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