *di;
5f39d397 3576 struct btrfs_key key;
aec7477b 3577 u64 index;
33345d01
LZ		 3578 u64 ino = btrfs_ino(inode);
3579 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3580
3581 path = btrfs_alloc_path();
54aa1f4d
CM		 3582 if (!path) {
3583 ret = -ENOMEM;
554233a6 3584 goto out;
54aa1f4d
CM		 3585 }
3586
b9473439 3587 path->leave_spinning = 1;
33345d01 3588 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3589 name, name_len, -1);
3590 if (IS_ERR(di)) {
3591 ret = PTR_ERR(di);
3592 goto err;
3593 }
3594 if (!di) {
3595 ret = -ENOENT;
3596 goto err;
3597 }
5f39d397
CM		 3598 leaf = path->nodes[0];
3599 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3600 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3601 if (ret)
3602 goto err;
b3b4aa74 3603 btrfs_release_path(path);
39279cc3 3604
33345d01
LZ		 3605 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3606 dir_ino, &index);
aec7477b 3607 if (ret) {
c2cf52eb
SK		 3608 btrfs_info(root->fs_info,
3609 "failed to delete reference to %.*s, inode %llu parent %llu",
3610 name_len, name,
3611 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 3612 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3613 goto err;
3614 }
3615
16cdcec7 3616 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3617 if (ret) {
3618 btrfs_abort_transaction(trans, root, ret);
39279cc3 3619 goto err;
79787eaa 3620 }
39279cc3 3621
e02119d5 3622 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3623 inode, dir_ino);
79787eaa
JM		 3624 if (ret != 0 && ret != -ENOENT) {
3625 btrfs_abort_transaction(trans, root, ret);
3626 goto err;
3627 }
e02119d5
CM		 3628
3629 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3630 dir, index);
6418c961
CM		 3631 if (ret == -ENOENT)
3632 ret = 0;
d4e3991b
ZB		 3633 else if (ret)
3634 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3635err:
3636 btrfs_free_path(path);
e02119d5
CM		 3637 if (ret)
3638 goto out;
3639
3640 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3641 inode_inc_iversion(inode);
3642 inode_inc_iversion(dir);
e02119d5 3643 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3644 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3645out:
39279cc3
CM		 3646 return ret;
3647}
3648
92986796
AV		 3649int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3650 struct btrfs_root *root,
3651 struct inode *dir, struct inode *inode,
3652 const char *name, int name_len)
3653{
3654 int ret;
3655 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3656 if (!ret) {
3657 btrfs_drop_nlink(inode);
3658 ret = btrfs_update_inode(trans, root, inode);
3659 }
3660 return ret;
3661}
3662
3663
a22285a6
YZ		 3664/* helper to check if there is any shared block in the path */
3665static int check_path_shared(struct btrfs_root *root,
3666 struct btrfs_path *path)
39279cc3 3667{
a22285a6
YZ		 3668 struct extent_buffer *eb;
3669 int level;
0e4dcbef 3670 u64 refs = 1;
5df6a9f6 3671
a22285a6 3672 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 3673 int ret;
3674
a22285a6
YZ		 3675 if (!path->nodes[level])
3676 break;
3677 eb = path->nodes[level];
3678 if (!btrfs_block_can_be_shared(root, eb))
3679 continue;
3173a18f 3680 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
a22285a6
YZ		 3681 &refs, NULL);
3682 if (refs > 1)
3683 return 1;
5df6a9f6 3684 }
dedefd72 3685 return 0;
39279cc3
CM		 3686}
3687
a22285a6
YZ		 3688/*
3689 * helper to start transaction for unlink and rmdir.
3690 *
3691 * unlink and rmdir are special in btrfs, they do not always free space.
3692 * so in enospc case, we should make sure they will free space before
3693 * allowing them to use the global metadata reservation.
3694 */
3695static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3696 struct dentry *dentry)
4df27c4d 3697{
39279cc3 3698 struct btrfs_trans_handle *trans;
a22285a6 3699 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 3700 struct btrfs_path *path;
4df27c4d 3701 struct btrfs_dir_item *di;
7b128766 3702 struct inode *inode = dentry->d_inode;
4df27c4d 3703 u64 index;
a22285a6
YZ		 3704 int check_link = 1;
3705 int err = -ENOSPC;
4df27c4d 3706 int ret;
33345d01
LZ		 3707 u64 ino = btrfs_ino(inode);
3708 u64 dir_ino = btrfs_ino(dir);
4df27c4d 3709
e70bea5f
JB		 3710 /*
3711 * 1 for the possible orphan item
3712 * 1 for the dir item
3713 * 1 for the dir index
3714 * 1 for the inode ref
e70bea5f
JB		 3715 * 1 for the inode
3716 */
6e137ed3 3717 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3718 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3719 return trans;
4df27c4d 3720
33345d01 3721 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 3722 return ERR_PTR(-ENOSPC);
4df27c4d 3723
a22285a6
YZ		 3724 /* check if there is someone else holds reference */
3725 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3726 return ERR_PTR(-ENOSPC);
4df27c4d 3727
a22285a6
YZ		 3728 if (atomic_read(&inode->i_count) > 2)
3729 return ERR_PTR(-ENOSPC);
4df27c4d 3730
a22285a6
YZ		 3731 if (xchg(&root->fs_info->enospc_unlink, 1))
3732 return ERR_PTR(-ENOSPC);
3733
3734 path = btrfs_alloc_path();
3735 if (!path) {
3736 root->fs_info->enospc_unlink = 0;
3737 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 3738 }
3739
3880a1b4
JB		 3740 /* 1 for the orphan item */
3741 trans = btrfs_start_transaction(root, 1);
5df6a9f6 3742 if (IS_ERR(trans)) {
a22285a6
YZ		 3743 btrfs_free_path(path);
3744 root->fs_info->enospc_unlink = 0;
3745 return trans;
3746 }
4df27c4d 3747
a22285a6
YZ		 3748 path->skip_locking = 1;
3749 path->search_commit_root = 1;
4df27c4d 3750
a22285a6
YZ		 3751 ret = btrfs_lookup_inode(trans, root, path,
3752 &BTRFS_I(dir)->location, 0);
3753 if (ret < 0) {
3754 err = ret;
3755 goto out;
3756 }
3757 if (ret == 0) {
3758 if (check_path_shared(root, path))
3759 goto out;
3760 } else {
3761 check_link = 0;
5df6a9f6 3762 }
b3b4aa74 3763 btrfs_release_path(path);
a22285a6
YZ		 3764
3765 ret = btrfs_lookup_inode(trans, root, path,
3766 &BTRFS_I(inode)->location, 0);
3767 if (ret < 0) {
3768 err = ret;
3769 goto out;
3770 }
3771 if (ret == 0) {
3772 if (check_path_shared(root, path))
3773 goto out;
3774 } else {
3775 check_link = 0;
3776 }
b3b4aa74 3777 btrfs_release_path(path);
a22285a6
YZ		 3778
3779 if (ret == 0 && S_ISREG(inode->i_mode)) {
3780 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 3781 ino, (u64)-1, 0);
a22285a6
YZ		 3782 if (ret < 0) {
3783 err = ret;
3784 goto out;
3785 }
79787eaa 3786 BUG_ON(ret == 0); /* Corruption */
a22285a6
YZ		 3787 if (check_path_shared(root, path))
3788 goto out;
b3b4aa74 3789 btrfs_release_path(path);
a22285a6
YZ		 3790 }
3791
3792 if (!check_link) {
3793 err = 0;
3794 goto out;
3795 }
3796
33345d01 3797 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 3798 dentry->d_name.name, dentry->d_name.len, 0);
3799 if (IS_ERR(di)) {
3800 err = PTR_ERR(di);
3801 goto out;
3802 }
3803 if (di) {
3804 if (check_path_shared(root, path))
3805 goto out;
3806 } else {
3807 err = 0;
3808 goto out;
3809 }
b3b4aa74 3810 btrfs_release_path(path);
a22285a6 3811
f186373f
MF		 3812 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3813 dentry->d_name.len, ino, dir_ino, 0,
3814 &index);
3815 if (ret) {
3816 err = ret;
a22285a6
YZ		 3817 goto out;
3818 }
f186373f 3819
a22285a6
YZ		 3820 if (check_path_shared(root, path))
3821 goto out;
f186373f 3822
b3b4aa74 3823 btrfs_release_path(path);
a22285a6 3824
16cdcec7
MX		 3825 /*
3826 * This is a commit root search, if we can lookup inode item and other
3827 * relative items in the commit root, it means the transaction of
3828 * dir/file creation has been committed, and the dir index item that we
3829 * delay to insert has also been inserted into the commit root. So
3830 * we needn't worry about the delayed insertion of the dir index item
3831 * here.
3832 */
33345d01 3833 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 3834 dentry->d_name.name, dentry->d_name.len, 0);
3835 if (IS_ERR(di)) {
3836 err = PTR_ERR(di);
3837 goto out;
3838 }
3839 BUG_ON(ret == -ENOENT);
3840 if (check_path_shared(root, path))
3841 goto out;
3842
3843 err = 0;
3844out:
3845 btrfs_free_path(path);
3880a1b4
JB		 3846 /* Migrate the orphan reservation over */
3847 if (!err)
3848 err = btrfs_block_rsv_migrate(trans->block_rsv,
3849 &root->fs_info->global_block_rsv,
5a77d76c 3850 trans->bytes_reserved);
3880a1b4 3851
a22285a6
YZ		 3852 if (err) {
3853 btrfs_end_transaction(trans, root);
3854 root->fs_info->enospc_unlink = 0;
3855 return ERR_PTR(err);
3856 }
3857
3858 trans->block_rsv = &root->fs_info->global_block_rsv;
3859 return trans;
3860}
3861
3862static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3863 struct btrfs_root *root)
3864{
66d8f3dd 3865 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
5a77d76c
JB		 3866 btrfs_block_rsv_release(root, trans->block_rsv,
3867 trans->bytes_reserved);
3868 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 3869 BUG_ON(!root->fs_info->enospc_unlink);
3870 root->fs_info->enospc_unlink = 0;
3871 }
7ad85bb7 3872 btrfs_end_transaction(trans, root);
a22285a6
YZ		 3873}
3874
3875static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3876{
3877 struct btrfs_root *root = BTRFS_I(dir)->root;
3878 struct btrfs_trans_handle *trans;
3879 struct inode *inode = dentry->d_inode;
3880 int ret;
a22285a6
YZ		 3881
3882 trans = __unlink_start_trans(dir, dentry);
3883 if (IS_ERR(trans))
3884 return PTR_ERR(trans);
5f39d397 3885
12fcfd22
CM		 3886 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3887
e02119d5
CM		 3888 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3889 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3890 if (ret)
3891 goto out;
7b128766 3892
a22285a6 3893 if (inode->i_nlink == 0) {
7b128766 3894 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3895 if (ret)
3896 goto out;
a22285a6 3897 }
7b128766 3898
b532402e 3899out:
a22285a6 3900 __unlink_end_trans(trans, root);
b53d3f5d 3901 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3902 return ret;
3903}
3904
4df27c4d
YZ		 3905int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct inode *dir, u64 objectid,
3908 const char *name, int name_len)
3909{
3910 struct btrfs_path *path;
3911 struct extent_buffer *leaf;
3912 struct btrfs_dir_item *di;
3913 struct btrfs_key key;
3914 u64 index;
3915 int ret;
33345d01 3916 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3917
3918 path = btrfs_alloc_path();
3919 if (!path)
3920 return -ENOMEM;
3921
33345d01 3922 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3923 name, name_len, -1);
79787eaa
JM		 3924 if (IS_ERR_OR_NULL(di)) {
3925 if (!di)
3926 ret = -ENOENT;
3927 else
3928 ret = PTR_ERR(di);
3929 goto out;
3930 }
4df27c4d
YZ		 3931
3932 leaf = path->nodes[0];
3933 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3934 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3935 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3936 if (ret) {
3937 btrfs_abort_transaction(trans, root, ret);
3938 goto out;
3939 }
b3b4aa74 3940 btrfs_release_path(path);
4df27c4d
YZ		 3941
3942 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3943 objectid, root->root_key.objectid,
33345d01 3944 dir_ino, &index, name, name_len);
4df27c4d 3945 if (ret < 0) {
79787eaa
JM		 3946 if (ret != -ENOENT) {
3947 btrfs_abort_transaction(trans, root, ret);
3948 goto out;
3949 }
33345d01 3950 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3951 name, name_len);
79787eaa
JM		 3952 if (IS_ERR_OR_NULL(di)) {
3953 if (!di)
3954 ret = -ENOENT;
3955 else
3956 ret = PTR_ERR(di);
3957 btrfs_abort_transaction(trans, root, ret);
3958 goto out;
3959 }
4df27c4d
YZ		 3960
3961 leaf = path->nodes[0];
3962 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3963 btrfs_release_path(path);
4df27c4d
YZ		 3964 index = key.offset;
3965 }
945d8962 3966 btrfs_release_path(path);
4df27c4d 3967
16cdcec7 3968 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3969 if (ret) {
3970 btrfs_abort_transaction(trans, root, ret);
3971 goto out;
3972 }
4df27c4d
YZ		 3973
3974 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3975 inode_inc_iversion(dir);
4df27c4d 3976 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3977 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3978 if (ret)
3979 btrfs_abort_transaction(trans, root, ret);
3980out:
71d7aed0 3981 btrfs_free_path(path);
79787eaa 3982 return ret;
4df27c4d
YZ		 3983}
3984
39279cc3
CM		 3985static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3986{
3987 struct inode *inode = dentry->d_inode;
1832a6d5 3988 int err = 0;
39279cc3 3989 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3990 struct btrfs_trans_handle *trans;
39279cc3 3991
b3ae244e 3992 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3993 return -ENOTEMPTY;
b3ae244e
DS		 3994 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3995 return -EPERM;
134d4512 3996
a22285a6
YZ		 3997 trans = __unlink_start_trans(dir, dentry);
3998 if (IS_ERR(trans))
5df6a9f6 3999 return PTR_ERR(trans);
5df6a9f6 4000
33345d01 4001 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4002 err = btrfs_unlink_subvol(trans, root, dir,
4003 BTRFS_I(inode)->location.objectid,
4004 dentry->d_name.name,
4005 dentry->d_name.len);
4006 goto out;
4007 }
4008
7b128766
JB		 4009 err = btrfs_orphan_add(trans, inode);
4010 if (err)
4df27c4d 4011 goto out;
7b128766 4012
39279cc3 4013 /* now the directory is empty */
e02119d5
CM		 4014 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4015 dentry->d_name.name, dentry->d_name.len);
d397712b 4016 if (!err)
dbe674a9 4017 btrfs_i_size_write(inode, 0);
4df27c4d 4018out:
a22285a6 4019 __unlink_end_trans(trans, root);
b53d3f5d 4020 btrfs_btree_balance_dirty(root);
3954401f 4021
39279cc3
CM		 4022 return err;
4023}
4024
39279cc3
CM		 4025/*
4026 * this can truncate away extent items, csum items and directory items.
4027 * It starts at a high offset and removes keys until it can't find
d352ac68 4028 * any higher than new_size
39279cc3
CM		 4029 *
4030 * csum items that cross the new i_size are truncated to the new size
4031 * as well.
7b128766
JB		 4032 *
4033 * min_type is the minimum key type to truncate down to. If set to 0, this
4034 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4035 */
8082510e
YZ		 4036int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4037 struct btrfs_root *root,
4038 struct inode *inode,
4039 u64 new_size, u32 min_type)
39279cc3 4040{
39279cc3 4041 struct btrfs_path *path;
5f39d397 4042 struct extent_buffer *leaf;
39279cc3 4043 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4044 struct btrfs_key key;
4045 struct btrfs_key found_key;
39279cc3 4046 u64 extent_start = 0;
db94535d 4047 u64 extent_num_bytes = 0;
5d4f98a2 4048 u64 extent_offset = 0;
39279cc3 4049 u64 item_end = 0;
8082510e 4050 u32 found_type = (u8)-1;
39279cc3
CM		 4051 int found_extent;
4052 int del_item;
85e21bac
CM		 4053 int pending_del_nr = 0;
4054 int pending_del_slot = 0;
179e29e4 4055 int extent_type = -1;
8082510e
YZ		 4056 int ret;
4057 int err = 0;
33345d01 4058 u64 ino = btrfs_ino(inode);
8082510e
YZ		 4059
4060 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4061
0eb0e19c
MF		 4062 path = btrfs_alloc_path();
4063 if (!path)
4064 return -ENOMEM;
4065 path->reada = -1;
4066
5dc562c5
JB		 4067 /*
4068 * We want to drop from the next block forward in case this new size is
4069 * not block aligned since we will be keeping the last block of the
4070 * extent just the way it is.
4071 */
0af3d00b 4072 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 4073 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4074 root->sectorsize), (u64)-1, 0);
8082510e 4075
16cdcec7
MX		 4076 /*
4077 * This function is also used to drop the items in the log tree before
4078 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4079 * it is used to drop the loged items. So we shouldn't kill the delayed
4080 * items.
4081 */
4082 if (min_type == 0 && root == BTRFS_I(inode)->root)
4083 btrfs_kill_delayed_inode_items(inode);
4084
33345d01 4085 key.objectid = ino;
39279cc3 4086 key.offset = (u64)-1;
5f39d397
CM		 4087 key.type = (u8)-1;
4088
85e21bac 4089search_again:
b9473439 4090 path->leave_spinning = 1;
85e21bac 4091 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4092 if (ret < 0) {
4093 err = ret;
4094 goto out;
4095 }
d397712b 4096
85e21bac 4097 if (ret > 0) {
e02119d5
CM		 4098 /* there are no items in the tree for us to truncate, we're
4099 * done
4100 */
8082510e
YZ		 4101 if (path->slots[0] == 0)
4102 goto out;
85e21bac
CM		 4103 path->slots[0]--;
4104 }
4105
d397712b 4106 while (1) {
39279cc3 4107 fi = NULL;
5f39d397
CM		 4108 leaf = path->nodes[0];
4109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4110 found_type = btrfs_key_type(&found_key);
39279cc3 4111
33345d01 4112 if (found_key.objectid != ino)
39279cc3 4113 break;
5f39d397 4114
85e21bac 4115 if (found_type < min_type)
39279cc3
CM		 4116 break;
4117
5f39d397 4118 item_end = found_key.offset;
39279cc3 4119 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4120 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4121 struct btrfs_file_extent_item);
179e29e4
CM		 4122 extent_type = btrfs_file_extent_type(leaf, fi);
4123 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4124 item_end +=
db94535d 4125 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4126 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4127 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 4128 fi);
39279cc3 4129 }
008630c1 4130 item_end--;
39279cc3 4131 }
8082510e
YZ		 4132 if (found_type > min_type) {
4133 del_item = 1;
4134 } else {
4135 if (item_end < new_size)
b888db2b 4136 break;
8082510e
YZ		 4137 if (found_key.offset >= new_size)
4138 del_item = 1;
4139 else
4140 del_item = 0;
39279cc3 4141 }
39279cc3 4142 found_extent = 0;
39279cc3 4143 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4144 if (found_type != BTRFS_EXTENT_DATA_KEY)
4145 goto delete;
4146
4147 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4148 u64 num_dec;
db94535d 4149 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4150 if (!del_item) {
db94535d
CM		 4151 u64 orig_num_bytes =
4152 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4153 extent_num_bytes = ALIGN(new_size -
4154 found_key.offset,
4155 root->sectorsize);
db94535d
CM		 4156 btrfs_set_file_extent_num_bytes(leaf, fi,
4157 extent_num_bytes);
4158 num_dec = (orig_num_bytes -
9069218d 4159 extent_num_bytes);
e02119d5 4160 if (root->ref_cows && extent_start != 0)
a76a3cd4 4161 inode_sub_bytes(inode, num_dec);
5f39d397 4162 btrfs_mark_buffer_dirty(leaf);
39279cc3 4163 } else {
db94535d
CM		 4164 extent_num_bytes =
4165 btrfs_file_extent_disk_num_bytes(leaf,
4166 fi);
5d4f98a2
YZ		 4167 extent_offset = found_key.offset -
4168 btrfs_file_extent_offset(leaf, fi);
4169
39279cc3 4170 /* FIXME blocksize != 4096 */
9069218d 4171 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4172 if (extent_start != 0) {
4173 found_extent = 1;
e02119d5 4174 if (root->ref_cows)
a76a3cd4 4175 inode_sub_bytes(inode, num_dec);
e02119d5 4176 }
39279cc3 4177 }
9069218d 4178 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4179 /*
4180 * we can't truncate inline items that have had
4181 * special encodings
4182 */
4183 if (!del_item &&
4184 btrfs_file_extent_compression(leaf, fi) == 0 &&
4185 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4186 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4187 u32 size = new_size - found_key.offset;
4188
4189 if (root->ref_cows) {
a76a3cd4
YZ		 4190 inode_sub_bytes(inode, item_end + 1 -
4191 new_size);
e02119d5
CM		 4192 }
4193 size =
4194 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4195 btrfs_truncate_item(root, path, size, 1);
e02119d5 4196 } else if (root->ref_cows) {
a76a3cd4
YZ		 4197 inode_sub_bytes(inode, item_end + 1 -
4198 found_key.offset);
9069218d 4199 }
39279cc3 4200 }
179e29e4 4201delete:
39279cc3 4202 if (del_item) {
85e21bac
CM		 4203 if (!pending_del_nr) {
4204 /* no pending yet, add ourselves */
4205 pending_del_slot = path->slots[0];
4206 pending_del_nr = 1;
4207 } else if (pending_del_nr &&
4208 path->slots[0] + 1 == pending_del_slot) {
4209 /* hop on the pending chunk */
4210 pending_del_nr++;
4211 pending_del_slot = path->slots[0];
4212 } else {
d397712b 4213 BUG();
85e21bac 4214 }
39279cc3
CM		 4215 } else {
4216 break;
4217 }
0af3d00b
JB		 4218 if (found_extent && (root->ref_cows ||
4219 root == root->fs_info->tree_root)) {
b9473439 4220 btrfs_set_path_blocking(path);
39279cc3 4221 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4222 extent_num_bytes, 0,
4223 btrfs_header_owner(leaf),
66d7e7f0 4224 ino, extent_offset, 0);
39279cc3
CM		 4225 BUG_ON(ret);
4226 }
85e21bac 4227
8082510e
YZ		 4228 if (found_type == BTRFS_INODE_ITEM_KEY)
4229 break;
4230
4231 if (path->slots[0] == 0 ||
4232 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4233 if (pending_del_nr) {
4234 ret = btrfs_del_items(trans, root, path,
4235 pending_del_slot,
4236 pending_del_nr);
79787eaa
JM		 4237 if (ret) {
4238 btrfs_abort_transaction(trans,
4239 root, ret);
4240 goto error;
4241 }
8082510e
YZ		 4242 pending_del_nr = 0;
4243 }
b3b4aa74 4244 btrfs_release_path(path);
85e21bac 4245 goto search_again;
8082510e
YZ		 4246 } else {
4247 path->slots[0]--;
85e21bac 4248 }
39279cc3 4249 }
8082510e 4250out:
85e21bac
CM		 4251 if (pending_del_nr) {
4252 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4253 pending_del_nr);
79787eaa
JM		 4254 if (ret)
4255 btrfs_abort_transaction(trans, root, ret);
85e21bac 4256 }
79787eaa 4257error:
39279cc3 4258 btrfs_free_path(path);
8082510e 4259 return err;
39279cc3
CM		 4260}
4261
4262/*
2aaa6655
JB		 4263 * btrfs_truncate_page - read, zero a chunk and write a page
4264 * @inode - inode that we're zeroing
4265 * @from - the offset to start zeroing
4266 * @len - the length to zero, 0 to zero the entire range respective to the
4267 * offset
4268 * @front - zero up to the offset instead of from the offset on
4269 *
4270 * This will find the page for the "from" offset and cow the page and zero the
4271 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4272 */
2aaa6655
JB		 4273int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4274 int front)
39279cc3 4275{
2aaa6655 4276 struct address_space *mapping = inode->i_mapping;
db94535d 4277 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4278 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4279 struct btrfs_ordered_extent *ordered;
2ac55d41 4280 struct extent_state *cached_state = NULL;
e6dcd2dc 4281 char *kaddr;
db94535d 4282 u32 blocksize = root->sectorsize;
39279cc3
CM		 4283 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4284 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4285 struct page *page;
3b16a4e3 4286 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4287 int ret = 0;
a52d9a80 4288 u64 page_start;
e6dcd2dc 4289 u64 page_end;
39279cc3 4290
2aaa6655
JB		 4291 if ((offset & (blocksize - 1)) == 0 &&
4292 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4293 goto out;
0ca1f7ce 4294 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4295 if (ret)
4296 goto out;
39279cc3 4297
211c17f5 4298again:
3b16a4e3 4299 page = find_or_create_page(mapping, index, mask);
5d5e103a 4300 if (!page) {
0ca1f7ce 4301 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4302 ret = -ENOMEM;
39279cc3 4303 goto out;
5d5e103a 4304 }
e6dcd2dc
CM		 4305
4306 page_start = page_offset(page);
4307 page_end = page_start + PAGE_CACHE_SIZE - 1;
4308
39279cc3 4309 if (!PageUptodate(page)) {
9ebefb18 4310 ret = btrfs_readpage(NULL, page);
39279cc3 4311 lock_page(page);
211c17f5
CM		 4312 if (page->mapping != mapping) {
4313 unlock_page(page);
4314 page_cache_release(page);
4315 goto again;
4316 }
39279cc3
CM		 4317 if (!PageUptodate(page)) {
4318 ret = -EIO;
89642229 4319 goto out_unlock;
39279cc3
CM		 4320 }
4321 }
211c17f5 4322 wait_on_page_writeback(page);
e6dcd2dc 4323
d0082371 4324 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4325 set_page_extent_mapped(page);
4326
4327 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4328 if (ordered) {
2ac55d41
JB		 4329 unlock_extent_cached(io_tree, page_start, page_end,
4330 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4331 unlock_page(page);
4332 page_cache_release(page);
eb84ae03 4333 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4334 btrfs_put_ordered_extent(ordered);
4335 goto again;
4336 }
4337
2ac55d41 4338 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4339 EXTENT_DIRTY | EXTENT_DELALLOC |
4340 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4341 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4342
2ac55d41
JB		 4343 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4344 &cached_state);
9ed74f2d 4345 if (ret) {
2ac55d41
JB		 4346 unlock_extent_cached(io_tree, page_start, page_end,
4347 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4348 goto out_unlock;
4349 }
4350
e6dcd2dc 4351 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4352 if (!len)
4353 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4354 kaddr = kmap(page);
2aaa6655
JB		 4355 if (front)
4356 memset(kaddr, 0, offset);
4357 else
4358 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4359 flush_dcache_page(page);
4360 kunmap(page);
4361 }
247e743c 4362 ClearPageChecked(page);
e6dcd2dc 4363 set_page_dirty(page);
2ac55d41
JB		 4364 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4365 GFP_NOFS);
39279cc3 4366
89642229 4367out_unlock:
5d5e103a 4368 if (ret)
0ca1f7ce 4369 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4370 unlock_page(page);
4371 page_cache_release(page);
4372out:
4373 return ret;
4374}
4375
695a0d0d
JB		 4376/*
4377 * This function puts in dummy file extents for the area we're creating a hole
4378 * for. So if we are truncating this file to a larger size we need to insert
4379 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4380 * the range between oldsize and size
4381 */
a41ad394 4382int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4383{
9036c102
YZ		 4384 struct btrfs_trans_handle *trans;
4385 struct btrfs_root *root = BTRFS_I(inode)->root;
4386 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4387 struct extent_map *em = NULL;
2ac55d41 4388 struct extent_state *cached_state = NULL;
5dc562c5 4389 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4390 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4391 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4392 u64 last_byte;
4393 u64 cur_offset;
4394 u64 hole_size;
9ed74f2d 4395 int err = 0;
39279cc3 4396
9036c102
YZ		 4397 if (size <= hole_start)
4398 return 0;
4399
9036c102
YZ		 4400 while (1) {
4401 struct btrfs_ordered_extent *ordered;
4402 btrfs_wait_ordered_range(inode, hole_start,
4403 block_end - hole_start);
2ac55d41 4404 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4405 &cached_state);
9036c102
YZ		 4406 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4407 if (!ordered)
4408 break;
2ac55d41
JB		 4409 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4410 &cached_state, GFP_NOFS);
9036c102
YZ		 4411 btrfs_put_ordered_extent(ordered);
4412 }
39279cc3 4413
9036c102
YZ		 4414 cur_offset = hole_start;
4415 while (1) {
4416 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4417 block_end - cur_offset, 0);
79787eaa
JM		 4418 if (IS_ERR(em)) {
4419 err = PTR_ERR(em);
f2767956 4420 em = NULL;
79787eaa
JM		 4421 break;
4422 }
9036c102 4423 last_byte = min(extent_map_end(em), block_end);
fda2832f 4424 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4425 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4426 struct extent_map *hole_em;
9036c102 4427 hole_size = last_byte - cur_offset;
9ed74f2d 4428
3642320e 4429 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4430 if (IS_ERR(trans)) {
4431 err = PTR_ERR(trans);
9ed74f2d 4432 break;
a22285a6 4433 }
8082510e 4434
5dc562c5
JB		 4435 err = btrfs_drop_extents(trans, root, inode,
4436 cur_offset,
2671485d 4437 cur_offset + hole_size, 1);
5b397377 4438 if (err) {
79787eaa 4439 btrfs_abort_transaction(trans, root, err);
5b397377 4440 btrfs_end_transaction(trans, root);
3893e33b 4441 break;
5b397377 4442 }
8082510e 4443
9036c102 4444 err = btrfs_insert_file_extent(trans, root,
33345d01 4445 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4446 0, hole_size, 0, hole_size,
4447 0, 0, 0);
5b397377 4448 if (err) {
79787eaa 4449 btrfs_abort_transaction(trans, root, err);
5b397377 4450 btrfs_end_transaction(trans, root);
3893e33b 4451 break;
5b397377 4452 }
8082510e 4453
5dc562c5
JB		 4454 btrfs_drop_extent_cache(inode, cur_offset,
4455 cur_offset + hole_size - 1, 0);
4456 hole_em = alloc_extent_map();
4457 if (!hole_em) {
4458 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4459 &BTRFS_I(inode)->runtime_flags);
4460 goto next;
4461 }
4462 hole_em->start = cur_offset;
4463 hole_em->len = hole_size;
4464 hole_em->orig_start = cur_offset;
8082510e 4465
5dc562c5
JB		 4466 hole_em->block_start = EXTENT_MAP_HOLE;
4467 hole_em->block_len = 0;
b4939680 4468 hole_em->orig_block_len = 0;
cc95bef6 4469 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4470 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4471 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4472 hole_em->generation = trans->transid;
8082510e 4473
5dc562c5
JB		 4474 while (1) {
4475 write_lock(&em_tree->lock);
09a2a8f9 4476 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4477 write_unlock(&em_tree->lock);
4478 if (err != -EEXIST)
4479 break;
4480 btrfs_drop_extent_cache(inode, cur_offset,
4481 cur_offset +
4482 hole_size - 1, 0);
4483 }
4484 free_extent_map(hole_em);
4485next:
3642320e 4486 btrfs_update_inode(trans, root, inode);
8082510e 4487 btrfs_end_transaction(trans, root);
9036c102
YZ		 4488 }
4489 free_extent_map(em);
a22285a6 4490 em = NULL;
9036c102 4491 cur_offset = last_byte;
8082510e 4492 if (cur_offset >= block_end)
9036c102
YZ		 4493 break;
4494 }
1832a6d5 4495
a22285a6 4496 free_extent_map(em);
2ac55d41
JB		 4497 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4498 GFP_NOFS);
9036c102
YZ		 4499 return err;
4500}
39279cc3 4501
3972f260 4502static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4503{
f4a2f4c5
MX		 4504 struct btrfs_root *root = BTRFS_I(inode)->root;
4505 struct btrfs_trans_handle *trans;
a41ad394 4506 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4507 loff_t newsize = attr->ia_size;
4508 int mask = attr->ia_valid;
8082510e
YZ		 4509 int ret;
4510
a41ad394 4511 if (newsize == oldsize)
8082510e
YZ		 4512 return 0;
4513
3972f260
ES		 4514 /*
4515 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4516 * special case where we need to update the times despite not having
4517 * these flags set. For all other operations the VFS set these flags
4518 * explicitly if it wants a timestamp update.
4519 */
4520 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4521 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4522
a41ad394 4523 if (newsize > oldsize) {
a41ad394
JB		 4524 truncate_pagecache(inode, oldsize, newsize);
4525 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4526 if (ret)
8082510e 4527 return ret;
8082510e 4528
f4a2f4c5
MX		 4529 trans = btrfs_start_transaction(root, 1);
4530 if (IS_ERR(trans))
4531 return PTR_ERR(trans);
4532
4533 i_size_write(inode, newsize);
4534 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4535 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4536 btrfs_end_transaction(trans, root);
a41ad394 4537 } else {
8082510e 4538
a41ad394
JB		 4539 /*
4540 * We're truncating a file that used to have good data down to
4541 * zero. Make sure it gets into the ordered flush list so that
4542 * any new writes get down to disk quickly.
4543 */
4544 if (newsize == 0)
72ac3c0d
JB		 4545 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4546 &BTRFS_I(inode)->runtime_flags);
8082510e 4547
f3fe820c
JB		 4548 /*
4549 * 1 for the orphan item we're going to add
4550 * 1 for the orphan item deletion.
4551 */
4552 trans = btrfs_start_transaction(root, 2);
4553 if (IS_ERR(trans))
4554 return PTR_ERR(trans);
4555
4556 /*
4557 * We need to do this in case we fail at _any_ point during the
4558 * actual truncate. Once we do the truncate_setsize we could
4559 * invalidate pages which forces any outstanding ordered io to
4560 * be instantly completed which will give us extents that need
4561 * to be truncated. If we fail to get an orphan inode down we
4562 * could have left over extents that were never meant to live,
4563 * so we need to garuntee from this point on that everything
4564 * will be consistent.
4565 */
4566 ret = btrfs_orphan_add(trans, inode);
4567 btrfs_end_transaction(trans, root);
4568 if (ret)
4569 return ret;
4570
a41ad394
JB		 4571 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4572 truncate_setsize(inode, newsize);
2e60a51e
MX		 4573
4574 /* Disable nonlocked read DIO to avoid the end less truncate */
4575 btrfs_inode_block_unlocked_dio(inode);
4576 inode_dio_wait(inode);
4577 btrfs_inode_resume_unlocked_dio(inode);
4578
a41ad394 4579 ret = btrfs_truncate(inode);
f3fe820c
JB		 4580 if (ret && inode->i_nlink)
4581 btrfs_orphan_del(NULL, inode);
8082510e
YZ		 4582 }
4583
a41ad394 4584 return ret;
8082510e
YZ		 4585}
4586
9036c102
YZ		 4587static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4588{
4589 struct inode *inode = dentry->d_inode;
b83cc969 4590 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4591 int err;
39279cc3 4592
b83cc969
LZ		 4593 if (btrfs_root_readonly(root))
4594 return -EROFS;
4595
9036c102
YZ		 4596 err = inode_change_ok(inode, attr);
4597 if (err)
4598 return err;
2bf5a725 4599
5a3f23d5 4600 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4601 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4602 if (err)
4603 return err;
39279cc3 4604 }
9036c102 4605
1025774c
CH		 4606 if (attr->ia_valid) {
4607 setattr_copy(inode, attr);
0c4d2d95 4608 inode_inc_iversion(inode);
22c44fe6 4609 err = btrfs_dirty_inode(inode);
1025774c 4610
22c44fe6 4611 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4612 err = btrfs_acl_chmod(inode);
4613 }
33268eaf 4614
39279cc3
CM		 4615 return err;
4616}
61295eb8 4617
bd555975 4618void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4619{
4620 struct btrfs_trans_handle *trans;
4621 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4622 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4623 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4624 int ret;
4625
1abe9b8a 4626 trace_btrfs_inode_evict(inode);
4627
39279cc3 4628 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 4629 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 4630 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4631 goto no_delete;
4632
39279cc3 4633 if (is_bad_inode(inode)) {
7b128766 4634 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4635 goto no_delete;
4636 }
bd555975 4637 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4638 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4639
c71bf099 4640 if (root->fs_info->log_root_recovering) {
6bf02314 4641 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4642 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4643 goto no_delete;
4644 }
4645
76dda93c
YZ		 4646 if (inode->i_nlink > 0) {
4647 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4648 goto no_delete;
4649 }
4650
0e8c36a9
MX		 4651 ret = btrfs_commit_inode_delayed_inode(inode);
4652 if (ret) {
4653 btrfs_orphan_del(NULL, inode);
4654 goto no_delete;
4655 }
4656
66d8f3dd 4657 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4658 if (!rsv) {
4659 btrfs_orphan_del(NULL, inode);
4660 goto no_delete;
4661 }
4a338542 4662 rsv->size = min_size;
ca7e70f5 4663 rsv->failfast = 1;
726c35fa 4664 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4665
dbe674a9 4666 btrfs_i_size_write(inode, 0);
5f39d397 4667
4289a667 4668 /*
8407aa46
MX		 4669 * This is a bit simpler than btrfs_truncate since we've already
4670 * reserved our space for our orphan item in the unlink, so we just
4671 * need to reserve some slack space in case we add bytes and update
4672 * inode item when doing the truncate.
4289a667 4673 */
8082510e 4674 while (1) {
08e007d2
MX		 4675 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4676 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4677
4678 /*
4679 * Try and steal from the global reserve since we will
4680 * likely not use this space anyway, we want to try as
4681 * hard as possible to get this to work.
4682 */
4683 if (ret)
4684 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4685
d68fc57b 4686 if (ret) {
c2cf52eb
SK		 4687 btrfs_warn(root->fs_info,
4688 "Could not get space for a delete, will truncate on mount %d",
4689 ret);
4289a667
JB		 4690 btrfs_orphan_del(NULL, inode);
4691 btrfs_free_block_rsv(root, rsv);
4692 goto no_delete;
d68fc57b 4693 }
7b128766 4694
0e8c36a9 4695 trans = btrfs_join_transaction(root);
4289a667
JB		 4696 if (IS_ERR(trans)) {
4697 btrfs_orphan_del(NULL, inode);
4698 btrfs_free_block_rsv(root, rsv);
4699 goto no_delete;
d68fc57b 4700 }
7b128766 4701
4289a667
JB		 4702 trans->block_rsv = rsv;
4703
d68fc57b 4704 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4705 if (ret != -ENOSPC)
8082510e 4706 break;
85e21bac 4707
8407aa46 4708 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4709 btrfs_end_transaction(trans, root);
4710 trans = NULL;
b53d3f5d 4711 btrfs_btree_balance_dirty(root);
8082510e 4712 }
5f39d397 4713
4289a667
JB		 4714 btrfs_free_block_rsv(root, rsv);
4715
8082510e 4716 if (ret == 0) {
4289a667 4717 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 4718 ret = btrfs_orphan_del(trans, inode);
4719 BUG_ON(ret);
4720 }
54aa1f4d 4721
4289a667 4722 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4723 if (!(root == root->fs_info->tree_root ||
4724 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4725 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4726
54aa1f4d 4727 btrfs_end_transaction(trans, root);
b53d3f5d 4728 btrfs_btree_balance_dirty(root);
39279cc3 4729no_delete:
89042e5a 4730 btrfs_remove_delayed_node(inode);
dbd5768f 4731 clear_inode(inode);
8082510e 4732 return;
39279cc3
CM		 4733}
4734
4735/*
4736 * this returns the key found in the dir entry in the location pointer.
4737 * If no dir entries were found, location->objectid is 0.
4738 */
4739static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4740 struct btrfs_key *location)
4741{
4742 const char *name = dentry->d_name.name;
4743 int namelen = dentry->d_name.len;
4744 struct btrfs_dir_item *di;
4745 struct btrfs_path *path;
4746 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4747 int ret = 0;
39279cc3
CM		 4748
4749 path = btrfs_alloc_path();
d8926bb3
MF		 4750 if (!path)
4751 return -ENOMEM;
3954401f 4752
33345d01 4753 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4754 namelen, 0);
0d9f7f3e
Y		 4755 if (IS_ERR(di))
4756 ret = PTR_ERR(di);
d397712b 4757
c704005d 4758 if (IS_ERR_OR_NULL(di))
3954401f 4759 goto out_err;
d397712b 4760
5f39d397 4761 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4762out:
39279cc3
CM		 4763 btrfs_free_path(path);
4764 return ret;
3954401f
CM		 4765out_err:
4766 location->objectid = 0;
4767 goto out;
39279cc3
CM		 4768}
4769
4770/*
4771 * when we hit a tree root in a directory, the btrfs part of the inode
4772 * needs to be changed to reflect the root directory of the tree root. This
4773 * is kind of like crossing a mount point.
4774 */
4775static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4776 struct inode *dir,
4777 struct dentry *dentry,
4778 struct btrfs_key *location,
4779 struct btrfs_root **sub_root)
39279cc3 4780{
4df27c4d
YZ		 4781 struct btrfs_path *path;
4782 struct btrfs_root *new_root;
4783 struct btrfs_root_ref *ref;
4784 struct extent_buffer *leaf;
4785 int ret;
4786 int err = 0;
39279cc3 4787
4df27c4d
YZ		 4788 path = btrfs_alloc_path();
4789 if (!path) {
4790 err = -ENOMEM;
4791 goto out;
4792 }
39279cc3 4793
4df27c4d
YZ		 4794 err = -ENOENT;
4795 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4796 BTRFS_I(dir)->root->root_key.objectid,
4797 location->objectid);
4798 if (ret) {
4799 if (ret < 0)
4800 err = ret;
4801 goto out;
4802 }
39279cc3 4803
4df27c4d
YZ		 4804 leaf = path->nodes[0];
4805 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4806 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4807 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4808 goto out;
39279cc3 4809
4df27c4d
YZ		 4810 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4811 (unsigned long)(ref + 1),
4812 dentry->d_name.len);
4813 if (ret)
4814 goto out;
4815
b3b4aa74 4816 btrfs_release_path(path);
4df27c4d
YZ		 4817
4818 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4819 if (IS_ERR(new_root)) {
4820 err = PTR_ERR(new_root);
4821 goto out;
4822 }
4823
4824 if (btrfs_root_refs(&new_root->root_item) == 0) {
4825 err = -ENOENT;
4826 goto out;
4827 }
4828
4829 *sub_root = new_root;
4830 location->objectid = btrfs_root_dirid(&new_root->root_item);
4831 location->type = BTRFS_INODE_ITEM_KEY;
4832 location->offset = 0;
4833 err = 0;
4834out:
4835 btrfs_free_path(path);
4836 return err;
39279cc3
CM		 4837}
4838
5d4f98a2
YZ		 4839static void inode_tree_add(struct inode *inode)
4840{
4841 struct btrfs_root *root = BTRFS_I(inode)->root;
4842 struct btrfs_inode *entry;
03e860bd
FNP		 4843 struct rb_node **p;
4844 struct rb_node *parent;
33345d01 4845 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 4846again:
4847 p = &root->inode_tree.rb_node;
4848 parent = NULL;
5d4f98a2 4849
1d3382cb 4850 if (inode_unhashed(inode))
76dda93c
YZ		 4851 return;
4852
5d4f98a2
YZ		 4853 spin_lock(&root->inode_lock);
4854 while (*p) {
4855 parent = *p;
4856 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4857
33345d01 4858 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4859 p = &parent->rb_left;
33345d01 4860 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4861 p = &parent->rb_right;
5d4f98a2
YZ		 4862 else {
4863 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4864 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 4865 rb_erase(parent, &root->inode_tree);
4866 RB_CLEAR_NODE(parent);
4867 spin_unlock(&root->inode_lock);
4868 goto again;
5d4f98a2
YZ		 4869 }
4870 }
4871 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4872 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873 spin_unlock(&root->inode_lock);
4874}
4875
4876static void inode_tree_del(struct inode *inode)
4877{
4878 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4879 int empty = 0;
5d4f98a2 4880
03e860bd 4881 spin_lock(&root->inode_lock);
5d4f98a2 4882 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4883 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4884 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4885 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4886 }
03e860bd 4887 spin_unlock(&root->inode_lock);
76dda93c 4888
0af3d00b
JB		 4889 /*
4890 * Free space cache has inodes in the tree root, but the tree root has a
4891 * root_refs of 0, so this could end up dropping the tree root as a
4892 * snapshot, so we need the extra !root->fs_info->tree_root check to
4893 * make sure we don't drop it.
4894 */
4895 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4896 root != root->fs_info->tree_root) {
76dda93c
YZ		 4897 synchronize_srcu(&root->fs_info->subvol_srcu);
4898 spin_lock(&root->inode_lock);
4899 empty = RB_EMPTY_ROOT(&root->inode_tree);
4900 spin_unlock(&root->inode_lock);
4901 if (empty)
4902 btrfs_add_dead_root(root);
4903 }
4904}
4905
143bede5 4906void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4907{
4908 struct rb_node *node;
4909 struct rb_node *prev;
4910 struct btrfs_inode *entry;
4911 struct inode *inode;
4912 u64 objectid = 0;
4913
4914 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4915
4916 spin_lock(&root->inode_lock);
4917again:
4918 node = root->inode_tree.rb_node;
4919 prev = NULL;
4920 while (node) {
4921 prev = node;
4922 entry = rb_entry(node, struct btrfs_inode, rb_node);
4923
33345d01 4924 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4925 node = node->rb_left;
33345d01 4926 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4927 node = node->rb_right;
4928 else
4929 break;
4930 }
4931 if (!node) {
4932 while (prev) {
4933 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4934 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4935 node = prev;
4936 break;
4937 }
4938 prev = rb_next(prev);
4939 }
4940 }
4941 while (node) {
4942 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4943 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4944 inode = igrab(&entry->vfs_inode);
4945 if (inode) {
4946 spin_unlock(&root->inode_lock);
4947 if (atomic_read(&inode->i_count) > 1)
4948 d_prune_aliases(inode);
4949 /*
45321ac5 4950 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4951 * the inode cache when its usage count
4952 * hits zero.
4953 */
4954 iput(inode);
4955 cond_resched();
4956 spin_lock(&root->inode_lock);
4957 goto again;
4958 }
4959
4960 if (cond_resched_lock(&root->inode_lock))
4961 goto again;
4962
4963 node = rb_next(node);
4964 }
4965 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4966}
4967
e02119d5
CM		 4968static int btrfs_init_locked_inode(struct inode *inode, void *p)
4969{
4970 struct btrfs_iget_args *args = p;
4971 inode->i_ino = args->ino;
e02119d5 4972 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4973 return 0;
4974}
4975
4976static int btrfs_find_actor(struct inode *inode, void *opaque)
4977{
4978 struct btrfs_iget_args *args = opaque;
33345d01 4979 return args->ino == btrfs_ino(inode) &&
d397712b 4980 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4981}
4982
5d4f98a2
YZ		 4983static struct inode *btrfs_iget_locked(struct super_block *s,
4984 u64 objectid,
4985 struct btrfs_root *root)
39279cc3
CM		 4986{
4987 struct inode *inode;
4988 struct btrfs_iget_args args;
4989 args.ino = objectid;
4990 args.root = root;
4991
4992 inode = iget5_locked(s, objectid, btrfs_find_actor,
4993 btrfs_init_locked_inode,
4994 (void *)&args);
4995 return inode;
4996}
4997
1a54ef8c
BR		 4998/* Get an inode object given its location and corresponding root.
4999 * Returns in *is_new if the inode was read from disk
5000 */
5001struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5002 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5003{
5004 struct inode *inode;
5005
5006 inode = btrfs_iget_locked(s, location->objectid, root);
5007 if (!inode)
5d4f98a2 5008 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5009
5010 if (inode->i_state & I_NEW) {
5011 BTRFS_I(inode)->root = root;
5012 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5013 btrfs_read_locked_inode(inode);
1748f843
MF		 5014 if (!is_bad_inode(inode)) {
5015 inode_tree_add(inode);
5016 unlock_new_inode(inode);
5017 if (new)
5018 *new = 1;
5019 } else {
e0b6d65b
ST		 5020 unlock_new_inode(inode);
5021 iput(inode);
5022 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5023 }
5024 }
5025
1a54ef8c
BR		 5026 return inode;
5027}
5028
4df27c4d
YZ		 5029static struct inode *new_simple_dir(struct super_block *s,
5030 struct btrfs_key *key,
5031 struct btrfs_root *root)
5032{
5033 struct inode *inode = new_inode(s);
5034
5035 if (!inode)
5036 return ERR_PTR(-ENOMEM);
5037
4df27c4d
YZ		 5038 BTRFS_I(inode)->root = root;
5039 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5040 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5041
5042 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5043 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5044 inode->i_fop = &simple_dir_operations;
5045 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5046 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5047
5048 return inode;
5049}
5050
3de4586c 5051struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5052{
d397712b 5053 struct inode *inode;
4df27c4d 5054 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5055 struct btrfs_root *sub_root = root;
5056 struct btrfs_key location;
76dda93c 5057 int index;
b4aff1f8 5058 int ret = 0;
39279cc3
CM		 5059
5060 if (dentry->d_name.len > BTRFS_NAME_LEN)
5061 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5062
39e3c955 5063 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5064 if (ret < 0)
5065 return ERR_PTR(ret);
5f39d397 5066
4df27c4d
YZ		 5067 if (location.objectid == 0)
5068 return NULL;
5069
5070 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5071 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5072 return inode;
5073 }
5074
5075 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5076
76dda93c 5077 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5078 ret = fixup_tree_root_location(root, dir, dentry,
5079 &location, &sub_root);
5080 if (ret < 0) {
5081 if (ret != -ENOENT)
5082 inode = ERR_PTR(ret);
5083 else
5084 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5085 } else {
73f73415 5086 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5087 }
76dda93c
YZ		 5088 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5089
34d19bad 5090 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5091 down_read(&root->fs_info->cleanup_work_sem);
5092 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5093 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5094 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 5095 if (ret)
5096 inode = ERR_PTR(ret);
c71bf099
YZ		 5097 }
5098
3de4586c
CM		 5099 return inode;
5100}
5101
fe15ce44 5102static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5103{
5104 struct btrfs_root *root;
848cce0d 5105 struct inode *inode = dentry->d_inode;
76dda93c 5106
848cce0d
LZ		 5107 if (!inode && !IS_ROOT(dentry))
5108 inode = dentry->d_parent->d_inode;
76dda93c 5109
848cce0d
LZ		 5110 if (inode) {
5111 root = BTRFS_I(inode)->root;
efefb143
YZ		 5112 if (btrfs_root_refs(&root->root_item) == 0)
5113 return 1;
848cce0d
LZ		 5114
5115 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5116 return 1;
efefb143 5117 }
76dda93c
YZ		 5118 return 0;
5119}
5120
b4aff1f8
JB		 5121static void btrfs_dentry_release(struct dentry *dentry)
5122{
5123 if (dentry->d_fsdata)
5124 kfree(dentry->d_fsdata);
5125}
5126
3de4586c 5127static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5128 unsigned int flags)
3de4586c 5129{
a66e7cc6
JB		 5130 struct dentry *ret;
5131
5132 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 5133 return ret;
39279cc3
CM		 5134}
5135
16cdcec7 5136unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5137 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5138};
5139
cbdf5a24
DW		 5140static int btrfs_real_readdir(struct file *filp, void *dirent,
5141 filldir_t filldir)
39279cc3 5142{
496ad9aa 5143 struct inode *inode = file_inode(filp);
39279cc3
CM		 5144 struct btrfs_root *root = BTRFS_I(inode)->root;
5145 struct btrfs_item *item;
5146 struct btrfs_dir_item *di;
5147 struct btrfs_key key;
5f39d397 5148 struct btrfs_key found_key;
39279cc3 5149 struct btrfs_path *path;
16cdcec7
MX		 5150 struct list_head ins_list;
5151 struct list_head del_list;
39279cc3 5152 int ret;
5f39d397 5153 struct extent_buffer *leaf;
39279cc3 5154 int slot;
39279cc3
CM		 5155 unsigned char d_type;
5156 int over = 0;
5157 u32 di_cur;
5158 u32 di_total;
5159 u32 di_len;
5160 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5161 char tmp_name[32];
5162 char *name_ptr;
5163 int name_len;
16cdcec7 5164 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 5165
5166 /* FIXME, use a real flag for deciding about the key type */
5167 if (root->fs_info->tree_root == root)
5168 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5169
3954401f
CM		 5170 /* special case for "." */
5171 if (filp->f_pos == 0) {
3765fefa
HS		 5172 over = filldir(dirent, ".", 1,
5173 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 5174 if (over)
5175 return 0;
5176 filp->f_pos = 1;
5177 }
3954401f
CM		 5178 /* special case for .., just use the back ref */
5179 if (filp->f_pos == 1) {
5ecc7e5d 5180 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 5181 over = filldir(dirent, "..", 2,
3765fefa 5182 filp->f_pos, pino, DT_DIR);
3954401f 5183 if (over)
49593bfa 5184 return 0;
3954401f
CM		 5185 filp->f_pos = 2;
5186 }
49593bfa 5187 path = btrfs_alloc_path();
16cdcec7
MX		 5188 if (!path)
5189 return -ENOMEM;
ff5714cc 5190
026fd317 5191 path->reada = 1;
49593bfa 5192
16cdcec7
MX		 5193 if (key_type == BTRFS_DIR_INDEX_KEY) {
5194 INIT_LIST_HEAD(&ins_list);
5195 INIT_LIST_HEAD(&del_list);
5196 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5197 }
5198
39279cc3
CM		 5199 btrfs_set_key_type(&key, key_type);
5200 key.offset = filp->f_pos;
33345d01 5201 key.objectid = btrfs_ino(inode);
5f39d397 5202
39279cc3
CM		 5203 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5204 if (ret < 0)
5205 goto err;
49593bfa
DW		 5206
5207 while (1) {
5f39d397 5208 leaf = path->nodes[0];
39279cc3 5209 slot = path->slots[0];
b9e03af0
LZ		 5210 if (slot >= btrfs_header_nritems(leaf)) {
5211 ret = btrfs_next_leaf(root, path);
5212 if (ret < 0)
5213 goto err;
5214 else if (ret > 0)
5215 break;
5216 continue;
39279cc3 5217 }
3de4586c 5218
5f39d397
CM		 5219 item = btrfs_item_nr(leaf, slot);
5220 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5221
5222 if (found_key.objectid != key.objectid)
39279cc3 5223 break;
5f39d397 5224 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5225 break;
5f39d397 5226 if (found_key.offset < filp->f_pos)
b9e03af0 5227 goto next;
16cdcec7
MX		 5228 if (key_type == BTRFS_DIR_INDEX_KEY &&
5229 btrfs_should_delete_dir_index(&del_list,
5230 found_key.offset))
5231 goto next;
5f39d397
CM		 5232
5233 filp->f_pos = found_key.offset;
16cdcec7 5234 is_curr = 1;
49593bfa 5235
39279cc3
CM		 5236 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5237 di_cur = 0;
5f39d397 5238 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5239
5240 while (di_cur < di_total) {
5f39d397
CM		 5241 struct btrfs_key location;
5242
22a94d44
JB		 5243 if (verify_dir_item(root, leaf, di))
5244 break;
5245
5f39d397 5246 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5247 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5248 name_ptr = tmp_name;
5249 } else {
5250 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5251 if (!name_ptr) {
5252 ret = -ENOMEM;
5253 goto err;
5254 }
5f39d397
CM		 5255 }
5256 read_extent_buffer(leaf, name_ptr,
5257 (unsigned long)(di + 1), name_len);
5258
5259 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5260 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5261
fede766f 5262
3de4586c 5263 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5264 * skip it.
5265 *
5266 * In contrast to old kernels, we insert the snapshot's
5267 * dir item and dir index after it has been created, so
5268 * we won't find a reference to our own snapshot. We
5269 * still keep the following code for backward
5270 * compatibility.
3de4586c
CM		 5271 */
5272 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5273 location.objectid == root->root_key.objectid) {
5274 over = 0;
5275 goto skip;
5276 }
5f39d397 5277 over = filldir(dirent, name_ptr, name_len,
49593bfa 5278 found_key.offset, location.objectid,
39279cc3 5279 d_type);
5f39d397 5280
3de4586c 5281skip:
5f39d397
CM		 5282 if (name_ptr != tmp_name)
5283 kfree(name_ptr);
5284
39279cc3
CM		 5285 if (over)
5286 goto nopos;
5103e947 5287 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5288 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5289 di_cur += di_len;
5290 di = (struct btrfs_dir_item *)((char *)di + di_len);
5291 }
b9e03af0
LZ		 5292next:
5293 path->slots[0]++;
39279cc3 5294 }
49593bfa 5295
16cdcec7
MX		 5296 if (key_type == BTRFS_DIR_INDEX_KEY) {
5297 if (is_curr)
5298 filp->f_pos++;
5299 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5300 &ins_list);
5301 if (ret)
5302 goto nopos;
5303 }
5304
49593bfa 5305 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 5306 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 5307 /*
5308 * 32-bit glibc will use getdents64, but then strtol -
5309 * so the last number we can serve is this.
5310 */
5311 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 5312 else
5313 filp->f_pos++;
39279cc3
CM		 5314nopos:
5315 ret = 0;
5316err:
16cdcec7
MX		 5317 if (key_type == BTRFS_DIR_INDEX_KEY)
5318 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5319 btrfs_free_path(path);
39279cc3
CM		 5320 return ret;
5321}
5322
a9185b41 5323int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5324{
5325 struct btrfs_root *root = BTRFS_I(inode)->root;
5326 struct btrfs_trans_handle *trans;
5327 int ret = 0;
0af3d00b 5328 bool nolock = false;
39279cc3 5329
72ac3c0d 5330 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5331 return 0;
5332
83eea1f1 5333 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5334 nolock = true;
0af3d00b 5335
a9185b41 5336 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5337 if (nolock)
7a7eaa40 5338 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5339 else
7a7eaa40 5340 trans = btrfs_join_transaction(root);
3612b495
TI		 5341 if (IS_ERR(trans))
5342 return PTR_ERR(trans);
a698d075 5343 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5344 }
5345 return ret;
5346}
5347
5348/*
54aa1f4d 5349 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5350 * inode changes. But, it is most likely to find the inode in cache.
5351 * FIXME, needs more benchmarking...there are no reasons other than performance
5352 * to keep or drop this code.
5353 */
48a3b636 5354static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5355{
5356 struct btrfs_root *root = BTRFS_I(inode)->root;
5357 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5358 int ret;
5359
72ac3c0d 5360 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5361 return 0;
39279cc3 5362
7a7eaa40 5363 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5364 if (IS_ERR(trans))
5365 return PTR_ERR(trans);
8929ecfa
YZ		 5366
5367 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5368 if (ret && ret == -ENOSPC) {
5369 /* whoops, lets try again with the full transaction */
5370 btrfs_end_transaction(trans, root);
5371 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5372 if (IS_ERR(trans))
5373 return PTR_ERR(trans);
8929ecfa 5374
94b60442 5375 ret = btrfs_update_inode(trans, root, inode);
94b60442 5376 }
39279cc3 5377 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5378 if (BTRFS_I(inode)->delayed_node)
5379 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5380
5381 return ret;
5382}
5383
5384/*
5385 * This is a copy of file_update_time. We need this so we can return error on
5386 * ENOSPC for updating the inode in the case of file write and mmap writes.
5387 */
e41f941a
JB		 5388static int btrfs_update_time(struct inode *inode, struct timespec *now,
5389 int flags)
22c44fe6 5390{
2bc55652
AB		 5391 struct btrfs_root *root = BTRFS_I(inode)->root;
5392
5393 if (btrfs_root_readonly(root))
5394 return -EROFS;
5395
e41f941a 5396 if (flags & S_VERSION)
22c44fe6 5397 inode_inc_iversion(inode);
e41f941a
JB		 5398 if (flags & S_CTIME)
5399 inode->i_ctime = *now;
5400 if (flags & S_MTIME)
5401 inode->i_mtime = *now;
5402 if (flags & S_ATIME)
5403 inode->i_atime = *now;
5404 return btrfs_dirty_inode(inode);
39279cc3
CM		 5405}
5406
d352ac68
CM		 5407/*
5408 * find the highest existing sequence number in a directory
5409 * and then set the in-memory index_cnt variable to reflect
5410 * free sequence numbers
5411 */
aec7477b
JB		 5412static int btrfs_set_inode_index_count(struct inode *inode)
5413{
5414 struct btrfs_root *root = BTRFS_I(inode)->root;
5415 struct btrfs_key key, found_key;
5416 struct btrfs_path *path;
5417 struct extent_buffer *leaf;
5418 int ret;
5419
33345d01 5420 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5421 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5422 key.offset = (u64)-1;
5423
5424 path = btrfs_alloc_path();
5425 if (!path)
5426 return -ENOMEM;
5427
5428 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5429 if (ret < 0)
5430 goto out;
5431 /* FIXME: we should be able to handle this */
5432 if (ret == 0)
5433 goto out;
5434 ret = 0;
5435
5436 /*
5437 * MAGIC NUMBER EXPLANATION:
5438 * since we search a directory based on f_pos we have to start at 2
5439 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5440 * else has to start at 2
5441 */
5442 if (path->slots[0] == 0) {
5443 BTRFS_I(inode)->index_cnt = 2;
5444 goto out;
5445 }
5446
5447 path->slots[0]--;
5448
5449 leaf = path->nodes[0];
5450 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5451
33345d01 5452 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5453 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5454 BTRFS_I(inode)->index_cnt = 2;
5455 goto out;
5456 }
5457
5458 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5459out:
5460 btrfs_free_path(path);
5461 return ret;
5462}
5463
d352ac68
CM		 5464/*
5465 * helper to find a free sequence number in a given directory. This current
5466 * code is very simple, later versions will do smarter things in the btree
5467 */
3de4586c 5468int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5469{
5470 int ret = 0;
5471
5472 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5473 ret = btrfs_inode_delayed_dir_index_count(dir);
5474 if (ret) {
5475 ret = btrfs_set_inode_index_count(dir);
5476 if (ret)
5477 return ret;
5478 }
aec7477b
JB		 5479 }
5480
00e4e6b3 5481 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5482 BTRFS_I(dir)->index_cnt++;
5483
5484 return ret;
5485}
5486
39279cc3
CM		 5487static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5488 struct btrfs_root *root,
aec7477b 5489 struct inode *dir,
9c58309d 5490 const char *name, int name_len,
175a4eb7
AV		 5491 u64 ref_objectid, u64 objectid,
5492 umode_t mode, u64 *index)
39279cc3
CM		 5493{
5494 struct inode *inode;
5f39d397 5495 struct btrfs_inode_item *inode_item;
39279cc3 5496 struct btrfs_key *location;
5f39d397 5497 struct btrfs_path *path;
9c58309d
CM		 5498 struct btrfs_inode_ref *ref;
5499 struct btrfs_key key[2];
5500 u32 sizes[2];
5501 unsigned long ptr;
39279cc3
CM		 5502 int ret;
5503 int owner;
5504
5f39d397 5505 path = btrfs_alloc_path();
d8926bb3
MF		 5506 if (!path)
5507 return ERR_PTR(-ENOMEM);
5f39d397 5508
39279cc3 5509 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5510 if (!inode) {
5511 btrfs_free_path(path);
39279cc3 5512 return ERR_PTR(-ENOMEM);
8fb27640 5513 }
39279cc3 5514
581bb050
LZ		 5515 /*
5516 * we have to initialize this early, so we can reclaim the inode
5517 * number if we fail afterwards in this function.
5518 */
5519 inode->i_ino = objectid;
5520
aec7477b 5521 if (dir) {
1abe9b8a 5522 trace_btrfs_inode_request(dir);
5523
3de4586c 5524 ret = btrfs_set_inode_index(dir, index);
09771430 5525 if (ret) {
8fb27640 5526 btrfs_free_path(path);
09771430 5527 iput(inode);
aec7477b 5528 return ERR_PTR(ret);
09771430 5529 }
aec7477b
JB		 5530 }
5531 /*
5532 * index_cnt is ignored for everything but a dir,
5533 * btrfs_get_inode_index_count has an explanation for the magic
5534 * number
5535 */
5536 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5537 BTRFS_I(inode)->root = root;
e02119d5 5538 BTRFS_I(inode)->generation = trans->transid;
76195853 5539 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5540
5dc562c5
JB		 5541 /*
5542 * We could have gotten an inode number from somebody who was fsynced
5543 * and then removed in this same transaction, so let's just set full
5544 * sync since it will be a full sync anyway and this will blow away the
5545 * old info in the log.
5546 */
5547 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5548
569254b0 5549 if (S_ISDIR(mode))
39279cc3
CM		 5550 owner = 0;
5551 else
5552 owner = 1;
9c58309d
CM		 5553
5554 key[0].objectid = objectid;
5555 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5556 key[0].offset = 0;
5557
f186373f
MF		 5558 /*
5559 * Start new inodes with an inode_ref. This is slightly more
5560 * efficient for small numbers of hard links since they will
5561 * be packed into one item. Extended refs will kick in if we
5562 * add more hard links than can fit in the ref item.
5563 */
9c58309d
CM		 5564 key[1].objectid = objectid;
5565 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5566 key[1].offset = ref_objectid;
5567
5568 sizes[0] = sizeof(struct btrfs_inode_item);
5569 sizes[1] = name_len + sizeof(*ref);
5570
b9473439 5571 path->leave_spinning = 1;
9c58309d
CM		 5572 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5573 if (ret != 0)
5f39d397
CM		 5574 goto fail;
5575
ecc11fab 5576 inode_init_owner(inode, dir, mode);
a76a3cd4 5577 inode_set_bytes(inode, 0);
39279cc3 5578 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5579 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5580 struct btrfs_inode_item);
293f7e07
LZ		 5581 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5582 sizeof(*inode_item));
e02119d5 5583 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5584
5585 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5586 struct btrfs_inode_ref);
5587 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5588 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5589 ptr = (unsigned long)(ref + 1);
5590 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5591
5f39d397
CM		 5592 btrfs_mark_buffer_dirty(path->nodes[0]);
5593 btrfs_free_path(path);
5594
39279cc3
CM		 5595 location = &BTRFS_I(inode)->location;
5596 location->objectid = objectid;
39279cc3
CM		 5597 location->offset = 0;
5598 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5599
6cbff00f
CH		 5600 btrfs_inherit_iflags(inode, dir);
5601
569254b0 5602 if (S_ISREG(mode)) {
94272164
CM		 5603 if (btrfs_test_opt(root, NODATASUM))
5604 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5605 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5606 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5607 BTRFS_INODE_NODATASUM;
94272164
CM		 5608 }
5609
39279cc3 5610 insert_inode_hash(inode);
5d4f98a2 5611 inode_tree_add(inode);
1abe9b8a 5612
5613 trace_btrfs_inode_new(inode);
1973f0fa 5614 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5615
8ea05e3a
AB		 5616 btrfs_update_root_times(trans, root);
5617
39279cc3 5618 return inode;
5f39d397 5619fail:
aec7477b
JB		 5620 if (dir)
5621 BTRFS_I(dir)->index_cnt--;
5f39d397 5622 btrfs_free_path(path);
09771430 5623 iput(inode);
5f39d397 5624 return ERR_PTR(ret);
39279cc3
CM		 5625}
5626
5627static inline u8 btrfs_inode_type(struct inode *inode)
5628{
5629 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5630}
5631
d352ac68
CM		 5632/*
5633 * utility function to add 'inode' into 'parent_inode' with
5634 * a give name and a given sequence number.
5635 * if 'add_backref' is true, also insert a backref from the
5636 * inode to the parent directory.
5637 */
e02119d5
CM		 5638int btrfs_add_link(struct btrfs_trans_handle *trans,
5639 struct inode *parent_inode, struct inode *inode,
5640 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5641{
4df27c4d 5642 int ret = 0;
39279cc3 5643 struct btrfs_key key;
e02119d5 5644 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5645 u64 ino = btrfs_ino(inode);
5646 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5647
33345d01 5648 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5649 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5650 } else {
33345d01 5651 key.objectid = ino;
4df27c4d
YZ		 5652 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5653 key.offset = 0;
5654 }
5655
33345d01 5656 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5657 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5658 key.objectid, root->root_key.objectid,
33345d01 5659 parent_ino, index, name, name_len);
4df27c4d 5660 } else if (add_backref) {
33345d01
LZ		 5661 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5662 parent_ino, index);
4df27c4d 5663 }
39279cc3 5664
79787eaa
JM		 5665 /* Nothing to clean up yet */
5666 if (ret)
5667 return ret;
4df27c4d 5668
79787eaa
JM		 5669 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5670 parent_inode, &key,
5671 btrfs_inode_type(inode), index);
9c52057c 5672 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5673 goto fail_dir_item;
5674 else if (ret) {
5675 btrfs_abort_transaction(trans, root, ret);
5676 return ret;
39279cc3 5677 }
79787eaa
JM		 5678
5679 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5680 name_len * 2);
0c4d2d95 5681 inode_inc_iversion(parent_inode);
79787eaa
JM		 5682 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5683 ret = btrfs_update_inode(trans, root, parent_inode);
5684 if (ret)
5685 btrfs_abort_transaction(trans, root, ret);
39279cc3 5686 return ret;
fe66a05a
CM		 5687
5688fail_dir_item:
5689 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5690 u64 local_index;
5691 int err;
5692 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5693 key.objectid, root->root_key.objectid,
5694 parent_ino, &local_index, name, name_len);
5695
5696 } else if (add_backref) {
5697 u64 local_index;
5698 int err;
5699
5700 err = btrfs_del_inode_ref(trans, root, name, name_len,
5701 ino, parent_ino, &local_index);
5702 }
5703 return ret;
39279cc3
CM		 5704}
5705
5706static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5707 struct inode *dir, struct dentry *dentry,
5708 struct inode *inode, int backref, u64 index)
39279cc3 5709{
a1b075d2
JB		 5710 int err = btrfs_add_link(trans, dir, inode,
5711 dentry->d_name.name, dentry->d_name.len,
5712 backref, index);
39279cc3
CM		 5713 if (err > 0)
5714 err = -EEXIST;
5715 return err;
5716}
5717
618e21d5 5718static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5719 umode_t mode, dev_t rdev)
618e21d5
JB		 5720{
5721 struct btrfs_trans_handle *trans;
5722 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5723 struct inode *inode = NULL;
618e21d5
JB		 5724 int err;
5725 int drop_inode = 0;
5726 u64 objectid;
00e4e6b3 5727 u64 index = 0;
618e21d5
JB		 5728
5729 if (!new_valid_dev(rdev))
5730 return -EINVAL;
5731
9ed74f2d
JB		 5732 /*
5733 * 2 for inode item and ref
5734 * 2 for dir items
5735 * 1 for xattr if selinux is on
5736 */
a22285a6
YZ		 5737 trans = btrfs_start_transaction(root, 5);
5738 if (IS_ERR(trans))
5739 return PTR_ERR(trans);
1832a6d5 5740
581bb050
LZ		 5741 err = btrfs_find_free_ino(root, &objectid);
5742 if (err)
5743 goto out_unlock;
5744
aec7477b 5745 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5746 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5747 mode, &index);
7cf96da3
TI		 5748 if (IS_ERR(inode)) {
5749 err = PTR_ERR(inode);
618e21d5 5750 goto out_unlock;
7cf96da3 5751 }
618e21d5 5752
2a7dba39 5753 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5754 if (err) {
5755 drop_inode = 1;
5756 goto out_unlock;
5757 }
5758
ad19db71
CS		 5759 /*
5760 * If the active LSM wants to access the inode during
5761 * d_instantiate it needs these. Smack checks to see
5762 * if the filesystem supports xattrs by looking at the
5763 * ops vector.
5764 */
5765
5766 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5767 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5768 if (err)
5769 drop_inode = 1;
5770 else {
618e21d5 5771 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5772 btrfs_update_inode(trans, root, inode);
08c422c2 5773 d_instantiate(dentry, inode);
618e21d5 5774 }
618e21d5 5775out_unlock:
7ad85bb7 5776 btrfs_end_transaction(trans, root);
b53d3f5d 5777 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5778 if (drop_inode) {
5779 inode_dec_link_count(inode);
5780 iput(inode);
5781 }
618e21d5
JB		 5782 return err;
5783}
5784
39279cc3 5785static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5786 umode_t mode, bool excl)
39279cc3
CM		 5787{
5788 struct btrfs_trans_handle *trans;
5789 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5790 struct inode *inode = NULL;
43baa579 5791 int drop_inode_on_err = 0;
a22285a6 5792 int err;
39279cc3 5793 u64 objectid;
00e4e6b3 5794 u64 index = 0;
39279cc3 5795
9ed74f2d
JB		 5796 /*
5797 * 2 for inode item and ref
5798 * 2 for dir items
5799 * 1 for xattr if selinux is on
5800 */
a22285a6
YZ		 5801 trans = btrfs_start_transaction(root, 5);
5802 if (IS_ERR(trans))
5803 return PTR_ERR(trans);
9ed74f2d 5804
581bb050
LZ		 5805 err = btrfs_find_free_ino(root, &objectid);
5806 if (err)
5807 goto out_unlock;
5808
aec7477b 5809 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5810 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5811 mode, &index);
7cf96da3
TI		 5812 if (IS_ERR(inode)) {
5813 err = PTR_ERR(inode);
39279cc3 5814 goto out_unlock;
7cf96da3 5815 }
43baa579 5816 drop_inode_on_err = 1;
39279cc3 5817
2a7dba39 5818 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5819 if (err)
33268eaf 5820 goto out_unlock;
33268eaf 5821
9185aa58
FB		 5822 err = btrfs_update_inode(trans, root, inode);
5823 if (err)
5824 goto out_unlock;
5825
ad19db71
CS		 5826 /*
5827 * If the active LSM wants to access the inode during
5828 * d_instantiate it needs these. Smack checks to see
5829 * if the filesystem supports xattrs by looking at the
5830 * ops vector.
5831 */
5832 inode->i_fop = &btrfs_file_operations;
5833 inode->i_op = &btrfs_file_inode_operations;
5834
a1b075d2 5835 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5836 if (err)
43baa579
FB		 5837 goto out_unlock;
5838
5839 inode->i_mapping->a_ops = &btrfs_aops;
5840 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5841 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5842 d_instantiate(dentry, inode);
5843
39279cc3 5844out_unlock:
7ad85bb7 5845 btrfs_end_transaction(trans, root);
43baa579 5846 if (err && drop_inode_on_err) {
39279cc3
CM		 5847 inode_dec_link_count(inode);
5848 iput(inode);
5849 }
b53d3f5d 5850 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5851 return err;
5852}
5853
5854static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5855 struct dentry *dentry)
5856{
5857 struct btrfs_trans_handle *trans;
5858 struct btrfs_root *root = BTRFS_I(dir)->root;
5859 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5860 u64 index;
39279cc3
CM		 5861 int err;
5862 int drop_inode = 0;
5863
4a8be425
TH		 5864 /* do not allow sys_link's with other subvols of the same device */
5865 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5866 return -EXDEV;
4a8be425 5867
f186373f 5868 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5869 return -EMLINK;
4a8be425 5870
3de4586c 5871 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5872 if (err)
5873 goto fail;
5874
a22285a6 5875 /*
7e6b6465 5876 * 2 items for inode and inode ref
a22285a6 5877 * 2 items for dir items
7e6b6465 5878 * 1 item for parent inode
a22285a6 5879 */
7e6b6465 5880 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5881 if (IS_ERR(trans)) {
5882 err = PTR_ERR(trans);
5883 goto fail;
5884 }
5f39d397 5885
3153495d 5886 btrfs_inc_nlink(inode);
0c4d2d95 5887 inode_inc_iversion(inode);
3153495d 5888 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5889 ihold(inode);
e9976151 5890 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5891
a1b075d2 5892 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5893
a5719521 5894 if (err) {
54aa1f4d 5895 drop_inode = 1;
a5719521 5896 } else {
10d9f309 5897 struct dentry *parent = dentry->d_parent;
a5719521 5898 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5899 if (err)
5900 goto fail;
08c422c2 5901 d_instantiate(dentry, inode);
6a912213 5902 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5903 }
39279cc3 5904
7ad85bb7 5905 btrfs_end_transaction(trans, root);
1832a6d5 5906fail:
39279cc3
CM		 5907 if (drop_inode) {
5908 inode_dec_link_count(inode);
5909 iput(inode);
5910 }
b53d3f5d 5911 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5912 return err;
5913}
5914
18bb1db3 5915static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5916{
b9d86667 5917 struct inode *inode = NULL;
39279cc3
CM		 5918 struct btrfs_trans_handle *trans;
5919 struct btrfs_root *root = BTRFS_I(dir)->root;
5920 int err = 0;
5921 int drop_on_err = 0;
b9d86667 5922 u64 objectid = 0;
00e4e6b3 5923 u64 index = 0;
39279cc3 5924
9ed74f2d
JB		 5925 /*
5926 * 2 items for inode and ref
5927 * 2 items for dir items
5928 * 1 for xattr if selinux is on
5929 */
a22285a6
YZ		 5930 trans = btrfs_start_transaction(root, 5);
5931 if (IS_ERR(trans))
5932 return PTR_ERR(trans);
39279cc3 5933
581bb050
LZ		 5934 err = btrfs_find_free_ino(root, &objectid);
5935 if (err)
5936 goto out_fail;
5937
aec7477b 5938 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5939 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5940 S_IFDIR | mode, &index);
39279cc3
CM		 5941 if (IS_ERR(inode)) {
5942 err = PTR_ERR(inode);
5943 goto out_fail;
5944 }
5f39d397 5945
39279cc3 5946 drop_on_err = 1;
33268eaf 5947
2a7dba39 5948 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5949 if (err)
5950 goto out_fail;
5951
39279cc3
CM		 5952 inode->i_op = &btrfs_dir_inode_operations;
5953 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5954
dbe674a9 5955 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5956 err = btrfs_update_inode(trans, root, inode);
5957 if (err)
5958 goto out_fail;
5f39d397 5959
a1b075d2
JB		 5960 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5961 dentry->d_name.len, 0, index);
39279cc3
CM		 5962 if (err)
5963 goto out_fail;
5f39d397 5964
39279cc3
CM		 5965 d_instantiate(dentry, inode);
5966 drop_on_err = 0;
39279cc3
CM		 5967
5968out_fail:
7ad85bb7 5969 btrfs_end_transaction(trans, root);
39279cc3
CM		 5970 if (drop_on_err)
5971 iput(inode);
b53d3f5d 5972 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5973 return err;
5974}
5975
d352ac68
CM		 5976/* helper for btfs_get_extent. Given an existing extent in the tree,
5977 * and an extent that you want to insert, deal with overlap and insert
5978 * the new extent into the tree.
5979 */
3b951516
CM		 5980static int merge_extent_mapping(struct extent_map_tree *em_tree,
5981 struct extent_map *existing,
e6dcd2dc
CM		 5982 struct extent_map *em,
5983 u64 map_start, u64 map_len)
3b951516
CM		 5984{
5985 u64 start_diff;
3b951516 5986
e6dcd2dc
CM		 5987 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5988 start_diff = map_start - em->start;
5989 em->start = map_start;
5990 em->len = map_len;
c8b97818
CM		 5991 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5992 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5993 em->block_start += start_diff;
c8b97818
CM		 5994 em->block_len -= start_diff;
5995 }
09a2a8f9 5996 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 5997}
5998
c8b97818
CM		 5999static noinline int uncompress_inline(struct btrfs_path *path,
6000 struct inode *inode, struct page *page,
6001 size_t pg_offset, u64 extent_offset,
6002 struct btrfs_file_extent_item *item)
6003{
6004 int ret;
6005 struct extent_buffer *leaf = path->nodes[0];
6006 char *tmp;
6007 size_t max_size;
6008 unsigned long inline_size;
6009 unsigned long ptr;
261507a0 6010 int compress_type;
c8b97818
CM		 6011
6012 WARN_ON(pg_offset != 0);
261507a0 6013 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6014 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6015 inline_size = btrfs_file_extent_inline_item_len(leaf,
6016 btrfs_item_nr(leaf, path->slots[0]));
6017 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6018 if (!tmp)
6019 return -ENOMEM;
c8b97818
CM		 6020 ptr = btrfs_file_extent_inline_start(item);
6021
6022 read_extent_buffer(leaf, tmp, ptr, inline_size);
6023
5b050f04 6024 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6025 ret = btrfs_decompress(compress_type, tmp, page,
6026 extent_offset, inline_size, max_size);
c8b97818 6027 if (ret) {
7ac687d9 6028 char *kaddr = kmap_atomic(page);
c8b97818
CM		 6029 unsigned long copy_size = min_t(u64,
6030 PAGE_CACHE_SIZE - pg_offset,
6031 max_size - extent_offset);
6032 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 6033 kunmap_atomic(kaddr);
c8b97818
CM		 6034 }
6035 kfree(tmp);
6036 return 0;
6037}
6038
d352ac68
CM		 6039/*
6040 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6041 * the ugly parts come from merging extents from the disk with the in-ram
6042 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6043 * where the in-ram extents might be locked pending data=ordered completion.
6044 *
6045 * This also copies inline extents directly into the page.
6046 */
d397712b 6047
a52d9a80 6048struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6049 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6050 int create)
6051{
6052 int ret;
6053 int err = 0;
db94535d 6054 u64 bytenr;
a52d9a80
CM		 6055 u64 extent_start = 0;
6056 u64 extent_end = 0;
33345d01 6057 u64 objectid = btrfs_ino(inode);
a52d9a80 6058 u32 found_type;
f421950f 6059 struct btrfs_path *path = NULL;
a52d9a80
CM		 6060 struct btrfs_root *root = BTRFS_I(inode)->root;
6061 struct btrfs_file_extent_item *item;
5f39d397
CM		 6062 struct extent_buffer *leaf;
6063 struct btrfs_key found_key;
a52d9a80
CM		 6064 struct extent_map *em = NULL;
6065 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6066 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6067 struct btrfs_trans_handle *trans = NULL;
261507a0 6068 int compress_type;
a52d9a80 6069
a52d9a80 6070again:
890871be 6071 read_lock(&em_tree->lock);
d1310b2e 6072 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6073 if (em)
6074 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6075 read_unlock(&em_tree->lock);
d1310b2e 6076
a52d9a80 6077 if (em) {
e1c4b745
CM		 6078 if (em->start > start || em->start + em->len <= start)
6079 free_extent_map(em);
6080 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6081 free_extent_map(em);
6082 else
6083 goto out;
a52d9a80 6084 }
172ddd60 6085 em = alloc_extent_map();
a52d9a80 6086 if (!em) {
d1310b2e
CM		 6087 err = -ENOMEM;
6088 goto out;
a52d9a80 6089 }
e6dcd2dc 6090 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6091 em->start = EXTENT_MAP_HOLE;
445a6944 6092 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6093 em->len = (u64)-1;
c8b97818 6094 em->block_len = (u64)-1;
f421950f
CM		 6095
6096 if (!path) {
6097 path = btrfs_alloc_path();
026fd317
JB		 6098 if (!path) {
6099 err = -ENOMEM;
6100 goto out;
6101 }
6102 /*
6103 * Chances are we'll be called again, so go ahead and do
6104 * readahead
6105 */
6106 path->reada = 1;
f421950f
CM		 6107 }
6108
179e29e4
CM		 6109 ret = btrfs_lookup_file_extent(trans, root, path,
6110 objectid, start, trans != NULL);
a52d9a80
CM		 6111 if (ret < 0) {
6112 err = ret;
6113 goto out;
6114 }
6115
6116 if (ret != 0) {
6117 if (path->slots[0] == 0)
6118 goto not_found;
6119 path->slots[0]--;
6120 }
6121
5f39d397
CM		 6122 leaf = path->nodes[0];
6123 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6124 struct btrfs_file_extent_item);
a52d9a80 6125 /* are we inside the extent that was found? */
5f39d397
CM		 6126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6127 found_type = btrfs_key_type(&found_key);
6128 if (found_key.objectid != objectid ||
a52d9a80
CM		 6129 found_type != BTRFS_EXTENT_DATA_KEY) {
6130 goto not_found;
6131 }
6132
5f39d397
CM		 6133 found_type = btrfs_file_extent_type(leaf, item);
6134 extent_start = found_key.offset;
261507a0 6135 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 6136 if (found_type == BTRFS_FILE_EXTENT_REG ||
6137 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6138 extent_end = extent_start +
db94535d 6139 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6140 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6141 size_t size;
6142 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 6143 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102
YZ		 6144 }
6145
6146 if (start >= extent_end) {
6147 path->slots[0]++;
6148 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6149 ret = btrfs_next_leaf(root, path);
6150 if (ret < 0) {
6151 err = ret;
6152 goto out;
a52d9a80 6153 }
9036c102
YZ		 6154 if (ret > 0)
6155 goto not_found;
6156 leaf = path->nodes[0];
a52d9a80 6157 }
9036c102
YZ		 6158 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6159 if (found_key.objectid != objectid ||
6160 found_key.type != BTRFS_EXTENT_DATA_KEY)
6161 goto not_found;
6162 if (start + len <= found_key.offset)
6163 goto not_found;
6164 em->start = start;
70c8a91c 6165 em->orig_start = start;
9036c102
YZ		 6166 em->len = found_key.offset - start;
6167 goto not_found_em;
6168 }
6169
cc95bef6 6170 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6171 if (found_type == BTRFS_FILE_EXTENT_REG ||
6172 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6173 em->start = extent_start;
6174 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6175 em->orig_start = extent_start -
6176 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6177 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6178 item);
db94535d
CM		 6179 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6180 if (bytenr == 0) {
5f39d397 6181 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6182 goto insert;
6183 }
261507a0 6184 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6185 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6186 em->compress_type = compress_type;
c8b97818 6187 em->block_start = bytenr;
b4939680 6188 em->block_len = em->orig_block_len;
c8b97818
CM		 6189 } else {
6190 bytenr += btrfs_file_extent_offset(leaf, item);
6191 em->block_start = bytenr;
6192 em->block_len = em->len;
d899e052
YZ		 6193 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6194 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6195 }
a52d9a80
CM		 6196 goto insert;
6197 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6198 unsigned long ptr;
a52d9a80 6199 char *map;
3326d1b0
CM		 6200 size_t size;
6201 size_t extent_offset;
6202 size_t copy_size;
a52d9a80 6203
689f9346 6204 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6205 if (!page || create) {
689f9346 6206 em->start = extent_start;
9036c102 6207 em->len = extent_end - extent_start;
689f9346
Y		 6208 goto out;
6209 }
5f39d397 6210
9036c102
YZ		 6211 size = btrfs_file_extent_inline_len(leaf, item);
6212 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6213 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6214 size - extent_offset);
3326d1b0 6215 em->start = extent_start + extent_offset;
fda2832f 6216 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6217 em->orig_block_len = em->len;
70c8a91c 6218 em->orig_start = em->start;
261507a0 6219 if (compress_type) {
c8b97818 6220 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6221 em->compress_type = compress_type;
6222 }
689f9346 6223 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6224 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6225 if (btrfs_file_extent_compression(leaf, item) !=
6226 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6227 ret = uncompress_inline(path, inode, page,
6228 pg_offset,
6229 extent_offset, item);
79787eaa 6230 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6231 } else {
6232 map = kmap(page);
6233 read_extent_buffer(leaf, map + pg_offset, ptr,
6234 copy_size);
93c82d57
CM		 6235 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6236 memset(map + pg_offset + copy_size, 0,
6237 PAGE_CACHE_SIZE - pg_offset -
6238 copy_size);
6239 }
c8b97818
CM		 6240 kunmap(page);
6241 }
179e29e4
CM		 6242 flush_dcache_page(page);
6243 } else if (create && PageUptodate(page)) {
6bf7e080 6244 BUG();
179e29e4
CM		 6245 if (!trans) {
6246 kunmap(page);
6247 free_extent_map(em);
6248 em = NULL;
ff5714cc 6249
b3b4aa74 6250 btrfs_release_path(path);
7a7eaa40 6251 trans = btrfs_join_transaction(root);
ff5714cc 6252
3612b495
TI		 6253 if (IS_ERR(trans))
6254 return ERR_CAST(trans);
179e29e4
CM		 6255 goto again;
6256 }
c8b97818 6257 map = kmap(page);
70dec807 6258 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6259 copy_size);
c8b97818 6260 kunmap(page);
179e29e4 6261 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6262 }
d1310b2e 6263 set_extent_uptodate(io_tree, em->start,
507903b8 6264 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6265 goto insert;
6266 } else {
31b1a2bd 6267 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6268 }
6269not_found:
6270 em->start = start;
70c8a91c 6271 em->orig_start = start;
d1310b2e 6272 em->len = len;
a52d9a80 6273not_found_em:
5f39d397 6274 em->block_start = EXTENT_MAP_HOLE;
9036c102 6275 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6276insert:
b3b4aa74 6277 btrfs_release_path(path);
d1310b2e 6278 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb
SK		 6279 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6280 (unsigned long long)em->start,
6281 (unsigned long long)em->len,
6282 (unsigned long long)start,
6283 (unsigned long long)len);
a52d9a80
CM		 6284 err = -EIO;
6285 goto out;
6286 }
d1310b2e
CM		 6287
6288 err = 0;
890871be 6289 write_lock(&em_tree->lock);
09a2a8f9 6290 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6291 /* it is possible that someone inserted the extent into the tree
6292 * while we had the lock dropped. It is also possible that
6293 * an overlapping map exists in the tree
6294 */
a52d9a80 6295 if (ret == -EEXIST) {
3b951516 6296 struct extent_map *existing;
e6dcd2dc
CM		 6297
6298 ret = 0;
6299
3b951516 6300 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6301 if (existing && (existing->start > start ||
6302 existing->start + existing->len <= start)) {
6303 free_extent_map(existing);
6304 existing = NULL;
6305 }
3b951516
CM		 6306 if (!existing) {
6307 existing = lookup_extent_mapping(em_tree, em->start,
6308 em->len);
6309 if (existing) {
6310 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6311 em, start,
6312 root->sectorsize);
3b951516
CM		 6313 free_extent_map(existing);
6314 if (err) {
6315 free_extent_map(em);
6316 em = NULL;
6317 }
6318 } else {
6319 err = -EIO;
3b951516
CM		 6320 free_extent_map(em);
6321 em = NULL;
6322 }
6323 } else {
6324 free_extent_map(em);
6325 em = existing;
e6dcd2dc 6326 err = 0;
a52d9a80 6327 }
a52d9a80 6328 }
890871be 6329 write_unlock(&em_tree->lock);
a52d9a80 6330out:
1abe9b8a 6331
f0bd95ea
TI		 6332 if (em)
6333 trace_btrfs_get_extent(root, em);
1abe9b8a 6334
f421950f
CM		 6335 if (path)
6336 btrfs_free_path(path);
a52d9a80
CM		 6337 if (trans) {
6338 ret = btrfs_end_transaction(trans, root);
d397712b 6339 if (!err)
a52d9a80
CM		 6340 err = ret;
6341 }
a52d9a80
CM		 6342 if (err) {
6343 free_extent_map(em);
a52d9a80
CM		 6344 return ERR_PTR(err);
6345 }
79787eaa 6346 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6347 return em;
6348}
6349
ec29ed5b
CM		 6350struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6351 size_t pg_offset, u64 start, u64 len,
6352 int create)
6353{
6354 struct extent_map *em;
6355 struct extent_map *hole_em = NULL;
6356 u64 range_start = start;
6357 u64 end;
6358 u64 found;
6359 u64 found_end;
6360 int err = 0;
6361
6362 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6363 if (IS_ERR(em))
6364 return em;
6365 if (em) {
6366 /*
f9e4fb53
LB		 6367 * if our em maps to
6368 * - a hole or
6369 * - a pre-alloc extent,
6370 * there might actually be delalloc bytes behind it.
ec29ed5b 6371 */
f9e4fb53
LB		 6372 if (em->block_start != EXTENT_MAP_HOLE &&
6373 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6374 return em;
6375 else
6376 hole_em = em;
6377 }
6378
6379 /* check to see if we've wrapped (len == -1 or similar) */
6380 end = start + len;
6381 if (end < start)
6382 end = (u64)-1;
6383 else
6384 end -= 1;
6385
6386 em = NULL;
6387
6388 /* ok, we didn't find anything, lets look for delalloc */
6389 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6390 end, len, EXTENT_DELALLOC, 1);
6391 found_end = range_start + found;
6392 if (found_end < range_start)
6393 found_end = (u64)-1;
6394
6395 /*
6396 * we didn't find anything useful, return
6397 * the original results from get_extent()
6398 */
6399 if (range_start > end || found_end <= start) {
6400 em = hole_em;
6401 hole_em = NULL;
6402 goto out;
6403 }
6404
6405 /* adjust the range_start to make sure it doesn't
6406 * go backwards from the start they passed in
6407 */
6408 range_start = max(start,range_start);
6409 found = found_end - range_start;
6410
6411 if (found > 0) {
6412 u64 hole_start = start;
6413 u64 hole_len = len;
6414
172ddd60 6415 em = alloc_extent_map();
ec29ed5b
CM		 6416 if (!em) {
6417 err = -ENOMEM;
6418 goto out;
6419 }
6420 /*
6421 * when btrfs_get_extent can't find anything it
6422 * returns one huge hole
6423 *
6424 * make sure what it found really fits our range, and
6425 * adjust to make sure it is based on the start from
6426 * the caller
6427 */
6428 if (hole_em) {
6429 u64 calc_end = extent_map_end(hole_em);
6430
6431 if (calc_end <= start || (hole_em->start > end)) {
6432 free_extent_map(hole_em);
6433 hole_em = NULL;
6434 } else {
6435 hole_start = max(hole_em->start, start);
6436 hole_len = calc_end - hole_start;
6437 }
6438 }
6439 em->bdev = NULL;
6440 if (hole_em && range_start > hole_start) {
6441 /* our hole starts before our delalloc, so we
6442 * have to return just the parts of the hole
6443 * that go until the delalloc starts
6444 */
6445 em->len = min(hole_len,
6446 range_start - hole_start);
6447 em->start = hole_start;
6448 em->orig_start = hole_start;
6449 /*
6450 * don't adjust block start at all,
6451 * it is fixed at EXTENT_MAP_HOLE
6452 */
6453 em->block_start = hole_em->block_start;
6454 em->block_len = hole_len;
f9e4fb53
LB		 6455 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6456 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6457 } else {
6458 em->start = range_start;
6459 em->len = found;
6460 em->orig_start = range_start;
6461 em->block_start = EXTENT_MAP_DELALLOC;
6462 em->block_len = found;
6463 }
6464 } else if (hole_em) {
6465 return hole_em;
6466 }
6467out:
6468
6469 free_extent_map(hole_em);
6470 if (err) {
6471 free_extent_map(em);
6472 return ERR_PTR(err);
6473 }
6474 return em;
6475}
6476
4b46fce2
JB		 6477static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6478 u64 start, u64 len)
6479{
6480 struct btrfs_root *root = BTRFS_I(inode)->root;
6481 struct btrfs_trans_handle *trans;
70c8a91c 6482 struct extent_map *em;
4b46fce2
JB		 6483 struct btrfs_key ins;
6484 u64 alloc_hint;
6485 int ret;
4b46fce2 6486
7a7eaa40 6487 trans = btrfs_join_transaction(root);
3612b495
TI		 6488 if (IS_ERR(trans))
6489 return ERR_CAST(trans);
4b46fce2
JB		 6490
6491 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6492
6493 alloc_hint = get_extent_allocation_hint(inode, start, len);
6494 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 6495 alloc_hint, &ins, 1);
4b46fce2
JB		 6496 if (ret) {
6497 em = ERR_PTR(ret);
6498 goto out;
6499 }
6500
70c8a91c 6501 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6502 ins.offset, ins.offset, ins.offset, 0);
70c8a91c
JB		 6503 if (IS_ERR(em))
6504 goto out;
4b46fce2
JB		 6505
6506 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6507 ins.offset, ins.offset, 0);
6508 if (ret) {
6509 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6510 em = ERR_PTR(ret);
6511 }
6512out:
6513 btrfs_end_transaction(trans, root);
6514 return em;
6515}
6516
46bfbb5c
CM		 6517/*
6518 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6519 * block must be cow'd
6520 */
6521static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
eb384b55
JB		 6522 struct inode *inode, u64 offset, u64 *len,
6523 u64 *orig_start, u64 *orig_block_len,
6524 u64 *ram_bytes)
46bfbb5c
CM		 6525{
6526 struct btrfs_path *path;
6527 int ret;
6528 struct extent_buffer *leaf;
6529 struct btrfs_root *root = BTRFS_I(inode)->root;
6530 struct btrfs_file_extent_item *fi;
6531 struct btrfs_key key;
6532 u64 disk_bytenr;
6533 u64 backref_offset;
6534 u64 extent_end;
6535 u64 num_bytes;
6536 int slot;
6537 int found_type;
6538
6539 path = btrfs_alloc_path();
6540 if (!path)
6541 return -ENOMEM;
6542
33345d01 6543 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6544 offset, 0);
6545 if (ret < 0)
6546 goto out;
6547
6548 slot = path->slots[0];
6549 if (ret == 1) {
6550 if (slot == 0) {
6551 /* can't find the item, must cow */
6552 ret = 0;
6553 goto out;
6554 }
6555 slot--;
6556 }
6557 ret = 0;
6558 leaf = path->nodes[0];
6559 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6560 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6561 key.type != BTRFS_EXTENT_DATA_KEY) {
6562 /* not our file or wrong item type, must cow */
6563 goto out;
6564 }
6565
6566 if (key.offset > offset) {
6567 /* Wrong offset, must cow */
6568 goto out;
6569 }
6570
6571 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6572 found_type = btrfs_file_extent_type(leaf, fi);
6573 if (found_type != BTRFS_FILE_EXTENT_REG &&
6574 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6575 /* not a regular extent, must cow */
6576 goto out;
6577 }
6578 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6579 backref_offset = btrfs_file_extent_offset(leaf, fi);
6580
eb384b55
JB		 6581 *orig_start = key.offset - backref_offset;
6582 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6583 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6584
46bfbb5c 6585 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
eb384b55 6586 if (extent_end < offset + *len) {
46bfbb5c
CM		 6587 /* extent doesn't include our full range, must cow */
6588 goto out;
6589 }
6590
6591 if (btrfs_extent_readonly(root, disk_bytenr))
6592 goto out;
6593
6594 /*
6595 * look for other files referencing this extent, if we
6596 * find any we must cow
6597 */
33345d01 6598 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 6599 key.offset - backref_offset, disk_bytenr))
6600 goto out;
6601
6602 /*
6603 * adjust disk_bytenr and num_bytes to cover just the bytes
6604 * in this extent we are about to write. If there
6605 * are any csums in that range we have to cow in order
6606 * to keep the csums correct
6607 */
6608 disk_bytenr += backref_offset;
6609 disk_bytenr += offset - key.offset;
eb384b55 6610 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6611 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6612 goto out;
6613 /*
6614 * all of the above have passed, it is safe to overwrite this extent
6615 * without cow
6616 */
eb384b55 6617 *len = num_bytes;
46bfbb5c
CM		 6618 ret = 1;
6619out:
6620 btrfs_free_path(path);
6621 return ret;
6622}
6623
eb838e73
JB		 6624static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6625 struct extent_state **cached_state, int writing)
6626{
6627 struct btrfs_ordered_extent *ordered;
6628 int ret = 0;
6629
6630 while (1) {
6631 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6632 0, cached_state);
6633 /*
6634 * We're concerned with the entire range that we're going to be
6635 * doing DIO to, so we need to make sure theres no ordered
6636 * extents in this range.
6637 */
6638 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6639 lockend - lockstart + 1);
6640
6641 /*
6642 * We need to make sure there are no buffered pages in this
6643 * range either, we could have raced between the invalidate in
6644 * generic_file_direct_write and locking the extent. The
6645 * invalidate needs to happen so that reads after a write do not
6646 * get stale data.
6647 */
6648 if (!ordered && (!writing ||
6649 !test_range_bit(&BTRFS_I(inode)->io_tree,
6650 lockstart, lockend, EXTENT_UPTODATE, 0,
6651 *cached_state)))
6652 break;
6653
6654 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6655 cached_state, GFP_NOFS);
6656
6657 if (ordered) {
6658 btrfs_start_ordered_extent(inode, ordered, 1);
6659 btrfs_put_ordered_extent(ordered);
6660 } else {
6661 /* Screw you mmap */
6662 ret = filemap_write_and_wait_range(inode->i_mapping,
6663 lockstart,
6664 lockend);
6665 if (ret)
6666 break;
6667
6668 /*
6669 * If we found a page that couldn't be invalidated just
6670 * fall back to buffered.
6671 */
6672 ret = invalidate_inode_pages2_range(inode->i_mapping,
6673 lockstart >> PAGE_CACHE_SHIFT,
6674 lockend >> PAGE_CACHE_SHIFT);
6675 if (ret)
6676 break;
6677 }
6678
6679 cond_resched();
6680 }
6681
6682 return ret;
6683}
6684
69ffb543
JB		 6685static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6686 u64 len, u64 orig_start,
6687 u64 block_start, u64 block_len,
cc95bef6
JB		 6688 u64 orig_block_len, u64 ram_bytes,
6689 int type)
69ffb543
JB		 6690{
6691 struct extent_map_tree *em_tree;
6692 struct extent_map *em;
6693 struct btrfs_root *root = BTRFS_I(inode)->root;
6694 int ret;
6695
6696 em_tree = &BTRFS_I(inode)->extent_tree;
6697 em = alloc_extent_map();
6698 if (!em)
6699 return ERR_PTR(-ENOMEM);
6700
6701 em->start = start;
6702 em->orig_start = orig_start;
2ab28f32
JB		 6703 em->mod_start = start;
6704 em->mod_len = len;
69ffb543
JB		 6705 em->len = len;
6706 em->block_len = block_len;
6707 em->block_start = block_start;
6708 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6709 em->orig_block_len = orig_block_len;
cc95bef6 6710 em->ram_bytes = ram_bytes;
70c8a91c 6711 em->generation = -1;
69ffb543
JB		 6712 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6713 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6714 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6715
6716 do {
6717 btrfs_drop_extent_cache(inode, em->start,
6718 em->start + em->len - 1, 0);
6719 write_lock(&em_tree->lock);
09a2a8f9 6720 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6721 write_unlock(&em_tree->lock);
6722 } while (ret == -EEXIST);
6723
6724 if (ret) {
6725 free_extent_map(em);
6726 return ERR_PTR(ret);
6727 }
6728
6729 return em;
6730}
6731
6732
4b46fce2
JB		 6733static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6734 struct buffer_head *bh_result, int create)
6735{
6736 struct extent_map *em;
6737 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6738 struct extent_state *cached_state = NULL;
4b46fce2 6739 u64 start = iblock << inode->i_blkbits;
eb838e73 6740 u64 lockstart, lockend;
4b46fce2 6741 u64 len = bh_result->b_size;
46bfbb5c 6742 struct btrfs_trans_handle *trans;
eb838e73 6743 int unlock_bits = EXTENT_LOCKED;
0934856d 6744 int ret = 0;
eb838e73 6745
172a5049 6746 if (create)
eb838e73 6747 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6748 else
c329861d 6749 len = min_t(u64, len, root->sectorsize);
eb838e73 6750
c329861d
JB		 6751 lockstart = start;
6752 lockend = start + len - 1;
6753
eb838e73
JB		 6754 /*
6755 * If this errors out it's because we couldn't invalidate pagecache for
6756 * this range and we need to fallback to buffered.
6757 */
6758 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6759 return -ENOTBLK;
6760
4b46fce2 6761 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6762 if (IS_ERR(em)) {
6763 ret = PTR_ERR(em);
6764 goto unlock_err;
6765 }
4b46fce2
JB		 6766
6767 /*
6768 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6769 * io. INLINE is special, and we could probably kludge it in here, but
6770 * it's still buffered so for safety lets just fall back to the generic
6771 * buffered path.
6772 *
6773 * For COMPRESSED we _have_ to read the entire extent in so we can
6774 * decompress it, so there will be buffering required no matter what we
6775 * do, so go ahead and fallback to buffered.
6776 *
6777 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6778 * to buffered IO. Don't blame me, this is the price we pay for using
6779 * the generic code.
6780 */
6781 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6782 em->block_start == EXTENT_MAP_INLINE) {
6783 free_extent_map(em);
eb838e73
JB		 6784 ret = -ENOTBLK;
6785 goto unlock_err;
4b46fce2
JB		 6786 }
6787
6788 /* Just a good old fashioned hole, return */
6789 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6790 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6791 free_extent_map(em);
eb838e73 6792 goto unlock_err;
4b46fce2
JB		 6793 }
6794
6795 /*
6796 * We don't allocate a new extent in the following cases
6797 *
6798 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6799 * existing extent.
6800 * 2) The extent is marked as PREALLOC. We're good to go here and can
6801 * just use the extent.
6802 *
6803 */
46bfbb5c 6804 if (!create) {
eb838e73
JB		 6805 len = min(len, em->len - (start - em->start));
6806 lockstart = start + len;
6807 goto unlock;
46bfbb5c 6808 }
4b46fce2
JB		 6809
6810 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6811 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6812 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6813 int type;
6814 int ret;
eb384b55 6815 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6816
6817 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6818 type = BTRFS_ORDERED_PREALLOC;
6819 else
6820 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6821 len = min(len, em->len - (start - em->start));
4b46fce2 6822 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 6823
6824 /*
6825 * we're not going to log anything, but we do need
6826 * to make sure the current transaction stays open
6827 * while we look for nocow cross refs
6828 */
7a7eaa40 6829 trans = btrfs_join_transaction(root);
3612b495 6830 if (IS_ERR(trans))
46bfbb5c
CM		 6831 goto must_cow;
6832
eb384b55
JB		 6833 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6834 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6835 if (type == BTRFS_ORDERED_PREALLOC) {
6836 free_extent_map(em);
6837 em = create_pinned_em(inode, start, len,
6838 orig_start,
b4939680 6839 block_start, len,
cc95bef6
JB		 6840 orig_block_len,
6841 ram_bytes, type);
69ffb543
JB		 6842 if (IS_ERR(em)) {
6843 btrfs_end_transaction(trans, root);
6844 goto unlock_err;
6845 }
6846 }
6847
46bfbb5c
CM		 6848 ret = btrfs_add_ordered_extent_dio(inode, start,
6849 block_start, len, len, type);
6850 btrfs_end_transaction(trans, root);
6851 if (ret) {
6852 free_extent_map(em);
eb838e73 6853 goto unlock_err;
46bfbb5c
CM		 6854 }
6855 goto unlock;
4b46fce2 6856 }
46bfbb5c 6857 btrfs_end_transaction(trans, root);
4b46fce2 6858 }
46bfbb5c
CM		 6859must_cow:
6860 /*
6861 * this will cow the extent, reset the len in case we changed
6862 * it above
6863 */
6864 len = bh_result->b_size;
70c8a91c
JB		 6865 free_extent_map(em);
6866 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6867 if (IS_ERR(em)) {
6868 ret = PTR_ERR(em);
6869 goto unlock_err;
6870 }
46bfbb5c
CM		 6871 len = min(len, em->len - (start - em->start));
6872unlock:
4b46fce2
JB		 6873 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6874 inode->i_blkbits;
46bfbb5c 6875 bh_result->b_size = len;
4b46fce2
JB		 6876 bh_result->b_bdev = em->bdev;
6877 set_buffer_mapped(bh_result);
c3473e83
JB		 6878 if (create) {
6879 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6880 set_buffer_new(bh_result);
6881
6882 /*
6883 * Need to update the i_size under the extent lock so buffered
6884 * readers will get the updated i_size when we unlock.
6885 */
6886 if (start + len > i_size_read(inode))
6887 i_size_write(inode, start + len);
0934856d 6888
172a5049
MX		 6889 spin_lock(&BTRFS_I(inode)->lock);
6890 BTRFS_I(inode)->outstanding_extents++;
6891 spin_unlock(&BTRFS_I(inode)->lock);
6892
0934856d
MX		 6893 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6894 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6895 &cached_state, GFP_NOFS);
6896 BUG_ON(ret);
c3473e83 6897 }
4b46fce2 6898
eb838e73
JB		 6899 /*
6900 * In the case of write we need to clear and unlock the entire range,
6901 * in the case of read we need to unlock only the end area that we
6902 * aren't using if there is any left over space.
6903 */
24c03fa5 6904 if (lockstart < lockend) {
0934856d
MX		 6905 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6906 lockend, unlock_bits, 1, 0,
6907 &cached_state, GFP_NOFS);
24c03fa5 6908 } else {
eb838e73 6909 free_extent_state(cached_state);
24c03fa5 6910 }
eb838e73 6911
4b46fce2
JB		 6912 free_extent_map(em);
6913
6914 return 0;
eb838e73
JB		 6915
6916unlock_err:
eb838e73
JB		 6917 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6918 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6919 return ret;
4b46fce2
JB		 6920}
6921
6922struct btrfs_dio_private {
6923 struct inode *inode;
6924 u64 logical_offset;
6925 u64 disk_bytenr;
6926 u64 bytes;
4b46fce2 6927 void *private;
e65e1535
MX		 6928
6929 /* number of bios pending for this dio */
6930 atomic_t pending_bios;
6931
6932 /* IO errors */
6933 int errors;
6934
6935 struct bio *orig_bio;
4b46fce2
JB		 6936};
6937
6938static void btrfs_endio_direct_read(struct bio *bio, int err)
6939{
e65e1535 6940 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6941 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6942 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6943 struct inode *inode = dip->inode;
c2cf52eb 6944 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6945 u64 start;
4b46fce2
JB		 6946
6947 start = dip->logical_offset;
6948 do {
6949 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6950 struct page *page = bvec->bv_page;
6951 char *kaddr;
6952 u32 csum = ~(u32)0;
c329861d 6953 u64 private = ~(u32)0;
4b46fce2
JB		 6954 unsigned long flags;
6955
c329861d
JB		 6956 if (get_state_private(&BTRFS_I(inode)->io_tree,
6957 start, &private))
6958 goto failed;
4b46fce2 6959 local_irq_save(flags);
7ac687d9 6960 kaddr = kmap_atomic(page);
b0496686 6961 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6962 csum, bvec->bv_len);
6963 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6964 kunmap_atomic(kaddr);
4b46fce2
JB		 6965 local_irq_restore(flags);
6966
6967 flush_dcache_page(bvec->bv_page);
c329861d
JB		 6968 if (csum != private) {
6969failed:
c2cf52eb
SK		 6970 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6971 (unsigned long long)btrfs_ino(inode),
6972 (unsigned long long)start,
6973 csum, (unsigned)private);
4b46fce2
JB		 6974 err = -EIO;
6975 }
6976 }
6977
6978 start += bvec->bv_len;
4b46fce2
JB		 6979 bvec++;
6980 } while (bvec <= bvec_end);
6981
6982 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6983 dip->logical_offset + dip->bytes - 1);
4b46fce2
JB		 6984 bio->bi_private = dip->private;
6985
4b46fce2 6986 kfree(dip);
c0da7aa1
JB		 6987
6988 /* If we had a csum failure make sure to clear the uptodate flag */
6989 if (err)
6990 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 6991 dio_end_io(bio, err);
6992}
6993
6994static void btrfs_endio_direct_write(struct bio *bio, int err)
6995{
6996 struct btrfs_dio_private *dip = bio->bi_private;
6997 struct inode *inode = dip->inode;
6998 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6999 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7000 u64 ordered_offset = dip->logical_offset;
7001 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 7002 int ret;
7003
7004 if (err)
7005 goto out_done;
163cf09c
CM		 7006again:
7007 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7008 &ordered_offset,
5fd02043 7009 ordered_bytes, !err);
4b46fce2 7010 if (!ret)
163cf09c 7011 goto out_test;
4b46fce2 7012
5fd02043
JB		 7013 ordered->work.func = finish_ordered_fn;
7014 ordered->work.flags = 0;
7015 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7016 &ordered->work);
163cf09c
CM		 7017out_test:
7018 /*
7019 * our bio might span multiple ordered extents. If we haven't
7020 * completed the accounting for the whole dio, go back and try again
7021 */
7022 if (ordered_offset < dip->logical_offset + dip->bytes) {
7023 ordered_bytes = dip->logical_offset + dip->bytes -
7024 ordered_offset;
5fd02043 7025 ordered = NULL;
163cf09c
CM		 7026 goto again;
7027 }
4b46fce2
JB		 7028out_done:
7029 bio->bi_private = dip->private;
7030
4b46fce2 7031 kfree(dip);
c0da7aa1
JB		 7032
7033 /* If we had an error make sure to clear the uptodate flag */
7034 if (err)
7035 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 7036 dio_end_io(bio, err);
7037}
7038
eaf25d93
CM		 7039static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7040 struct bio *bio, int mirror_num,
7041 unsigned long bio_flags, u64 offset)
7042{
7043 int ret;
7044 struct btrfs_root *root = BTRFS_I(inode)->root;
7045 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7046 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7047 return 0;
7048}
7049
e65e1535
MX		 7050static void btrfs_end_dio_bio(struct bio *bio, int err)
7051{
7052 struct btrfs_dio_private *dip = bio->bi_private;
7053
7054 if (err) {
33345d01 7055 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 7056 "sector %#Lx len %u err no %d\n",
33345d01 7057 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 7058 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 7059 dip->errors = 1;
7060
7061 /*
7062 * before atomic variable goto zero, we must make sure
7063 * dip->errors is perceived to be set.
7064 */
7065 smp_mb__before_atomic_dec();
7066 }
7067
7068 /* if there are more bios still pending for this dio, just exit */
7069 if (!atomic_dec_and_test(&dip->pending_bios))
7070 goto out;
7071
7072 if (dip->errors)
7073 bio_io_error(dip->orig_bio);
7074 else {
7075 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
7076 bio_endio(dip->orig_bio, 0);
7077 }
7078out:
7079 bio_put(bio);
7080}
7081
7082static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7083 u64 first_sector, gfp_t gfp_flags)
7084{
7085 int nr_vecs = bio_get_nr_vecs(bdev);
7086 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7087}
7088
7089static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7090 int rw, u64 file_offset, int skip_sum,
c329861d 7091 int async_submit)
e65e1535
MX		 7092{
7093 int write = rw & REQ_WRITE;
7094 struct btrfs_root *root = BTRFS_I(inode)->root;
7095 int ret;
7096
b812ce28
JB		 7097 if (async_submit)
7098 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7099
e65e1535 7100 bio_get(bio);
5fd02043
JB		 7101
7102 if (!write) {
7103 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7104 if (ret)
7105 goto err;
7106 }
e65e1535 7107
1ae39938
JB		 7108 if (skip_sum)
7109 goto map;
7110
7111 if (write && async_submit) {
e65e1535
MX		 7112 ret = btrfs_wq_submit_bio(root->fs_info,
7113 inode, rw, bio, 0, 0,
7114 file_offset,
7115 __btrfs_submit_bio_start_direct_io,
7116 __btrfs_submit_bio_done);
7117 goto err;
1ae39938
JB		 7118 } else if (write) {
7119 /*
7120 * If we aren't doing async submit, calculate the csum of the
7121 * bio now.
7122 */
7123 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7124 if (ret)
7125 goto err;
c2db1073 7126 } else if (!skip_sum) {
c329861d 7127 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI		 7128 if (ret)
7129 goto err;
7130 }
e65e1535 7131
1ae39938
JB		 7132map:
7133 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7134err:
7135 bio_put(bio);
7136 return ret;
7137}
7138
7139static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7140 int skip_sum)
7141{
7142 struct inode *inode = dip->inode;
7143 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7144 struct bio *bio;
7145 struct bio *orig_bio = dip->orig_bio;
7146 struct bio_vec *bvec = orig_bio->bi_io_vec;
7147 u64 start_sector = orig_bio->bi_sector;
7148 u64 file_offset = dip->logical_offset;
7149 u64 submit_len = 0;
7150 u64 map_length;
7151 int nr_pages = 0;
e65e1535 7152 int ret = 0;
1ae39938 7153 int async_submit = 0;
e65e1535 7154
e65e1535 7155 map_length = orig_bio->bi_size;
53b381b3 7156 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 7157 &map_length, NULL, 0);
7158 if (ret) {
64728bbb 7159 bio_put(orig_bio);
e65e1535
MX		 7160 return -EIO;
7161 }
02f57c7a
JB		 7162 if (map_length >= orig_bio->bi_size) {
7163 bio = orig_bio;
7164 goto submit;
7165 }
7166
53b381b3
DW		 7167 /* async crcs make it difficult to collect full stripe writes. */
7168 if (btrfs_get_alloc_profile(root, 1) &
7169 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7170 async_submit = 0;
7171 else
7172 async_submit = 1;
7173
02f57c7a
JB		 7174 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7175 if (!bio)
7176 return -ENOMEM;
7177 bio->bi_private = dip;
7178 bio->bi_end_io = btrfs_end_dio_bio;
7179 atomic_inc(&dip->pending_bios);
7180
e65e1535
MX		 7181 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7182 if (unlikely(map_length < submit_len + bvec->bv_len ||
7183 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7184 bvec->bv_offset) < bvec->bv_len)) {
7185 /*
7186 * inc the count before we submit the bio so
7187 * we know the end IO handler won't happen before
7188 * we inc the count. Otherwise, the dip might get freed
7189 * before we're done setting it up
7190 */
7191 atomic_inc(&dip->pending_bios);
7192 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7193 file_offset, skip_sum,
c329861d 7194 async_submit);
e65e1535
MX		 7195 if (ret) {
7196 bio_put(bio);
7197 atomic_dec(&dip->pending_bios);
7198 goto out_err;
7199 }
7200
e65e1535
MX		 7201 start_sector += submit_len >> 9;
7202 file_offset += submit_len;
7203
7204 submit_len = 0;
7205 nr_pages = 0;
7206
7207 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7208 start_sector, GFP_NOFS);
7209 if (!bio)
7210 goto out_err;
7211 bio->bi_private = dip;
7212 bio->bi_end_io = btrfs_end_dio_bio;
7213
7214 map_length = orig_bio->bi_size;
53b381b3 7215 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7216 start_sector << 9,
e65e1535
MX		 7217 &map_length, NULL, 0);
7218 if (ret) {
7219 bio_put(bio);
7220 goto out_err;
7221 }
7222 } else {
7223 submit_len += bvec->bv_len;
7224 nr_pages ++;
7225 bvec++;
7226 }
7227 }
7228
02f57c7a 7229submit:
e65e1535 7230 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7231 async_submit);
e65e1535
MX		 7232 if (!ret)
7233 return 0;
7234
7235 bio_put(bio);
7236out_err:
7237 dip->errors = 1;
7238 /*
7239 * before atomic variable goto zero, we must
7240 * make sure dip->errors is perceived to be set.
7241 */
7242 smp_mb__before_atomic_dec();
7243 if (atomic_dec_and_test(&dip->pending_bios))
7244 bio_io_error(dip->orig_bio);
7245
7246 /* bio_end_io() will handle error, so we needn't return it */
7247 return 0;
7248}
7249
4b46fce2
JB		 7250static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
7251 loff_t file_offset)
7252{
7253 struct btrfs_root *root = BTRFS_I(inode)->root;
7254 struct btrfs_dio_private *dip;
7255 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 7256 int skip_sum;
7b6d91da 7257 int write = rw & REQ_WRITE;
4b46fce2
JB		 7258 int ret = 0;
7259
7260 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7261
7262 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7263 if (!dip) {
7264 ret = -ENOMEM;
7265 goto free_ordered;
7266 }
4b46fce2
JB		 7267
7268 dip->private = bio->bi_private;
7269 dip->inode = inode;
7270 dip->logical_offset = file_offset;
7271
4b46fce2
JB		 7272 dip->bytes = 0;
7273 do {
7274 dip->bytes += bvec->bv_len;
7275 bvec++;
7276 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
7277
46bfbb5c 7278 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 7279 bio->bi_private = dip;
e65e1535
MX		 7280 dip->errors = 0;
7281 dip->orig_bio = bio;
7282 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7283
7284 if (write)
7285 bio->bi_end_io = btrfs_endio_direct_write;
7286 else
7287 bio->bi_end_io = btrfs_endio_direct_read;
7288
e65e1535
MX		 7289 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7290 if (!ret)
eaf25d93 7291 return;
4b46fce2
JB		 7292free_ordered:
7293 /*
7294 * If this is a write, we need to clean up the reserved space and kill
7295 * the ordered extent.
7296 */
7297 if (write) {
7298 struct btrfs_ordered_extent *ordered;
955256f2 7299 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7300 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7301 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7302 btrfs_free_reserved_extent(root, ordered->start,
7303 ordered->disk_len);
7304 btrfs_put_ordered_extent(ordered);
7305 btrfs_put_ordered_extent(ordered);
7306 }
7307 bio_endio(bio, ret);
7308}
7309
5a5f79b5
CM		 7310static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7311 const struct iovec *iov, loff_t offset,
7312 unsigned long nr_segs)
7313{
7314 int seg;
a1b75f7d 7315 int i;
5a5f79b5
CM		 7316 size_t size;
7317 unsigned long addr;
7318 unsigned blocksize_mask = root->sectorsize - 1;
7319 ssize_t retval = -EINVAL;
7320 loff_t end = offset;
7321
7322 if (offset & blocksize_mask)
7323 goto out;
7324
7325 /* Check the memory alignment. Blocks cannot straddle pages */
7326 for (seg = 0; seg < nr_segs; seg++) {
7327 addr = (unsigned long)iov[seg].iov_base;
7328 size = iov[seg].iov_len;
7329 end += size;
a1b75f7d 7330 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7331 goto out;
a1b75f7d
JB		 7332
7333 /* If this is a write we don't need to check anymore */
7334 if (rw & WRITE)
7335 continue;
7336
7337 /*
7338 * Check to make sure we don't have duplicate iov_base's in this
7339 * iovec, if so return EINVAL, otherwise we'll get csum errors
7340 * when reading back.
7341 */
7342 for (i = seg + 1; i < nr_segs; i++) {
7343 if (iov[seg].iov_base == iov[i].iov_base)
7344 goto out;
7345 }
5a5f79b5
CM		 7346 }
7347 retval = 0;
7348out:
7349 return retval;
7350}
eb838e73 7351
16432985
CM		 7352static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7353 const struct iovec *iov, loff_t offset,
7354 unsigned long nr_segs)
7355{
4b46fce2
JB		 7356 struct file *file = iocb->ki_filp;
7357 struct inode *inode = file->f_mapping->host;
0934856d 7358 size_t count = 0;
2e60a51e 7359 int flags = 0;
38851cc1
MX		 7360 bool wakeup = true;
7361 bool relock = false;
0934856d 7362 ssize_t ret;
4b46fce2 7363
5a5f79b5 7364 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7365 offset, nr_segs))
5a5f79b5 7366 return 0;
3f7c579c 7367
38851cc1
MX		 7368 atomic_inc(&inode->i_dio_count);
7369 smp_mb__after_atomic_inc();
7370
0934856d
MX		 7371 if (rw & WRITE) {
7372 count = iov_length(iov, nr_segs);
38851cc1
MX		 7373 /*
7374 * If the write DIO is beyond the EOF, we need update
7375 * the isize, but it is protected by i_mutex. So we can
7376 * not unlock the i_mutex at this case.
7377 */
7378 if (offset + count <= inode->i_size) {
7379 mutex_unlock(&inode->i_mutex);
7380 relock = true;
7381 }
0934856d
MX		 7382 ret = btrfs_delalloc_reserve_space(inode, count);
7383 if (ret)
38851cc1
MX		 7384 goto out;
7385 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7386 &BTRFS_I(inode)->runtime_flags))) {
7387 inode_dio_done(inode);
7388 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7389 wakeup = false;
0934856d
MX		 7390 }
7391
7392 ret = __blockdev_direct_IO(rw, iocb, inode,
7393 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7394 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7395 btrfs_submit_direct, flags);
0934856d
MX		 7396 if (rw & WRITE) {
7397 if (ret < 0 && ret != -EIOCBQUEUED)
7398 btrfs_delalloc_release_space(inode, count);
172a5049 7399 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7400 btrfs_delalloc_release_space(inode,
7401 count - (size_t)ret);
172a5049
MX		 7402 else
7403 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7404 }
38851cc1 7405out:
2e60a51e
MX		 7406 if (wakeup)
7407 inode_dio_done(inode);
38851cc1
MX		 7408 if (relock)
7409 mutex_lock(&inode->i_mutex);
0934856d
MX		 7410
7411 return ret;
16432985
CM		 7412}
7413
05dadc09
TI		 7414#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7415
1506fcc8
YS		 7416static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7417 __u64 start, __u64 len)
7418{
05dadc09
TI		 7419 int ret;
7420
7421 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7422 if (ret)
7423 return ret;
7424
ec29ed5b 7425 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7426}
7427
a52d9a80 7428int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7429{
d1310b2e
CM		 7430 struct extent_io_tree *tree;
7431 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7432 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7433}
1832a6d5 7434
a52d9a80 7435static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7436{
d1310b2e 7437 struct extent_io_tree *tree;
b888db2b
CM		 7438
7439
7440 if (current->flags & PF_MEMALLOC) {
7441 redirty_page_for_writepage(wbc, page);
7442 unlock_page(page);
7443 return 0;
7444 }
d1310b2e 7445 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7446 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7447}
7448
48a3b636
ES		 7449static int btrfs_writepages(struct address_space *mapping,
7450 struct writeback_control *wbc)
b293f02e 7451{
d1310b2e 7452 struct extent_io_tree *tree;
771ed689 7453
d1310b2e 7454 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7455 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7456}
7457
3ab2fb5a
CM		 7458static int
7459btrfs_readpages(struct file *file, struct address_space *mapping,
7460 struct list_head *pages, unsigned nr_pages)
7461{
d1310b2e
CM		 7462 struct extent_io_tree *tree;
7463 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7464 return extent_readpages(tree, mapping, pages, nr_pages,
7465 btrfs_get_extent);
7466}
e6dcd2dc 7467static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7468{
d1310b2e
CM		 7469 struct extent_io_tree *tree;
7470 struct extent_map_tree *map;
a52d9a80 7471 int ret;
8c2383c3 7472
d1310b2e
CM		 7473 tree = &BTRFS_I(page->mapping->host)->io_tree;
7474 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7475 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7476 if (ret == 1) {
7477 ClearPagePrivate(page);
7478 set_page_private(page, 0);
7479 page_cache_release(page);
39279cc3 7480 }
a52d9a80 7481 return ret;
39279cc3
CM		 7482}
7483
e6dcd2dc
CM		 7484static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7485{
98509cfc
CM		 7486 if (PageWriteback(page) || PageDirty(page))
7487 return 0;
b335b003 7488 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7489}
7490
a52d9a80 7491static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 7492{
5fd02043 7493 struct inode *inode = page->mapping->host;
d1310b2e 7494 struct extent_io_tree *tree;
e6dcd2dc 7495 struct btrfs_ordered_extent *ordered;
2ac55d41 7496 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7497 u64 page_start = page_offset(page);
7498 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7499
8b62b72b
CM		 7500 /*
7501 * we have the page locked, so new writeback can't start,
7502 * and the dirty bit won't be cleared while we are here.
7503 *
7504 * Wait for IO on this page so that we can safely clear
7505 * the PagePrivate2 bit and do ordered accounting
7506 */
e6dcd2dc 7507 wait_on_page_writeback(page);
8b62b72b 7508
5fd02043 7509 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7510 if (offset) {
7511 btrfs_releasepage(page, GFP_NOFS);
7512 return;
7513 }
d0082371 7514 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7515 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7516 if (ordered) {
eb84ae03
CM		 7517 /*
7518 * IO on this page will never be started, so we need
7519 * to account for any ordered extents now
7520 */
e6dcd2dc
CM		 7521 clear_extent_bit(tree, page_start, page_end,
7522 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7523 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7524 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7525 /*
7526 * whoever cleared the private bit is responsible
7527 * for the finish_ordered_io
7528 */
5fd02043
JB		 7529 if (TestClearPagePrivate2(page) &&
7530 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7531 PAGE_CACHE_SIZE, 1)) {
7532 btrfs_finish_ordered_io(ordered);
8b62b72b 7533 }
e6dcd2dc 7534 btrfs_put_ordered_extent(ordered);
2ac55d41 7535 cached_state = NULL;
d0082371 7536 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7537 }
7538 clear_extent_bit(tree, page_start, page_end,
32c00aff 7539 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7540 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7541 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7542 __btrfs_releasepage(page, GFP_NOFS);
7543
4a096752 7544 ClearPageChecked(page);
9ad6b7bc 7545 if (PagePrivate(page)) {
9ad6b7bc
CM		 7546 ClearPagePrivate(page);
7547 set_page_private(page, 0);
7548 page_cache_release(page);
7549 }
39279cc3
CM		 7550}
7551
9ebefb18
CM		 7552/*
7553 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7554 * called from a page fault handler when a page is first dirtied. Hence we must
7555 * be careful to check for EOF conditions here. We set the page up correctly
7556 * for a written page which means we get ENOSPC checking when writing into
7557 * holes and correct delalloc and unwritten extent mapping on filesystems that
7558 * support these features.
7559 *
7560 * We are not allowed to take the i_mutex here so we have to play games to
7561 * protect against truncate races as the page could now be beyond EOF. Because
7562 * vmtruncate() writes the inode size before removing pages, once we have the
7563 * page lock we can determine safely if the page is beyond EOF. If it is not
7564 * beyond EOF, then the page is guaranteed safe against truncation until we
7565 * unlock the page.
7566 */
c2ec175c 7567int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7568{
c2ec175c 7569 struct page *page = vmf->page;
496ad9aa 7570 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7571 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7572 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7573 struct btrfs_ordered_extent *ordered;
2ac55d41 7574 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7575 char *kaddr;
7576 unsigned long zero_start;
9ebefb18 7577 loff_t size;
1832a6d5 7578 int ret;
9998eb70 7579 int reserved = 0;
a52d9a80 7580 u64 page_start;
e6dcd2dc 7581 u64 page_end;
9ebefb18 7582
b2b5ef5c 7583 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7584 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7585 if (!ret) {
e41f941a 7586 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7587 reserved = 1;
7588 }
56a76f82
NP		 7589 if (ret) {
7590 if (ret == -ENOMEM)
7591 ret = VM_FAULT_OOM;
7592 else /* -ENOSPC, -EIO, etc */
7593 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7594 if (reserved)
7595 goto out;
7596 goto out_noreserve;
56a76f82 7597 }
1832a6d5 7598
56a76f82 7599 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7600again:
9ebefb18 7601 lock_page(page);
9ebefb18 7602 size = i_size_read(inode);
e6dcd2dc
CM		 7603 page_start = page_offset(page);
7604 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7605
9ebefb18 7606 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7607 (page_start >= size)) {
9ebefb18
CM		 7608 /* page got truncated out from underneath us */
7609 goto out_unlock;
7610 }
e6dcd2dc
CM		 7611 wait_on_page_writeback(page);
7612
d0082371 7613 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7614 set_page_extent_mapped(page);
7615
eb84ae03
CM		 7616 /*
7617 * we can't set the delalloc bits if there are pending ordered
7618 * extents. Drop our locks and wait for them to finish
7619 */
e6dcd2dc
CM		 7620 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7621 if (ordered) {
2ac55d41
JB		 7622 unlock_extent_cached(io_tree, page_start, page_end,
7623 &cached_state, GFP_NOFS);
e6dcd2dc 7624 unlock_page(page);
eb84ae03 7625 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7626 btrfs_put_ordered_extent(ordered);
7627 goto again;
7628 }
7629
fbf19087
JB		 7630 /*
7631 * XXX - page_mkwrite gets called every time the page is dirtied, even
7632 * if it was already dirty, so for space accounting reasons we need to
7633 * clear any delalloc bits for the range we are fixing to save. There
7634 * is probably a better way to do this, but for now keep consistent with
7635 * prepare_pages in the normal write path.
7636 */
2ac55d41 7637 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7638 EXTENT_DIRTY | EXTENT_DELALLOC |
7639 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7640 0, 0, &cached_state, GFP_NOFS);
fbf19087 7641
2ac55d41
JB		 7642 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7643 &cached_state);
9ed74f2d 7644 if (ret) {
2ac55d41
JB		 7645 unlock_extent_cached(io_tree, page_start, page_end,
7646 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7647 ret = VM_FAULT_SIGBUS;
7648 goto out_unlock;
7649 }
e6dcd2dc 7650 ret = 0;
9ebefb18
CM		 7651
7652 /* page is wholly or partially inside EOF */
a52d9a80 7653 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7654 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7655 else
e6dcd2dc 7656 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7657
e6dcd2dc
CM		 7658 if (zero_start != PAGE_CACHE_SIZE) {
7659 kaddr = kmap(page);
7660 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7661 flush_dcache_page(page);
7662 kunmap(page);
7663 }
247e743c 7664 ClearPageChecked(page);
e6dcd2dc 7665 set_page_dirty(page);
50a9b214 7666 SetPageUptodate(page);
5a3f23d5 7667
257c62e1
CM		 7668 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7669 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7670 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7671
2ac55d41 7672 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7673
7674out_unlock:
b2b5ef5c
JK		 7675 if (!ret) {
7676 sb_end_pagefault(inode->i_sb);
50a9b214 7677 return VM_FAULT_LOCKED;
b2b5ef5c 7678 }
9ebefb18 7679 unlock_page(page);
1832a6d5 7680out:
ec39e180 7681 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7682out_noreserve:
b2b5ef5c 7683 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7684 return ret;
7685}
7686
a41ad394 7687static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7688{
7689 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7690 struct btrfs_block_rsv *rsv;
39279cc3 7691 int ret;
3893e33b 7692 int err = 0;
39279cc3 7693 struct btrfs_trans_handle *trans;
dbe674a9 7694 u64 mask = root->sectorsize - 1;
07127184 7695 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7696
2aaa6655 7697 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
5d5e103a 7698 if (ret)
a41ad394 7699 return ret;
8082510e 7700
4a096752 7701 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 7702 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 7703
fcb80c2a
JB		 7704 /*
7705 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7706 * 3 things going on here
7707 *
7708 * 1) We need to reserve space for our orphan item and the space to
7709 * delete our orphan item. Lord knows we don't want to have a dangling
7710 * orphan item because we didn't reserve space to remove it.
7711 *
7712 * 2) We need to reserve space to update our inode.
7713 *
7714 * 3) We need to have something to cache all the space that is going to
7715 * be free'd up by the truncate operation, but also have some slack
7716 * space reserved in case it uses space during the truncate (thank you
7717 * very much snapshotting).
7718 *
7719 * And we need these to all be seperate. The fact is we can use alot of
7720 * space doing the truncate, and we have no earthly idea how much space
7721 * we will use, so we need the truncate reservation to be seperate so it
7722 * doesn't end up using space reserved for updating the inode or
7723 * removing the orphan item. We also need to be able to stop the
7724 * transaction and start a new one, which means we need to be able to
7725 * update the inode several times, and we have no idea of knowing how
7726 * many times that will be, so we can't just reserve 1 item for the
7727 * entirety of the opration, so that has to be done seperately as well.
7728 * Then there is the orphan item, which does indeed need to be held on
7729 * to for the whole operation, and we need nobody to touch this reserved
7730 * space except the orphan code.
7731 *
7732 * So that leaves us with
7733 *
7734 * 1) root->orphan_block_rsv - for the orphan deletion.
7735 * 2) rsv - for the truncate reservation, which we will steal from the
7736 * transaction reservation.
7737 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7738 * updating the inode.
7739 */
66d8f3dd 7740 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 7741 if (!rsv)
7742 return -ENOMEM;
4a338542 7743 rsv->size = min_size;
ca7e70f5 7744 rsv->failfast = 1;
f0cd846e 7745
907cbceb 7746 /*
07127184 7747 * 1 for the truncate slack space
907cbceb
JB		 7748 * 1 for updating the inode.
7749 */
f3fe820c 7750 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 7751 if (IS_ERR(trans)) {
7752 err = PTR_ERR(trans);
7753 goto out;
7754 }
f0cd846e 7755
907cbceb
JB		 7756 /* Migrate the slack space for the truncate to our reserve */
7757 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7758 min_size);
fcb80c2a 7759 BUG_ON(ret);
f0cd846e 7760
5a3f23d5
CM		 7761 /*
7762 * setattr is responsible for setting the ordered_data_close flag,
7763 * but that is only tested during the last file release. That
7764 * could happen well after the next commit, leaving a great big
7765 * window where new writes may get lost if someone chooses to write
7766 * to this file after truncating to zero
7767 *
7768 * The inode doesn't have any dirty data here, and so if we commit
7769 * this is a noop. If someone immediately starts writing to the inode
7770 * it is very likely we'll catch some of their writes in this
7771 * transaction, and the commit will find this file on the ordered
7772 * data list with good things to send down.
7773 *
7774 * This is a best effort solution, there is still a window where
7775 * using truncate to replace the contents of the file will
7776 * end up with a zero length file after a crash.
7777 */
72ac3c0d
JB		 7778 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7779 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 7780 btrfs_add_ordered_operation(trans, root, inode);
7781
5dc562c5
JB		 7782 /*
7783 * So if we truncate and then write and fsync we normally would just
7784 * write the extents that changed, which is a problem if we need to
7785 * first truncate that entire inode. So set this flag so we write out
7786 * all of the extents in the inode to the sync log so we're completely
7787 * safe.
7788 */
7789 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7790 trans->block_rsv = rsv;
907cbceb 7791
8082510e
YZ		 7792 while (1) {
7793 ret = btrfs_truncate_inode_items(trans, root, inode,
7794 inode->i_size,
7795 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7796 if (ret != -ENOSPC) {
3893e33b 7797 err = ret;
8082510e 7798 break;
3893e33b 7799 }
39279cc3 7800
fcb80c2a 7801 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7802 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 7803 if (ret) {
7804 err = ret;
7805 break;
7806 }
ca7e70f5 7807
8082510e 7808 btrfs_end_transaction(trans, root);
b53d3f5d 7809 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 7810
7811 trans = btrfs_start_transaction(root, 2);
7812 if (IS_ERR(trans)) {
7813 ret = err = PTR_ERR(trans);
7814 trans = NULL;
7815 break;
7816 }
7817
7818 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7819 rsv, min_size);
7820 BUG_ON(ret); /* shouldn't happen */
7821 trans->block_rsv = rsv;
8082510e
YZ		 7822 }
7823
7824 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7825 trans->block_rsv = root->orphan_block_rsv;
8082510e 7826 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7827 if (ret)
7828 err = ret;
8082510e
YZ		 7829 }
7830
917c16b2
CM		 7831 if (trans) {
7832 trans->block_rsv = &root->fs_info->trans_block_rsv;
7833 ret = btrfs_update_inode(trans, root, inode);
7834 if (ret && !err)
7835 err = ret;
7b128766 7836
7ad85bb7 7837 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7838 btrfs_btree_balance_dirty(root);
917c16b2 7839 }
fcb80c2a
JB		 7840
7841out:
7842 btrfs_free_block_rsv(root, rsv);
7843
3893e33b
JB		 7844 if (ret && !err)
7845 err = ret;
a41ad394 7846
3893e33b 7847 return err;
39279cc3
CM		 7848}
7849
d352ac68
CM		 7850/*
7851 * create a new subvolume directory/inode (helper for the ioctl).
7852 */
d2fb3437 7853int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7854 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7855{
39279cc3 7856 struct inode *inode;
76dda93c 7857 int err;
00e4e6b3 7858 u64 index = 0;
39279cc3 7859
12fc9d09
FA		 7860 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7861 new_dirid, new_dirid,
7862 S_IFDIR | (~current_umask() & S_IRWXUGO),
7863 &index);
54aa1f4d 7864 if (IS_ERR(inode))
f46b5a66 7865 return PTR_ERR(inode);
39279cc3
CM		 7866 inode->i_op = &btrfs_dir_inode_operations;
7867 inode->i_fop = &btrfs_dir_file_operations;
7868
bfe86848 7869 set_nlink(inode, 1);
dbe674a9 7870 btrfs_i_size_write(inode, 0);
3b96362c 7871
76dda93c 7872 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7873
76dda93c 7874 iput(inode);
ce598979 7875 return err;
39279cc3
CM		 7876}
7877
39279cc3
CM		 7878struct inode *btrfs_alloc_inode(struct super_block *sb)
7879{
7880 struct btrfs_inode *ei;
2ead6ae7 7881 struct inode *inode;
39279cc3
CM		 7882
7883 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7884 if (!ei)
7885 return NULL;
2ead6ae7
YZ		 7886
7887 ei->root = NULL;
2ead6ae7 7888 ei->generation = 0;
15ee9bc7 7889 ei->last_trans = 0;
257c62e1 7890 ei->last_sub_trans = 0;
e02119d5 7891 ei->logged_trans = 0;
2ead6ae7 7892 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7893 ei->disk_i_size = 0;
7894 ei->flags = 0;
7709cde3 7895 ei->csum_bytes = 0;
2ead6ae7
YZ		 7896 ei->index_cnt = (u64)-1;
7897 ei->last_unlink_trans = 0;
46d8bc34 7898 ei->last_log_commit = 0;
2ead6ae7 7899
9e0baf60
JB		 7900 spin_lock_init(&ei->lock);
7901 ei->outstanding_extents = 0;
7902 ei->reserved_extents = 0;
2ead6ae7 7903
72ac3c0d 7904 ei->runtime_flags = 0;
261507a0 7905 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7906
16cdcec7
MX		 7907 ei->delayed_node = NULL;
7908
2ead6ae7 7909 inode = &ei->vfs_inode;
a8067e02 7910 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7911 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7912 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7913 ei->io_tree.track_uptodate = 1;
7914 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7915 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7916 mutex_init(&ei->log_mutex);
f248679e 7917 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7918 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7919 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7920 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7921 RB_CLEAR_NODE(&ei->rb_node);
7922
7923 return inode;
39279cc3
CM		 7924}
7925
fa0d7e3d
NP		 7926static void btrfs_i_callback(struct rcu_head *head)
7927{
7928 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7929 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7930}
7931
39279cc3
CM		 7932void btrfs_destroy_inode(struct inode *inode)
7933{
e6dcd2dc 7934 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7935 struct btrfs_root *root = BTRFS_I(inode)->root;
7936
b3d9b7a3 7937 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7938 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7939 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7940 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7941 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7942 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7943
a6dbd429
JB		 7944 /*
7945 * This can happen where we create an inode, but somebody else also
7946 * created the same inode and we need to destroy the one we already
7947 * created.
7948 */
7949 if (!root)
7950 goto free;
7951
5a3f23d5
CM		 7952 /*
7953 * Make sure we're properly removed from the ordered operation
7954 * lists.
7955 */
7956 smp_mb();
7957 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7958 spin_lock(&root->fs_info->ordered_extent_lock);
7959 list_del_init(&BTRFS_I(inode)->ordered_operations);
7960 spin_unlock(&root->fs_info->ordered_extent_lock);
7961 }
7962
8a35d95f
JB		 7963 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7964 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb
SK		 7965 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7966 (unsigned long long)btrfs_ino(inode));
8a35d95f 7967 atomic_dec(&root->orphan_inodes);
7b128766 7968 }
7b128766 7969
d397712b 7970 while (1) {
e6dcd2dc
CM		 7971 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7972 if (!ordered)
7973 break;
7974 else {
c2cf52eb
SK		 7975 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7976 (unsigned long long)ordered->file_offset,
7977 (unsigned long long)ordered->len);
e6dcd2dc
CM		 7978 btrfs_remove_ordered_extent(inode, ordered);
7979 btrfs_put_ordered_extent(ordered);
7980 btrfs_put_ordered_extent(ordered);
7981 }
7982 }
5d4f98a2 7983 inode_tree_del(inode);
5b21f2ed 7984 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7985free:
fa0d7e3d 7986 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7987}
7988
45321ac5 7989int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7990{
7991 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7992
fa6ac876 7993 /* the snap/subvol tree is on deleting */
0af3d00b 7994 if (btrfs_root_refs(&root->root_item) == 0 &&
fa6ac876 7995 root != root->fs_info->tree_root)
45321ac5 7996 return 1;
76dda93c 7997 else
45321ac5 7998 return generic_drop_inode(inode);
76dda93c
YZ		 7999}
8000
0ee0fda0 8001static void init_once(void *foo)
39279cc3
CM		 8002{
8003 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8004
8005 inode_init_once(&ei->vfs_inode);
8006}
8007
8008void btrfs_destroy_cachep(void)
8009{
8c0a8537
KS		 8010 /*
8011 * Make sure all delayed rcu free inodes are flushed before we
8012 * destroy cache.
8013 */
8014 rcu_barrier();
39279cc3
CM		 8015 if (btrfs_inode_cachep)
8016 kmem_cache_destroy(btrfs_inode_cachep);
8017 if (btrfs_trans_handle_cachep)
8018 kmem_cache_destroy(btrfs_trans_handle_cachep);
8019 if (btrfs_transaction_cachep)
8020 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 8021 if (btrfs_path_cachep)
8022 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 8023 if (btrfs_free_space_cachep)
8024 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 8025 if (btrfs_delalloc_work_cachep)
8026 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 8027}
8028
8029int btrfs_init_cachep(void)
8030{
837e1972 8031 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 8032 sizeof(struct btrfs_inode), 0,
8033 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 8034 if (!btrfs_inode_cachep)
8035 goto fail;
9601e3f6 8036
837e1972 8037 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 8038 sizeof(struct btrfs_trans_handle), 0,
8039 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8040 if (!btrfs_trans_handle_cachep)
8041 goto fail;
9601e3f6 8042
837e1972 8043 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 8044 sizeof(struct btrfs_transaction), 0,
8045 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8046 if (!btrfs_transaction_cachep)
8047 goto fail;
9601e3f6 8048
837e1972 8049 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 8050 sizeof(struct btrfs_path), 0,
8051 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8052 if (!btrfs_path_cachep)
8053 goto fail;
9601e3f6 8054
837e1972 8055 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 8056 sizeof(struct btrfs_free_space), 0,
8057 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8058 if (!btrfs_free_space_cachep)
8059 goto fail;
8060
8ccf6f19
MX		 8061 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8062 sizeof(struct btrfs_delalloc_work), 0,
8063 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8064 NULL);
8065 if (!btrfs_delalloc_work_cachep)
8066 goto fail;
8067
39279cc3
CM		 8068 return 0;
8069fail:
8070 btrfs_destroy_cachep();
8071 return -ENOMEM;
8072}
8073
8074static int btrfs_getattr(struct vfsmount *mnt,
8075 struct dentry *dentry, struct kstat *stat)
8076{
df0af1a5 8077 u64 delalloc_bytes;
39279cc3 8078 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 8079 u32 blocksize = inode->i_sb->s_blocksize;
8080
39279cc3 8081 generic_fillattr(inode, stat);
0ee5dc67 8082 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 8083 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 8084
8085 spin_lock(&BTRFS_I(inode)->lock);
8086 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8087 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 8088 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 8089 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 8090 return 0;
8091}
8092
d397712b
CM		 8093static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8094 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 8095{
8096 struct btrfs_trans_handle *trans;
8097 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 8098 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 8099 struct inode *new_inode = new_dentry->d_inode;
8100 struct inode *old_inode = old_dentry->d_inode;
8101 struct timespec ctime = CURRENT_TIME;
00e4e6b3 8102 u64 index = 0;
4df27c4d 8103 u64 root_objectid;
39279cc3 8104 int ret;
33345d01 8105 u64 old_ino = btrfs_ino(old_inode);
39279cc3 8106
33345d01 8107 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 8108 return -EPERM;
8109
4df27c4d 8110 /* we only allow rename subvolume link between subvolumes */
33345d01 8111 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8112 return -EXDEV;
8113
33345d01
LZ		 8114 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8115 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8116 return -ENOTEMPTY;
5f39d397 8117
4df27c4d
YZ		 8118 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8119 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8120 return -ENOTEMPTY;
9c52057c
CM		 8121
8122
8123 /* check for collisions, even if the name isn't there */
8124 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8125 new_dentry->d_name.name,
8126 new_dentry->d_name.len);
8127
8128 if (ret) {
8129 if (ret == -EEXIST) {
8130 /* we shouldn't get
8131 * eexist without a new_inode */
8132 if (!new_inode) {
8133 WARN_ON(1);
8134 return ret;
8135 }
8136 } else {
8137 /* maybe -EOVERFLOW */
8138 return ret;
8139 }
8140 }
8141 ret = 0;
8142
5a3f23d5
CM		 8143 /*
8144 * we're using rename to replace one file with another.
8145 * and the replacement file is large. Start IO on it now so
8146 * we don't add too much work to the end of the transaction
8147 */
4baf8c92 8148 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8149 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8150 filemap_flush(old_inode->i_mapping);
8151
76dda93c 8152 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8153 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8154 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8155 /*
8156 * We want to reserve the absolute worst case amount of items. So if
8157 * both inodes are subvols and we need to unlink them then that would
8158 * require 4 item modifications, but if they are both normal inodes it
8159 * would require 5 item modifications, so we'll assume their normal
8160 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8161 * should cover the worst case number of items we'll modify.
8162 */
6e137ed3 8163 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8164 if (IS_ERR(trans)) {
8165 ret = PTR_ERR(trans);
8166 goto out_notrans;
8167 }
76dda93c 8168
4df27c4d
YZ		 8169 if (dest != root)
8170 btrfs_record_root_in_trans(trans, dest);
5f39d397 8171
a5719521
YZ		 8172 ret = btrfs_set_inode_index(new_dir, &index);
8173 if (ret)
8174 goto out_fail;
5a3f23d5 8175
33345d01 8176 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8177 /* force full log commit if subvolume involved. */
8178 root->fs_info->last_trans_log_full_commit = trans->transid;
8179 } else {
a5719521
YZ		 8180 ret = btrfs_insert_inode_ref(trans, dest,
8181 new_dentry->d_name.name,
8182 new_dentry->d_name.len,
33345d01
LZ		 8183 old_ino,
8184 btrfs_ino(new_dir), index);
a5719521
YZ		 8185 if (ret)
8186 goto out_fail;
4df27c4d
YZ		 8187 /*
8188 * this is an ugly little race, but the rename is required
8189 * to make sure that if we crash, the inode is either at the
8190 * old name or the new one. pinning the log transaction lets
8191 * us make sure we don't allow a log commit to come in after
8192 * we unlink the name but before we add the new name back in.
8193 */
8194 btrfs_pin_log_trans(root);
8195 }
5a3f23d5
CM		 8196 /*
8197 * make sure the inode gets flushed if it is replacing
8198 * something.
8199 */
33345d01 8200 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8201 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8202
0c4d2d95
JB		 8203 inode_inc_iversion(old_dir);
8204 inode_inc_iversion(new_dir);
8205 inode_inc_iversion(old_inode);
39279cc3
CM		 8206 old_dir->i_ctime = old_dir->i_mtime = ctime;
8207 new_dir->i_ctime = new_dir->i_mtime = ctime;
8208 old_inode->i_ctime = ctime;
5f39d397 8209
12fcfd22
CM		 8210 if (old_dentry->d_parent != new_dentry->d_parent)
8211 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8212
33345d01 8213 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8214 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8215 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8216 old_dentry->d_name.name,
8217 old_dentry->d_name.len);
8218 } else {
92986796
AV		 8219 ret = __btrfs_unlink_inode(trans, root, old_dir,
8220 old_dentry->d_inode,
8221 old_dentry->d_name.name,
8222 old_dentry->d_name.len);
8223 if (!ret)
8224 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8225 }
79787eaa
JM		 8226 if (ret) {
8227 btrfs_abort_transaction(trans, root, ret);
8228 goto out_fail;
8229 }
39279cc3
CM		 8230
8231 if (new_inode) {
0c4d2d95 8232 inode_inc_iversion(new_inode);
39279cc3 8233 new_inode->i_ctime = CURRENT_TIME;
33345d01 8234 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8235 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8236 root_objectid = BTRFS_I(new_inode)->location.objectid;
8237 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8238 root_objectid,
8239 new_dentry->d_name.name,
8240 new_dentry->d_name.len);
8241 BUG_ON(new_inode->i_nlink == 0);
8242 } else {
8243 ret = btrfs_unlink_inode(trans, dest, new_dir,
8244 new_dentry->d_inode,
8245 new_dentry->d_name.name,
8246 new_dentry->d_name.len);
8247 }
79787eaa 8248 if (!ret && new_inode->i_nlink == 0) {
e02119d5 8249 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 8250 BUG_ON(ret);
7b128766 8251 }
79787eaa
JM		 8252 if (ret) {
8253 btrfs_abort_transaction(trans, root, ret);
8254 goto out_fail;
8255 }
39279cc3 8256 }
aec7477b 8257
4df27c4d
YZ		 8258 ret = btrfs_add_link(trans, new_dir, old_inode,
8259 new_dentry->d_name.name,
a5719521 8260 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8261 if (ret) {
8262 btrfs_abort_transaction(trans, root, ret);
8263 goto out_fail;
8264 }
39279cc3 8265
33345d01 8266 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8267 struct dentry *parent = new_dentry->d_parent;
6a912213 8268 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8269 btrfs_end_log_trans(root);
8270 }
39279cc3 8271out_fail:
7ad85bb7 8272 btrfs_end_transaction(trans, root);
b44c59a8 8273out_notrans:
33345d01 8274 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8275 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8276
39279cc3
CM		 8277 return ret;
8278}
8279
8ccf6f19
MX		 8280static void btrfs_run_delalloc_work(struct btrfs_work *work)
8281{
8282 struct btrfs_delalloc_work *delalloc_work;
8283
8284 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8285 work);
8286 if (delalloc_work->wait)
8287 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8288 else
8289 filemap_flush(delalloc_work->inode->i_mapping);
8290
8291 if (delalloc_work->delay_iput)
8292 btrfs_add_delayed_iput(delalloc_work->inode);
8293 else
8294 iput(delalloc_work->inode);
8295 complete(&delalloc_work->completion);
8296}
8297
8298struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8299 int wait, int delay_iput)
8300{
8301 struct btrfs_delalloc_work *work;
8302
8303 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8304 if (!work)
8305 return NULL;
8306
8307 init_completion(&work->completion);
8308 INIT_LIST_HEAD(&work->list);
8309 work->inode = inode;
8310 work->wait = wait;
8311 work->delay_iput = delay_iput;
8312 work->work.func = btrfs_run_delalloc_work;
8313
8314 return work;
8315}
8316
8317void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8318{
8319 wait_for_completion(&work->completion);
8320 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8321}
8322
d352ac68
CM		 8323/*
8324 * some fairly slow code that needs optimization. This walks the list
8325 * of all the inodes with pending delalloc and forces them to disk.
8326 */
24bbcf04 8327int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8328{
ea8c2819 8329 struct btrfs_inode *binode;
5b21f2ed 8330 struct inode *inode;
8ccf6f19
MX		 8331 struct btrfs_delalloc_work *work, *next;
8332 struct list_head works;
1eafa6c7 8333 struct list_head splice;
8ccf6f19 8334 int ret = 0;
ea8c2819 8335
c146afad
YZ		 8336 if (root->fs_info->sb->s_flags & MS_RDONLY)
8337 return -EROFS;
8338
8ccf6f19 8339 INIT_LIST_HEAD(&works);
1eafa6c7 8340 INIT_LIST_HEAD(&splice);
63607cc8 8341
75eff68e 8342 spin_lock(&root->fs_info->delalloc_lock);
1eafa6c7
MX		 8343 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8344 while (!list_empty(&splice)) {
8345 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8346 delalloc_inodes);
1eafa6c7
MX		 8347
8348 list_del_init(&binode->delalloc_inodes);
8349
5b21f2ed 8350 inode = igrab(&binode->vfs_inode);
df0af1a5
MX		 8351 if (!inode) {
8352 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8353 &binode->runtime_flags);
1eafa6c7 8354 continue;
df0af1a5 8355 }
1eafa6c7
MX		 8356
8357 list_add_tail(&binode->delalloc_inodes,
8358 &root->fs_info->delalloc_inodes);
75eff68e 8359 spin_unlock(&root->fs_info->delalloc_lock);
1eafa6c7
MX		 8360
8361 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8362 if (unlikely(!work)) {
8363 ret = -ENOMEM;
8364 goto out;
5b21f2ed 8365 }
1eafa6c7
MX		 8366 list_add_tail(&work->list, &works);
8367 btrfs_queue_worker(&root->fs_info->flush_workers,
8368 &work->work);
8369
5b21f2ed 8370 cond_resched();
75eff68e 8371 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 8372 }
75eff68e 8373 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d 8374
1eafa6c7
MX		 8375 list_for_each_entry_safe(work, next, &works, list) {
8376 list_del_init(&work->list);
8377 btrfs_wait_and_free_delalloc_work(work);
8378 }
8379
8c8bee1d
CM		 8380 /* the filemap_flush will queue IO into the worker threads, but
8381 * we have to make sure the IO is actually started and that
8382 * ordered extents get created before we return
8383 */
8384 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8385 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8386 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8387 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8388 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8389 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8390 }
8391 atomic_dec(&root->fs_info->async_submit_draining);
1eafa6c7 8392 return 0;
8ccf6f19
MX		 8393out:
8394 list_for_each_entry_safe(work, next, &works, list) {
8395 list_del_init(&work->list);
8396 btrfs_wait_and_free_delalloc_work(work);
8397 }
1eafa6c7
MX		 8398
8399 if (!list_empty_careful(&splice)) {
8400 spin_lock(&root->fs_info->delalloc_lock);
8401 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8402 spin_unlock(&root->fs_info->delalloc_lock);
8403 }
8ccf6f19 8404 return ret;
ea8c2819
CM		 8405}
8406
39279cc3
CM		 8407static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8408 const char *symname)
8409{
8410 struct btrfs_trans_handle *trans;
8411 struct btrfs_root *root = BTRFS_I(dir)->root;
8412 struct btrfs_path *path;
8413 struct btrfs_key key;
1832a6d5 8414 struct inode *inode = NULL;
39279cc3
CM		 8415 int err;
8416 int drop_inode = 0;
8417 u64 objectid;
00e4e6b3 8418 u64 index = 0 ;
39279cc3
CM		 8419 int name_len;
8420 int datasize;
5f39d397 8421 unsigned long ptr;
39279cc3 8422 struct btrfs_file_extent_item *ei;
5f39d397 8423 struct extent_buffer *leaf;
39279cc3
CM		 8424
8425 name_len = strlen(symname) + 1;
8426 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8427 return -ENAMETOOLONG;
1832a6d5 8428
9ed74f2d
JB		 8429 /*
8430 * 2 items for inode item and ref
8431 * 2 items for dir items
8432 * 1 item for xattr if selinux is on
8433 */
a22285a6
YZ		 8434 trans = btrfs_start_transaction(root, 5);
8435 if (IS_ERR(trans))
8436 return PTR_ERR(trans);
1832a6d5 8437
581bb050
LZ		 8438 err = btrfs_find_free_ino(root, &objectid);
8439 if (err)
8440 goto out_unlock;
8441
aec7477b 8442 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8443 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8444 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8445 if (IS_ERR(inode)) {
8446 err = PTR_ERR(inode);
39279cc3 8447 goto out_unlock;
7cf96da3 8448 }
39279cc3 8449
2a7dba39 8450 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8451 if (err) {
8452 drop_inode = 1;
8453 goto out_unlock;
8454 }
8455
ad19db71
CS		 8456 /*
8457 * If the active LSM wants to access the inode during
8458 * d_instantiate it needs these. Smack checks to see
8459 * if the filesystem supports xattrs by looking at the
8460 * ops vector.
8461 */
8462 inode->i_fop = &btrfs_file_operations;
8463 inode->i_op = &btrfs_file_inode_operations;
8464
a1b075d2 8465 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8466 if (err)
8467 drop_inode = 1;
8468 else {
8469 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8470 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8471 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8472 }
39279cc3
CM		 8473 if (drop_inode)
8474 goto out_unlock;
8475
8476 path = btrfs_alloc_path();
d8926bb3
MF		 8477 if (!path) {
8478 err = -ENOMEM;
8479 drop_inode = 1;
8480 goto out_unlock;
8481 }
33345d01 8482 key.objectid = btrfs_ino(inode);
39279cc3 8483 key.offset = 0;
39279cc3
CM		 8484 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8485 datasize = btrfs_file_extent_calc_inline_size(name_len);
8486 err = btrfs_insert_empty_item(trans, root, path, &key,
8487 datasize);
54aa1f4d
CM		 8488 if (err) {
8489 drop_inode = 1;
b0839166 8490 btrfs_free_path(path);
54aa1f4d
CM		 8491 goto out_unlock;
8492 }
5f39d397
CM		 8493 leaf = path->nodes[0];
8494 ei = btrfs_item_ptr(leaf, path->slots[0],
8495 struct btrfs_file_extent_item);
8496 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8497 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8498 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8499 btrfs_set_file_extent_encryption(leaf, ei, 0);
8500 btrfs_set_file_extent_compression(leaf, ei, 0);
8501 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8502 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8503
39279cc3 8504 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8505 write_extent_buffer(leaf, symname, ptr, name_len);
8506 btrfs_mark_buffer_dirty(leaf);
39279cc3 8507 btrfs_free_path(path);
5f39d397 8508
39279cc3
CM		 8509 inode->i_op = &btrfs_symlink_inode_operations;
8510 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8511 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8512 inode_set_bytes(inode, name_len);
dbe674a9 8513 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 8514 err = btrfs_update_inode(trans, root, inode);
8515 if (err)
8516 drop_inode = 1;
39279cc3
CM		 8517
8518out_unlock:
08c422c2
AV		 8519 if (!err)
8520 d_instantiate(dentry, inode);
7ad85bb7 8521 btrfs_end_transaction(trans, root);
39279cc3
CM		 8522 if (drop_inode) {
8523 inode_dec_link_count(inode);
8524 iput(inode);
8525 }
b53d3f5d 8526 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8527 return err;
8528}
16432985 8529
0af3d00b
JB		 8530static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8531 u64 start, u64 num_bytes, u64 min_size,
8532 loff_t actual_len, u64 *alloc_hint,
8533 struct btrfs_trans_handle *trans)
d899e052 8534{
5dc562c5
JB		 8535 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8536 struct extent_map *em;
d899e052
YZ		 8537 struct btrfs_root *root = BTRFS_I(inode)->root;
8538 struct btrfs_key ins;
d899e052 8539 u64 cur_offset = start;
55a61d1d 8540 u64 i_size;
154ea289 8541 u64 cur_bytes;
d899e052 8542 int ret = 0;
0af3d00b 8543 bool own_trans = true;
d899e052 8544
0af3d00b
JB		 8545 if (trans)
8546 own_trans = false;
d899e052 8547 while (num_bytes > 0) {
0af3d00b
JB		 8548 if (own_trans) {
8549 trans = btrfs_start_transaction(root, 3);
8550 if (IS_ERR(trans)) {
8551 ret = PTR_ERR(trans);
8552 break;
8553 }
5a303d5d
YZ		 8554 }
8555
154ea289
CM		 8556 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8557 cur_bytes = max(cur_bytes, min_size);
8558 ret = btrfs_reserve_extent(trans, root, cur_bytes,
24542bf7 8559 min_size, 0, *alloc_hint, &ins, 1);
5a303d5d 8560 if (ret) {
0af3d00b
JB		 8561 if (own_trans)
8562 btrfs_end_transaction(trans, root);
a22285a6 8563 break;
d899e052 8564 }
5a303d5d 8565
d899e052
YZ		 8566 ret = insert_reserved_file_extent(trans, inode,
8567 cur_offset, ins.objectid,
8568 ins.offset, ins.offset,
920bbbfb 8569 ins.offset, 0, 0, 0,
d899e052 8570 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 8571 if (ret) {
8572 btrfs_abort_transaction(trans, root, ret);
8573 if (own_trans)
8574 btrfs_end_transaction(trans, root);
8575 break;
8576 }
a1ed835e
CM		 8577 btrfs_drop_extent_cache(inode, cur_offset,
8578 cur_offset + ins.offset -1, 0);
5a303d5d 8579
5dc562c5
JB		 8580 em = alloc_extent_map();
8581 if (!em) {
8582 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8583 &BTRFS_I(inode)->runtime_flags);
8584 goto next;
8585 }
8586
8587 em->start = cur_offset;
8588 em->orig_start = cur_offset;
8589 em->len = ins.offset;
8590 em->block_start = ins.objectid;
8591 em->block_len = ins.offset;
b4939680 8592 em->orig_block_len = ins.offset;
cc95bef6 8593 em->ram_bytes = ins.offset;
5dc562c5
JB		 8594 em->bdev = root->fs_info->fs_devices->latest_bdev;
8595 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8596 em->generation = trans->transid;
8597
8598 while (1) {
8599 write_lock(&em_tree->lock);
09a2a8f9 8600 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8601 write_unlock(&em_tree->lock);
8602 if (ret != -EEXIST)
8603 break;
8604 btrfs_drop_extent_cache(inode, cur_offset,
8605 cur_offset + ins.offset - 1,
8606 0);
8607 }
8608 free_extent_map(em);
8609next:
d899e052
YZ		 8610 num_bytes -= ins.offset;
8611 cur_offset += ins.offset;
efa56464 8612 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8613
0c4d2d95 8614 inode_inc_iversion(inode);
d899e052 8615 inode->i_ctime = CURRENT_TIME;
6cbff00f 8616 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8617 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8618 (actual_len > inode->i_size) &&
8619 (cur_offset > inode->i_size)) {
d1ea6a61 8620 if (cur_offset > actual_len)
55a61d1d 8621 i_size = actual_len;
d1ea6a61 8622 else
55a61d1d
JB		 8623 i_size = cur_offset;
8624 i_size_write(inode, i_size);
8625 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8626 }
8627
d899e052 8628 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8629
8630 if (ret) {
8631 btrfs_abort_transaction(trans, root, ret);
8632 if (own_trans)
8633 btrfs_end_transaction(trans, root);
8634 break;
8635 }
d899e052 8636
0af3d00b
JB		 8637 if (own_trans)
8638 btrfs_end_transaction(trans, root);
5a303d5d 8639 }
d899e052
YZ		 8640 return ret;
8641}
8642
0af3d00b
JB		 8643int btrfs_prealloc_file_range(struct inode *inode, int mode,
8644 u64 start, u64 num_bytes, u64 min_size,
8645 loff_t actual_len, u64 *alloc_hint)
8646{
8647 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8648 min_size, actual_len, alloc_hint,
8649 NULL);
8650}
8651
8652int btrfs_prealloc_file_range_trans(struct inode *inode,
8653 struct btrfs_trans_handle *trans, int mode,
8654 u64 start, u64 num_bytes, u64 min_size,
8655 loff_t actual_len, u64 *alloc_hint)
8656{
8657 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8658 min_size, actual_len, alloc_hint, trans);
8659}
8660
e6dcd2dc
CM		 8661static int btrfs_set_page_dirty(struct page *page)
8662{
e6dcd2dc
CM		 8663 return __set_page_dirty_nobuffers(page);
8664}
8665
10556cb2 8666static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8667{
b83cc969 8668 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8669 umode_t mode = inode->i_mode;
b83cc969 8670
cb6db4e5
JM		 8671 if (mask & MAY_WRITE &&
8672 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8673 if (btrfs_root_readonly(root))
8674 return -EROFS;
8675 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8676 return -EACCES;
8677 }
2830ba7f 8678 return generic_permission(inode, mask);
fdebe2bd 8679}
39279cc3 8680
6e1d5dcc 8681static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8682 .getattr = btrfs_getattr,
39279cc3
CM		 8683 .lookup = btrfs_lookup,
8684 .create = btrfs_create,
8685 .unlink = btrfs_unlink,
8686 .link = btrfs_link,
8687 .mkdir = btrfs_mkdir,
8688 .rmdir = btrfs_rmdir,
8689 .rename = btrfs_rename,
8690 .symlink = btrfs_symlink,
8691 .setattr = btrfs_setattr,
618e21d5 8692 .mknod = btrfs_mknod,
95819c05
CH		 8693 .setxattr = btrfs_setxattr,
8694 .getxattr = btrfs_getxattr,
5103e947 8695 .listxattr = btrfs_listxattr,
95819c05 8696 .removexattr = btrfs_removexattr,
fdebe2bd 8697 .permission = btrfs_permission,
4e34e719 8698 .get_acl = btrfs_get_acl,
39279cc3 8699};
6e1d5dcc 8700static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8701 .lookup = btrfs_lookup,
fdebe2bd 8702 .permission = btrfs_permission,
4e34e719 8703 .get_acl = btrfs_get_acl,
39279cc3 8704};
76dda93c 8705
828c0950 8706static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8707 .llseek = generic_file_llseek,
8708 .read = generic_read_dir,
cbdf5a24 8709 .readdir = btrfs_real_readdir,
34287aa3 8710 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8711#ifdef CONFIG_COMPAT
34287aa3 8712 .compat_ioctl = btrfs_ioctl,
39279cc3 8713#endif
6bf13c0c 8714 .release = btrfs_release_file,
e02119d5 8715 .fsync = btrfs_sync_file,
39279cc3
CM		 8716};
8717
d1310b2e 8718static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8719 .fill_delalloc = run_delalloc_range,
065631f6 8720 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8721 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8722 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8723 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8724 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8725 .set_bit_hook = btrfs_set_bit_hook,
8726 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8727 .merge_extent_hook = btrfs_merge_extent_hook,
8728 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8729};
8730
35054394
CM		 8731/*
8732 * btrfs doesn't support the bmap operation because swapfiles
8733 * use bmap to make a mapping of extents in the file. They assume
8734 * these extents won't change over the life of the file and they
8735 * use the bmap result to do IO directly to the drive.
8736 *
8737 * the btrfs bmap call would return logical addresses that aren't
8738 * suitable for IO and they also will change frequently as COW
8739 * operations happen. So, swapfile + btrfs == corruption.
8740 *
8741 * For now we're avoiding this by dropping bmap.
8742 */
7f09410b 8743static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8744 .readpage = btrfs_readpage,
8745 .writepage = btrfs_writepage,
b293f02e 8746 .writepages = btrfs_writepages,
3ab2fb5a 8747 .readpages = btrfs_readpages,
16432985 8748 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8749 .invalidatepage = btrfs_invalidatepage,
8750 .releasepage = btrfs_releasepage,
e6dcd2dc 8751 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8752 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8753};
8754
7f09410b 8755static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8756 .readpage = btrfs_readpage,
8757 .writepage = btrfs_writepage,
2bf5a725
CM		 8758 .invalidatepage = btrfs_invalidatepage,
8759 .releasepage = btrfs_releasepage,
39279cc3
CM		 8760};
8761
6e1d5dcc 8762static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8763 .getattr = btrfs_getattr,
8764 .setattr = btrfs_setattr,
95819c05
CH		 8765 .setxattr = btrfs_setxattr,
8766 .getxattr = btrfs_getxattr,
5103e947 8767 .listxattr = btrfs_listxattr,
95819c05 8768 .removexattr = btrfs_removexattr,
fdebe2bd 8769 .permission = btrfs_permission,
1506fcc8 8770 .fiemap = btrfs_fiemap,
4e34e719 8771 .get_acl = btrfs_get_acl,
e41f941a 8772 .update_time = btrfs_update_time,
39279cc3 8773};
6e1d5dcc 8774static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8775 .getattr = btrfs_getattr,
8776 .setattr = btrfs_setattr,
fdebe2bd 8777 .permission = btrfs_permission,
95819c05
CH		 8778 .setxattr = btrfs_setxattr,
8779 .getxattr = btrfs_getxattr,
33268eaf 8780 .listxattr = btrfs_listxattr,
95819c05 8781 .removexattr = btrfs_removexattr,
4e34e719 8782 .get_acl = btrfs_get_acl,
e41f941a 8783 .update_time = btrfs_update_time,
618e21d5 8784};
6e1d5dcc 8785static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8786 .readlink = generic_readlink,
8787 .follow_link = page_follow_link_light,
8788 .put_link = page_put_link,
f209561a 8789 .getattr = btrfs_getattr,
22c44fe6 8790 .setattr = btrfs_setattr,
fdebe2bd 8791 .permission = btrfs_permission,
0279b4cd
JO		 8792 .setxattr = btrfs_setxattr,
8793 .getxattr = btrfs_getxattr,
8794 .listxattr = btrfs_listxattr,
8795 .removexattr = btrfs_removexattr,
4e34e719 8796 .get_acl = btrfs_get_acl,
e41f941a 8797 .update_time = btrfs_update_time,
39279cc3 8798};
76dda93c 8799
82d339d9 8800const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8801 .d_delete = btrfs_dentry_delete,
b4aff1f8 8802 .d_release = btrfs_dentry_release,
76dda93c 8803};
Linux 6.x block layer and io_uring tree(s)