projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: kill the durable block rsv stuff
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
4b4e25f2 41#include "compat.h"
39279cc3
CM		 42#include "ctree.h"
43#include "disk-io.h"
44#include "transaction.h"
45#include "btrfs_inode.h"
46#include "ioctl.h"
47#include "print-tree.h"
0b86a832 48#include "volumes.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 51#include "tree-log.h"
c8b97818 52#include "compression.h"
b4ce94de 53#include "locking.h"
dc89e982 54#include "free-space-cache.h"
581bb050 55#include "inode-map.h"
39279cc3
CM		 56
57struct btrfs_iget_args {
58 u64 ino;
59 struct btrfs_root *root;
60};
61
6e1d5dcc
AD		 62static const struct inode_operations btrfs_dir_inode_operations;
63static const struct inode_operations btrfs_symlink_inode_operations;
64static const struct inode_operations btrfs_dir_ro_inode_operations;
65static const struct inode_operations btrfs_special_inode_operations;
66static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 67static const struct address_space_operations btrfs_aops;
68static const struct address_space_operations btrfs_symlink_aops;
828c0950 69static const struct file_operations btrfs_dir_file_operations;
d1310b2e 70static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 71
72static struct kmem_cache *btrfs_inode_cachep;
73struct kmem_cache *btrfs_trans_handle_cachep;
74struct kmem_cache *btrfs_transaction_cachep;
39279cc3 75struct kmem_cache *btrfs_path_cachep;
dc89e982 76struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 77
78#define S_SHIFT 12
79static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87};
88
a41ad394
JB		 89static int btrfs_setsize(struct inode *inode, loff_t newsize);
90static int btrfs_truncate(struct inode *inode);
c8b97818 91static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 92static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
7b128766 96
f34f57a3 97static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 98 struct inode *inode, struct inode *dir,
99 const struct qstr *qstr)
0279b4cd
JO		 100{
101 int err;
102
f34f57a3 103 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 104 if (!err)
2a7dba39 105 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 106 return err;
107}
108
c8b97818
CM		 109/*
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
113 */
d397712b 114static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 115 struct btrfs_root *root, struct inode *inode,
116 u64 start, size_t size, size_t compressed_size,
fe3f566c 117 int compress_type,
c8b97818
CM		 118 struct page **compressed_pages)
119{
120 struct btrfs_key key;
121 struct btrfs_path *path;
122 struct extent_buffer *leaf;
123 struct page *page = NULL;
124 char *kaddr;
125 unsigned long ptr;
126 struct btrfs_file_extent_item *ei;
127 int err = 0;
128 int ret;
129 size_t cur_size = size;
130 size_t datasize;
131 unsigned long offset;
c8b97818 132
fe3f566c 133 if (compressed_size && compressed_pages)
c8b97818 134 cur_size = compressed_size;
c8b97818 135
d397712b
CM		 136 path = btrfs_alloc_path();
137 if (!path)
c8b97818
CM		 138 return -ENOMEM;
139
b9473439 140 path->leave_spinning = 1;
c8b97818 141
33345d01 142 key.objectid = btrfs_ino(inode);
c8b97818
CM		 143 key.offset = start;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
149 datasize);
150 BUG_ON(ret);
151 if (ret) {
152 err = ret;
c8b97818
CM		 153 goto fail;
154 }
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
164
261507a0 165 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 166 struct page *cpage;
167 int i = 0;
d397712b 168 while (compressed_size > 0) {
c8b97818 169 cpage = compressed_pages[i];
5b050f04 170 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 171 PAGE_CACHE_SIZE);
172
b9473439 173 kaddr = kmap_atomic(cpage, KM_USER0);
c8b97818 174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
b9473439 175 kunmap_atomic(kaddr, KM_USER0);
c8b97818
CM		 176
177 i++;
178 ptr += cur_size;
179 compressed_size -= cur_size;
180 }
181 btrfs_set_file_extent_compression(leaf, ei,
261507a0 182 compress_type);
c8b97818
CM		 183 } else {
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
192 }
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
195
c2167754
YZ		 196 /*
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
200 *
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
204 */
c8b97818
CM		 205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
c2167754 207
c8b97818
CM		 208 return 0;
209fail:
210 btrfs_free_path(path);
211 return err;
212}
213
214
215/*
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
219 */
7f366cfe 220static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
fe3f566c 223 size_t compressed_size, int compress_type,
c8b97818
CM		 224 struct page **compressed_pages)
225{
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
231 u64 hint_byte;
232 u64 data_len = inline_len;
233 int ret;
234
235 if (compressed_size)
236 data_len = compressed_size;
237
238 if (start > 0 ||
70b99e69 239 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (!compressed_size &&
242 (actual_end & (root->sectorsize - 1)) == 0) ||
243 end + 1 < isize ||
244 data_len > root->fs_info->max_inline) {
245 return 1;
246 }
247
920bbbfb 248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
a1ed835e 249 &hint_byte, 1);
c8b97818
CM		 250 BUG_ON(ret);
251
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
fe3f566c 256 compress_type, compressed_pages);
c8b97818 257 BUG_ON(ret);
0ca1f7ce 258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 260 return 0;
261}
262
771ed689
CM		 263struct async_extent {
264 u64 start;
265 u64 ram_size;
266 u64 compressed_size;
267 struct page **pages;
268 unsigned long nr_pages;
261507a0 269 int compress_type;
771ed689
CM		 270 struct list_head list;
271};
272
273struct async_cow {
274 struct inode *inode;
275 struct btrfs_root *root;
276 struct page *locked_page;
277 u64 start;
278 u64 end;
279 struct list_head extents;
280 struct btrfs_work work;
281};
282
283static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
285 u64 compressed_size,
286 struct page **pages,
261507a0
LZ		 287 unsigned long nr_pages,
288 int compress_type)
771ed689
CM		 289{
290 struct async_extent *async_extent;
291
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
dac97e51 293 BUG_ON(!async_extent);
771ed689
CM		 294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
261507a0 299 async_extent->compress_type = compress_type;
771ed689
CM		 300 list_add_tail(&async_extent->list, &cow->extents);
301 return 0;
302}
303
d352ac68 304/*
771ed689
CM		 305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
c8b97818 308 *
771ed689
CM		 309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
c8b97818 314 *
771ed689
CM		 315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
d352ac68 319 */
771ed689
CM		 320static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
322 u64 start, u64 end,
323 struct async_cow *async_cow,
324 int *num_added)
b888db2b
CM		 325{
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
db94535d 328 u64 num_bytes;
db94535d 329 u64 blocksize = root->sectorsize;
c8b97818 330 u64 actual_end;
42dc7bab 331 u64 isize = i_size_read(inode);
e6dcd2dc 332 int ret = 0;
c8b97818
CM		 333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
771ed689 339 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 340 int i;
341 int will_compress;
261507a0 342 int compress_type = root->fs_info->compress_type;
b888db2b 343
4cb5300b
CM		 344 /* if this is a small write inside eof, kick off a defragbot */
345 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL, inode);
347
42dc7bab 348 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 349again:
350 will_compress = 0;
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 353
f03d9301
CM		 354 /*
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
360 *
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
363 */
364 if (actual_end <= start)
365 goto cleanup_and_bail_uncompressed;
366
c8b97818
CM		 367 total_compressed = actual_end - start;
368
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
374 *
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
c8b97818
CM		 378 */
379 total_compressed = min(total_compressed, max_uncompressed);
db94535d 380 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 381 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 382 total_in = 0;
383 ret = 0;
db94535d 384
771ed689
CM		 385 /*
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
c8b97818 389 */
6cbff00f 390 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 391 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 392 (BTRFS_I(inode)->force_compress) ||
393 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 394 WARN_ON(pages);
cfbc246e 395 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
dac97e51 396 BUG_ON(!pages);
c8b97818 397
261507a0
LZ		 398 if (BTRFS_I(inode)->force_compress)
399 compress_type = BTRFS_I(inode)->force_compress;
400
401 ret = btrfs_compress_pages(compress_type,
402 inode->i_mapping, start,
403 total_compressed, pages,
404 nr_pages, &nr_pages_ret,
405 &total_in,
406 &total_compressed,
407 max_compressed);
c8b97818
CM		 408
409 if (!ret) {
410 unsigned long offset = total_compressed &
411 (PAGE_CACHE_SIZE - 1);
412 struct page *page = pages[nr_pages_ret - 1];
413 char *kaddr;
414
415 /* zero the tail end of the last page, we might be
416 * sending it down to disk
417 */
418 if (offset) {
419 kaddr = kmap_atomic(page, KM_USER0);
420 memset(kaddr + offset, 0,
421 PAGE_CACHE_SIZE - offset);
422 kunmap_atomic(kaddr, KM_USER0);
423 }
424 will_compress = 1;
425 }
426 }
427 if (start == 0) {
7a7eaa40 428 trans = btrfs_join_transaction(root);
3612b495 429 BUG_ON(IS_ERR(trans));
0ca1f7ce 430 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 431
c8b97818 432 /* lets try to make an inline extent */
771ed689 433 if (ret || total_in < (actual_end - start)) {
c8b97818 434 /* we didn't compress the entire range, try
771ed689 435 * to make an uncompressed inline extent.
c8b97818
CM		 436 */
437 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 438 start, end, 0, 0, NULL);
c8b97818 439 } else {
771ed689 440 /* try making a compressed inline extent */
c8b97818
CM		 441 ret = cow_file_range_inline(trans, root, inode,
442 start, end,
fe3f566c
LZ		 443 total_compressed,
444 compress_type, pages);
c8b97818
CM		 445 }
446 if (ret == 0) {
771ed689
CM		 447 /*
448 * inline extent creation worked, we don't need
449 * to create any more async work items. Unlock
450 * and free up our temp pages.
451 */
c8b97818 452 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 453 &BTRFS_I(inode)->io_tree,
454 start, end, NULL,
455 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 456 EXTENT_CLEAR_DELALLOC |
a791e35e 457 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 458
459 btrfs_end_transaction(trans, root);
c8b97818
CM		 460 goto free_pages_out;
461 }
c2167754 462 btrfs_end_transaction(trans, root);
c8b97818
CM		 463 }
464
465 if (will_compress) {
466 /*
467 * we aren't doing an inline extent round the compressed size
468 * up to a block size boundary so the allocator does sane
469 * things
470 */
471 total_compressed = (total_compressed + blocksize - 1) &
472 ~(blocksize - 1);
473
474 /*
475 * one last check to make sure the compression is really a
476 * win, compare the page count read with the blocks on disk
477 */
478 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
479 ~(PAGE_CACHE_SIZE - 1);
480 if (total_compressed >= total_in) {
481 will_compress = 0;
482 } else {
c8b97818
CM		 483 num_bytes = total_in;
484 }
485 }
486 if (!will_compress && pages) {
487 /*
488 * the compression code ran but failed to make things smaller,
489 * free any pages it allocated and our page pointer array
490 */
491 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 492 WARN_ON(pages[i]->mapping);
c8b97818
CM		 493 page_cache_release(pages[i]);
494 }
495 kfree(pages);
496 pages = NULL;
497 total_compressed = 0;
498 nr_pages_ret = 0;
499
500 /* flag the file so we don't compress in the future */
1e701a32
CM		 501 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
502 !(BTRFS_I(inode)->force_compress)) {
a555f810 503 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 504 }
c8b97818 505 }
771ed689
CM		 506 if (will_compress) {
507 *num_added += 1;
c8b97818 508
771ed689
CM		 509 /* the async work queues will take care of doing actual
510 * allocation on disk for these compressed pages,
511 * and will submit them to the elevator.
512 */
513 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 514 total_compressed, pages, nr_pages_ret,
515 compress_type);
179e29e4 516
24ae6365 517 if (start + num_bytes < end) {
771ed689
CM		 518 start += num_bytes;
519 pages = NULL;
520 cond_resched();
521 goto again;
522 }
523 } else {
f03d9301 524cleanup_and_bail_uncompressed:
771ed689
CM		 525 /*
526 * No compression, but we still need to write the pages in
527 * the file we've been given so far. redirty the locked
528 * page if it corresponds to our extent and set things up
529 * for the async work queue to run cow_file_range to do
530 * the normal delalloc dance
531 */
532 if (page_offset(locked_page) >= start &&
533 page_offset(locked_page) <= end) {
534 __set_page_dirty_nobuffers(locked_page);
535 /* unlocked later on in the async handlers */
536 }
261507a0
LZ		 537 add_async_extent(async_cow, start, end - start + 1,
538 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 539 *num_added += 1;
540 }
3b951516 541
771ed689
CM		 542out:
543 return 0;
544
545free_pages_out:
546 for (i = 0; i < nr_pages_ret; i++) {
547 WARN_ON(pages[i]->mapping);
548 page_cache_release(pages[i]);
549 }
d397712b 550 kfree(pages);
771ed689
CM		 551
552 goto out;
553}
554
555/*
556 * phase two of compressed writeback. This is the ordered portion
557 * of the code, which only gets called in the order the work was
558 * queued. We walk all the async extents created by compress_file_range
559 * and send them down to the disk.
560 */
561static noinline int submit_compressed_extents(struct inode *inode,
562 struct async_cow *async_cow)
563{
564 struct async_extent *async_extent;
565 u64 alloc_hint = 0;
566 struct btrfs_trans_handle *trans;
567 struct btrfs_key ins;
568 struct extent_map *em;
569 struct btrfs_root *root = BTRFS_I(inode)->root;
570 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
571 struct extent_io_tree *io_tree;
f5a84ee3 572 int ret = 0;
771ed689
CM		 573
574 if (list_empty(&async_cow->extents))
575 return 0;
576
771ed689 577
d397712b 578 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 579 async_extent = list_entry(async_cow->extents.next,
580 struct async_extent, list);
581 list_del(&async_extent->list);
c8b97818 582
771ed689
CM		 583 io_tree = &BTRFS_I(inode)->io_tree;
584
f5a84ee3 585retry:
771ed689
CM		 586 /* did the compression code fall back to uncompressed IO? */
587 if (!async_extent->pages) {
588 int page_started = 0;
589 unsigned long nr_written = 0;
590
591 lock_extent(io_tree, async_extent->start,
2ac55d41
JB		 592 async_extent->start +
593 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 594
595 /* allocate blocks */
f5a84ee3
JB		 596 ret = cow_file_range(inode, async_cow->locked_page,
597 async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1,
600 &page_started, &nr_written, 0);
771ed689
CM		 601
602 /*
603 * if page_started, cow_file_range inserted an
604 * inline extent and took care of all the unlocking
605 * and IO for us. Otherwise, we need to submit
606 * all those pages down to the drive.
607 */
f5a84ee3 608 if (!page_started && !ret)
771ed689
CM		 609 extent_write_locked_range(io_tree,
610 inode, async_extent->start,
d397712b 611 async_extent->start +
771ed689
CM		 612 async_extent->ram_size - 1,
613 btrfs_get_extent,
614 WB_SYNC_ALL);
615 kfree(async_extent);
616 cond_resched();
617 continue;
618 }
619
620 lock_extent(io_tree, async_extent->start,
621 async_extent->start + async_extent->ram_size - 1,
622 GFP_NOFS);
771ed689 623
7a7eaa40 624 trans = btrfs_join_transaction(root);
3612b495 625 BUG_ON(IS_ERR(trans));
74b21075 626 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689
CM		 627 ret = btrfs_reserve_extent(trans, root,
628 async_extent->compressed_size,
629 async_extent->compressed_size,
630 0, alloc_hint,
631 (u64)-1, &ins, 1);
c2167754
YZ		 632 btrfs_end_transaction(trans, root);
633
f5a84ee3
JB		 634 if (ret) {
635 int i;
636 for (i = 0; i < async_extent->nr_pages; i++) {
637 WARN_ON(async_extent->pages[i]->mapping);
638 page_cache_release(async_extent->pages[i]);
639 }
640 kfree(async_extent->pages);
641 async_extent->nr_pages = 0;
642 async_extent->pages = NULL;
643 unlock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1, GFP_NOFS);
646 goto retry;
647 }
648
c2167754
YZ		 649 /*
650 * here we're doing allocation and writeback of the
651 * compressed pages
652 */
653 btrfs_drop_extent_cache(inode, async_extent->start,
654 async_extent->start +
655 async_extent->ram_size - 1, 0);
656
172ddd60 657 em = alloc_extent_map();
c26a9203 658 BUG_ON(!em);
771ed689
CM		 659 em->start = async_extent->start;
660 em->len = async_extent->ram_size;
445a6944 661 em->orig_start = em->start;
c8b97818 662
771ed689
CM		 663 em->block_start = ins.objectid;
664 em->block_len = ins.offset;
665 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 666 em->compress_type = async_extent->compress_type;
771ed689
CM		 667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
668 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
669
d397712b 670 while (1) {
890871be 671 write_lock(&em_tree->lock);
771ed689 672 ret = add_extent_mapping(em_tree, em);
890871be 673 write_unlock(&em_tree->lock);
771ed689
CM		 674 if (ret != -EEXIST) {
675 free_extent_map(em);
676 break;
677 }
678 btrfs_drop_extent_cache(inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1, 0);
681 }
682
261507a0
LZ		 683 ret = btrfs_add_ordered_extent_compress(inode,
684 async_extent->start,
685 ins.objectid,
686 async_extent->ram_size,
687 ins.offset,
688 BTRFS_ORDERED_COMPRESSED,
689 async_extent->compress_type);
771ed689
CM		 690 BUG_ON(ret);
691
771ed689
CM		 692 /*
693 * clear dirty, set writeback and unlock the pages.
694 */
695 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 696 &BTRFS_I(inode)->io_tree,
697 async_extent->start,
698 async_extent->start +
699 async_extent->ram_size - 1,
700 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
701 EXTENT_CLEAR_UNLOCK |
a3429ab7 702 EXTENT_CLEAR_DELALLOC |
a791e35e 703 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 704
705 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 706 async_extent->start,
707 async_extent->ram_size,
708 ins.objectid,
709 ins.offset, async_extent->pages,
710 async_extent->nr_pages);
771ed689
CM		 711
712 BUG_ON(ret);
771ed689
CM		 713 alloc_hint = ins.objectid + ins.offset;
714 kfree(async_extent);
715 cond_resched();
716 }
717
771ed689
CM		 718 return 0;
719}
720
4b46fce2
JB		 721static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
722 u64 num_bytes)
723{
724 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
725 struct extent_map *em;
726 u64 alloc_hint = 0;
727
728 read_lock(&em_tree->lock);
729 em = search_extent_mapping(em_tree, start, num_bytes);
730 if (em) {
731 /*
732 * if block start isn't an actual block number then find the
733 * first block in this inode and use that as a hint. If that
734 * block is also bogus then just don't worry about it.
735 */
736 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
737 free_extent_map(em);
738 em = search_extent_mapping(em_tree, 0, 0);
739 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
740 alloc_hint = em->block_start;
741 if (em)
742 free_extent_map(em);
743 } else {
744 alloc_hint = em->block_start;
745 free_extent_map(em);
746 }
747 }
748 read_unlock(&em_tree->lock);
749
750 return alloc_hint;
751}
752
771ed689
CM		 753/*
754 * when extent_io.c finds a delayed allocation range in the file,
755 * the call backs end up in this code. The basic idea is to
756 * allocate extents on disk for the range, and create ordered data structs
757 * in ram to track those extents.
758 *
759 * locked_page is the page that writepage had locked already. We use
760 * it to make sure we don't do extra locks or unlocks.
761 *
762 * *page_started is set to one if we unlock locked_page and do everything
763 * required to start IO on it. It may be clean and already done with
764 * IO when we return.
765 */
766static noinline int cow_file_range(struct inode *inode,
767 struct page *locked_page,
768 u64 start, u64 end, int *page_started,
769 unsigned long *nr_written,
770 int unlock)
771{
772 struct btrfs_root *root = BTRFS_I(inode)->root;
773 struct btrfs_trans_handle *trans;
774 u64 alloc_hint = 0;
775 u64 num_bytes;
776 unsigned long ram_size;
777 u64 disk_num_bytes;
778 u64 cur_alloc_size;
779 u64 blocksize = root->sectorsize;
771ed689
CM		 780 struct btrfs_key ins;
781 struct extent_map *em;
782 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
783 int ret = 0;
784
2cf8572d 785 BUG_ON(btrfs_is_free_space_inode(root, inode));
7a7eaa40 786 trans = btrfs_join_transaction(root);
3612b495 787 BUG_ON(IS_ERR(trans));
0ca1f7ce 788 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 789
771ed689
CM		 790 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
791 num_bytes = max(blocksize, num_bytes);
792 disk_num_bytes = num_bytes;
793 ret = 0;
794
4cb5300b
CM		 795 /* if this is a small write inside eof, kick off defrag */
796 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
797 btrfs_add_inode_defrag(trans, inode);
798
771ed689
CM		 799 if (start == 0) {
800 /* lets try to make an inline extent */
801 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 802 start, end, 0, 0, NULL);
771ed689
CM		 803 if (ret == 0) {
804 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 805 &BTRFS_I(inode)->io_tree,
806 start, end, NULL,
807 EXTENT_CLEAR_UNLOCK_PAGE |
808 EXTENT_CLEAR_UNLOCK |
809 EXTENT_CLEAR_DELALLOC |
810 EXTENT_CLEAR_DIRTY |
811 EXTENT_SET_WRITEBACK |
812 EXTENT_END_WRITEBACK);
c2167754 813
771ed689
CM		 814 *nr_written = *nr_written +
815 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
816 *page_started = 1;
817 ret = 0;
818 goto out;
819 }
820 }
821
822 BUG_ON(disk_num_bytes >
823 btrfs_super_total_bytes(&root->fs_info->super_copy));
824
4b46fce2 825 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 826 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
827
d397712b 828 while (disk_num_bytes > 0) {
a791e35e
CM		 829 unsigned long op;
830
287a0ab9 831 cur_alloc_size = disk_num_bytes;
e6dcd2dc 832 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 833 root->sectorsize, 0, alloc_hint,
e6dcd2dc 834 (u64)-1, &ins, 1);
d397712b
CM		 835 BUG_ON(ret);
836
172ddd60 837 em = alloc_extent_map();
c26a9203 838 BUG_ON(!em);
e6dcd2dc 839 em->start = start;
445a6944 840 em->orig_start = em->start;
771ed689
CM		 841 ram_size = ins.offset;
842 em->len = ins.offset;
c8b97818 843
e6dcd2dc 844 em->block_start = ins.objectid;
c8b97818 845 em->block_len = ins.offset;
e6dcd2dc 846 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 847 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 848
d397712b 849 while (1) {
890871be 850 write_lock(&em_tree->lock);
e6dcd2dc 851 ret = add_extent_mapping(em_tree, em);
890871be 852 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 853 if (ret != -EEXIST) {
854 free_extent_map(em);
855 break;
856 }
857 btrfs_drop_extent_cache(inode, start,
c8b97818 858 start + ram_size - 1, 0);
e6dcd2dc
CM		 859 }
860
98d20f67 861 cur_alloc_size = ins.offset;
e6dcd2dc 862 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 863 ram_size, cur_alloc_size, 0);
e6dcd2dc 864 BUG_ON(ret);
c8b97818 865
17d217fe
YZ		 866 if (root->root_key.objectid ==
867 BTRFS_DATA_RELOC_TREE_OBJECTID) {
868 ret = btrfs_reloc_clone_csums(inode, start,
869 cur_alloc_size);
870 BUG_ON(ret);
871 }
872
d397712b 873 if (disk_num_bytes < cur_alloc_size)
3b951516 874 break;
d397712b 875
c8b97818
CM		 876 /* we're not doing compressed IO, don't unlock the first
877 * page (which the caller expects to stay locked), don't
878 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 879 *
880 * Do set the Private2 bit so we know this page was properly
881 * setup for writepage
c8b97818 882 */
a791e35e
CM		 883 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
884 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
885 EXTENT_SET_PRIVATE2;
886
c8b97818
CM		 887 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
888 start, start + ram_size - 1,
a791e35e 889 locked_page, op);
c8b97818 890 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 891 num_bytes -= cur_alloc_size;
892 alloc_hint = ins.objectid + ins.offset;
893 start += cur_alloc_size;
b888db2b 894 }
b888db2b 895out:
771ed689 896 ret = 0;
b888db2b 897 btrfs_end_transaction(trans, root);
c8b97818 898
be20aa9d 899 return ret;
771ed689 900}
c8b97818 901
771ed689
CM		 902/*
903 * work queue call back to started compression on a file and pages
904 */
905static noinline void async_cow_start(struct btrfs_work *work)
906{
907 struct async_cow *async_cow;
908 int num_added = 0;
909 async_cow = container_of(work, struct async_cow, work);
910
911 compress_file_range(async_cow->inode, async_cow->locked_page,
912 async_cow->start, async_cow->end, async_cow,
913 &num_added);
914 if (num_added == 0)
915 async_cow->inode = NULL;
916}
917
918/*
919 * work queue call back to submit previously compressed pages
920 */
921static noinline void async_cow_submit(struct btrfs_work *work)
922{
923 struct async_cow *async_cow;
924 struct btrfs_root *root;
925 unsigned long nr_pages;
926
927 async_cow = container_of(work, struct async_cow, work);
928
929 root = async_cow->root;
930 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
931 PAGE_CACHE_SHIFT;
932
933 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
934
935 if (atomic_read(&root->fs_info->async_delalloc_pages) <
936 5 * 1042 * 1024 &&
937 waitqueue_active(&root->fs_info->async_submit_wait))
938 wake_up(&root->fs_info->async_submit_wait);
939
d397712b 940 if (async_cow->inode)
771ed689 941 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 942}
c8b97818 943
771ed689
CM		 944static noinline void async_cow_free(struct btrfs_work *work)
945{
946 struct async_cow *async_cow;
947 async_cow = container_of(work, struct async_cow, work);
948 kfree(async_cow);
949}
950
951static int cow_file_range_async(struct inode *inode, struct page *locked_page,
952 u64 start, u64 end, int *page_started,
953 unsigned long *nr_written)
954{
955 struct async_cow *async_cow;
956 struct btrfs_root *root = BTRFS_I(inode)->root;
957 unsigned long nr_pages;
958 u64 cur_end;
959 int limit = 10 * 1024 * 1042;
960
a3429ab7
CM		 961 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
962 1, 0, NULL, GFP_NOFS);
d397712b 963 while (start < end) {
771ed689 964 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
8d413713 965 BUG_ON(!async_cow);
771ed689
CM		 966 async_cow->inode = inode;
967 async_cow->root = root;
968 async_cow->locked_page = locked_page;
969 async_cow->start = start;
970
6cbff00f 971 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 972 cur_end = end;
973 else
974 cur_end = min(end, start + 512 * 1024 - 1);
975
976 async_cow->end = cur_end;
977 INIT_LIST_HEAD(&async_cow->extents);
978
979 async_cow->work.func = async_cow_start;
980 async_cow->work.ordered_func = async_cow_submit;
981 async_cow->work.ordered_free = async_cow_free;
982 async_cow->work.flags = 0;
983
771ed689
CM		 984 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
985 PAGE_CACHE_SHIFT;
986 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
987
988 btrfs_queue_worker(&root->fs_info->delalloc_workers,
989 &async_cow->work);
990
991 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
992 wait_event(root->fs_info->async_submit_wait,
993 (atomic_read(&root->fs_info->async_delalloc_pages) <
994 limit));
995 }
996
d397712b 997 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 998 atomic_read(&root->fs_info->async_delalloc_pages)) {
999 wait_event(root->fs_info->async_submit_wait,
1000 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1001 0));
1002 }
1003
1004 *nr_written += nr_pages;
1005 start = cur_end + 1;
1006 }
1007 *page_started = 1;
1008 return 0;
be20aa9d
CM		 1009}
1010
d397712b 1011static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1012 u64 bytenr, u64 num_bytes)
1013{
1014 int ret;
1015 struct btrfs_ordered_sum *sums;
1016 LIST_HEAD(list);
1017
07d400a6 1018 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1019 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1020 if (ret == 0 && list_empty(&list))
1021 return 0;
1022
1023 while (!list_empty(&list)) {
1024 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025 list_del(&sums->list);
1026 kfree(sums);
1027 }
1028 return 1;
1029}
1030
d352ac68
CM		 1031/*
1032 * when nowcow writeback call back. This checks for snapshots or COW copies
1033 * of the extents that exist in the file, and COWs the file as required.
1034 *
1035 * If no cow copies or snapshots exist, we write directly to the existing
1036 * blocks on disk
1037 */
7f366cfe
CM		 1038static noinline int run_delalloc_nocow(struct inode *inode,
1039 struct page *locked_page,
771ed689
CM		 1040 u64 start, u64 end, int *page_started, int force,
1041 unsigned long *nr_written)
be20aa9d 1042{
be20aa9d 1043 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1044 struct btrfs_trans_handle *trans;
be20aa9d 1045 struct extent_buffer *leaf;
be20aa9d 1046 struct btrfs_path *path;
80ff3856 1047 struct btrfs_file_extent_item *fi;
be20aa9d 1048 struct btrfs_key found_key;
80ff3856
YZ		 1049 u64 cow_start;
1050 u64 cur_offset;
1051 u64 extent_end;
5d4f98a2 1052 u64 extent_offset;
80ff3856
YZ		 1053 u64 disk_bytenr;
1054 u64 num_bytes;
1055 int extent_type;
1056 int ret;
d899e052 1057 int type;
80ff3856
YZ		 1058 int nocow;
1059 int check_prev = 1;
82d5902d 1060 bool nolock;
33345d01 1061 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1062
1063 path = btrfs_alloc_path();
d8926bb3
MF		 1064 if (!path)
1065 return -ENOMEM;
82d5902d 1066
2cf8572d 1067 nolock = btrfs_is_free_space_inode(root, inode);
82d5902d
LZ		 1068
1069 if (nolock)
7a7eaa40 1070 trans = btrfs_join_transaction_nolock(root);
82d5902d 1071 else
7a7eaa40 1072 trans = btrfs_join_transaction(root);
ff5714cc 1073
3612b495 1074 BUG_ON(IS_ERR(trans));
74b21075 1075 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1076
80ff3856
YZ		 1077 cow_start = (u64)-1;
1078 cur_offset = start;
1079 while (1) {
33345d01 1080 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856
YZ		 1081 cur_offset, 0);
1082 BUG_ON(ret < 0);
1083 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084 leaf = path->nodes[0];
1085 btrfs_item_key_to_cpu(leaf, &found_key,
1086 path->slots[0] - 1);
33345d01 1087 if (found_key.objectid == ino &&
80ff3856
YZ		 1088 found_key.type == BTRFS_EXTENT_DATA_KEY)
1089 path->slots[0]--;
1090 }
1091 check_prev = 0;
1092next_slot:
1093 leaf = path->nodes[0];
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1096 if (ret < 0)
1097 BUG_ON(1);
1098 if (ret > 0)
1099 break;
1100 leaf = path->nodes[0];
1101 }
be20aa9d 1102
80ff3856
YZ		 1103 nocow = 0;
1104 disk_bytenr = 0;
17d217fe 1105 num_bytes = 0;
80ff3856
YZ		 1106 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
33345d01 1108 if (found_key.objectid > ino ||
80ff3856
YZ		 1109 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110 found_key.offset > end)
1111 break;
1112
1113 if (found_key.offset > cur_offset) {
1114 extent_end = found_key.offset;
e9061e21 1115 extent_type = 0;
80ff3856
YZ		 1116 goto out_check;
1117 }
1118
1119 fi = btrfs_item_ptr(leaf, path->slots[0],
1120 struct btrfs_file_extent_item);
1121 extent_type = btrfs_file_extent_type(leaf, fi);
1122
d899e052
YZ		 1123 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1125 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1126 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1127 extent_end = found_key.offset +
1128 btrfs_file_extent_num_bytes(leaf, fi);
1129 if (extent_end <= start) {
1130 path->slots[0]++;
1131 goto next_slot;
1132 }
17d217fe
YZ		 1133 if (disk_bytenr == 0)
1134 goto out_check;
80ff3856
YZ		 1135 if (btrfs_file_extent_compression(leaf, fi) ||
1136 btrfs_file_extent_encryption(leaf, fi) ||
1137 btrfs_file_extent_other_encoding(leaf, fi))
1138 goto out_check;
d899e052
YZ		 1139 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140 goto out_check;
d2fb3437 1141 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1142 goto out_check;
33345d01 1143 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1144 found_key.offset -
1145 extent_offset, disk_bytenr))
17d217fe 1146 goto out_check;
5d4f98a2 1147 disk_bytenr += extent_offset;
17d217fe
YZ		 1148 disk_bytenr += cur_offset - found_key.offset;
1149 num_bytes = min(end + 1, extent_end) - cur_offset;
1150 /*
1151 * force cow if csum exists in the range.
1152 * this ensure that csum for a given extent are
1153 * either valid or do not exist.
1154 */
1155 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156 goto out_check;
80ff3856
YZ		 1157 nocow = 1;
1158 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159 extent_end = found_key.offset +
1160 btrfs_file_extent_inline_len(leaf, fi);
1161 extent_end = ALIGN(extent_end, root->sectorsize);
1162 } else {
1163 BUG_ON(1);
1164 }
1165out_check:
1166 if (extent_end <= start) {
1167 path->slots[0]++;
1168 goto next_slot;
1169 }
1170 if (!nocow) {
1171 if (cow_start == (u64)-1)
1172 cow_start = cur_offset;
1173 cur_offset = extent_end;
1174 if (cur_offset > end)
1175 break;
1176 path->slots[0]++;
1177 goto next_slot;
7ea394f1
YZ		 1178 }
1179
b3b4aa74 1180 btrfs_release_path(path);
80ff3856
YZ		 1181 if (cow_start != (u64)-1) {
1182 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1183 found_key.offset - 1, page_started,
1184 nr_written, 1);
80ff3856
YZ		 1185 BUG_ON(ret);
1186 cow_start = (u64)-1;
7ea394f1 1187 }
80ff3856 1188
d899e052
YZ		 1189 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190 struct extent_map *em;
1191 struct extent_map_tree *em_tree;
1192 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1193 em = alloc_extent_map();
c26a9203 1194 BUG_ON(!em);
d899e052 1195 em->start = cur_offset;
445a6944 1196 em->orig_start = em->start;
d899e052
YZ		 1197 em->len = num_bytes;
1198 em->block_len = num_bytes;
1199 em->block_start = disk_bytenr;
1200 em->bdev = root->fs_info->fs_devices->latest_bdev;
1201 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202 while (1) {
890871be 1203 write_lock(&em_tree->lock);
d899e052 1204 ret = add_extent_mapping(em_tree, em);
890871be 1205 write_unlock(&em_tree->lock);
d899e052
YZ		 1206 if (ret != -EEXIST) {
1207 free_extent_map(em);
1208 break;
1209 }
1210 btrfs_drop_extent_cache(inode, em->start,
1211 em->start + em->len - 1, 0);
1212 }
1213 type = BTRFS_ORDERED_PREALLOC;
1214 } else {
1215 type = BTRFS_ORDERED_NOCOW;
1216 }
80ff3856
YZ		 1217
1218 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1219 num_bytes, num_bytes, type);
1220 BUG_ON(ret);
771ed689 1221
efa56464
YZ		 1222 if (root->root_key.objectid ==
1223 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225 num_bytes);
1226 BUG_ON(ret);
1227 }
1228
d899e052 1229 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1230 cur_offset, cur_offset + num_bytes - 1,
1231 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1234 cur_offset = extent_end;
1235 if (cur_offset > end)
1236 break;
be20aa9d 1237 }
b3b4aa74 1238 btrfs_release_path(path);
80ff3856
YZ		 1239
1240 if (cur_offset <= end && cow_start == (u64)-1)
1241 cow_start = cur_offset;
1242 if (cow_start != (u64)-1) {
1243 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1244 page_started, nr_written, 1);
80ff3856
YZ		 1245 BUG_ON(ret);
1246 }
1247
0cb59c99
JB		 1248 if (nolock) {
1249 ret = btrfs_end_transaction_nolock(trans, root);
1250 BUG_ON(ret);
1251 } else {
1252 ret = btrfs_end_transaction(trans, root);
1253 BUG_ON(ret);
1254 }
7ea394f1 1255 btrfs_free_path(path);
80ff3856 1256 return 0;
be20aa9d
CM		 1257}
1258
d352ac68
CM		 1259/*
1260 * extent_io.c call back to do delayed allocation processing
1261 */
c8b97818 1262static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1263 u64 start, u64 end, int *page_started,
1264 unsigned long *nr_written)
be20aa9d 1265{
be20aa9d 1266 int ret;
7f366cfe 1267 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1268
6cbff00f 1269 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
c8b97818 1270 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1271 page_started, 1, nr_written);
6cbff00f 1272 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
d899e052 1273 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1274 page_started, 0, nr_written);
1e701a32 1275 else if (!btrfs_test_opt(root, COMPRESS) &&
75e7cb7f
LB		 1276 !(BTRFS_I(inode)->force_compress) &&
1277 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
7f366cfe
CM		 1278 ret = cow_file_range(inode, locked_page, start, end,
1279 page_started, nr_written, 1);
be20aa9d 1280 else
771ed689 1281 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1282 page_started, nr_written);
b888db2b
CM		 1283 return ret;
1284}
1285
1bf85046
JM		 1286static void btrfs_split_extent_hook(struct inode *inode,
1287 struct extent_state *orig, u64 split)
9ed74f2d 1288{
0ca1f7ce 1289 /* not delalloc, ignore it */
9ed74f2d 1290 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1291 return;
9ed74f2d 1292
9e0baf60
JB		 1293 spin_lock(&BTRFS_I(inode)->lock);
1294 BTRFS_I(inode)->outstanding_extents++;
1295 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1296}
1297
1298/*
1299 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300 * extents so we can keep track of new extents that are just merged onto old
1301 * extents, such as when we are doing sequential writes, so we can properly
1302 * account for the metadata space we'll need.
1303 */
1bf85046
JM		 1304static void btrfs_merge_extent_hook(struct inode *inode,
1305 struct extent_state *new,
1306 struct extent_state *other)
9ed74f2d 1307{
9ed74f2d
JB		 1308 /* not delalloc, ignore it */
1309 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1310 return;
9ed74f2d 1311
9e0baf60
JB		 1312 spin_lock(&BTRFS_I(inode)->lock);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1315}
1316
d352ac68
CM		 1317/*
1318 * extent_io.c set_bit_hook, used to track delayed allocation
1319 * bytes in this file, and to maintain the list of inodes that
1320 * have pending delalloc work to be done.
1321 */
1bf85046
JM		 1322static void btrfs_set_bit_hook(struct inode *inode,
1323 struct extent_state *state, int *bits)
291d673e 1324{
9ed74f2d 1325
75eff68e
CM		 1326 /*
1327 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1328 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1329 * bit, which is only set or cleared with irqs on
1330 */
0ca1f7ce 1331 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1332 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1333 u64 len = state->end + 1 - state->start;
2cf8572d 1334 bool do_list = !btrfs_is_free_space_inode(root, inode);
9ed74f2d 1335
9e0baf60 1336 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1337 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1338 } else {
1339 spin_lock(&BTRFS_I(inode)->lock);
1340 BTRFS_I(inode)->outstanding_extents++;
1341 spin_unlock(&BTRFS_I(inode)->lock);
1342 }
287a0ab9 1343
75eff68e 1344 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1345 BTRFS_I(inode)->delalloc_bytes += len;
1346 root->fs_info->delalloc_bytes += len;
0cb59c99 1347 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
ea8c2819
CM		 1348 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349 &root->fs_info->delalloc_inodes);
1350 }
75eff68e 1351 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1352 }
291d673e
CM		 1353}
1354
d352ac68
CM		 1355/*
1356 * extent_io.c clear_bit_hook, see set_bit_hook for why
1357 */
1bf85046
JM		 1358static void btrfs_clear_bit_hook(struct inode *inode,
1359 struct extent_state *state, int *bits)
291d673e 1360{
75eff68e
CM		 1361 /*
1362 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1363 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1364 * bit, which is only set or cleared with irqs on
1365 */
0ca1f7ce 1366 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1367 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1368 u64 len = state->end + 1 - state->start;
2cf8572d 1369 bool do_list = !btrfs_is_free_space_inode(root, inode);
bcbfce8a 1370
9e0baf60 1371 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1372 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1373 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374 spin_lock(&BTRFS_I(inode)->lock);
1375 BTRFS_I(inode)->outstanding_extents--;
1376 spin_unlock(&BTRFS_I(inode)->lock);
1377 }
0ca1f7ce
YZ		 1378
1379 if (*bits & EXTENT_DO_ACCOUNTING)
1380 btrfs_delalloc_release_metadata(inode, len);
1381
0cb59c99
JB		 1382 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1383 && do_list)
0ca1f7ce 1384 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1385
75eff68e 1386 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1387 root->fs_info->delalloc_bytes -= len;
1388 BTRFS_I(inode)->delalloc_bytes -= len;
1389
0cb59c99 1390 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
ea8c2819
CM		 1391 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1393 }
75eff68e 1394 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1395 }
291d673e
CM		 1396}
1397
d352ac68
CM		 1398/*
1399 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400 * we don't create bios that span stripes or chunks
1401 */
239b14b3 1402int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1403 size_t size, struct bio *bio,
1404 unsigned long bio_flags)
239b14b3
CM		 1405{
1406 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407 struct btrfs_mapping_tree *map_tree;
a62b9401 1408 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1409 u64 length = 0;
1410 u64 map_length;
239b14b3
CM		 1411 int ret;
1412
771ed689
CM		 1413 if (bio_flags & EXTENT_BIO_COMPRESSED)
1414 return 0;
1415
f2d8d74d 1416 length = bio->bi_size;
239b14b3
CM		 1417 map_tree = &root->fs_info->mapping_tree;
1418 map_length = length;
cea9e445 1419 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1420 &map_length, NULL, 0);
cea9e445 1421
d397712b 1422 if (map_length < length + size)
239b14b3 1423 return 1;
411fc6bc 1424 return ret;
239b14b3
CM		 1425}
1426
d352ac68
CM		 1427/*
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1431 *
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1434 */
d397712b
CM		 1435static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436 struct bio *bio, int mirror_num,
eaf25d93
CM		 1437 unsigned long bio_flags,
1438 u64 bio_offset)
065631f6 1439{
065631f6 1440 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1441 int ret = 0;
e015640f 1442
d20f7043 1443 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1444 BUG_ON(ret);
4a69a410
CM		 1445 return 0;
1446}
e015640f 1447
4a69a410
CM		 1448/*
1449 * in order to insert checksums into the metadata in large chunks,
1450 * we wait until bio submission time. All the pages in the bio are
1451 * checksummed and sums are attached onto the ordered extent record.
1452 *
1453 * At IO completion time the cums attached on the ordered extent record
1454 * are inserted into the btree
1455 */
b2950863 1456static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1457 int mirror_num, unsigned long bio_flags,
1458 u64 bio_offset)
4a69a410
CM		 1459{
1460 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1461 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1462}
1463
d352ac68 1464/*
cad321ad
CM		 1465 * extent_io.c submission hook. This does the right thing for csum calculation
1466 * on write, or reading the csums from the tree before a read
d352ac68 1467 */
b2950863 1468static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1469 int mirror_num, unsigned long bio_flags,
1470 u64 bio_offset)
44b8bd7e
CM		 1471{
1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1473 int ret = 0;
19b9bdb0 1474 int skip_sum;
44b8bd7e 1475
6cbff00f 1476 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1477
2cf8572d 1478 if (btrfs_is_free_space_inode(root, inode))
0cb59c99
JB		 1479 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1480 else
1481 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
e6dcd2dc 1482 BUG_ON(ret);
065631f6 1483
7b6d91da 1484 if (!(rw & REQ_WRITE)) {
d20f7043 1485 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1486 return btrfs_submit_compressed_read(inode, bio,
1487 mirror_num, bio_flags);
c2db1073
TI		 1488 } else if (!skip_sum) {
1489 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1490 if (ret)
1491 return ret;
1492 }
4d1b5fb4 1493 goto mapit;
19b9bdb0 1494 } else if (!skip_sum) {
17d217fe
YZ		 1495 /* csum items have already been cloned */
1496 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1497 goto mapit;
19b9bdb0
CM		 1498 /* we're doing a write, do the async checksumming */
1499 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1500 inode, rw, bio, mirror_num,
eaf25d93
CM		 1501 bio_flags, bio_offset,
1502 __btrfs_submit_bio_start,
4a69a410 1503 __btrfs_submit_bio_done);
19b9bdb0
CM		 1504 }
1505
0b86a832 1506mapit:
8b712842 1507 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1508}
6885f308 1509
d352ac68
CM		 1510/*
1511 * given a list of ordered sums record them in the inode. This happens
1512 * at IO completion time based on sums calculated at bio submission time.
1513 */
ba1da2f4 1514static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1515 struct inode *inode, u64 file_offset,
1516 struct list_head *list)
1517{
e6dcd2dc
CM		 1518 struct btrfs_ordered_sum *sum;
1519
c6e30871 1520 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1521 btrfs_csum_file_blocks(trans,
1522 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1523 }
1524 return 0;
1525}
1526
2ac55d41
JB		 1527int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528 struct extent_state **cached_state)
ea8c2819 1529{
d397712b 1530 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1531 WARN_ON(1);
ea8c2819 1532 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1533 cached_state, GFP_NOFS);
ea8c2819
CM		 1534}
1535
d352ac68 1536/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1537struct btrfs_writepage_fixup {
1538 struct page *page;
1539 struct btrfs_work work;
1540};
1541
b2950863 1542static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1543{
1544 struct btrfs_writepage_fixup *fixup;
1545 struct btrfs_ordered_extent *ordered;
2ac55d41 1546 struct extent_state *cached_state = NULL;
247e743c
CM		 1547 struct page *page;
1548 struct inode *inode;
1549 u64 page_start;
1550 u64 page_end;
1551
1552 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1553 page = fixup->page;
4a096752 1554again:
247e743c
CM		 1555 lock_page(page);
1556 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557 ClearPageChecked(page);
1558 goto out_page;
1559 }
1560
1561 inode = page->mapping->host;
1562 page_start = page_offset(page);
1563 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1564
2ac55d41
JB		 1565 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566 &cached_state, GFP_NOFS);
4a096752
CM		 1567
1568 /* already ordered? We're done */
8b62b72b 1569 if (PagePrivate2(page))
247e743c 1570 goto out;
4a096752
CM		 1571
1572 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1573 if (ordered) {
2ac55d41
JB		 1574 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1576 unlock_page(page);
1577 btrfs_start_ordered_extent(inode, ordered, 1);
1578 goto again;
1579 }
247e743c 1580
0ca1f7ce 1581 BUG();
2ac55d41 1582 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c
CM		 1583 ClearPageChecked(page);
1584out:
2ac55d41
JB		 1585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586 &cached_state, GFP_NOFS);
247e743c
CM		 1587out_page:
1588 unlock_page(page);
1589 page_cache_release(page);
b897abec 1590 kfree(fixup);
247e743c
CM		 1591}
1592
1593/*
1594 * There are a few paths in the higher layers of the kernel that directly
1595 * set the page dirty bit without asking the filesystem if it is a
1596 * good idea. This causes problems because we want to make sure COW
1597 * properly happens and the data=ordered rules are followed.
1598 *
c8b97818 1599 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1600 * hasn't been properly setup for IO. We kick off an async process
1601 * to fix it up. The async helper will wait for ordered extents, set
1602 * the delalloc bit and make it safe to write the page.
1603 */
b2950863 1604static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1605{
1606 struct inode *inode = page->mapping->host;
1607 struct btrfs_writepage_fixup *fixup;
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1609
8b62b72b
CM		 1610 /* this page is properly in the ordered list */
1611 if (TestClearPagePrivate2(page))
247e743c
CM		 1612 return 0;
1613
1614 if (PageChecked(page))
1615 return -EAGAIN;
1616
1617 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1618 if (!fixup)
1619 return -EAGAIN;
f421950f 1620
247e743c
CM		 1621 SetPageChecked(page);
1622 page_cache_get(page);
1623 fixup->work.func = btrfs_writepage_fixup_worker;
1624 fixup->page = page;
1625 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1626 return -EAGAIN;
1627}
1628
d899e052
YZ		 1629static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630 struct inode *inode, u64 file_pos,
1631 u64 disk_bytenr, u64 disk_num_bytes,
1632 u64 num_bytes, u64 ram_bytes,
1633 u8 compression, u8 encryption,
1634 u16 other_encoding, int extent_type)
1635{
1636 struct btrfs_root *root = BTRFS_I(inode)->root;
1637 struct btrfs_file_extent_item *fi;
1638 struct btrfs_path *path;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key ins;
1641 u64 hint;
1642 int ret;
1643
1644 path = btrfs_alloc_path();
d8926bb3
MF		 1645 if (!path)
1646 return -ENOMEM;
d899e052 1647
b9473439 1648 path->leave_spinning = 1;
a1ed835e
CM		 1649
1650 /*
1651 * we may be replacing one extent in the tree with another.
1652 * The new extent is pinned in the extent map, and we don't want
1653 * to drop it from the cache until it is completely in the btree.
1654 *
1655 * So, tell btrfs_drop_extents to leave this extent in the cache.
1656 * the caller is expected to unpin it and allow it to be merged
1657 * with the others.
1658 */
920bbbfb
YZ		 1659 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1660 &hint, 0);
d899e052
YZ		 1661 BUG_ON(ret);
1662
33345d01 1663 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1664 ins.offset = file_pos;
1665 ins.type = BTRFS_EXTENT_DATA_KEY;
1666 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1667 BUG_ON(ret);
1668 leaf = path->nodes[0];
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672 btrfs_set_file_extent_type(leaf, fi, extent_type);
1673 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675 btrfs_set_file_extent_offset(leaf, fi, 0);
1676 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678 btrfs_set_file_extent_compression(leaf, fi, compression);
1679 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1681
1682 btrfs_unlock_up_safe(path, 1);
1683 btrfs_set_lock_blocking(leaf);
1684
d899e052
YZ		 1685 btrfs_mark_buffer_dirty(leaf);
1686
1687 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1688
1689 ins.objectid = disk_bytenr;
1690 ins.offset = disk_num_bytes;
1691 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1692 ret = btrfs_alloc_reserved_file_extent(trans, root,
1693 root->root_key.objectid,
33345d01 1694 btrfs_ino(inode), file_pos, &ins);
d899e052 1695 BUG_ON(ret);
d899e052 1696 btrfs_free_path(path);
b9473439 1697
d899e052
YZ		 1698 return 0;
1699}
1700
5d13a98f
CM		 1701/*
1702 * helper function for btrfs_finish_ordered_io, this
1703 * just reads in some of the csum leaves to prime them into ram
1704 * before we start the transaction. It limits the amount of btree
1705 * reads required while inside the transaction.
1706 */
d352ac68
CM		 1707/* as ordered data IO finishes, this gets called so we can finish
1708 * an ordered extent if the range of bytes in the file it covers are
1709 * fully written.
1710 */
211f90e6 1711static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1712{
e6dcd2dc 1713 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1714 struct btrfs_trans_handle *trans = NULL;
5d13a98f 1715 struct btrfs_ordered_extent *ordered_extent = NULL;
e6dcd2dc 1716 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1717 struct extent_state *cached_state = NULL;
261507a0 1718 int compress_type = 0;
e6dcd2dc 1719 int ret;
82d5902d 1720 bool nolock;
e6dcd2dc 1721
5a1a3df1
JB		 1722 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1723 end - start + 1);
ba1da2f4 1724 if (!ret)
e6dcd2dc 1725 return 0;
e6dcd2dc 1726 BUG_ON(!ordered_extent);
efd049fb 1727
2cf8572d 1728 nolock = btrfs_is_free_space_inode(root, inode);
0cb59c99 1729
c2167754
YZ		 1730 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731 BUG_ON(!list_empty(&ordered_extent->list));
1732 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1733 if (!ret) {
0cb59c99 1734 if (nolock)
7a7eaa40 1735 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1736 else
7a7eaa40 1737 trans = btrfs_join_transaction(root);
3612b495 1738 BUG_ON(IS_ERR(trans));
0ca1f7ce 1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754
YZ		 1740 ret = btrfs_update_inode(trans, root, inode);
1741 BUG_ON(ret);
c2167754
YZ		 1742 }
1743 goto out;
1744 }
e6dcd2dc 1745
2ac55d41
JB		 1746 lock_extent_bits(io_tree, ordered_extent->file_offset,
1747 ordered_extent->file_offset + ordered_extent->len - 1,
1748 0, &cached_state, GFP_NOFS);
e6dcd2dc 1749
0cb59c99 1750 if (nolock)
7a7eaa40 1751 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1752 else
7a7eaa40 1753 trans = btrfs_join_transaction(root);
3612b495 1754 BUG_ON(IS_ERR(trans));
0ca1f7ce 1755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1756
c8b97818 1757 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1758 compress_type = ordered_extent->compress_type;
d899e052 1759 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1760 BUG_ON(compress_type);
920bbbfb 1761 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1762 ordered_extent->file_offset,
1763 ordered_extent->file_offset +
1764 ordered_extent->len);
1765 BUG_ON(ret);
1766 } else {
0af3d00b 1767 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1768 ret = insert_reserved_file_extent(trans, inode,
1769 ordered_extent->file_offset,
1770 ordered_extent->start,
1771 ordered_extent->disk_len,
1772 ordered_extent->len,
1773 ordered_extent->len,
261507a0 1774 compress_type, 0, 0,
d899e052 1775 BTRFS_FILE_EXTENT_REG);
a1ed835e
CM		 1776 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777 ordered_extent->file_offset,
1778 ordered_extent->len);
d899e052
YZ		 1779 BUG_ON(ret);
1780 }
2ac55d41
JB		 1781 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1784
e6dcd2dc
CM		 1785 add_pending_csums(trans, inode, ordered_extent->file_offset,
1786 &ordered_extent->list);
1787
1ef30be1 1788 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1789 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1ef30be1
JB		 1790 ret = btrfs_update_inode(trans, root, inode);
1791 BUG_ON(ret);
1792 }
1793 ret = 0;
c2167754 1794out:
0cb59c99
JB		 1795 if (nolock) {
1796 if (trans)
1797 btrfs_end_transaction_nolock(trans, root);
1798 } else {
1799 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1800 if (trans)
1801 btrfs_end_transaction(trans, root);
1802 }
1803
e6dcd2dc
CM		 1804 /* once for us */
1805 btrfs_put_ordered_extent(ordered_extent);
1806 /* once for the tree */
1807 btrfs_put_ordered_extent(ordered_extent);
1808
e6dcd2dc
CM		 1809 return 0;
1810}
1811
b2950863 1812static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1813 struct extent_state *state, int uptodate)
1814{
1abe9b8a 1815 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1816
8b62b72b 1817 ClearPagePrivate2(page);
211f90e6
CM		 1818 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1819}
1820
d352ac68
CM		 1821/*
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1828 */
7e38326f
CM		 1829struct io_failure_record {
1830 struct page *page;
1831 u64 start;
1832 u64 len;
1833 u64 logical;
d20f7043 1834 unsigned long bio_flags;
7e38326f
CM		 1835 int last_mirror;
1836};
1837
b2950863 1838static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1839 struct page *page, u64 start, u64 end,
1840 struct extent_state *state)
7e38326f
CM		 1841{
1842 struct io_failure_record *failrec = NULL;
1843 u64 private;
1844 struct extent_map *em;
1845 struct inode *inode = page->mapping->host;
1846 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1848 struct bio *bio;
1849 int num_copies;
1850 int ret;
1259ab75 1851 int rw;
7e38326f
CM		 1852 u64 logical;
1853
1854 ret = get_state_private(failure_tree, start, &private);
1855 if (ret) {
7e38326f
CM		 1856 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1857 if (!failrec)
1858 return -ENOMEM;
1859 failrec->start = start;
1860 failrec->len = end - start + 1;
1861 failrec->last_mirror = 0;
d20f7043 1862 failrec->bio_flags = 0;
7e38326f 1863
890871be 1864 read_lock(&em_tree->lock);
3b951516
CM		 1865 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866 if (em->start > start || em->start + em->len < start) {
1867 free_extent_map(em);
1868 em = NULL;
1869 }
890871be 1870 read_unlock(&em_tree->lock);
7e38326f 1871
c704005d 1872 if (IS_ERR_OR_NULL(em)) {
7e38326f
CM		 1873 kfree(failrec);
1874 return -EIO;
1875 }
1876 logical = start - em->start;
1877 logical = em->block_start + logical;
d20f7043
CM		 1878 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879 logical = em->block_start;
1880 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
261507a0
LZ		 1881 extent_set_compress_type(&failrec->bio_flags,
1882 em->compress_type);
d20f7043 1883 }
7e38326f
CM		 1884 failrec->logical = logical;
1885 free_extent_map(em);
1886 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1888 set_state_private(failure_tree, start,
1889 (u64)(unsigned long)failrec);
7e38326f 1890 } else {
587f7704 1891 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1892 }
1893 num_copies = btrfs_num_copies(
1894 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895 failrec->logical, failrec->len);
1896 failrec->last_mirror++;
1897 if (!state) {
cad321ad 1898 spin_lock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1899 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900 failrec->start,
1901 EXTENT_LOCKED);
1902 if (state && state->start != failrec->start)
1903 state = NULL;
cad321ad 1904 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1905 }
1906 if (!state || failrec->last_mirror > num_copies) {
1907 set_state_private(failure_tree, failrec->start, 0);
1908 clear_extent_bits(failure_tree, failrec->start,
1909 failrec->start + failrec->len - 1,
1910 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911 kfree(failrec);
1912 return -EIO;
1913 }
1914 bio = bio_alloc(GFP_NOFS, 1);
1915 bio->bi_private = state;
1916 bio->bi_end_io = failed_bio->bi_end_io;
1917 bio->bi_sector = failrec->logical >> 9;
1918 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1919 bio->bi_size = 0;
d20f7043 1920
7e38326f 1921 bio_add_page(bio, page, failrec->len, start - page_offset(page));
7b6d91da 1922 if (failed_bio->bi_rw & REQ_WRITE)
1259ab75
CM		 1923 rw = WRITE;
1924 else
1925 rw = READ;
1926
c2db1073 1927 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1928 failrec->last_mirror,
eaf25d93 1929 failrec->bio_flags, 0);
c2db1073 1930 return ret;
1259ab75
CM		 1931}
1932
d352ac68
CM		 1933/*
1934 * each time an IO finishes, we do a fast check in the IO failure tree
1935 * to see if we need to process or clean up an io_failure_record
1936 */
b2950863 1937static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1938{
1939 u64 private;
1940 u64 private_failure;
1941 struct io_failure_record *failure;
1942 int ret;
1943
1944 private = 0;
1945 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
ec29ed5b 1946 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1259ab75
CM		 1947 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948 start, &private_failure);
1949 if (ret == 0) {
1950 failure = (struct io_failure_record *)(unsigned long)
1951 private_failure;
1952 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953 failure->start, 0);
1954 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955 failure->start,
1956 failure->start + failure->len - 1,
1957 EXTENT_DIRTY | EXTENT_LOCKED,
1958 GFP_NOFS);
1959 kfree(failure);
1960 }
1961 }
7e38326f
CM		 1962 return 0;
1963}
1964
d352ac68
CM		 1965/*
1966 * when reads are done, we need to check csums to verify the data is correct
1967 * if there's a match, we allow the bio to finish. If not, we go through
1968 * the io_failure_record routines to find good copies
1969 */
b2950863 1970static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1971 struct extent_state *state)
07157aac 1972{
35ebb934 1973 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1974 struct inode *inode = page->mapping->host;
d1310b2e 1975 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1976 char *kaddr;
aadfeb6e 1977 u64 private = ~(u32)0;
07157aac 1978 int ret;
ff79f819
CM		 1979 struct btrfs_root *root = BTRFS_I(inode)->root;
1980 u32 csum = ~(u32)0;
d1310b2e 1981
d20f7043
CM		 1982 if (PageChecked(page)) {
1983 ClearPageChecked(page);
1984 goto good;
1985 }
6cbff00f
CH		 1986
1987 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 1988 goto good;
17d217fe
YZ		 1989
1990 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 1991 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 1992 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993 GFP_NOFS);
b6cda9bc 1994 return 0;
17d217fe 1995 }
d20f7043 1996
c2e639f0 1997 if (state && state->start == start) {
70dec807
CM		 1998 private = state->private;
1999 ret = 0;
2000 } else {
2001 ret = get_state_private(io_tree, start, &private);
2002 }
9ab86c8e 2003 kaddr = kmap_atomic(page, KM_USER0);
d397712b 2004 if (ret)
07157aac 2005 goto zeroit;
d397712b 2006
ff79f819
CM		 2007 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2008 btrfs_csum_final(csum, (char *)&csum);
d397712b 2009 if (csum != private)
07157aac 2010 goto zeroit;
d397712b 2011
9ab86c8e 2012 kunmap_atomic(kaddr, KM_USER0);
d20f7043 2013good:
7e38326f
CM		 2014 /* if the io failure tree for this inode is non-empty,
2015 * check to see if we've recovered from a failed IO
2016 */
1259ab75 2017 btrfs_clean_io_failures(inode, start);
07157aac
CM		 2018 return 0;
2019
2020zeroit:
945d8962 2021 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 2022 "private %llu\n",
2023 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 2024 (unsigned long long)start, csum,
2025 (unsigned long long)private);
db94535d
CM		 2026 memset(kaddr + offset, 1, end - start + 1);
2027 flush_dcache_page(page);
9ab86c8e 2028 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 2029 if (private == 0)
2030 return 0;
7e38326f 2031 return -EIO;
07157aac 2032}
b888db2b 2033
24bbcf04
YZ		 2034struct delayed_iput {
2035 struct list_head list;
2036 struct inode *inode;
2037};
2038
2039void btrfs_add_delayed_iput(struct inode *inode)
2040{
2041 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042 struct delayed_iput *delayed;
2043
2044 if (atomic_add_unless(&inode->i_count, -1, 1))
2045 return;
2046
2047 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048 delayed->inode = inode;
2049
2050 spin_lock(&fs_info->delayed_iput_lock);
2051 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052 spin_unlock(&fs_info->delayed_iput_lock);
2053}
2054
2055void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056{
2057 LIST_HEAD(list);
2058 struct btrfs_fs_info *fs_info = root->fs_info;
2059 struct delayed_iput *delayed;
2060 int empty;
2061
2062 spin_lock(&fs_info->delayed_iput_lock);
2063 empty = list_empty(&fs_info->delayed_iputs);
2064 spin_unlock(&fs_info->delayed_iput_lock);
2065 if (empty)
2066 return;
2067
2068 down_read(&root->fs_info->cleanup_work_sem);
2069 spin_lock(&fs_info->delayed_iput_lock);
2070 list_splice_init(&fs_info->delayed_iputs, &list);
2071 spin_unlock(&fs_info->delayed_iput_lock);
2072
2073 while (!list_empty(&list)) {
2074 delayed = list_entry(list.next, struct delayed_iput, list);
2075 list_del(&delayed->list);
2076 iput(delayed->inode);
2077 kfree(delayed);
2078 }
2079 up_read(&root->fs_info->cleanup_work_sem);
2080}
2081
d68fc57b
YZ		 2082enum btrfs_orphan_cleanup_state {
2083 ORPHAN_CLEANUP_STARTED = 1,
2084 ORPHAN_CLEANUP_DONE = 2,
2085};
2086
2087/*
2088 * This is called in transaction commmit time. If there are no orphan
2089 * files in the subvolume, it removes orphan item and frees block_rsv
2090 * structure.
2091 */
2092void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2093 struct btrfs_root *root)
2094{
2095 int ret;
2096
2097 if (!list_empty(&root->orphan_list) ||
2098 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2099 return;
2100
2101 if (root->orphan_item_inserted &&
2102 btrfs_root_refs(&root->root_item) > 0) {
2103 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2104 root->root_key.objectid);
2105 BUG_ON(ret);
2106 root->orphan_item_inserted = 0;
2107 }
2108
2109 if (root->orphan_block_rsv) {
2110 WARN_ON(root->orphan_block_rsv->size > 0);
2111 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2112 root->orphan_block_rsv = NULL;
2113 }
2114}
2115
7b128766
JB		 2116/*
2117 * This creates an orphan entry for the given inode in case something goes
2118 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2119 *
2120 * NOTE: caller of this function should reserve 5 units of metadata for
2121 * this function.
7b128766
JB		 2122 */
2123int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2124{
2125 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2126 struct btrfs_block_rsv *block_rsv = NULL;
2127 int reserve = 0;
2128 int insert = 0;
2129 int ret;
7b128766 2130
d68fc57b
YZ		 2131 if (!root->orphan_block_rsv) {
2132 block_rsv = btrfs_alloc_block_rsv(root);
b532402e
TI		 2133 if (!block_rsv)
2134 return -ENOMEM;
d68fc57b 2135 }
7b128766 2136
d68fc57b
YZ		 2137 spin_lock(&root->orphan_lock);
2138 if (!root->orphan_block_rsv) {
2139 root->orphan_block_rsv = block_rsv;
2140 } else if (block_rsv) {
2141 btrfs_free_block_rsv(root, block_rsv);
2142 block_rsv = NULL;
7b128766 2143 }
7b128766 2144
d68fc57b
YZ		 2145 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2146 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2147#if 0
2148 /*
2149 * For proper ENOSPC handling, we should do orphan
2150 * cleanup when mounting. But this introduces backward
2151 * compatibility issue.
2152 */
2153 if (!xchg(&root->orphan_item_inserted, 1))
2154 insert = 2;
2155 else
2156 insert = 1;
2157#endif
2158 insert = 1;
7b128766
JB		 2159 }
2160
d68fc57b
YZ		 2161 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2162 BTRFS_I(inode)->orphan_meta_reserved = 1;
2163 reserve = 1;
2164 }
2165 spin_unlock(&root->orphan_lock);
7b128766 2166
d68fc57b
YZ		 2167 /* grab metadata reservation from transaction handle */
2168 if (reserve) {
2169 ret = btrfs_orphan_reserve_metadata(trans, inode);
2170 BUG_ON(ret);
2171 }
7b128766 2172
d68fc57b
YZ		 2173 /* insert an orphan item to track this unlinked/truncated file */
2174 if (insert >= 1) {
33345d01 2175 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2176 BUG_ON(ret);
2177 }
2178
2179 /* insert an orphan item to track subvolume contains orphan files */
2180 if (insert >= 2) {
2181 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2182 root->root_key.objectid);
2183 BUG_ON(ret);
2184 }
2185 return 0;
7b128766
JB		 2186}
2187
2188/*
2189 * We have done the truncate/delete so we can go ahead and remove the orphan
2190 * item for this particular inode.
2191 */
2192int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2193{
2194 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2195 int delete_item = 0;
2196 int release_rsv = 0;
7b128766
JB		 2197 int ret = 0;
2198
d68fc57b
YZ		 2199 spin_lock(&root->orphan_lock);
2200 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2201 list_del_init(&BTRFS_I(inode)->i_orphan);
2202 delete_item = 1;
7b128766
JB		 2203 }
2204
d68fc57b
YZ		 2205 if (BTRFS_I(inode)->orphan_meta_reserved) {
2206 BTRFS_I(inode)->orphan_meta_reserved = 0;
2207 release_rsv = 1;
7b128766 2208 }
d68fc57b 2209 spin_unlock(&root->orphan_lock);
7b128766 2210
d68fc57b 2211 if (trans && delete_item) {
33345d01 2212 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2213 BUG_ON(ret);
2214 }
7b128766 2215
d68fc57b
YZ		 2216 if (release_rsv)
2217 btrfs_orphan_release_metadata(inode);
7b128766 2218
d68fc57b 2219 return 0;
7b128766
JB		 2220}
2221
2222/*
2223 * this cleans up any orphans that may be left on the list from the last use
2224 * of this root.
2225 */
66b4ffd1 2226int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2227{
2228 struct btrfs_path *path;
2229 struct extent_buffer *leaf;
7b128766
JB		 2230 struct btrfs_key key, found_key;
2231 struct btrfs_trans_handle *trans;
2232 struct inode *inode;
2233 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2234
d68fc57b 2235 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2236 return 0;
c71bf099
YZ		 2237
2238 path = btrfs_alloc_path();
66b4ffd1
JB		 2239 if (!path) {
2240 ret = -ENOMEM;
2241 goto out;
2242 }
7b128766
JB		 2243 path->reada = -1;
2244
2245 key.objectid = BTRFS_ORPHAN_OBJECTID;
2246 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2247 key.offset = (u64)-1;
2248
7b128766
JB		 2249 while (1) {
2250 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2251 if (ret < 0)
2252 goto out;
7b128766
JB		 2253
2254 /*
2255 * if ret == 0 means we found what we were searching for, which
25985edc 2256 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2257 * find the key and see if we have stuff that matches
2258 */
2259 if (ret > 0) {
66b4ffd1 2260 ret = 0;
7b128766
JB		 2261 if (path->slots[0] == 0)
2262 break;
2263 path->slots[0]--;
2264 }
2265
2266 /* pull out the item */
2267 leaf = path->nodes[0];
7b128766
JB		 2268 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2269
2270 /* make sure the item matches what we want */
2271 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2272 break;
2273 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2274 break;
2275
2276 /* release the path since we're done with it */
b3b4aa74 2277 btrfs_release_path(path);
7b128766
JB		 2278
2279 /*
2280 * this is where we are basically btrfs_lookup, without the
2281 * crossing root thing. we store the inode number in the
2282 * offset of the orphan item.
2283 */
5d4f98a2
YZ		 2284 found_key.objectid = found_key.offset;
2285 found_key.type = BTRFS_INODE_ITEM_KEY;
2286 found_key.offset = 0;
73f73415 2287 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
66b4ffd1
JB		 2288 if (IS_ERR(inode)) {
2289 ret = PTR_ERR(inode);
2290 goto out;
2291 }
7b128766 2292
7b128766
JB		 2293 /*
2294 * add this inode to the orphan list so btrfs_orphan_del does
2295 * the proper thing when we hit it
2296 */
d68fc57b 2297 spin_lock(&root->orphan_lock);
7b128766 2298 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
d68fc57b 2299 spin_unlock(&root->orphan_lock);
7b128766
JB		 2300
2301 /*
2302 * if this is a bad inode, means we actually succeeded in
2303 * removing the inode, but not the orphan record, which means
2304 * we need to manually delete the orphan since iput will just
2305 * do a destroy_inode
2306 */
2307 if (is_bad_inode(inode)) {
a22285a6 2308 trans = btrfs_start_transaction(root, 0);
66b4ffd1
JB		 2309 if (IS_ERR(trans)) {
2310 ret = PTR_ERR(trans);
2311 goto out;
2312 }
7b128766 2313 btrfs_orphan_del(trans, inode);
5b21f2ed 2314 btrfs_end_transaction(trans, root);
7b128766
JB		 2315 iput(inode);
2316 continue;
2317 }
2318
2319 /* if we have links, this was a truncate, lets do that */
2320 if (inode->i_nlink) {
a41ad394
JB		 2321 if (!S_ISREG(inode->i_mode)) {
2322 WARN_ON(1);
2323 iput(inode);
2324 continue;
2325 }
7b128766 2326 nr_truncate++;
66b4ffd1 2327 ret = btrfs_truncate(inode);
7b128766
JB		 2328 } else {
2329 nr_unlink++;
2330 }
2331
2332 /* this will do delete_inode and everything for us */
2333 iput(inode);
66b4ffd1
JB		 2334 if (ret)
2335 goto out;
7b128766 2336 }
d68fc57b
YZ		 2337 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2338
2339 if (root->orphan_block_rsv)
2340 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2341 (u64)-1);
2342
2343 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2344 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2345 if (!IS_ERR(trans))
2346 btrfs_end_transaction(trans, root);
d68fc57b 2347 }
7b128766
JB		 2348
2349 if (nr_unlink)
2350 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2351 if (nr_truncate)
2352 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2353
2354out:
2355 if (ret)
2356 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2357 btrfs_free_path(path);
2358 return ret;
7b128766
JB		 2359}
2360
46a53cca
CM		 2361/*
2362 * very simple check to peek ahead in the leaf looking for xattrs. If we
2363 * don't find any xattrs, we know there can't be any acls.
2364 *
2365 * slot is the slot the inode is in, objectid is the objectid of the inode
2366 */
2367static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2368 int slot, u64 objectid)
2369{
2370 u32 nritems = btrfs_header_nritems(leaf);
2371 struct btrfs_key found_key;
2372 int scanned = 0;
2373
2374 slot++;
2375 while (slot < nritems) {
2376 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2377
2378 /* we found a different objectid, there must not be acls */
2379 if (found_key.objectid != objectid)
2380 return 0;
2381
2382 /* we found an xattr, assume we've got an acl */
2383 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2384 return 1;
2385
2386 /*
2387 * we found a key greater than an xattr key, there can't
2388 * be any acls later on
2389 */
2390 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2391 return 0;
2392
2393 slot++;
2394 scanned++;
2395
2396 /*
2397 * it goes inode, inode backrefs, xattrs, extents,
2398 * so if there are a ton of hard links to an inode there can
2399 * be a lot of backrefs. Don't waste time searching too hard,
2400 * this is just an optimization
2401 */
2402 if (scanned >= 8)
2403 break;
2404 }
2405 /* we hit the end of the leaf before we found an xattr or
2406 * something larger than an xattr. We have to assume the inode
2407 * has acls
2408 */
2409 return 1;
2410}
2411
d352ac68
CM		 2412/*
2413 * read an inode from the btree into the in-memory inode
2414 */
5d4f98a2 2415static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2416{
2417 struct btrfs_path *path;
5f39d397 2418 struct extent_buffer *leaf;
39279cc3 2419 struct btrfs_inode_item *inode_item;
0b86a832 2420 struct btrfs_timespec *tspec;
39279cc3
CM		 2421 struct btrfs_root *root = BTRFS_I(inode)->root;
2422 struct btrfs_key location;
46a53cca 2423 int maybe_acls;
618e21d5 2424 u32 rdev;
39279cc3 2425 int ret;
2f7e33d4
MX		 2426 bool filled = false;
2427
2428 ret = btrfs_fill_inode(inode, &rdev);
2429 if (!ret)
2430 filled = true;
39279cc3
CM		 2431
2432 path = btrfs_alloc_path();
1748f843
MF		 2433 if (!path)
2434 goto make_bad;
2435
d90c7321 2436 path->leave_spinning = 1;
39279cc3 2437 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2438
39279cc3 2439 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2440 if (ret)
39279cc3 2441 goto make_bad;
39279cc3 2442
5f39d397 2443 leaf = path->nodes[0];
2f7e33d4
MX		 2444
2445 if (filled)
2446 goto cache_acl;
2447
5f39d397
CM		 2448 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2449 struct btrfs_inode_item);
5f39d397
CM		 2450 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2451 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2452 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2453 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2454 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2455
2456 tspec = btrfs_inode_atime(inode_item);
2457 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2458 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2459
2460 tspec = btrfs_inode_mtime(inode_item);
2461 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2462 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2463
2464 tspec = btrfs_inode_ctime(inode_item);
2465 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2466 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2467
a76a3cd4 2468 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2469 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2470 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2471 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2472 inode->i_rdev = 0;
5f39d397
CM		 2473 rdev = btrfs_inode_rdev(leaf, inode_item);
2474
aec7477b 2475 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2476 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2477cache_acl:
46a53cca
CM		 2478 /*
2479 * try to precache a NULL acl entry for files that don't have
2480 * any xattrs or acls
2481 */
33345d01
LZ		 2482 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2483 btrfs_ino(inode));
72c04902
AV		 2484 if (!maybe_acls)
2485 cache_no_acl(inode);
46a53cca 2486
39279cc3 2487 btrfs_free_path(path);
39279cc3 2488
39279cc3 2489 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2490 case S_IFREG:
2491 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2492 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2493 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2494 inode->i_fop = &btrfs_file_operations;
2495 inode->i_op = &btrfs_file_inode_operations;
2496 break;
2497 case S_IFDIR:
2498 inode->i_fop = &btrfs_dir_file_operations;
2499 if (root == root->fs_info->tree_root)
2500 inode->i_op = &btrfs_dir_ro_inode_operations;
2501 else
2502 inode->i_op = &btrfs_dir_inode_operations;
2503 break;
2504 case S_IFLNK:
2505 inode->i_op = &btrfs_symlink_inode_operations;
2506 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2507 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2508 break;
618e21d5 2509 default:
0279b4cd 2510 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2511 init_special_inode(inode, inode->i_mode, rdev);
2512 break;
39279cc3 2513 }
6cbff00f
CH		 2514
2515 btrfs_update_iflags(inode);
39279cc3
CM		 2516 return;
2517
2518make_bad:
39279cc3 2519 btrfs_free_path(path);
39279cc3
CM		 2520 make_bad_inode(inode);
2521}
2522
d352ac68
CM		 2523/*
2524 * given a leaf and an inode, copy the inode fields into the leaf
2525 */
e02119d5
CM		 2526static void fill_inode_item(struct btrfs_trans_handle *trans,
2527 struct extent_buffer *leaf,
5f39d397 2528 struct btrfs_inode_item *item,
39279cc3
CM		 2529 struct inode *inode)
2530{
5f39d397
CM		 2531 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2532 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2533 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2534 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2535 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2536
2537 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2538 inode->i_atime.tv_sec);
2539 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2540 inode->i_atime.tv_nsec);
2541
2542 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2543 inode->i_mtime.tv_sec);
2544 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2545 inode->i_mtime.tv_nsec);
2546
2547 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2548 inode->i_ctime.tv_sec);
2549 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2550 inode->i_ctime.tv_nsec);
2551
a76a3cd4 2552 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2553 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2554 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2555 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2556 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2557 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2558 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2559}
2560
d352ac68
CM		 2561/*
2562 * copy everything in the in-memory inode into the btree.
2563 */
d397712b
CM		 2564noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2565 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2566{
2567 struct btrfs_inode_item *inode_item;
2568 struct btrfs_path *path;
5f39d397 2569 struct extent_buffer *leaf;
39279cc3
CM		 2570 int ret;
2571
16cdcec7 2572 /*
149e2d76
MX		 2573 * If the inode is a free space inode, we can deadlock during commit
2574 * if we put it into the delayed code.
2575 *
2576 * The data relocation inode should also be directly updated
2577 * without delay
16cdcec7 2578 */
2cf8572d 2579 if (!btrfs_is_free_space_inode(root, inode)
149e2d76 2580 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
16cdcec7
MX		 2581 ret = btrfs_delayed_update_inode(trans, root, inode);
2582 if (!ret)
2583 btrfs_set_inode_last_trans(trans, inode);
2584 return ret;
2585 }
2586
39279cc3 2587 path = btrfs_alloc_path();
16cdcec7
MX		 2588 if (!path)
2589 return -ENOMEM;
2590
b9473439 2591 path->leave_spinning = 1;
16cdcec7
MX		 2592 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2593 1);
39279cc3
CM		 2594 if (ret) {
2595 if (ret > 0)
2596 ret = -ENOENT;
2597 goto failed;
2598 }
2599
b4ce94de 2600 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2601 leaf = path->nodes[0];
2602 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2603 struct btrfs_inode_item);
39279cc3 2604
e02119d5 2605 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2606 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2607 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2608 ret = 0;
2609failed:
39279cc3
CM		 2610 btrfs_free_path(path);
2611 return ret;
2612}
2613
d352ac68
CM		 2614/*
2615 * unlink helper that gets used here in inode.c and in the tree logging
2616 * recovery code. It remove a link in a directory with a given name, and
2617 * also drops the back refs in the inode to the directory
2618 */
92986796
AV		 2619static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2620 struct btrfs_root *root,
2621 struct inode *dir, struct inode *inode,
2622 const char *name, int name_len)
39279cc3
CM		 2623{
2624 struct btrfs_path *path;
39279cc3 2625 int ret = 0;
5f39d397 2626 struct extent_buffer *leaf;
39279cc3 2627 struct btrfs_dir_item *di;
5f39d397 2628 struct btrfs_key key;
aec7477b 2629 u64 index;
33345d01
LZ		 2630 u64 ino = btrfs_ino(inode);
2631 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2632
2633 path = btrfs_alloc_path();
54aa1f4d
CM		 2634 if (!path) {
2635 ret = -ENOMEM;
554233a6 2636 goto out;
54aa1f4d
CM		 2637 }
2638
b9473439 2639 path->leave_spinning = 1;
33345d01 2640 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2641 name, name_len, -1);
2642 if (IS_ERR(di)) {
2643 ret = PTR_ERR(di);
2644 goto err;
2645 }
2646 if (!di) {
2647 ret = -ENOENT;
2648 goto err;
2649 }
5f39d397
CM		 2650 leaf = path->nodes[0];
2651 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2652 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2653 if (ret)
2654 goto err;
b3b4aa74 2655 btrfs_release_path(path);
39279cc3 2656
33345d01
LZ		 2657 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2658 dir_ino, &index);
aec7477b 2659 if (ret) {
d397712b 2660 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2661 "inode %llu parent %llu\n", name_len, name,
2662 (unsigned long long)ino, (unsigned long long)dir_ino);
aec7477b
JB		 2663 goto err;
2664 }
2665
16cdcec7
MX		 2666 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2667 if (ret)
39279cc3 2668 goto err;
39279cc3 2669
e02119d5 2670 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2671 inode, dir_ino);
49eb7e46 2672 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2673
2674 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2675 dir, index);
6418c961
CM		 2676 if (ret == -ENOENT)
2677 ret = 0;
39279cc3
CM		 2678err:
2679 btrfs_free_path(path);
e02119d5
CM		 2680 if (ret)
2681 goto out;
2682
2683 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2684 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2685 btrfs_update_inode(trans, root, dir);
e02119d5 2686out:
39279cc3
CM		 2687 return ret;
2688}
2689
92986796
AV		 2690int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root,
2692 struct inode *dir, struct inode *inode,
2693 const char *name, int name_len)
2694{
2695 int ret;
2696 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2697 if (!ret) {
2698 btrfs_drop_nlink(inode);
2699 ret = btrfs_update_inode(trans, root, inode);
2700 }
2701 return ret;
2702}
2703
2704
a22285a6
YZ		 2705/* helper to check if there is any shared block in the path */
2706static int check_path_shared(struct btrfs_root *root,
2707 struct btrfs_path *path)
39279cc3 2708{
a22285a6
YZ		 2709 struct extent_buffer *eb;
2710 int level;
0e4dcbef 2711 u64 refs = 1;
5df6a9f6 2712
a22285a6 2713 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2714 int ret;
2715
a22285a6
YZ		 2716 if (!path->nodes[level])
2717 break;
2718 eb = path->nodes[level];
2719 if (!btrfs_block_can_be_shared(root, eb))
2720 continue;
2721 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2722 &refs, NULL);
2723 if (refs > 1)
2724 return 1;
5df6a9f6 2725 }
dedefd72 2726 return 0;
39279cc3
CM		 2727}
2728
a22285a6
YZ		 2729/*
2730 * helper to start transaction for unlink and rmdir.
2731 *
2732 * unlink and rmdir are special in btrfs, they do not always free space.
2733 * so in enospc case, we should make sure they will free space before
2734 * allowing them to use the global metadata reservation.
2735 */
2736static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2737 struct dentry *dentry)
4df27c4d 2738{
39279cc3 2739 struct btrfs_trans_handle *trans;
a22285a6 2740 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2741 struct btrfs_path *path;
a22285a6 2742 struct btrfs_inode_ref *ref;
4df27c4d 2743 struct btrfs_dir_item *di;
7b128766 2744 struct inode *inode = dentry->d_inode;
4df27c4d 2745 u64 index;
a22285a6
YZ		 2746 int check_link = 1;
2747 int err = -ENOSPC;
4df27c4d 2748 int ret;
33345d01
LZ		 2749 u64 ino = btrfs_ino(inode);
2750 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2751
a22285a6
YZ		 2752 trans = btrfs_start_transaction(root, 10);
2753 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2754 return trans;
4df27c4d 2755
33345d01 2756 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2757 return ERR_PTR(-ENOSPC);
4df27c4d 2758
a22285a6
YZ		 2759 /* check if there is someone else holds reference */
2760 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2761 return ERR_PTR(-ENOSPC);
4df27c4d 2762
a22285a6
YZ		 2763 if (atomic_read(&inode->i_count) > 2)
2764 return ERR_PTR(-ENOSPC);
4df27c4d 2765
a22285a6
YZ		 2766 if (xchg(&root->fs_info->enospc_unlink, 1))
2767 return ERR_PTR(-ENOSPC);
2768
2769 path = btrfs_alloc_path();
2770 if (!path) {
2771 root->fs_info->enospc_unlink = 0;
2772 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2773 }
2774
a22285a6 2775 trans = btrfs_start_transaction(root, 0);
5df6a9f6 2776 if (IS_ERR(trans)) {
a22285a6
YZ		 2777 btrfs_free_path(path);
2778 root->fs_info->enospc_unlink = 0;
2779 return trans;
2780 }
4df27c4d 2781
a22285a6
YZ		 2782 path->skip_locking = 1;
2783 path->search_commit_root = 1;
4df27c4d 2784
a22285a6
YZ		 2785 ret = btrfs_lookup_inode(trans, root, path,
2786 &BTRFS_I(dir)->location, 0);
2787 if (ret < 0) {
2788 err = ret;
2789 goto out;
2790 }
2791 if (ret == 0) {
2792 if (check_path_shared(root, path))
2793 goto out;
2794 } else {
2795 check_link = 0;
5df6a9f6 2796 }
b3b4aa74 2797 btrfs_release_path(path);
a22285a6
YZ		 2798
2799 ret = btrfs_lookup_inode(trans, root, path,
2800 &BTRFS_I(inode)->location, 0);
2801 if (ret < 0) {
2802 err = ret;
2803 goto out;
2804 }
2805 if (ret == 0) {
2806 if (check_path_shared(root, path))
2807 goto out;
2808 } else {
2809 check_link = 0;
2810 }
b3b4aa74 2811 btrfs_release_path(path);
a22285a6
YZ		 2812
2813 if (ret == 0 && S_ISREG(inode->i_mode)) {
2814 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2815 ino, (u64)-1, 0);
a22285a6
YZ		 2816 if (ret < 0) {
2817 err = ret;
2818 goto out;
2819 }
2820 BUG_ON(ret == 0);
2821 if (check_path_shared(root, path))
2822 goto out;
b3b4aa74 2823 btrfs_release_path(path);
a22285a6
YZ		 2824 }
2825
2826 if (!check_link) {
2827 err = 0;
2828 goto out;
2829 }
2830
33345d01 2831 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2832 dentry->d_name.name, dentry->d_name.len, 0);
2833 if (IS_ERR(di)) {
2834 err = PTR_ERR(di);
2835 goto out;
2836 }
2837 if (di) {
2838 if (check_path_shared(root, path))
2839 goto out;
2840 } else {
2841 err = 0;
2842 goto out;
2843 }
b3b4aa74 2844 btrfs_release_path(path);
a22285a6
YZ		 2845
2846 ref = btrfs_lookup_inode_ref(trans, root, path,
2847 dentry->d_name.name, dentry->d_name.len,
33345d01 2848 ino, dir_ino, 0);
a22285a6
YZ		 2849 if (IS_ERR(ref)) {
2850 err = PTR_ERR(ref);
2851 goto out;
2852 }
2853 BUG_ON(!ref);
2854 if (check_path_shared(root, path))
2855 goto out;
2856 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 2857 btrfs_release_path(path);
a22285a6 2858
16cdcec7
MX		 2859 /*
2860 * This is a commit root search, if we can lookup inode item and other
2861 * relative items in the commit root, it means the transaction of
2862 * dir/file creation has been committed, and the dir index item that we
2863 * delay to insert has also been inserted into the commit root. So
2864 * we needn't worry about the delayed insertion of the dir index item
2865 * here.
2866 */
33345d01 2867 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 2868 dentry->d_name.name, dentry->d_name.len, 0);
2869 if (IS_ERR(di)) {
2870 err = PTR_ERR(di);
2871 goto out;
2872 }
2873 BUG_ON(ret == -ENOENT);
2874 if (check_path_shared(root, path))
2875 goto out;
2876
2877 err = 0;
2878out:
2879 btrfs_free_path(path);
2880 if (err) {
2881 btrfs_end_transaction(trans, root);
2882 root->fs_info->enospc_unlink = 0;
2883 return ERR_PTR(err);
2884 }
2885
2886 trans->block_rsv = &root->fs_info->global_block_rsv;
2887 return trans;
2888}
2889
2890static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2891 struct btrfs_root *root)
2892{
2893 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2894 BUG_ON(!root->fs_info->enospc_unlink);
2895 root->fs_info->enospc_unlink = 0;
2896 }
2897 btrfs_end_transaction_throttle(trans, root);
2898}
2899
2900static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2901{
2902 struct btrfs_root *root = BTRFS_I(dir)->root;
2903 struct btrfs_trans_handle *trans;
2904 struct inode *inode = dentry->d_inode;
2905 int ret;
2906 unsigned long nr = 0;
2907
2908 trans = __unlink_start_trans(dir, dentry);
2909 if (IS_ERR(trans))
2910 return PTR_ERR(trans);
5f39d397 2911
12fcfd22
CM		 2912 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2913
e02119d5
CM		 2914 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2915 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 2916 if (ret)
2917 goto out;
7b128766 2918
a22285a6 2919 if (inode->i_nlink == 0) {
7b128766 2920 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 2921 if (ret)
2922 goto out;
a22285a6 2923 }
7b128766 2924
b532402e 2925out:
d3c2fdcf 2926 nr = trans->blocks_used;
a22285a6 2927 __unlink_end_trans(trans, root);
d3c2fdcf 2928 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2929 return ret;
2930}
2931
4df27c4d
YZ		 2932int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root,
2934 struct inode *dir, u64 objectid,
2935 const char *name, int name_len)
2936{
2937 struct btrfs_path *path;
2938 struct extent_buffer *leaf;
2939 struct btrfs_dir_item *di;
2940 struct btrfs_key key;
2941 u64 index;
2942 int ret;
33345d01 2943 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 2944
2945 path = btrfs_alloc_path();
2946 if (!path)
2947 return -ENOMEM;
2948
33345d01 2949 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 2950 name, name_len, -1);
c704005d 2951 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2952
2953 leaf = path->nodes[0];
2954 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2955 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2956 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2957 BUG_ON(ret);
b3b4aa74 2958 btrfs_release_path(path);
4df27c4d
YZ		 2959
2960 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2961 objectid, root->root_key.objectid,
33345d01 2962 dir_ino, &index, name, name_len);
4df27c4d
YZ		 2963 if (ret < 0) {
2964 BUG_ON(ret != -ENOENT);
33345d01 2965 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 2966 name, name_len);
c704005d 2967 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2968
2969 leaf = path->nodes[0];
2970 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2971 btrfs_release_path(path);
4df27c4d
YZ		 2972 index = key.offset;
2973 }
945d8962 2974 btrfs_release_path(path);
4df27c4d 2975
16cdcec7 2976 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4df27c4d 2977 BUG_ON(ret);
4df27c4d
YZ		 2978
2979 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2980 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2981 ret = btrfs_update_inode(trans, root, dir);
2982 BUG_ON(ret);
4df27c4d 2983
71d7aed0 2984 btrfs_free_path(path);
4df27c4d
YZ		 2985 return 0;
2986}
2987
39279cc3
CM		 2988static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2989{
2990 struct inode *inode = dentry->d_inode;
1832a6d5 2991 int err = 0;
39279cc3 2992 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2993 struct btrfs_trans_handle *trans;
1832a6d5 2994 unsigned long nr = 0;
39279cc3 2995
3394e160 2996 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 2997 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 2998 return -ENOTEMPTY;
2999
a22285a6
YZ		 3000 trans = __unlink_start_trans(dir, dentry);
3001 if (IS_ERR(trans))
5df6a9f6 3002 return PTR_ERR(trans);
5df6a9f6 3003
33345d01 3004 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3005 err = btrfs_unlink_subvol(trans, root, dir,
3006 BTRFS_I(inode)->location.objectid,
3007 dentry->d_name.name,
3008 dentry->d_name.len);
3009 goto out;
3010 }
3011
7b128766
JB		 3012 err = btrfs_orphan_add(trans, inode);
3013 if (err)
4df27c4d 3014 goto out;
7b128766 3015
39279cc3 3016 /* now the directory is empty */
e02119d5
CM		 3017 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3018 dentry->d_name.name, dentry->d_name.len);
d397712b 3019 if (!err)
dbe674a9 3020 btrfs_i_size_write(inode, 0);
4df27c4d 3021out:
d3c2fdcf 3022 nr = trans->blocks_used;
a22285a6 3023 __unlink_end_trans(trans, root);
d3c2fdcf 3024 btrfs_btree_balance_dirty(root, nr);
3954401f 3025
39279cc3
CM		 3026 return err;
3027}
3028
39279cc3
CM		 3029/*
3030 * this can truncate away extent items, csum items and directory items.
3031 * It starts at a high offset and removes keys until it can't find
d352ac68 3032 * any higher than new_size
39279cc3
CM		 3033 *
3034 * csum items that cross the new i_size are truncated to the new size
3035 * as well.
7b128766
JB		 3036 *
3037 * min_type is the minimum key type to truncate down to. If set to 0, this
3038 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3039 */
8082510e
YZ		 3040int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3041 struct btrfs_root *root,
3042 struct inode *inode,
3043 u64 new_size, u32 min_type)
39279cc3 3044{
39279cc3 3045 struct btrfs_path *path;
5f39d397 3046 struct extent_buffer *leaf;
39279cc3 3047 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3048 struct btrfs_key key;
3049 struct btrfs_key found_key;
39279cc3 3050 u64 extent_start = 0;
db94535d 3051 u64 extent_num_bytes = 0;
5d4f98a2 3052 u64 extent_offset = 0;
39279cc3 3053 u64 item_end = 0;
8082510e
YZ		 3054 u64 mask = root->sectorsize - 1;
3055 u32 found_type = (u8)-1;
39279cc3
CM		 3056 int found_extent;
3057 int del_item;
85e21bac
CM		 3058 int pending_del_nr = 0;
3059 int pending_del_slot = 0;
179e29e4 3060 int extent_type = -1;
771ed689 3061 int encoding;
8082510e
YZ		 3062 int ret;
3063 int err = 0;
33345d01 3064 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3065
3066 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3067
0eb0e19c
MF		 3068 path = btrfs_alloc_path();
3069 if (!path)
3070 return -ENOMEM;
3071 path->reada = -1;
3072
0af3d00b 3073 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 3074 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 3075
16cdcec7
MX		 3076 /*
3077 * This function is also used to drop the items in the log tree before
3078 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3079 * it is used to drop the loged items. So we shouldn't kill the delayed
3080 * items.
3081 */
3082 if (min_type == 0 && root == BTRFS_I(inode)->root)
3083 btrfs_kill_delayed_inode_items(inode);
3084
33345d01 3085 key.objectid = ino;
39279cc3 3086 key.offset = (u64)-1;
5f39d397
CM		 3087 key.type = (u8)-1;
3088
85e21bac 3089search_again:
b9473439 3090 path->leave_spinning = 1;
85e21bac 3091 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3092 if (ret < 0) {
3093 err = ret;
3094 goto out;
3095 }
d397712b 3096
85e21bac 3097 if (ret > 0) {
e02119d5
CM		 3098 /* there are no items in the tree for us to truncate, we're
3099 * done
3100 */
8082510e
YZ		 3101 if (path->slots[0] == 0)
3102 goto out;
85e21bac
CM		 3103 path->slots[0]--;
3104 }
3105
d397712b 3106 while (1) {
39279cc3 3107 fi = NULL;
5f39d397
CM		 3108 leaf = path->nodes[0];
3109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3110 found_type = btrfs_key_type(&found_key);
771ed689 3111 encoding = 0;
39279cc3 3112
33345d01 3113 if (found_key.objectid != ino)
39279cc3 3114 break;
5f39d397 3115
85e21bac 3116 if (found_type < min_type)
39279cc3
CM		 3117 break;
3118
5f39d397 3119 item_end = found_key.offset;
39279cc3 3120 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3121 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3122 struct btrfs_file_extent_item);
179e29e4 3123 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 3124 encoding = btrfs_file_extent_compression(leaf, fi);
3125 encoding |= btrfs_file_extent_encryption(leaf, fi);
3126 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3127
179e29e4 3128 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3129 item_end +=
db94535d 3130 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3131 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3132 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3133 fi);
39279cc3 3134 }
008630c1 3135 item_end--;
39279cc3 3136 }
8082510e
YZ		 3137 if (found_type > min_type) {
3138 del_item = 1;
3139 } else {
3140 if (item_end < new_size)
b888db2b 3141 break;
8082510e
YZ		 3142 if (found_key.offset >= new_size)
3143 del_item = 1;
3144 else
3145 del_item = 0;
39279cc3 3146 }
39279cc3 3147 found_extent = 0;
39279cc3 3148 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3149 if (found_type != BTRFS_EXTENT_DATA_KEY)
3150 goto delete;
3151
3152 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3153 u64 num_dec;
db94535d 3154 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 3155 if (!del_item && !encoding) {
db94535d
CM		 3156 u64 orig_num_bytes =
3157 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3158 extent_num_bytes = new_size -
5f39d397 3159 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3160 extent_num_bytes = extent_num_bytes &
3161 ~((u64)root->sectorsize - 1);
db94535d
CM		 3162 btrfs_set_file_extent_num_bytes(leaf, fi,
3163 extent_num_bytes);
3164 num_dec = (orig_num_bytes -
9069218d 3165 extent_num_bytes);
e02119d5 3166 if (root->ref_cows && extent_start != 0)
a76a3cd4 3167 inode_sub_bytes(inode, num_dec);
5f39d397 3168 btrfs_mark_buffer_dirty(leaf);
39279cc3 3169 } else {
db94535d
CM		 3170 extent_num_bytes =
3171 btrfs_file_extent_disk_num_bytes(leaf,
3172 fi);
5d4f98a2
YZ		 3173 extent_offset = found_key.offset -
3174 btrfs_file_extent_offset(leaf, fi);
3175
39279cc3 3176 /* FIXME blocksize != 4096 */
9069218d 3177 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3178 if (extent_start != 0) {
3179 found_extent = 1;
e02119d5 3180 if (root->ref_cows)
a76a3cd4 3181 inode_sub_bytes(inode, num_dec);
e02119d5 3182 }
39279cc3 3183 }
9069218d 3184 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3185 /*
3186 * we can't truncate inline items that have had
3187 * special encodings
3188 */
3189 if (!del_item &&
3190 btrfs_file_extent_compression(leaf, fi) == 0 &&
3191 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3192 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3193 u32 size = new_size - found_key.offset;
3194
3195 if (root->ref_cows) {
a76a3cd4
YZ		 3196 inode_sub_bytes(inode, item_end + 1 -
3197 new_size);
e02119d5
CM		 3198 }
3199 size =
3200 btrfs_file_extent_calc_inline_size(size);
9069218d 3201 ret = btrfs_truncate_item(trans, root, path,
e02119d5 3202 size, 1);
e02119d5 3203 } else if (root->ref_cows) {
a76a3cd4
YZ		 3204 inode_sub_bytes(inode, item_end + 1 -
3205 found_key.offset);
9069218d 3206 }
39279cc3 3207 }
179e29e4 3208delete:
39279cc3 3209 if (del_item) {
85e21bac
CM		 3210 if (!pending_del_nr) {
3211 /* no pending yet, add ourselves */
3212 pending_del_slot = path->slots[0];
3213 pending_del_nr = 1;
3214 } else if (pending_del_nr &&
3215 path->slots[0] + 1 == pending_del_slot) {
3216 /* hop on the pending chunk */
3217 pending_del_nr++;
3218 pending_del_slot = path->slots[0];
3219 } else {
d397712b 3220 BUG();
85e21bac 3221 }
39279cc3
CM		 3222 } else {
3223 break;
3224 }
0af3d00b
JB		 3225 if (found_extent && (root->ref_cows ||
3226 root == root->fs_info->tree_root)) {
b9473439 3227 btrfs_set_path_blocking(path);
39279cc3 3228 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3229 extent_num_bytes, 0,
3230 btrfs_header_owner(leaf),
33345d01 3231 ino, extent_offset);
39279cc3
CM		 3232 BUG_ON(ret);
3233 }
85e21bac 3234
8082510e
YZ		 3235 if (found_type == BTRFS_INODE_ITEM_KEY)
3236 break;
3237
3238 if (path->slots[0] == 0 ||
3239 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3240 if (root->ref_cows &&
3241 BTRFS_I(inode)->location.objectid !=
3242 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3243 err = -EAGAIN;
3244 goto out;
3245 }
3246 if (pending_del_nr) {
3247 ret = btrfs_del_items(trans, root, path,
3248 pending_del_slot,
3249 pending_del_nr);
3250 BUG_ON(ret);
3251 pending_del_nr = 0;
3252 }
b3b4aa74 3253 btrfs_release_path(path);
85e21bac 3254 goto search_again;
8082510e
YZ		 3255 } else {
3256 path->slots[0]--;
85e21bac 3257 }
39279cc3 3258 }
8082510e 3259out:
85e21bac
CM		 3260 if (pending_del_nr) {
3261 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3262 pending_del_nr);
d68fc57b 3263 BUG_ON(ret);
85e21bac 3264 }
39279cc3 3265 btrfs_free_path(path);
8082510e 3266 return err;
39279cc3
CM		 3267}
3268
3269/*
3270 * taken from block_truncate_page, but does cow as it zeros out
3271 * any bytes left in the last page in the file.
3272 */
3273static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3274{
3275 struct inode *inode = mapping->host;
db94535d 3276 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3277 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3278 struct btrfs_ordered_extent *ordered;
2ac55d41 3279 struct extent_state *cached_state = NULL;
e6dcd2dc 3280 char *kaddr;
db94535d 3281 u32 blocksize = root->sectorsize;
39279cc3
CM		 3282 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3283 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3284 struct page *page;
39279cc3 3285 int ret = 0;
a52d9a80 3286 u64 page_start;
e6dcd2dc 3287 u64 page_end;
39279cc3
CM		 3288
3289 if ((offset & (blocksize - 1)) == 0)
3290 goto out;
0ca1f7ce 3291 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3292 if (ret)
3293 goto out;
39279cc3
CM		 3294
3295 ret = -ENOMEM;
211c17f5 3296again:
a94733d0 3297 page = find_or_create_page(mapping, index, GFP_NOFS);
5d5e103a 3298 if (!page) {
0ca1f7ce 3299 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3300 goto out;
5d5e103a 3301 }
e6dcd2dc
CM		 3302
3303 page_start = page_offset(page);
3304 page_end = page_start + PAGE_CACHE_SIZE - 1;
3305
39279cc3 3306 if (!PageUptodate(page)) {
9ebefb18 3307 ret = btrfs_readpage(NULL, page);
39279cc3 3308 lock_page(page);
211c17f5
CM		 3309 if (page->mapping != mapping) {
3310 unlock_page(page);
3311 page_cache_release(page);
3312 goto again;
3313 }
39279cc3
CM		 3314 if (!PageUptodate(page)) {
3315 ret = -EIO;
89642229 3316 goto out_unlock;
39279cc3
CM		 3317 }
3318 }
211c17f5 3319 wait_on_page_writeback(page);
e6dcd2dc 3320
2ac55d41
JB		 3321 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3322 GFP_NOFS);
e6dcd2dc
CM		 3323 set_page_extent_mapped(page);
3324
3325 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3326 if (ordered) {
2ac55d41
JB		 3327 unlock_extent_cached(io_tree, page_start, page_end,
3328 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3329 unlock_page(page);
3330 page_cache_release(page);
eb84ae03 3331 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3332 btrfs_put_ordered_extent(ordered);
3333 goto again;
3334 }
3335
2ac55d41 3336 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3337 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3338 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3339
2ac55d41
JB		 3340 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3341 &cached_state);
9ed74f2d 3342 if (ret) {
2ac55d41
JB		 3343 unlock_extent_cached(io_tree, page_start, page_end,
3344 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3345 goto out_unlock;
3346 }
3347
e6dcd2dc
CM		 3348 ret = 0;
3349 if (offset != PAGE_CACHE_SIZE) {
3350 kaddr = kmap(page);
3351 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3352 flush_dcache_page(page);
3353 kunmap(page);
3354 }
247e743c 3355 ClearPageChecked(page);
e6dcd2dc 3356 set_page_dirty(page);
2ac55d41
JB		 3357 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3358 GFP_NOFS);
39279cc3 3359
89642229 3360out_unlock:
5d5e103a 3361 if (ret)
0ca1f7ce 3362 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3363 unlock_page(page);
3364 page_cache_release(page);
3365out:
3366 return ret;
3367}
3368
695a0d0d
JB		 3369/*
3370 * This function puts in dummy file extents for the area we're creating a hole
3371 * for. So if we are truncating this file to a larger size we need to insert
3372 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3373 * the range between oldsize and size
3374 */
a41ad394 3375int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3376{
9036c102
YZ		 3377 struct btrfs_trans_handle *trans;
3378 struct btrfs_root *root = BTRFS_I(inode)->root;
3379 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3380 struct extent_map *em = NULL;
2ac55d41 3381 struct extent_state *cached_state = NULL;
9036c102 3382 u64 mask = root->sectorsize - 1;
a41ad394 3383 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3384 u64 block_end = (size + mask) & ~mask;
3385 u64 last_byte;
3386 u64 cur_offset;
3387 u64 hole_size;
9ed74f2d 3388 int err = 0;
39279cc3 3389
9036c102
YZ		 3390 if (size <= hole_start)
3391 return 0;
3392
9036c102
YZ		 3393 while (1) {
3394 struct btrfs_ordered_extent *ordered;
3395 btrfs_wait_ordered_range(inode, hole_start,
3396 block_end - hole_start);
2ac55d41
JB		 3397 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3398 &cached_state, GFP_NOFS);
9036c102
YZ		 3399 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3400 if (!ordered)
3401 break;
2ac55d41
JB		 3402 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3403 &cached_state, GFP_NOFS);
9036c102
YZ		 3404 btrfs_put_ordered_extent(ordered);
3405 }
39279cc3 3406
9036c102
YZ		 3407 cur_offset = hole_start;
3408 while (1) {
3409 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3410 block_end - cur_offset, 0);
c704005d 3411 BUG_ON(IS_ERR_OR_NULL(em));
9036c102
YZ		 3412 last_byte = min(extent_map_end(em), block_end);
3413 last_byte = (last_byte + mask) & ~mask;
8082510e 3414 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3415 u64 hint_byte = 0;
9036c102 3416 hole_size = last_byte - cur_offset;
9ed74f2d 3417
a22285a6
YZ		 3418 trans = btrfs_start_transaction(root, 2);
3419 if (IS_ERR(trans)) {
3420 err = PTR_ERR(trans);
9ed74f2d 3421 break;
a22285a6 3422 }
8082510e
YZ		 3423
3424 err = btrfs_drop_extents(trans, inode, cur_offset,
3425 cur_offset + hole_size,
3426 &hint_byte, 1);
5b397377
MX		 3427 if (err) {
3428 btrfs_end_transaction(trans, root);
3893e33b 3429 break;
5b397377 3430 }
8082510e 3431
9036c102 3432 err = btrfs_insert_file_extent(trans, root,
33345d01 3433 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3434 0, hole_size, 0, hole_size,
3435 0, 0, 0);
5b397377
MX		 3436 if (err) {
3437 btrfs_end_transaction(trans, root);
3893e33b 3438 break;
5b397377 3439 }
8082510e 3440
9036c102
YZ		 3441 btrfs_drop_extent_cache(inode, hole_start,
3442 last_byte - 1, 0);
8082510e
YZ		 3443
3444 btrfs_end_transaction(trans, root);
9036c102
YZ		 3445 }
3446 free_extent_map(em);
a22285a6 3447 em = NULL;
9036c102 3448 cur_offset = last_byte;
8082510e 3449 if (cur_offset >= block_end)
9036c102
YZ		 3450 break;
3451 }
1832a6d5 3452
a22285a6 3453 free_extent_map(em);
2ac55d41
JB		 3454 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3455 GFP_NOFS);
9036c102
YZ		 3456 return err;
3457}
39279cc3 3458
a41ad394 3459static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3460{
a41ad394 3461 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3462 int ret;
3463
a41ad394 3464 if (newsize == oldsize)
8082510e
YZ		 3465 return 0;
3466
a41ad394
JB		 3467 if (newsize > oldsize) {
3468 i_size_write(inode, newsize);
3469 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3470 truncate_pagecache(inode, oldsize, newsize);
3471 ret = btrfs_cont_expand(inode, oldsize, newsize);
8082510e 3472 if (ret) {
a41ad394 3473 btrfs_setsize(inode, oldsize);
8082510e
YZ		 3474 return ret;
3475 }
3476
930f028a 3477 mark_inode_dirty(inode);
a41ad394 3478 } else {
8082510e 3479
a41ad394
JB		 3480 /*
3481 * We're truncating a file that used to have good data down to
3482 * zero. Make sure it gets into the ordered flush list so that
3483 * any new writes get down to disk quickly.
3484 */
3485 if (newsize == 0)
3486 BTRFS_I(inode)->ordered_data_close = 1;
8082510e 3487
a41ad394
JB		 3488 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3489 truncate_setsize(inode, newsize);
3490 ret = btrfs_truncate(inode);
8082510e
YZ		 3491 }
3492
a41ad394 3493 return ret;
8082510e
YZ		 3494}
3495
9036c102
YZ		 3496static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3497{
3498 struct inode *inode = dentry->d_inode;
b83cc969 3499 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3500 int err;
39279cc3 3501
b83cc969
LZ		 3502 if (btrfs_root_readonly(root))
3503 return -EROFS;
3504
9036c102
YZ		 3505 err = inode_change_ok(inode, attr);
3506 if (err)
3507 return err;
2bf5a725 3508
5a3f23d5 3509 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3510 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3511 if (err)
3512 return err;
39279cc3 3513 }
9036c102 3514
1025774c
CH		 3515 if (attr->ia_valid) {
3516 setattr_copy(inode, attr);
3517 mark_inode_dirty(inode);
3518
3519 if (attr->ia_valid & ATTR_MODE)
3520 err = btrfs_acl_chmod(inode);
3521 }
33268eaf 3522
39279cc3
CM		 3523 return err;
3524}
61295eb8 3525
bd555975 3526void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3527{
3528 struct btrfs_trans_handle *trans;
3529 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 3530 unsigned long nr;
39279cc3
CM		 3531 int ret;
3532
1abe9b8a 3533 trace_btrfs_inode_evict(inode);
3534
39279cc3 3535 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3536 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
2cf8572d 3537 btrfs_is_free_space_inode(root, inode)))
bd555975
AV		 3538 goto no_delete;
3539
39279cc3 3540 if (is_bad_inode(inode)) {
7b128766 3541 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3542 goto no_delete;
3543 }
bd555975 3544 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3545 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3546
c71bf099
YZ		 3547 if (root->fs_info->log_root_recovering) {
3548 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3549 goto no_delete;
3550 }
3551
76dda93c
YZ		 3552 if (inode->i_nlink > 0) {
3553 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3554 goto no_delete;
3555 }
3556
dbe674a9 3557 btrfs_i_size_write(inode, 0);
5f39d397 3558
8082510e 3559 while (1) {
30b4caf5 3560 trans = btrfs_join_transaction(root);
d68fc57b 3561 BUG_ON(IS_ERR(trans));
d68fc57b
YZ		 3562 trans->block_rsv = root->orphan_block_rsv;
3563
3564 ret = btrfs_block_rsv_check(trans, root,
3565 root->orphan_block_rsv, 0, 5);
3566 if (ret) {
3567 BUG_ON(ret != -EAGAIN);
3568 ret = btrfs_commit_transaction(trans, root);
3569 BUG_ON(ret);
3570 continue;
3571 }
7b128766 3572
d68fc57b 3573 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3574 if (ret != -EAGAIN)
3575 break;
85e21bac 3576
8082510e
YZ		 3577 nr = trans->blocks_used;
3578 btrfs_end_transaction(trans, root);
3579 trans = NULL;
3580 btrfs_btree_balance_dirty(root, nr);
d68fc57b 3581
8082510e 3582 }
5f39d397 3583
8082510e
YZ		 3584 if (ret == 0) {
3585 ret = btrfs_orphan_del(trans, inode);
3586 BUG_ON(ret);
3587 }
54aa1f4d 3588
581bb050
LZ		 3589 if (!(root == root->fs_info->tree_root ||
3590 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3591 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3592
d3c2fdcf 3593 nr = trans->blocks_used;
54aa1f4d 3594 btrfs_end_transaction(trans, root);
d3c2fdcf 3595 btrfs_btree_balance_dirty(root, nr);
39279cc3 3596no_delete:
bd555975 3597 end_writeback(inode);
8082510e 3598 return;
39279cc3
CM		 3599}
3600
3601/*
3602 * this returns the key found in the dir entry in the location pointer.
3603 * If no dir entries were found, location->objectid is 0.
3604 */
3605static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3606 struct btrfs_key *location)
3607{
3608 const char *name = dentry->d_name.name;
3609 int namelen = dentry->d_name.len;
3610 struct btrfs_dir_item *di;
3611 struct btrfs_path *path;
3612 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3613 int ret = 0;
39279cc3
CM		 3614
3615 path = btrfs_alloc_path();
d8926bb3
MF		 3616 if (!path)
3617 return -ENOMEM;
3954401f 3618
33345d01 3619 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3620 namelen, 0);
0d9f7f3e
Y		 3621 if (IS_ERR(di))
3622 ret = PTR_ERR(di);
d397712b 3623
c704005d 3624 if (IS_ERR_OR_NULL(di))
3954401f 3625 goto out_err;
d397712b 3626
5f39d397 3627 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3628out:
39279cc3
CM		 3629 btrfs_free_path(path);
3630 return ret;
3954401f
CM		 3631out_err:
3632 location->objectid = 0;
3633 goto out;
39279cc3
CM		 3634}
3635
3636/*
3637 * when we hit a tree root in a directory, the btrfs part of the inode
3638 * needs to be changed to reflect the root directory of the tree root. This
3639 * is kind of like crossing a mount point.
3640 */
3641static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3642 struct inode *dir,
3643 struct dentry *dentry,
3644 struct btrfs_key *location,
3645 struct btrfs_root **sub_root)
39279cc3 3646{
4df27c4d
YZ		 3647 struct btrfs_path *path;
3648 struct btrfs_root *new_root;
3649 struct btrfs_root_ref *ref;
3650 struct extent_buffer *leaf;
3651 int ret;
3652 int err = 0;
39279cc3 3653
4df27c4d
YZ		 3654 path = btrfs_alloc_path();
3655 if (!path) {
3656 err = -ENOMEM;
3657 goto out;
3658 }
39279cc3 3659
4df27c4d
YZ		 3660 err = -ENOENT;
3661 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3662 BTRFS_I(dir)->root->root_key.objectid,
3663 location->objectid);
3664 if (ret) {
3665 if (ret < 0)
3666 err = ret;
3667 goto out;
3668 }
39279cc3 3669
4df27c4d
YZ		 3670 leaf = path->nodes[0];
3671 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3672 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3673 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3674 goto out;
39279cc3 3675
4df27c4d
YZ		 3676 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3677 (unsigned long)(ref + 1),
3678 dentry->d_name.len);
3679 if (ret)
3680 goto out;
3681
b3b4aa74 3682 btrfs_release_path(path);
4df27c4d
YZ		 3683
3684 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3685 if (IS_ERR(new_root)) {
3686 err = PTR_ERR(new_root);
3687 goto out;
3688 }
3689
3690 if (btrfs_root_refs(&new_root->root_item) == 0) {
3691 err = -ENOENT;
3692 goto out;
3693 }
3694
3695 *sub_root = new_root;
3696 location->objectid = btrfs_root_dirid(&new_root->root_item);
3697 location->type = BTRFS_INODE_ITEM_KEY;
3698 location->offset = 0;
3699 err = 0;
3700out:
3701 btrfs_free_path(path);
3702 return err;
39279cc3
CM		 3703}
3704
5d4f98a2
YZ		 3705static void inode_tree_add(struct inode *inode)
3706{
3707 struct btrfs_root *root = BTRFS_I(inode)->root;
3708 struct btrfs_inode *entry;
03e860bd
FNP		 3709 struct rb_node **p;
3710 struct rb_node *parent;
33345d01 3711 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3712again:
3713 p = &root->inode_tree.rb_node;
3714 parent = NULL;
5d4f98a2 3715
1d3382cb 3716 if (inode_unhashed(inode))
76dda93c
YZ		 3717 return;
3718
5d4f98a2
YZ		 3719 spin_lock(&root->inode_lock);
3720 while (*p) {
3721 parent = *p;
3722 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3723
33345d01 3724 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3725 p = &parent->rb_left;
33345d01 3726 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3727 p = &parent->rb_right;
5d4f98a2
YZ		 3728 else {
3729 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3730 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3731 rb_erase(parent, &root->inode_tree);
3732 RB_CLEAR_NODE(parent);
3733 spin_unlock(&root->inode_lock);
3734 goto again;
5d4f98a2
YZ		 3735 }
3736 }
3737 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3738 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3739 spin_unlock(&root->inode_lock);
3740}
3741
3742static void inode_tree_del(struct inode *inode)
3743{
3744 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3745 int empty = 0;
5d4f98a2 3746
03e860bd 3747 spin_lock(&root->inode_lock);
5d4f98a2 3748 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3749 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3750 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3751 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3752 }
03e860bd 3753 spin_unlock(&root->inode_lock);
76dda93c 3754
0af3d00b
JB		 3755 /*
3756 * Free space cache has inodes in the tree root, but the tree root has a
3757 * root_refs of 0, so this could end up dropping the tree root as a
3758 * snapshot, so we need the extra !root->fs_info->tree_root check to
3759 * make sure we don't drop it.
3760 */
3761 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3762 root != root->fs_info->tree_root) {
76dda93c
YZ		 3763 synchronize_srcu(&root->fs_info->subvol_srcu);
3764 spin_lock(&root->inode_lock);
3765 empty = RB_EMPTY_ROOT(&root->inode_tree);
3766 spin_unlock(&root->inode_lock);
3767 if (empty)
3768 btrfs_add_dead_root(root);
3769 }
3770}
3771
3772int btrfs_invalidate_inodes(struct btrfs_root *root)
3773{
3774 struct rb_node *node;
3775 struct rb_node *prev;
3776 struct btrfs_inode *entry;
3777 struct inode *inode;
3778 u64 objectid = 0;
3779
3780 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3781
3782 spin_lock(&root->inode_lock);
3783again:
3784 node = root->inode_tree.rb_node;
3785 prev = NULL;
3786 while (node) {
3787 prev = node;
3788 entry = rb_entry(node, struct btrfs_inode, rb_node);
3789
33345d01 3790 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 3791 node = node->rb_left;
33345d01 3792 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 3793 node = node->rb_right;
3794 else
3795 break;
3796 }
3797 if (!node) {
3798 while (prev) {
3799 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 3800 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 3801 node = prev;
3802 break;
3803 }
3804 prev = rb_next(prev);
3805 }
3806 }
3807 while (node) {
3808 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 3809 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 3810 inode = igrab(&entry->vfs_inode);
3811 if (inode) {
3812 spin_unlock(&root->inode_lock);
3813 if (atomic_read(&inode->i_count) > 1)
3814 d_prune_aliases(inode);
3815 /*
45321ac5 3816 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 3817 * the inode cache when its usage count
3818 * hits zero.
3819 */
3820 iput(inode);
3821 cond_resched();
3822 spin_lock(&root->inode_lock);
3823 goto again;
3824 }
3825
3826 if (cond_resched_lock(&root->inode_lock))
3827 goto again;
3828
3829 node = rb_next(node);
3830 }
3831 spin_unlock(&root->inode_lock);
3832 return 0;
5d4f98a2
YZ		 3833}
3834
e02119d5
CM		 3835static int btrfs_init_locked_inode(struct inode *inode, void *p)
3836{
3837 struct btrfs_iget_args *args = p;
3838 inode->i_ino = args->ino;
e02119d5 3839 BTRFS_I(inode)->root = args->root;
6a63209f 3840 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3841 return 0;
3842}
3843
3844static int btrfs_find_actor(struct inode *inode, void *opaque)
3845{
3846 struct btrfs_iget_args *args = opaque;
33345d01 3847 return args->ino == btrfs_ino(inode) &&
d397712b 3848 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3849}
3850
5d4f98a2
YZ		 3851static struct inode *btrfs_iget_locked(struct super_block *s,
3852 u64 objectid,
3853 struct btrfs_root *root)
39279cc3
CM		 3854{
3855 struct inode *inode;
3856 struct btrfs_iget_args args;
3857 args.ino = objectid;
3858 args.root = root;
3859
3860 inode = iget5_locked(s, objectid, btrfs_find_actor,
3861 btrfs_init_locked_inode,
3862 (void *)&args);
3863 return inode;
3864}
3865
1a54ef8c
BR		 3866/* Get an inode object given its location and corresponding root.
3867 * Returns in *is_new if the inode was read from disk
3868 */
3869struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 3870 struct btrfs_root *root, int *new)
1a54ef8c
BR		 3871{
3872 struct inode *inode;
3873
3874 inode = btrfs_iget_locked(s, location->objectid, root);
3875 if (!inode)
5d4f98a2 3876 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3877
3878 if (inode->i_state & I_NEW) {
3879 BTRFS_I(inode)->root = root;
3880 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3881 btrfs_read_locked_inode(inode);
1748f843
MF		 3882 if (!is_bad_inode(inode)) {
3883 inode_tree_add(inode);
3884 unlock_new_inode(inode);
3885 if (new)
3886 *new = 1;
3887 } else {
e0b6d65b
ST		 3888 unlock_new_inode(inode);
3889 iput(inode);
3890 inode = ERR_PTR(-ESTALE);
1748f843
MF		 3891 }
3892 }
3893
1a54ef8c
BR		 3894 return inode;
3895}
3896
4df27c4d
YZ		 3897static struct inode *new_simple_dir(struct super_block *s,
3898 struct btrfs_key *key,
3899 struct btrfs_root *root)
3900{
3901 struct inode *inode = new_inode(s);
3902
3903 if (!inode)
3904 return ERR_PTR(-ENOMEM);
3905
4df27c4d
YZ		 3906 BTRFS_I(inode)->root = root;
3907 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3908 BTRFS_I(inode)->dummy_inode = 1;
3909
3910 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3911 inode->i_op = &simple_dir_inode_operations;
3912 inode->i_fop = &simple_dir_operations;
3913 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3914 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3915
3916 return inode;
3917}
3918
3de4586c 3919struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3920{
d397712b 3921 struct inode *inode;
4df27c4d 3922 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 3923 struct btrfs_root *sub_root = root;
3924 struct btrfs_key location;
76dda93c 3925 int index;
b4aff1f8 3926 int ret = 0;
39279cc3
CM		 3927
3928 if (dentry->d_name.len > BTRFS_NAME_LEN)
3929 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3930
b4aff1f8
JB		 3931 if (unlikely(d_need_lookup(dentry))) {
3932 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3933 kfree(dentry->d_fsdata);
3934 dentry->d_fsdata = NULL;
a66e7cc6
JB		 3935 /* This thing is hashed, drop it for now */
3936 d_drop(dentry);
b4aff1f8
JB		 3937 } else {
3938 ret = btrfs_inode_by_name(dir, dentry, &location);
3939 }
5f39d397 3940
39279cc3
CM		 3941 if (ret < 0)
3942 return ERR_PTR(ret);
5f39d397 3943
4df27c4d
YZ		 3944 if (location.objectid == 0)
3945 return NULL;
3946
3947 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 3948 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 3949 return inode;
3950 }
3951
3952 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3953
76dda93c 3954 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 3955 ret = fixup_tree_root_location(root, dir, dentry,
3956 &location, &sub_root);
3957 if (ret < 0) {
3958 if (ret != -ENOENT)
3959 inode = ERR_PTR(ret);
3960 else
3961 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3962 } else {
73f73415 3963 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 3964 }
76dda93c
YZ		 3965 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3966
34d19bad 3967 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 3968 down_read(&root->fs_info->cleanup_work_sem);
3969 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 3970 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 3971 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 3972 if (ret)
3973 inode = ERR_PTR(ret);
c71bf099
YZ		 3974 }
3975
3de4586c
CM		 3976 return inode;
3977}
3978
fe15ce44 3979static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 3980{
3981 struct btrfs_root *root;
3982
efefb143
YZ		 3983 if (!dentry->d_inode && !IS_ROOT(dentry))
3984 dentry = dentry->d_parent;
76dda93c 3985
efefb143
YZ		 3986 if (dentry->d_inode) {
3987 root = BTRFS_I(dentry->d_inode)->root;
3988 if (btrfs_root_refs(&root->root_item) == 0)
3989 return 1;
3990 }
76dda93c
YZ		 3991 return 0;
3992}
3993
b4aff1f8
JB		 3994static void btrfs_dentry_release(struct dentry *dentry)
3995{
3996 if (dentry->d_fsdata)
3997 kfree(dentry->d_fsdata);
3998}
3999
3de4586c
CM		 4000static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4001 struct nameidata *nd)
4002{
a66e7cc6
JB		 4003 struct dentry *ret;
4004
4005 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4006 if (unlikely(d_need_lookup(dentry))) {
4007 spin_lock(&dentry->d_lock);
4008 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4009 spin_unlock(&dentry->d_lock);
4010 }
4011 return ret;
39279cc3
CM		 4012}
4013
16cdcec7 4014unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4015 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4016};
4017
cbdf5a24
DW		 4018static int btrfs_real_readdir(struct file *filp, void *dirent,
4019 filldir_t filldir)
39279cc3 4020{
6da6abae 4021 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4022 struct btrfs_root *root = BTRFS_I(inode)->root;
4023 struct btrfs_item *item;
4024 struct btrfs_dir_item *di;
4025 struct btrfs_key key;
5f39d397 4026 struct btrfs_key found_key;
39279cc3 4027 struct btrfs_path *path;
16cdcec7
MX		 4028 struct list_head ins_list;
4029 struct list_head del_list;
b4aff1f8 4030 struct qstr q;
39279cc3 4031 int ret;
5f39d397 4032 struct extent_buffer *leaf;
39279cc3 4033 int slot;
39279cc3
CM		 4034 unsigned char d_type;
4035 int over = 0;
4036 u32 di_cur;
4037 u32 di_total;
4038 u32 di_len;
4039 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4040 char tmp_name[32];
4041 char *name_ptr;
4042 int name_len;
16cdcec7 4043 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4044
4045 /* FIXME, use a real flag for deciding about the key type */
4046 if (root->fs_info->tree_root == root)
4047 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4048
3954401f
CM		 4049 /* special case for "." */
4050 if (filp->f_pos == 0) {
3765fefa
HS		 4051 over = filldir(dirent, ".", 1,
4052 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4053 if (over)
4054 return 0;
4055 filp->f_pos = 1;
4056 }
3954401f
CM		 4057 /* special case for .., just use the back ref */
4058 if (filp->f_pos == 1) {
5ecc7e5d 4059 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4060 over = filldir(dirent, "..", 2,
3765fefa 4061 filp->f_pos, pino, DT_DIR);
3954401f 4062 if (over)
49593bfa 4063 return 0;
3954401f
CM		 4064 filp->f_pos = 2;
4065 }
49593bfa 4066 path = btrfs_alloc_path();
16cdcec7
MX		 4067 if (!path)
4068 return -ENOMEM;
ff5714cc 4069
026fd317 4070 path->reada = 1;
49593bfa 4071
16cdcec7
MX		 4072 if (key_type == BTRFS_DIR_INDEX_KEY) {
4073 INIT_LIST_HEAD(&ins_list);
4074 INIT_LIST_HEAD(&del_list);
4075 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4076 }
4077
39279cc3
CM		 4078 btrfs_set_key_type(&key, key_type);
4079 key.offset = filp->f_pos;
33345d01 4080 key.objectid = btrfs_ino(inode);
5f39d397 4081
39279cc3
CM		 4082 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4083 if (ret < 0)
4084 goto err;
49593bfa
DW		 4085
4086 while (1) {
5f39d397 4087 leaf = path->nodes[0];
39279cc3 4088 slot = path->slots[0];
b9e03af0
LZ		 4089 if (slot >= btrfs_header_nritems(leaf)) {
4090 ret = btrfs_next_leaf(root, path);
4091 if (ret < 0)
4092 goto err;
4093 else if (ret > 0)
4094 break;
4095 continue;
39279cc3 4096 }
3de4586c 4097
5f39d397
CM		 4098 item = btrfs_item_nr(leaf, slot);
4099 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4100
4101 if (found_key.objectid != key.objectid)
39279cc3 4102 break;
5f39d397 4103 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4104 break;
5f39d397 4105 if (found_key.offset < filp->f_pos)
b9e03af0 4106 goto next;
16cdcec7
MX		 4107 if (key_type == BTRFS_DIR_INDEX_KEY &&
4108 btrfs_should_delete_dir_index(&del_list,
4109 found_key.offset))
4110 goto next;
5f39d397
CM		 4111
4112 filp->f_pos = found_key.offset;
16cdcec7 4113 is_curr = 1;
49593bfa 4114
39279cc3
CM		 4115 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4116 di_cur = 0;
5f39d397 4117 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4118
4119 while (di_cur < di_total) {
5f39d397 4120 struct btrfs_key location;
b4aff1f8 4121 struct dentry *tmp;
5f39d397 4122
22a94d44
JB		 4123 if (verify_dir_item(root, leaf, di))
4124 break;
4125
5f39d397 4126 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4127 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4128 name_ptr = tmp_name;
4129 } else {
4130 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4131 if (!name_ptr) {
4132 ret = -ENOMEM;
4133 goto err;
4134 }
5f39d397
CM		 4135 }
4136 read_extent_buffer(leaf, name_ptr,
4137 (unsigned long)(di + 1), name_len);
4138
4139 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4140 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4141
b4aff1f8
JB		 4142 q.name = name_ptr;
4143 q.len = name_len;
4144 q.hash = full_name_hash(q.name, q.len);
4145 tmp = d_lookup(filp->f_dentry, &q);
4146 if (!tmp) {
4147 struct btrfs_key *newkey;
4148
4149 newkey = kzalloc(sizeof(struct btrfs_key),
4150 GFP_NOFS);
4151 if (!newkey)
4152 goto no_dentry;
4153 tmp = d_alloc(filp->f_dentry, &q);
4154 if (!tmp) {
4155 kfree(newkey);
4156 dput(tmp);
4157 goto no_dentry;
4158 }
4159 memcpy(newkey, &location,
4160 sizeof(struct btrfs_key));
4161 tmp->d_fsdata = newkey;
4162 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4163 d_rehash(tmp);
4164 dput(tmp);
4165 } else {
4166 dput(tmp);
4167 }
4168no_dentry:
3de4586c
CM		 4169 /* is this a reference to our own snapshot? If so
4170 * skip it
4171 */
4172 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4173 location.objectid == root->root_key.objectid) {
4174 over = 0;
4175 goto skip;
4176 }
5f39d397 4177 over = filldir(dirent, name_ptr, name_len,
49593bfa 4178 found_key.offset, location.objectid,
39279cc3 4179 d_type);
5f39d397 4180
3de4586c 4181skip:
5f39d397
CM		 4182 if (name_ptr != tmp_name)
4183 kfree(name_ptr);
4184
39279cc3
CM		 4185 if (over)
4186 goto nopos;
5103e947 4187 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4188 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4189 di_cur += di_len;
4190 di = (struct btrfs_dir_item *)((char *)di + di_len);
4191 }
b9e03af0
LZ		 4192next:
4193 path->slots[0]++;
39279cc3 4194 }
49593bfa 4195
16cdcec7
MX		 4196 if (key_type == BTRFS_DIR_INDEX_KEY) {
4197 if (is_curr)
4198 filp->f_pos++;
4199 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4200 &ins_list);
4201 if (ret)
4202 goto nopos;
4203 }
4204
49593bfa 4205 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4206 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4207 /*
4208 * 32-bit glibc will use getdents64, but then strtol -
4209 * so the last number we can serve is this.
4210 */
4211 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4212 else
4213 filp->f_pos++;
39279cc3
CM		 4214nopos:
4215 ret = 0;
4216err:
16cdcec7
MX		 4217 if (key_type == BTRFS_DIR_INDEX_KEY)
4218 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4219 btrfs_free_path(path);
39279cc3
CM		 4220 return ret;
4221}
4222
a9185b41 4223int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4224{
4225 struct btrfs_root *root = BTRFS_I(inode)->root;
4226 struct btrfs_trans_handle *trans;
4227 int ret = 0;
0af3d00b 4228 bool nolock = false;
39279cc3 4229
8929ecfa 4230 if (BTRFS_I(inode)->dummy_inode)
4ca8b41e
CM		 4231 return 0;
4232
2cf8572d 4233 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
82d5902d 4234 nolock = true;
0af3d00b 4235
a9185b41 4236 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4237 if (nolock)
7a7eaa40 4238 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4239 else
7a7eaa40 4240 trans = btrfs_join_transaction(root);
3612b495
TI		 4241 if (IS_ERR(trans))
4242 return PTR_ERR(trans);
0af3d00b
JB		 4243 if (nolock)
4244 ret = btrfs_end_transaction_nolock(trans, root);
4245 else
4246 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4247 }
4248 return ret;
4249}
4250
4251/*
54aa1f4d 4252 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4253 * inode changes. But, it is most likely to find the inode in cache.
4254 * FIXME, needs more benchmarking...there are no reasons other than performance
4255 * to keep or drop this code.
4256 */
aa385729 4257void btrfs_dirty_inode(struct inode *inode, int flags)
39279cc3
CM		 4258{
4259 struct btrfs_root *root = BTRFS_I(inode)->root;
4260 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4261 int ret;
4262
4263 if (BTRFS_I(inode)->dummy_inode)
4264 return;
39279cc3 4265
7a7eaa40 4266 trans = btrfs_join_transaction(root);
3612b495 4267 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 4268
4269 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4270 if (ret && ret == -ENOSPC) {
4271 /* whoops, lets try again with the full transaction */
4272 btrfs_end_transaction(trans, root);
4273 trans = btrfs_start_transaction(root, 1);
9aeead73 4274 if (IS_ERR(trans)) {
7a36ddec 4275 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4276 "dirty inode %llu error %ld\n",
4277 (unsigned long long)btrfs_ino(inode),
4278 PTR_ERR(trans));
9aeead73
CM		 4279 return;
4280 }
8929ecfa 4281
94b60442
CM		 4282 ret = btrfs_update_inode(trans, root, inode);
4283 if (ret) {
7a36ddec 4284 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4285 "dirty inode %llu error %d\n",
4286 (unsigned long long)btrfs_ino(inode),
4287 ret);
94b60442
CM		 4288 }
4289 }
39279cc3 4290 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4291 if (BTRFS_I(inode)->delayed_node)
4292 btrfs_balance_delayed_items(root);
39279cc3
CM		 4293}
4294
d352ac68
CM		 4295/*
4296 * find the highest existing sequence number in a directory
4297 * and then set the in-memory index_cnt variable to reflect
4298 * free sequence numbers
4299 */
aec7477b
JB		 4300static int btrfs_set_inode_index_count(struct inode *inode)
4301{
4302 struct btrfs_root *root = BTRFS_I(inode)->root;
4303 struct btrfs_key key, found_key;
4304 struct btrfs_path *path;
4305 struct extent_buffer *leaf;
4306 int ret;
4307
33345d01 4308 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4309 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4310 key.offset = (u64)-1;
4311
4312 path = btrfs_alloc_path();
4313 if (!path)
4314 return -ENOMEM;
4315
4316 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4317 if (ret < 0)
4318 goto out;
4319 /* FIXME: we should be able to handle this */
4320 if (ret == 0)
4321 goto out;
4322 ret = 0;
4323
4324 /*
4325 * MAGIC NUMBER EXPLANATION:
4326 * since we search a directory based on f_pos we have to start at 2
4327 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4328 * else has to start at 2
4329 */
4330 if (path->slots[0] == 0) {
4331 BTRFS_I(inode)->index_cnt = 2;
4332 goto out;
4333 }
4334
4335 path->slots[0]--;
4336
4337 leaf = path->nodes[0];
4338 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4339
33345d01 4340 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4341 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4342 BTRFS_I(inode)->index_cnt = 2;
4343 goto out;
4344 }
4345
4346 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4347out:
4348 btrfs_free_path(path);
4349 return ret;
4350}
4351
d352ac68
CM		 4352/*
4353 * helper to find a free sequence number in a given directory. This current
4354 * code is very simple, later versions will do smarter things in the btree
4355 */
3de4586c 4356int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4357{
4358 int ret = 0;
4359
4360 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4361 ret = btrfs_inode_delayed_dir_index_count(dir);
4362 if (ret) {
4363 ret = btrfs_set_inode_index_count(dir);
4364 if (ret)
4365 return ret;
4366 }
aec7477b
JB		 4367 }
4368
00e4e6b3 4369 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4370 BTRFS_I(dir)->index_cnt++;
4371
4372 return ret;
4373}
4374
39279cc3
CM		 4375static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root,
aec7477b 4377 struct inode *dir,
9c58309d 4378 const char *name, int name_len,
d82a6f1d
JB		 4379 u64 ref_objectid, u64 objectid, int mode,
4380 u64 *index)
39279cc3
CM		 4381{
4382 struct inode *inode;
5f39d397 4383 struct btrfs_inode_item *inode_item;
39279cc3 4384 struct btrfs_key *location;
5f39d397 4385 struct btrfs_path *path;
9c58309d
CM		 4386 struct btrfs_inode_ref *ref;
4387 struct btrfs_key key[2];
4388 u32 sizes[2];
4389 unsigned long ptr;
39279cc3
CM		 4390 int ret;
4391 int owner;
4392
5f39d397 4393 path = btrfs_alloc_path();
d8926bb3
MF		 4394 if (!path)
4395 return ERR_PTR(-ENOMEM);
5f39d397 4396
39279cc3 4397 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4398 if (!inode) {
4399 btrfs_free_path(path);
39279cc3 4400 return ERR_PTR(-ENOMEM);
8fb27640 4401 }
39279cc3 4402
581bb050
LZ		 4403 /*
4404 * we have to initialize this early, so we can reclaim the inode
4405 * number if we fail afterwards in this function.
4406 */
4407 inode->i_ino = objectid;
4408
aec7477b 4409 if (dir) {
1abe9b8a 4410 trace_btrfs_inode_request(dir);
4411
3de4586c 4412 ret = btrfs_set_inode_index(dir, index);
09771430 4413 if (ret) {
8fb27640 4414 btrfs_free_path(path);
09771430 4415 iput(inode);
aec7477b 4416 return ERR_PTR(ret);
09771430 4417 }
aec7477b
JB		 4418 }
4419 /*
4420 * index_cnt is ignored for everything but a dir,
4421 * btrfs_get_inode_index_count has an explanation for the magic
4422 * number
4423 */
4424 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4425 BTRFS_I(inode)->root = root;
e02119d5 4426 BTRFS_I(inode)->generation = trans->transid;
76195853 4427 inode->i_generation = BTRFS_I(inode)->generation;
6a63209f 4428 btrfs_set_inode_space_info(root, inode);
b888db2b 4429
569254b0 4430 if (S_ISDIR(mode))
39279cc3
CM		 4431 owner = 0;
4432 else
4433 owner = 1;
9c58309d
CM		 4434
4435 key[0].objectid = objectid;
4436 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4437 key[0].offset = 0;
4438
4439 key[1].objectid = objectid;
4440 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4441 key[1].offset = ref_objectid;
4442
4443 sizes[0] = sizeof(struct btrfs_inode_item);
4444 sizes[1] = name_len + sizeof(*ref);
4445
b9473439 4446 path->leave_spinning = 1;
9c58309d
CM		 4447 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4448 if (ret != 0)
5f39d397
CM		 4449 goto fail;
4450
ecc11fab 4451 inode_init_owner(inode, dir, mode);
a76a3cd4 4452 inode_set_bytes(inode, 0);
39279cc3 4453 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4454 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4455 struct btrfs_inode_item);
e02119d5 4456 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4457
4458 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4459 struct btrfs_inode_ref);
4460 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4461 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4462 ptr = (unsigned long)(ref + 1);
4463 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4464
5f39d397
CM		 4465 btrfs_mark_buffer_dirty(path->nodes[0]);
4466 btrfs_free_path(path);
4467
39279cc3
CM		 4468 location = &BTRFS_I(inode)->location;
4469 location->objectid = objectid;
39279cc3
CM		 4470 location->offset = 0;
4471 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4472
6cbff00f
CH		 4473 btrfs_inherit_iflags(inode, dir);
4474
569254b0 4475 if (S_ISREG(mode)) {
94272164
CM		 4476 if (btrfs_test_opt(root, NODATASUM))
4477 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4478 if (btrfs_test_opt(root, NODATACOW) ||
4479 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4480 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4481 }
4482
39279cc3 4483 insert_inode_hash(inode);
5d4f98a2 4484 inode_tree_add(inode);
1abe9b8a 4485
4486 trace_btrfs_inode_new(inode);
1973f0fa 4487 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4488
39279cc3 4489 return inode;
5f39d397 4490fail:
aec7477b
JB		 4491 if (dir)
4492 BTRFS_I(dir)->index_cnt--;
5f39d397 4493 btrfs_free_path(path);
09771430 4494 iput(inode);
5f39d397 4495 return ERR_PTR(ret);
39279cc3
CM		 4496}
4497
4498static inline u8 btrfs_inode_type(struct inode *inode)
4499{
4500 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4501}
4502
d352ac68
CM		 4503/*
4504 * utility function to add 'inode' into 'parent_inode' with
4505 * a give name and a given sequence number.
4506 * if 'add_backref' is true, also insert a backref from the
4507 * inode to the parent directory.
4508 */
e02119d5
CM		 4509int btrfs_add_link(struct btrfs_trans_handle *trans,
4510 struct inode *parent_inode, struct inode *inode,
4511 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4512{
4df27c4d 4513 int ret = 0;
39279cc3 4514 struct btrfs_key key;
e02119d5 4515 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4516 u64 ino = btrfs_ino(inode);
4517 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4518
33345d01 4519 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4520 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4521 } else {
33345d01 4522 key.objectid = ino;
4df27c4d
YZ		 4523 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4524 key.offset = 0;
4525 }
4526
33345d01 4527 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4528 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4529 key.objectid, root->root_key.objectid,
33345d01 4530 parent_ino, index, name, name_len);
4df27c4d 4531 } else if (add_backref) {
33345d01
LZ		 4532 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4533 parent_ino, index);
4df27c4d 4534 }
39279cc3 4535
39279cc3 4536 if (ret == 0) {
4df27c4d 4537 ret = btrfs_insert_dir_item(trans, root, name, name_len,
16cdcec7 4538 parent_inode, &key,
4df27c4d
YZ		 4539 btrfs_inode_type(inode), index);
4540 BUG_ON(ret);
4541
dbe674a9 4542 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 4543 name_len * 2);
79c44584 4544 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 4545 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 4546 }
4547 return ret;
4548}
4549
4550static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4551 struct inode *dir, struct dentry *dentry,
4552 struct inode *inode, int backref, u64 index)
39279cc3 4553{
a1b075d2
JB		 4554 int err = btrfs_add_link(trans, dir, inode,
4555 dentry->d_name.name, dentry->d_name.len,
4556 backref, index);
39279cc3
CM		 4557 if (!err) {
4558 d_instantiate(dentry, inode);
4559 return 0;
4560 }
4561 if (err > 0)
4562 err = -EEXIST;
4563 return err;
4564}
4565
618e21d5
JB		 4566static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4567 int mode, dev_t rdev)
4568{
4569 struct btrfs_trans_handle *trans;
4570 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4571 struct inode *inode = NULL;
618e21d5
JB		 4572 int err;
4573 int drop_inode = 0;
4574 u64 objectid;
1832a6d5 4575 unsigned long nr = 0;
00e4e6b3 4576 u64 index = 0;
618e21d5
JB		 4577
4578 if (!new_valid_dev(rdev))
4579 return -EINVAL;
4580
9ed74f2d
JB		 4581 /*
4582 * 2 for inode item and ref
4583 * 2 for dir items
4584 * 1 for xattr if selinux is on
4585 */
a22285a6
YZ		 4586 trans = btrfs_start_transaction(root, 5);
4587 if (IS_ERR(trans))
4588 return PTR_ERR(trans);
1832a6d5 4589
581bb050
LZ		 4590 err = btrfs_find_free_ino(root, &objectid);
4591 if (err)
4592 goto out_unlock;
4593
aec7477b 4594 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4595 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4596 mode, &index);
7cf96da3
TI		 4597 if (IS_ERR(inode)) {
4598 err = PTR_ERR(inode);
618e21d5 4599 goto out_unlock;
7cf96da3 4600 }
618e21d5 4601
2a7dba39 4602 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4603 if (err) {
4604 drop_inode = 1;
4605 goto out_unlock;
4606 }
4607
a1b075d2 4608 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4609 if (err)
4610 drop_inode = 1;
4611 else {
4612 inode->i_op = &btrfs_special_inode_operations;
4613 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4614 btrfs_update_inode(trans, root, inode);
618e21d5 4615 }
618e21d5 4616out_unlock:
d3c2fdcf 4617 nr = trans->blocks_used;
89ce8a63 4618 btrfs_end_transaction_throttle(trans, root);
a22285a6 4619 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4620 if (drop_inode) {
4621 inode_dec_link_count(inode);
4622 iput(inode);
4623 }
618e21d5
JB		 4624 return err;
4625}
4626
39279cc3
CM		 4627static int btrfs_create(struct inode *dir, struct dentry *dentry,
4628 int mode, struct nameidata *nd)
4629{
4630 struct btrfs_trans_handle *trans;
4631 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4632 struct inode *inode = NULL;
39279cc3 4633 int drop_inode = 0;
a22285a6 4634 int err;
1832a6d5 4635 unsigned long nr = 0;
39279cc3 4636 u64 objectid;
00e4e6b3 4637 u64 index = 0;
39279cc3 4638
9ed74f2d
JB		 4639 /*
4640 * 2 for inode item and ref
4641 * 2 for dir items
4642 * 1 for xattr if selinux is on
4643 */
a22285a6
YZ		 4644 trans = btrfs_start_transaction(root, 5);
4645 if (IS_ERR(trans))
4646 return PTR_ERR(trans);
9ed74f2d 4647
581bb050
LZ		 4648 err = btrfs_find_free_ino(root, &objectid);
4649 if (err)
4650 goto out_unlock;
4651
aec7477b 4652 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4653 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4654 mode, &index);
7cf96da3
TI		 4655 if (IS_ERR(inode)) {
4656 err = PTR_ERR(inode);
39279cc3 4657 goto out_unlock;
7cf96da3 4658 }
39279cc3 4659
2a7dba39 4660 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4661 if (err) {
4662 drop_inode = 1;
4663 goto out_unlock;
4664 }
4665
a1b075d2 4666 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4667 if (err)
4668 drop_inode = 1;
4669 else {
4670 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4671 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4672 inode->i_fop = &btrfs_file_operations;
4673 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4674 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 4675 }
39279cc3 4676out_unlock:
d3c2fdcf 4677 nr = trans->blocks_used;
ab78c84d 4678 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4679 if (drop_inode) {
4680 inode_dec_link_count(inode);
4681 iput(inode);
4682 }
d3c2fdcf 4683 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4684 return err;
4685}
4686
4687static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4688 struct dentry *dentry)
4689{
4690 struct btrfs_trans_handle *trans;
4691 struct btrfs_root *root = BTRFS_I(dir)->root;
4692 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4693 u64 index;
1832a6d5 4694 unsigned long nr = 0;
39279cc3
CM		 4695 int err;
4696 int drop_inode = 0;
4697
4a8be425
TH		 4698 /* do not allow sys_link's with other subvols of the same device */
4699 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4700 return -EXDEV;
4a8be425 4701
c055e99e
AV		 4702 if (inode->i_nlink == ~0U)
4703 return -EMLINK;
4a8be425 4704
3de4586c 4705 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4706 if (err)
4707 goto fail;
4708
a22285a6 4709 /*
7e6b6465 4710 * 2 items for inode and inode ref
a22285a6 4711 * 2 items for dir items
7e6b6465 4712 * 1 item for parent inode
a22285a6 4713 */
7e6b6465 4714 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4715 if (IS_ERR(trans)) {
4716 err = PTR_ERR(trans);
4717 goto fail;
4718 }
5f39d397 4719
3153495d
MX		 4720 btrfs_inc_nlink(inode);
4721 inode->i_ctime = CURRENT_TIME;
7de9c6ee 4722 ihold(inode);
aec7477b 4723
a1b075d2 4724 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 4725
a5719521 4726 if (err) {
54aa1f4d 4727 drop_inode = 1;
a5719521 4728 } else {
10d9f309 4729 struct dentry *parent = dentry->d_parent;
a5719521
YZ		 4730 err = btrfs_update_inode(trans, root, inode);
4731 BUG_ON(err);
6a912213 4732 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 4733 }
39279cc3 4734
d3c2fdcf 4735 nr = trans->blocks_used;
ab78c84d 4736 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4737fail:
39279cc3
CM		 4738 if (drop_inode) {
4739 inode_dec_link_count(inode);
4740 iput(inode);
4741 }
d3c2fdcf 4742 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4743 return err;
4744}
4745
39279cc3
CM		 4746static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4747{
b9d86667 4748 struct inode *inode = NULL;
39279cc3
CM		 4749 struct btrfs_trans_handle *trans;
4750 struct btrfs_root *root = BTRFS_I(dir)->root;
4751 int err = 0;
4752 int drop_on_err = 0;
b9d86667 4753 u64 objectid = 0;
00e4e6b3 4754 u64 index = 0;
d3c2fdcf 4755 unsigned long nr = 1;
39279cc3 4756
9ed74f2d
JB		 4757 /*
4758 * 2 items for inode and ref
4759 * 2 items for dir items
4760 * 1 for xattr if selinux is on
4761 */
a22285a6
YZ		 4762 trans = btrfs_start_transaction(root, 5);
4763 if (IS_ERR(trans))
4764 return PTR_ERR(trans);
39279cc3 4765
581bb050
LZ		 4766 err = btrfs_find_free_ino(root, &objectid);
4767 if (err)
4768 goto out_fail;
4769
aec7477b 4770 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4771 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4772 S_IFDIR | mode, &index);
39279cc3
CM		 4773 if (IS_ERR(inode)) {
4774 err = PTR_ERR(inode);
4775 goto out_fail;
4776 }
5f39d397 4777
39279cc3 4778 drop_on_err = 1;
33268eaf 4779
2a7dba39 4780 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4781 if (err)
4782 goto out_fail;
4783
39279cc3
CM		 4784 inode->i_op = &btrfs_dir_inode_operations;
4785 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 4786
dbe674a9 4787 btrfs_i_size_write(inode, 0);
39279cc3
CM		 4788 err = btrfs_update_inode(trans, root, inode);
4789 if (err)
4790 goto out_fail;
5f39d397 4791
a1b075d2
JB		 4792 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4793 dentry->d_name.len, 0, index);
39279cc3
CM		 4794 if (err)
4795 goto out_fail;
5f39d397 4796
39279cc3
CM		 4797 d_instantiate(dentry, inode);
4798 drop_on_err = 0;
39279cc3
CM		 4799
4800out_fail:
d3c2fdcf 4801 nr = trans->blocks_used;
ab78c84d 4802 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4803 if (drop_on_err)
4804 iput(inode);
d3c2fdcf 4805 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4806 return err;
4807}
4808
d352ac68
CM		 4809/* helper for btfs_get_extent. Given an existing extent in the tree,
4810 * and an extent that you want to insert, deal with overlap and insert
4811 * the new extent into the tree.
4812 */
3b951516
CM		 4813static int merge_extent_mapping(struct extent_map_tree *em_tree,
4814 struct extent_map *existing,
e6dcd2dc
CM		 4815 struct extent_map *em,
4816 u64 map_start, u64 map_len)
3b951516
CM		 4817{
4818 u64 start_diff;
3b951516 4819
e6dcd2dc
CM		 4820 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4821 start_diff = map_start - em->start;
4822 em->start = map_start;
4823 em->len = map_len;
c8b97818
CM		 4824 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4825 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 4826 em->block_start += start_diff;
c8b97818
CM		 4827 em->block_len -= start_diff;
4828 }
e6dcd2dc 4829 return add_extent_mapping(em_tree, em);
3b951516
CM		 4830}
4831
c8b97818
CM		 4832static noinline int uncompress_inline(struct btrfs_path *path,
4833 struct inode *inode, struct page *page,
4834 size_t pg_offset, u64 extent_offset,
4835 struct btrfs_file_extent_item *item)
4836{
4837 int ret;
4838 struct extent_buffer *leaf = path->nodes[0];
4839 char *tmp;
4840 size_t max_size;
4841 unsigned long inline_size;
4842 unsigned long ptr;
261507a0 4843 int compress_type;
c8b97818
CM		 4844
4845 WARN_ON(pg_offset != 0);
261507a0 4846 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 4847 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4848 inline_size = btrfs_file_extent_inline_item_len(leaf,
4849 btrfs_item_nr(leaf, path->slots[0]));
4850 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 4851 if (!tmp)
4852 return -ENOMEM;
c8b97818
CM		 4853 ptr = btrfs_file_extent_inline_start(item);
4854
4855 read_extent_buffer(leaf, tmp, ptr, inline_size);
4856
5b050f04 4857 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 4858 ret = btrfs_decompress(compress_type, tmp, page,
4859 extent_offset, inline_size, max_size);
c8b97818
CM		 4860 if (ret) {
4861 char *kaddr = kmap_atomic(page, KM_USER0);
4862 unsigned long copy_size = min_t(u64,
4863 PAGE_CACHE_SIZE - pg_offset,
4864 max_size - extent_offset);
4865 memset(kaddr + pg_offset, 0, copy_size);
4866 kunmap_atomic(kaddr, KM_USER0);
4867 }
4868 kfree(tmp);
4869 return 0;
4870}
4871
d352ac68
CM		 4872/*
4873 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4874 * the ugly parts come from merging extents from the disk with the in-ram
4875 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4876 * where the in-ram extents might be locked pending data=ordered completion.
4877 *
4878 * This also copies inline extents directly into the page.
4879 */
d397712b 4880
a52d9a80 4881struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4882 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4883 int create)
4884{
4885 int ret;
4886 int err = 0;
db94535d 4887 u64 bytenr;
a52d9a80
CM		 4888 u64 extent_start = 0;
4889 u64 extent_end = 0;
33345d01 4890 u64 objectid = btrfs_ino(inode);
a52d9a80 4891 u32 found_type;
f421950f 4892 struct btrfs_path *path = NULL;
a52d9a80
CM		 4893 struct btrfs_root *root = BTRFS_I(inode)->root;
4894 struct btrfs_file_extent_item *item;
5f39d397
CM		 4895 struct extent_buffer *leaf;
4896 struct btrfs_key found_key;
a52d9a80
CM		 4897 struct extent_map *em = NULL;
4898 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4899 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4900 struct btrfs_trans_handle *trans = NULL;
261507a0 4901 int compress_type;
a52d9a80 4902
a52d9a80 4903again:
890871be 4904 read_lock(&em_tree->lock);
d1310b2e 4905 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4906 if (em)
4907 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 4908 read_unlock(&em_tree->lock);
d1310b2e 4909
a52d9a80 4910 if (em) {
e1c4b745
CM		 4911 if (em->start > start || em->start + em->len <= start)
4912 free_extent_map(em);
4913 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4914 free_extent_map(em);
4915 else
4916 goto out;
a52d9a80 4917 }
172ddd60 4918 em = alloc_extent_map();
a52d9a80 4919 if (!em) {
d1310b2e
CM		 4920 err = -ENOMEM;
4921 goto out;
a52d9a80 4922 }
e6dcd2dc 4923 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 4924 em->start = EXTENT_MAP_HOLE;
445a6944 4925 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 4926 em->len = (u64)-1;
c8b97818 4927 em->block_len = (u64)-1;
f421950f
CM		 4928
4929 if (!path) {
4930 path = btrfs_alloc_path();
026fd317
JB		 4931 if (!path) {
4932 err = -ENOMEM;
4933 goto out;
4934 }
4935 /*
4936 * Chances are we'll be called again, so go ahead and do
4937 * readahead
4938 */
4939 path->reada = 1;
f421950f
CM		 4940 }
4941
179e29e4
CM		 4942 ret = btrfs_lookup_file_extent(trans, root, path,
4943 objectid, start, trans != NULL);
a52d9a80
CM		 4944 if (ret < 0) {
4945 err = ret;
4946 goto out;
4947 }
4948
4949 if (ret != 0) {
4950 if (path->slots[0] == 0)
4951 goto not_found;
4952 path->slots[0]--;
4953 }
4954
5f39d397
CM		 4955 leaf = path->nodes[0];
4956 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 4957 struct btrfs_file_extent_item);
a52d9a80 4958 /* are we inside the extent that was found? */
5f39d397
CM		 4959 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4960 found_type = btrfs_key_type(&found_key);
4961 if (found_key.objectid != objectid ||
a52d9a80
CM		 4962 found_type != BTRFS_EXTENT_DATA_KEY) {
4963 goto not_found;
4964 }
4965
5f39d397
CM		 4966 found_type = btrfs_file_extent_type(leaf, item);
4967 extent_start = found_key.offset;
261507a0 4968 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 4969 if (found_type == BTRFS_FILE_EXTENT_REG ||
4970 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 4971 extent_end = extent_start +
db94535d 4972 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 4973 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4974 size_t size;
4975 size = btrfs_file_extent_inline_len(leaf, item);
4976 extent_end = (extent_start + size + root->sectorsize - 1) &
4977 ~((u64)root->sectorsize - 1);
4978 }
4979
4980 if (start >= extent_end) {
4981 path->slots[0]++;
4982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4983 ret = btrfs_next_leaf(root, path);
4984 if (ret < 0) {
4985 err = ret;
4986 goto out;
a52d9a80 4987 }
9036c102
YZ		 4988 if (ret > 0)
4989 goto not_found;
4990 leaf = path->nodes[0];
a52d9a80 4991 }
9036c102
YZ		 4992 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4993 if (found_key.objectid != objectid ||
4994 found_key.type != BTRFS_EXTENT_DATA_KEY)
4995 goto not_found;
4996 if (start + len <= found_key.offset)
4997 goto not_found;
4998 em->start = start;
4999 em->len = found_key.offset - start;
5000 goto not_found_em;
5001 }
5002
d899e052
YZ		 5003 if (found_type == BTRFS_FILE_EXTENT_REG ||
5004 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5005 em->start = extent_start;
5006 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5007 em->orig_start = extent_start -
5008 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5009 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5010 if (bytenr == 0) {
5f39d397 5011 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5012 goto insert;
5013 }
261507a0 5014 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5015 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5016 em->compress_type = compress_type;
c8b97818
CM		 5017 em->block_start = bytenr;
5018 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5019 item);
5020 } else {
5021 bytenr += btrfs_file_extent_offset(leaf, item);
5022 em->block_start = bytenr;
5023 em->block_len = em->len;
d899e052
YZ		 5024 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5025 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5026 }
a52d9a80
CM		 5027 goto insert;
5028 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5029 unsigned long ptr;
a52d9a80 5030 char *map;
3326d1b0
CM		 5031 size_t size;
5032 size_t extent_offset;
5033 size_t copy_size;
a52d9a80 5034
689f9346 5035 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5036 if (!page || create) {
689f9346 5037 em->start = extent_start;
9036c102 5038 em->len = extent_end - extent_start;
689f9346
Y		 5039 goto out;
5040 }
5f39d397 5041
9036c102
YZ		 5042 size = btrfs_file_extent_inline_len(leaf, item);
5043 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5044 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5045 size - extent_offset);
3326d1b0 5046 em->start = extent_start + extent_offset;
70dec807
CM		 5047 em->len = (copy_size + root->sectorsize - 1) &
5048 ~((u64)root->sectorsize - 1);
ff5b7ee3 5049 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5050 if (compress_type) {
c8b97818 5051 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5052 em->compress_type = compress_type;
5053 }
689f9346 5054 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5055 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5056 if (btrfs_file_extent_compression(leaf, item) !=
5057 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5058 ret = uncompress_inline(path, inode, page,
5059 pg_offset,
5060 extent_offset, item);
5061 BUG_ON(ret);
5062 } else {
5063 map = kmap(page);
5064 read_extent_buffer(leaf, map + pg_offset, ptr,
5065 copy_size);
93c82d57
CM		 5066 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5067 memset(map + pg_offset + copy_size, 0,
5068 PAGE_CACHE_SIZE - pg_offset -
5069 copy_size);
5070 }
c8b97818
CM		 5071 kunmap(page);
5072 }
179e29e4
CM		 5073 flush_dcache_page(page);
5074 } else if (create && PageUptodate(page)) {
0ca1f7ce 5075 WARN_ON(1);
179e29e4
CM		 5076 if (!trans) {
5077 kunmap(page);
5078 free_extent_map(em);
5079 em = NULL;
ff5714cc 5080
b3b4aa74 5081 btrfs_release_path(path);
7a7eaa40 5082 trans = btrfs_join_transaction(root);
ff5714cc 5083
3612b495
TI		 5084 if (IS_ERR(trans))
5085 return ERR_CAST(trans);
179e29e4
CM		 5086 goto again;
5087 }
c8b97818 5088 map = kmap(page);
70dec807 5089 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5090 copy_size);
c8b97818 5091 kunmap(page);
179e29e4 5092 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5093 }
d1310b2e 5094 set_extent_uptodate(io_tree, em->start,
507903b8 5095 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5096 goto insert;
5097 } else {
d397712b 5098 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5099 WARN_ON(1);
5100 }
5101not_found:
5102 em->start = start;
d1310b2e 5103 em->len = len;
a52d9a80 5104not_found_em:
5f39d397 5105 em->block_start = EXTENT_MAP_HOLE;
9036c102 5106 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5107insert:
b3b4aa74 5108 btrfs_release_path(path);
d1310b2e 5109 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5110 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5111 "[%llu %llu]\n", (unsigned long long)em->start,
5112 (unsigned long long)em->len,
5113 (unsigned long long)start,
5114 (unsigned long long)len);
a52d9a80
CM		 5115 err = -EIO;
5116 goto out;
5117 }
d1310b2e
CM		 5118
5119 err = 0;
890871be 5120 write_lock(&em_tree->lock);
a52d9a80 5121 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5122 /* it is possible that someone inserted the extent into the tree
5123 * while we had the lock dropped. It is also possible that
5124 * an overlapping map exists in the tree
5125 */
a52d9a80 5126 if (ret == -EEXIST) {
3b951516 5127 struct extent_map *existing;
e6dcd2dc
CM		 5128
5129 ret = 0;
5130
3b951516 5131 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5132 if (existing && (existing->start > start ||
5133 existing->start + existing->len <= start)) {
5134 free_extent_map(existing);
5135 existing = NULL;
5136 }
3b951516
CM		 5137 if (!existing) {
5138 existing = lookup_extent_mapping(em_tree, em->start,
5139 em->len);
5140 if (existing) {
5141 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5142 em, start,
5143 root->sectorsize);
3b951516
CM		 5144 free_extent_map(existing);
5145 if (err) {
5146 free_extent_map(em);
5147 em = NULL;
5148 }
5149 } else {
5150 err = -EIO;
3b951516
CM		 5151 free_extent_map(em);
5152 em = NULL;
5153 }
5154 } else {
5155 free_extent_map(em);
5156 em = existing;
e6dcd2dc 5157 err = 0;
a52d9a80 5158 }
a52d9a80 5159 }
890871be 5160 write_unlock(&em_tree->lock);
a52d9a80 5161out:
1abe9b8a 5162
5163 trace_btrfs_get_extent(root, em);
5164
f421950f
CM		 5165 if (path)
5166 btrfs_free_path(path);
a52d9a80
CM		 5167 if (trans) {
5168 ret = btrfs_end_transaction(trans, root);
d397712b 5169 if (!err)
a52d9a80
CM		 5170 err = ret;
5171 }
a52d9a80
CM		 5172 if (err) {
5173 free_extent_map(em);
a52d9a80
CM		 5174 return ERR_PTR(err);
5175 }
5176 return em;
5177}
5178
ec29ed5b
CM		 5179struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5180 size_t pg_offset, u64 start, u64 len,
5181 int create)
5182{
5183 struct extent_map *em;
5184 struct extent_map *hole_em = NULL;
5185 u64 range_start = start;
5186 u64 end;
5187 u64 found;
5188 u64 found_end;
5189 int err = 0;
5190
5191 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5192 if (IS_ERR(em))
5193 return em;
5194 if (em) {
5195 /*
5196 * if our em maps to a hole, there might
5197 * actually be delalloc bytes behind it
5198 */
5199 if (em->block_start != EXTENT_MAP_HOLE)
5200 return em;
5201 else
5202 hole_em = em;
5203 }
5204
5205 /* check to see if we've wrapped (len == -1 or similar) */
5206 end = start + len;
5207 if (end < start)
5208 end = (u64)-1;
5209 else
5210 end -= 1;
5211
5212 em = NULL;
5213
5214 /* ok, we didn't find anything, lets look for delalloc */
5215 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5216 end, len, EXTENT_DELALLOC, 1);
5217 found_end = range_start + found;
5218 if (found_end < range_start)
5219 found_end = (u64)-1;
5220
5221 /*
5222 * we didn't find anything useful, return
5223 * the original results from get_extent()
5224 */
5225 if (range_start > end || found_end <= start) {
5226 em = hole_em;
5227 hole_em = NULL;
5228 goto out;
5229 }
5230
5231 /* adjust the range_start to make sure it doesn't
5232 * go backwards from the start they passed in
5233 */
5234 range_start = max(start,range_start);
5235 found = found_end - range_start;
5236
5237 if (found > 0) {
5238 u64 hole_start = start;
5239 u64 hole_len = len;
5240
172ddd60 5241 em = alloc_extent_map();
ec29ed5b
CM		 5242 if (!em) {
5243 err = -ENOMEM;
5244 goto out;
5245 }
5246 /*
5247 * when btrfs_get_extent can't find anything it
5248 * returns one huge hole
5249 *
5250 * make sure what it found really fits our range, and
5251 * adjust to make sure it is based on the start from
5252 * the caller
5253 */
5254 if (hole_em) {
5255 u64 calc_end = extent_map_end(hole_em);
5256
5257 if (calc_end <= start || (hole_em->start > end)) {
5258 free_extent_map(hole_em);
5259 hole_em = NULL;
5260 } else {
5261 hole_start = max(hole_em->start, start);
5262 hole_len = calc_end - hole_start;
5263 }
5264 }
5265 em->bdev = NULL;
5266 if (hole_em && range_start > hole_start) {
5267 /* our hole starts before our delalloc, so we
5268 * have to return just the parts of the hole
5269 * that go until the delalloc starts
5270 */
5271 em->len = min(hole_len,
5272 range_start - hole_start);
5273 em->start = hole_start;
5274 em->orig_start = hole_start;
5275 /*
5276 * don't adjust block start at all,
5277 * it is fixed at EXTENT_MAP_HOLE
5278 */
5279 em->block_start = hole_em->block_start;
5280 em->block_len = hole_len;
5281 } else {
5282 em->start = range_start;
5283 em->len = found;
5284 em->orig_start = range_start;
5285 em->block_start = EXTENT_MAP_DELALLOC;
5286 em->block_len = found;
5287 }
5288 } else if (hole_em) {
5289 return hole_em;
5290 }
5291out:
5292
5293 free_extent_map(hole_em);
5294 if (err) {
5295 free_extent_map(em);
5296 return ERR_PTR(err);
5297 }
5298 return em;
5299}
5300
4b46fce2 5301static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5302 struct extent_map *em,
4b46fce2
JB		 5303 u64 start, u64 len)
5304{
5305 struct btrfs_root *root = BTRFS_I(inode)->root;
5306 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5307 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5308 struct btrfs_key ins;
5309 u64 alloc_hint;
5310 int ret;
16d299ac 5311 bool insert = false;
4b46fce2 5312
16d299ac
JB		 5313 /*
5314 * Ok if the extent map we looked up is a hole and is for the exact
5315 * range we want, there is no reason to allocate a new one, however if
5316 * it is not right then we need to free this one and drop the cache for
5317 * our range.
5318 */
5319 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5320 em->len != len) {
5321 free_extent_map(em);
5322 em = NULL;
5323 insert = true;
5324 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5325 }
4b46fce2 5326
7a7eaa40 5327 trans = btrfs_join_transaction(root);
3612b495
TI		 5328 if (IS_ERR(trans))
5329 return ERR_CAST(trans);
4b46fce2 5330
4cb5300b
CM		 5331 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5332 btrfs_add_inode_defrag(trans, inode);
5333
4b46fce2
JB		 5334 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5335
5336 alloc_hint = get_extent_allocation_hint(inode, start, len);
5337 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5338 alloc_hint, (u64)-1, &ins, 1);
5339 if (ret) {
5340 em = ERR_PTR(ret);
5341 goto out;
5342 }
5343
4b46fce2 5344 if (!em) {
172ddd60 5345 em = alloc_extent_map();
16d299ac
JB		 5346 if (!em) {
5347 em = ERR_PTR(-ENOMEM);
5348 goto out;
5349 }
4b46fce2
JB		 5350 }
5351
5352 em->start = start;
5353 em->orig_start = em->start;
5354 em->len = ins.offset;
5355
5356 em->block_start = ins.objectid;
5357 em->block_len = ins.offset;
5358 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5359
5360 /*
5361 * We need to do this because if we're using the original em we searched
5362 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5363 */
5364 em->flags = 0;
4b46fce2
JB		 5365 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5366
16d299ac 5367 while (insert) {
4b46fce2
JB		 5368 write_lock(&em_tree->lock);
5369 ret = add_extent_mapping(em_tree, em);
5370 write_unlock(&em_tree->lock);
5371 if (ret != -EEXIST)
5372 break;
5373 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5374 }
5375
5376 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5377 ins.offset, ins.offset, 0);
5378 if (ret) {
5379 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5380 em = ERR_PTR(ret);
5381 }
5382out:
5383 btrfs_end_transaction(trans, root);
5384 return em;
5385}
5386
46bfbb5c
CM		 5387/*
5388 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5389 * block must be cow'd
5390 */
5391static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5392 struct inode *inode, u64 offset, u64 len)
5393{
5394 struct btrfs_path *path;
5395 int ret;
5396 struct extent_buffer *leaf;
5397 struct btrfs_root *root = BTRFS_I(inode)->root;
5398 struct btrfs_file_extent_item *fi;
5399 struct btrfs_key key;
5400 u64 disk_bytenr;
5401 u64 backref_offset;
5402 u64 extent_end;
5403 u64 num_bytes;
5404 int slot;
5405 int found_type;
5406
5407 path = btrfs_alloc_path();
5408 if (!path)
5409 return -ENOMEM;
5410
33345d01 5411 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5412 offset, 0);
5413 if (ret < 0)
5414 goto out;
5415
5416 slot = path->slots[0];
5417 if (ret == 1) {
5418 if (slot == 0) {
5419 /* can't find the item, must cow */
5420 ret = 0;
5421 goto out;
5422 }
5423 slot--;
5424 }
5425 ret = 0;
5426 leaf = path->nodes[0];
5427 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5428 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5429 key.type != BTRFS_EXTENT_DATA_KEY) {
5430 /* not our file or wrong item type, must cow */
5431 goto out;
5432 }
5433
5434 if (key.offset > offset) {
5435 /* Wrong offset, must cow */
5436 goto out;
5437 }
5438
5439 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5440 found_type = btrfs_file_extent_type(leaf, fi);
5441 if (found_type != BTRFS_FILE_EXTENT_REG &&
5442 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5443 /* not a regular extent, must cow */
5444 goto out;
5445 }
5446 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5447 backref_offset = btrfs_file_extent_offset(leaf, fi);
5448
5449 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5450 if (extent_end < offset + len) {
5451 /* extent doesn't include our full range, must cow */
5452 goto out;
5453 }
5454
5455 if (btrfs_extent_readonly(root, disk_bytenr))
5456 goto out;
5457
5458 /*
5459 * look for other files referencing this extent, if we
5460 * find any we must cow
5461 */
33345d01 5462 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5463 key.offset - backref_offset, disk_bytenr))
5464 goto out;
5465
5466 /*
5467 * adjust disk_bytenr and num_bytes to cover just the bytes
5468 * in this extent we are about to write. If there
5469 * are any csums in that range we have to cow in order
5470 * to keep the csums correct
5471 */
5472 disk_bytenr += backref_offset;
5473 disk_bytenr += offset - key.offset;
5474 num_bytes = min(offset + len, extent_end) - offset;
5475 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5476 goto out;
5477 /*
5478 * all of the above have passed, it is safe to overwrite this extent
5479 * without cow
5480 */
5481 ret = 1;
5482out:
5483 btrfs_free_path(path);
5484 return ret;
5485}
5486
4b46fce2
JB		 5487static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5488 struct buffer_head *bh_result, int create)
5489{
5490 struct extent_map *em;
5491 struct btrfs_root *root = BTRFS_I(inode)->root;
5492 u64 start = iblock << inode->i_blkbits;
5493 u64 len = bh_result->b_size;
46bfbb5c 5494 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5495
5496 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5497 if (IS_ERR(em))
5498 return PTR_ERR(em);
5499
5500 /*
5501 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5502 * io. INLINE is special, and we could probably kludge it in here, but
5503 * it's still buffered so for safety lets just fall back to the generic
5504 * buffered path.
5505 *
5506 * For COMPRESSED we _have_ to read the entire extent in so we can
5507 * decompress it, so there will be buffering required no matter what we
5508 * do, so go ahead and fallback to buffered.
5509 *
5510 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5511 * to buffered IO. Don't blame me, this is the price we pay for using
5512 * the generic code.
5513 */
5514 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5515 em->block_start == EXTENT_MAP_INLINE) {
5516 free_extent_map(em);
5517 return -ENOTBLK;
5518 }
5519
5520 /* Just a good old fashioned hole, return */
5521 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5522 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5523 free_extent_map(em);
5524 /* DIO will do one hole at a time, so just unlock a sector */
5525 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5526 start + root->sectorsize - 1, GFP_NOFS);
5527 return 0;
5528 }
5529
5530 /*
5531 * We don't allocate a new extent in the following cases
5532 *
5533 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5534 * existing extent.
5535 * 2) The extent is marked as PREALLOC. We're good to go here and can
5536 * just use the extent.
5537 *
5538 */
46bfbb5c
CM		 5539 if (!create) {
5540 len = em->len - (start - em->start);
4b46fce2 5541 goto map;
46bfbb5c 5542 }
4b46fce2
JB		 5543
5544 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5545 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5546 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5547 int type;
5548 int ret;
46bfbb5c 5549 u64 block_start;
4b46fce2
JB		 5550
5551 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5552 type = BTRFS_ORDERED_PREALLOC;
5553 else
5554 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5555 len = min(len, em->len - (start - em->start));
4b46fce2 5556 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5557
5558 /*
5559 * we're not going to log anything, but we do need
5560 * to make sure the current transaction stays open
5561 * while we look for nocow cross refs
5562 */
7a7eaa40 5563 trans = btrfs_join_transaction(root);
3612b495 5564 if (IS_ERR(trans))
46bfbb5c
CM		 5565 goto must_cow;
5566
5567 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5568 ret = btrfs_add_ordered_extent_dio(inode, start,
5569 block_start, len, len, type);
5570 btrfs_end_transaction(trans, root);
5571 if (ret) {
5572 free_extent_map(em);
5573 return ret;
5574 }
5575 goto unlock;
4b46fce2 5576 }
46bfbb5c 5577 btrfs_end_transaction(trans, root);
4b46fce2 5578 }
46bfbb5c
CM		 5579must_cow:
5580 /*
5581 * this will cow the extent, reset the len in case we changed
5582 * it above
5583 */
5584 len = bh_result->b_size;
16d299ac 5585 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5586 if (IS_ERR(em))
5587 return PTR_ERR(em);
5588 len = min(len, em->len - (start - em->start));
5589unlock:
4845e44f
CM		 5590 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5591 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5592 0, NULL, GFP_NOFS);
4b46fce2
JB		 5593map:
5594 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5595 inode->i_blkbits;
46bfbb5c 5596 bh_result->b_size = len;
4b46fce2
JB		 5597 bh_result->b_bdev = em->bdev;
5598 set_buffer_mapped(bh_result);
5599 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5600 set_buffer_new(bh_result);
5601
5602 free_extent_map(em);
5603
5604 return 0;
5605}
5606
5607struct btrfs_dio_private {
5608 struct inode *inode;
5609 u64 logical_offset;
5610 u64 disk_bytenr;
5611 u64 bytes;
5612 u32 *csums;
5613 void *private;
e65e1535
MX		 5614
5615 /* number of bios pending for this dio */
5616 atomic_t pending_bios;
5617
5618 /* IO errors */
5619 int errors;
5620
5621 struct bio *orig_bio;
4b46fce2
JB		 5622};
5623
5624static void btrfs_endio_direct_read(struct bio *bio, int err)
5625{
e65e1535 5626 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5627 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5628 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5629 struct inode *inode = dip->inode;
5630 struct btrfs_root *root = BTRFS_I(inode)->root;
5631 u64 start;
5632 u32 *private = dip->csums;
5633
5634 start = dip->logical_offset;
5635 do {
5636 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5637 struct page *page = bvec->bv_page;
5638 char *kaddr;
5639 u32 csum = ~(u32)0;
5640 unsigned long flags;
5641
5642 local_irq_save(flags);
5643 kaddr = kmap_atomic(page, KM_IRQ0);
5644 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5645 csum, bvec->bv_len);
5646 btrfs_csum_final(csum, (char *)&csum);
5647 kunmap_atomic(kaddr, KM_IRQ0);
5648 local_irq_restore(flags);
5649
5650 flush_dcache_page(bvec->bv_page);
5651 if (csum != *private) {
33345d01 5652 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5653 " %llu csum %u private %u\n",
33345d01
LZ		 5654 (unsigned long long)btrfs_ino(inode),
5655 (unsigned long long)start,
4b46fce2
JB		 5656 csum, *private);
5657 err = -EIO;
5658 }
5659 }
5660
5661 start += bvec->bv_len;
5662 private++;
5663 bvec++;
5664 } while (bvec <= bvec_end);
5665
5666 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5667 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5668 bio->bi_private = dip->private;
5669
5670 kfree(dip->csums);
5671 kfree(dip);
c0da7aa1
JB		 5672
5673 /* If we had a csum failure make sure to clear the uptodate flag */
5674 if (err)
5675 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5676 dio_end_io(bio, err);
5677}
5678
5679static void btrfs_endio_direct_write(struct bio *bio, int err)
5680{
5681 struct btrfs_dio_private *dip = bio->bi_private;
5682 struct inode *inode = dip->inode;
5683 struct btrfs_root *root = BTRFS_I(inode)->root;
5684 struct btrfs_trans_handle *trans;
5685 struct btrfs_ordered_extent *ordered = NULL;
5686 struct extent_state *cached_state = NULL;
163cf09c
CM		 5687 u64 ordered_offset = dip->logical_offset;
5688 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5689 int ret;
5690
5691 if (err)
5692 goto out_done;
163cf09c
CM		 5693again:
5694 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5695 &ordered_offset,
5696 ordered_bytes);
4b46fce2 5697 if (!ret)
163cf09c 5698 goto out_test;
4b46fce2
JB		 5699
5700 BUG_ON(!ordered);
5701
7a7eaa40 5702 trans = btrfs_join_transaction(root);
3612b495 5703 if (IS_ERR(trans)) {
4b46fce2
JB		 5704 err = -ENOMEM;
5705 goto out;
5706 }
5707 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5708
5709 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5710 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5711 if (!ret)
5712 ret = btrfs_update_inode(trans, root, inode);
5713 err = ret;
5714 goto out;
5715 }
5716
5717 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5718 ordered->file_offset + ordered->len - 1, 0,
5719 &cached_state, GFP_NOFS);
5720
5721 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5722 ret = btrfs_mark_extent_written(trans, inode,
5723 ordered->file_offset,
5724 ordered->file_offset +
5725 ordered->len);
5726 if (ret) {
5727 err = ret;
5728 goto out_unlock;
5729 }
5730 } else {
5731 ret = insert_reserved_file_extent(trans, inode,
5732 ordered->file_offset,
5733 ordered->start,
5734 ordered->disk_len,
5735 ordered->len,
5736 ordered->len,
5737 0, 0, 0,
5738 BTRFS_FILE_EXTENT_REG);
5739 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5740 ordered->file_offset, ordered->len);
5741 if (ret) {
5742 err = ret;
5743 WARN_ON(1);
5744 goto out_unlock;
5745 }
5746 }
5747
5748 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
1ef30be1 5749 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
a39f7521 5750 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
1ef30be1
JB		 5751 btrfs_update_inode(trans, root, inode);
5752 ret = 0;
4b46fce2
JB		 5753out_unlock:
5754 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5755 ordered->file_offset + ordered->len - 1,
5756 &cached_state, GFP_NOFS);
5757out:
5758 btrfs_delalloc_release_metadata(inode, ordered->len);
5759 btrfs_end_transaction(trans, root);
163cf09c 5760 ordered_offset = ordered->file_offset + ordered->len;
4b46fce2
JB		 5761 btrfs_put_ordered_extent(ordered);
5762 btrfs_put_ordered_extent(ordered);
163cf09c
CM		 5763
5764out_test:
5765 /*
5766 * our bio might span multiple ordered extents. If we haven't
5767 * completed the accounting for the whole dio, go back and try again
5768 */
5769 if (ordered_offset < dip->logical_offset + dip->bytes) {
5770 ordered_bytes = dip->logical_offset + dip->bytes -
5771 ordered_offset;
5772 goto again;
5773 }
4b46fce2
JB		 5774out_done:
5775 bio->bi_private = dip->private;
5776
5777 kfree(dip->csums);
5778 kfree(dip);
c0da7aa1
JB		 5779
5780 /* If we had an error make sure to clear the uptodate flag */
5781 if (err)
5782 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5783 dio_end_io(bio, err);
5784}
5785
eaf25d93
CM		 5786static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5787 struct bio *bio, int mirror_num,
5788 unsigned long bio_flags, u64 offset)
5789{
5790 int ret;
5791 struct btrfs_root *root = BTRFS_I(inode)->root;
5792 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5793 BUG_ON(ret);
5794 return 0;
5795}
5796
e65e1535
MX		 5797static void btrfs_end_dio_bio(struct bio *bio, int err)
5798{
5799 struct btrfs_dio_private *dip = bio->bi_private;
5800
5801 if (err) {
33345d01 5802 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 5803 "sector %#Lx len %u err no %d\n",
33345d01 5804 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 5805 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 5806 dip->errors = 1;
5807
5808 /*
5809 * before atomic variable goto zero, we must make sure
5810 * dip->errors is perceived to be set.
5811 */
5812 smp_mb__before_atomic_dec();
5813 }
5814
5815 /* if there are more bios still pending for this dio, just exit */
5816 if (!atomic_dec_and_test(&dip->pending_bios))
5817 goto out;
5818
5819 if (dip->errors)
5820 bio_io_error(dip->orig_bio);
5821 else {
5822 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5823 bio_endio(dip->orig_bio, 0);
5824 }
5825out:
5826 bio_put(bio);
5827}
5828
5829static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5830 u64 first_sector, gfp_t gfp_flags)
5831{
5832 int nr_vecs = bio_get_nr_vecs(bdev);
5833 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5834}
5835
5836static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5837 int rw, u64 file_offset, int skip_sum,
1ae39938 5838 u32 *csums, int async_submit)
e65e1535
MX		 5839{
5840 int write = rw & REQ_WRITE;
5841 struct btrfs_root *root = BTRFS_I(inode)->root;
5842 int ret;
5843
5844 bio_get(bio);
5845 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5846 if (ret)
5847 goto err;
5848
1ae39938
JB		 5849 if (skip_sum)
5850 goto map;
5851
5852 if (write && async_submit) {
e65e1535
MX		 5853 ret = btrfs_wq_submit_bio(root->fs_info,
5854 inode, rw, bio, 0, 0,
5855 file_offset,
5856 __btrfs_submit_bio_start_direct_io,
5857 __btrfs_submit_bio_done);
5858 goto err;
1ae39938
JB		 5859 } else if (write) {
5860 /*
5861 * If we aren't doing async submit, calculate the csum of the
5862 * bio now.
5863 */
5864 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5865 if (ret)
5866 goto err;
c2db1073
TI		 5867 } else if (!skip_sum) {
5868 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 5869 file_offset, csums);
c2db1073
TI		 5870 if (ret)
5871 goto err;
5872 }
e65e1535 5873
1ae39938
JB		 5874map:
5875 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 5876err:
5877 bio_put(bio);
5878 return ret;
5879}
5880
5881static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5882 int skip_sum)
5883{
5884 struct inode *inode = dip->inode;
5885 struct btrfs_root *root = BTRFS_I(inode)->root;
5886 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5887 struct bio *bio;
5888 struct bio *orig_bio = dip->orig_bio;
5889 struct bio_vec *bvec = orig_bio->bi_io_vec;
5890 u64 start_sector = orig_bio->bi_sector;
5891 u64 file_offset = dip->logical_offset;
5892 u64 submit_len = 0;
5893 u64 map_length;
5894 int nr_pages = 0;
5895 u32 *csums = dip->csums;
5896 int ret = 0;
1ae39938 5897 int async_submit = 0;
98bc3149 5898 int write = rw & REQ_WRITE;
e65e1535 5899
e65e1535
MX		 5900 map_length = orig_bio->bi_size;
5901 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5902 &map_length, NULL, 0);
5903 if (ret) {
64728bbb 5904 bio_put(orig_bio);
e65e1535
MX		 5905 return -EIO;
5906 }
5907
02f57c7a
JB		 5908 if (map_length >= orig_bio->bi_size) {
5909 bio = orig_bio;
5910 goto submit;
5911 }
5912
1ae39938 5913 async_submit = 1;
02f57c7a
JB		 5914 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5915 if (!bio)
5916 return -ENOMEM;
5917 bio->bi_private = dip;
5918 bio->bi_end_io = btrfs_end_dio_bio;
5919 atomic_inc(&dip->pending_bios);
5920
e65e1535
MX		 5921 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5922 if (unlikely(map_length < submit_len + bvec->bv_len ||
5923 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5924 bvec->bv_offset) < bvec->bv_len)) {
5925 /*
5926 * inc the count before we submit the bio so
5927 * we know the end IO handler won't happen before
5928 * we inc the count. Otherwise, the dip might get freed
5929 * before we're done setting it up
5930 */
5931 atomic_inc(&dip->pending_bios);
5932 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5933 file_offset, skip_sum,
1ae39938 5934 csums, async_submit);
e65e1535
MX		 5935 if (ret) {
5936 bio_put(bio);
5937 atomic_dec(&dip->pending_bios);
5938 goto out_err;
5939 }
5940
98bc3149
JB		 5941 /* Write's use the ordered csums */
5942 if (!write && !skip_sum)
e65e1535
MX		 5943 csums = csums + nr_pages;
5944 start_sector += submit_len >> 9;
5945 file_offset += submit_len;
5946
5947 submit_len = 0;
5948 nr_pages = 0;
5949
5950 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5951 start_sector, GFP_NOFS);
5952 if (!bio)
5953 goto out_err;
5954 bio->bi_private = dip;
5955 bio->bi_end_io = btrfs_end_dio_bio;
5956
5957 map_length = orig_bio->bi_size;
5958 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5959 &map_length, NULL, 0);
5960 if (ret) {
5961 bio_put(bio);
5962 goto out_err;
5963 }
5964 } else {
5965 submit_len += bvec->bv_len;
5966 nr_pages ++;
5967 bvec++;
5968 }
5969 }
5970
02f57c7a 5971submit:
e65e1535 5972 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 5973 csums, async_submit);
e65e1535
MX		 5974 if (!ret)
5975 return 0;
5976
5977 bio_put(bio);
5978out_err:
5979 dip->errors = 1;
5980 /*
5981 * before atomic variable goto zero, we must
5982 * make sure dip->errors is perceived to be set.
5983 */
5984 smp_mb__before_atomic_dec();
5985 if (atomic_dec_and_test(&dip->pending_bios))
5986 bio_io_error(dip->orig_bio);
5987
5988 /* bio_end_io() will handle error, so we needn't return it */
5989 return 0;
5990}
5991
4b46fce2
JB		 5992static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5993 loff_t file_offset)
5994{
5995 struct btrfs_root *root = BTRFS_I(inode)->root;
5996 struct btrfs_dio_private *dip;
5997 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 5998 int skip_sum;
7b6d91da 5999 int write = rw & REQ_WRITE;
4b46fce2
JB		 6000 int ret = 0;
6001
6002 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6003
6004 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6005 if (!dip) {
6006 ret = -ENOMEM;
6007 goto free_ordered;
6008 }
6009 dip->csums = NULL;
6010
98bc3149
JB		 6011 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6012 if (!write && !skip_sum) {
4b46fce2
JB		 6013 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6014 if (!dip->csums) {
b4966b77 6015 kfree(dip);
4b46fce2
JB		 6016 ret = -ENOMEM;
6017 goto free_ordered;
6018 }
6019 }
6020
6021 dip->private = bio->bi_private;
6022 dip->inode = inode;
6023 dip->logical_offset = file_offset;
6024
4b46fce2
JB		 6025 dip->bytes = 0;
6026 do {
6027 dip->bytes += bvec->bv_len;
6028 bvec++;
6029 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6030
46bfbb5c 6031 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6032 bio->bi_private = dip;
e65e1535
MX		 6033 dip->errors = 0;
6034 dip->orig_bio = bio;
6035 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6036
6037 if (write)
6038 bio->bi_end_io = btrfs_endio_direct_write;
6039 else
6040 bio->bi_end_io = btrfs_endio_direct_read;
6041
e65e1535
MX		 6042 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6043 if (!ret)
eaf25d93 6044 return;
4b46fce2
JB		 6045free_ordered:
6046 /*
6047 * If this is a write, we need to clean up the reserved space and kill
6048 * the ordered extent.
6049 */
6050 if (write) {
6051 struct btrfs_ordered_extent *ordered;
955256f2 6052 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6053 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6054 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6055 btrfs_free_reserved_extent(root, ordered->start,
6056 ordered->disk_len);
6057 btrfs_put_ordered_extent(ordered);
6058 btrfs_put_ordered_extent(ordered);
6059 }
6060 bio_endio(bio, ret);
6061}
6062
5a5f79b5
CM		 6063static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6064 const struct iovec *iov, loff_t offset,
6065 unsigned long nr_segs)
6066{
6067 int seg;
a1b75f7d 6068 int i;
5a5f79b5
CM		 6069 size_t size;
6070 unsigned long addr;
6071 unsigned blocksize_mask = root->sectorsize - 1;
6072 ssize_t retval = -EINVAL;
6073 loff_t end = offset;
6074
6075 if (offset & blocksize_mask)
6076 goto out;
6077
6078 /* Check the memory alignment. Blocks cannot straddle pages */
6079 for (seg = 0; seg < nr_segs; seg++) {
6080 addr = (unsigned long)iov[seg].iov_base;
6081 size = iov[seg].iov_len;
6082 end += size;
a1b75f7d 6083 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6084 goto out;
a1b75f7d
JB		 6085
6086 /* If this is a write we don't need to check anymore */
6087 if (rw & WRITE)
6088 continue;
6089
6090 /*
6091 * Check to make sure we don't have duplicate iov_base's in this
6092 * iovec, if so return EINVAL, otherwise we'll get csum errors
6093 * when reading back.
6094 */
6095 for (i = seg + 1; i < nr_segs; i++) {
6096 if (iov[seg].iov_base == iov[i].iov_base)
6097 goto out;
6098 }
5a5f79b5
CM		 6099 }
6100 retval = 0;
6101out:
6102 return retval;
6103}
16432985
CM		 6104static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6105 const struct iovec *iov, loff_t offset,
6106 unsigned long nr_segs)
6107{
4b46fce2
JB		 6108 struct file *file = iocb->ki_filp;
6109 struct inode *inode = file->f_mapping->host;
6110 struct btrfs_ordered_extent *ordered;
4845e44f 6111 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6112 u64 lockstart, lockend;
6113 ssize_t ret;
4845e44f
CM		 6114 int writing = rw & WRITE;
6115 int write_bits = 0;
3f7c579c 6116 size_t count = iov_length(iov, nr_segs);
4b46fce2 6117
5a5f79b5
CM		 6118 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6119 offset, nr_segs)) {
6120 return 0;
6121 }
6122
4b46fce2 6123 lockstart = offset;
3f7c579c
CM		 6124 lockend = offset + count - 1;
6125
6126 if (writing) {
6127 ret = btrfs_delalloc_reserve_space(inode, count);
6128 if (ret)
6129 goto out;
6130 }
4845e44f 6131
4b46fce2 6132 while (1) {
4845e44f
CM		 6133 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6134 0, &cached_state, GFP_NOFS);
4b46fce2
JB		 6135 /*
6136 * We're concerned with the entire range that we're going to be
6137 * doing DIO to, so we need to make sure theres no ordered
6138 * extents in this range.
6139 */
6140 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6141 lockend - lockstart + 1);
6142 if (!ordered)
6143 break;
4845e44f
CM		 6144 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6145 &cached_state, GFP_NOFS);
4b46fce2
JB		 6146 btrfs_start_ordered_extent(inode, ordered, 1);
6147 btrfs_put_ordered_extent(ordered);
6148 cond_resched();
6149 }
6150
4845e44f
CM		 6151 /*
6152 * we don't use btrfs_set_extent_delalloc because we don't want
6153 * the dirty or uptodate bits
6154 */
6155 if (writing) {
6156 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6157 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6158 EXTENT_DELALLOC, 0, NULL, &cached_state,
6159 GFP_NOFS);
6160 if (ret) {
6161 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6162 lockend, EXTENT_LOCKED | write_bits,
6163 1, 0, &cached_state, GFP_NOFS);
6164 goto out;
6165 }
6166 }
6167
6168 free_extent_state(cached_state);
6169 cached_state = NULL;
6170
5a5f79b5
CM		 6171 ret = __blockdev_direct_IO(rw, iocb, inode,
6172 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6173 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6174 btrfs_submit_direct, 0);
4b46fce2
JB		 6175
6176 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6177 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6178 offset + iov_length(iov, nr_segs) - 1,
6179 EXTENT_LOCKED | write_bits, 1, 0,
6180 &cached_state, GFP_NOFS);
4b46fce2
JB		 6181 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6182 /*
6183 * We're falling back to buffered, unlock the section we didn't
6184 * do IO on.
6185 */
4845e44f
CM		 6186 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6187 offset + iov_length(iov, nr_segs) - 1,
6188 EXTENT_LOCKED | write_bits, 1, 0,
6189 &cached_state, GFP_NOFS);
4b46fce2 6190 }
4845e44f
CM		 6191out:
6192 free_extent_state(cached_state);
4b46fce2 6193 return ret;
16432985
CM		 6194}
6195
1506fcc8
YS		 6196static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6197 __u64 start, __u64 len)
6198{
ec29ed5b 6199 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6200}
6201
a52d9a80 6202int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6203{
d1310b2e
CM		 6204 struct extent_io_tree *tree;
6205 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6206 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 6207}
1832a6d5 6208
a52d9a80 6209static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6210{
d1310b2e 6211 struct extent_io_tree *tree;
b888db2b
CM		 6212
6213
6214 if (current->flags & PF_MEMALLOC) {
6215 redirty_page_for_writepage(wbc, page);
6216 unlock_page(page);
6217 return 0;
6218 }
d1310b2e 6219 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6220 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6221}
6222
f421950f
CM		 6223int btrfs_writepages(struct address_space *mapping,
6224 struct writeback_control *wbc)
b293f02e 6225{
d1310b2e 6226 struct extent_io_tree *tree;
771ed689 6227
d1310b2e 6228 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6229 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6230}
6231
3ab2fb5a
CM		 6232static int
6233btrfs_readpages(struct file *file, struct address_space *mapping,
6234 struct list_head *pages, unsigned nr_pages)
6235{
d1310b2e
CM		 6236 struct extent_io_tree *tree;
6237 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6238 return extent_readpages(tree, mapping, pages, nr_pages,
6239 btrfs_get_extent);
6240}
e6dcd2dc 6241static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6242{
d1310b2e
CM		 6243 struct extent_io_tree *tree;
6244 struct extent_map_tree *map;
a52d9a80 6245 int ret;
8c2383c3 6246
d1310b2e
CM		 6247 tree = &BTRFS_I(page->mapping->host)->io_tree;
6248 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6249 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6250 if (ret == 1) {
6251 ClearPagePrivate(page);
6252 set_page_private(page, 0);
6253 page_cache_release(page);
39279cc3 6254 }
a52d9a80 6255 return ret;
39279cc3
CM		 6256}
6257
e6dcd2dc
CM		 6258static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6259{
98509cfc
CM		 6260 if (PageWriteback(page) || PageDirty(page))
6261 return 0;
b335b003 6262 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6263}
6264
a52d9a80 6265static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6266{
d1310b2e 6267 struct extent_io_tree *tree;
e6dcd2dc 6268 struct btrfs_ordered_extent *ordered;
2ac55d41 6269 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6270 u64 page_start = page_offset(page);
6271 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6272
8b62b72b
CM		 6273
6274 /*
6275 * we have the page locked, so new writeback can't start,
6276 * and the dirty bit won't be cleared while we are here.
6277 *
6278 * Wait for IO on this page so that we can safely clear
6279 * the PagePrivate2 bit and do ordered accounting
6280 */
e6dcd2dc 6281 wait_on_page_writeback(page);
8b62b72b 6282
d1310b2e 6283 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 6284 if (offset) {
6285 btrfs_releasepage(page, GFP_NOFS);
6286 return;
6287 }
2ac55d41
JB		 6288 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6289 GFP_NOFS);
e6dcd2dc
CM		 6290 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6291 page_offset(page));
6292 if (ordered) {
eb84ae03
CM		 6293 /*
6294 * IO on this page will never be started, so we need
6295 * to account for any ordered extents now
6296 */
e6dcd2dc
CM		 6297 clear_extent_bit(tree, page_start, page_end,
6298 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6299 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6300 &cached_state, GFP_NOFS);
8b62b72b
CM		 6301 /*
6302 * whoever cleared the private bit is responsible
6303 * for the finish_ordered_io
6304 */
6305 if (TestClearPagePrivate2(page)) {
6306 btrfs_finish_ordered_io(page->mapping->host,
6307 page_start, page_end);
6308 }
e6dcd2dc 6309 btrfs_put_ordered_extent(ordered);
2ac55d41
JB		 6310 cached_state = NULL;
6311 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6312 GFP_NOFS);
e6dcd2dc
CM		 6313 }
6314 clear_extent_bit(tree, page_start, page_end,
32c00aff 6315 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6316 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6317 __btrfs_releasepage(page, GFP_NOFS);
6318
4a096752 6319 ClearPageChecked(page);
9ad6b7bc 6320 if (PagePrivate(page)) {
9ad6b7bc
CM		 6321 ClearPagePrivate(page);
6322 set_page_private(page, 0);
6323 page_cache_release(page);
6324 }
39279cc3
CM		 6325}
6326
9ebefb18
CM		 6327/*
6328 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6329 * called from a page fault handler when a page is first dirtied. Hence we must
6330 * be careful to check for EOF conditions here. We set the page up correctly
6331 * for a written page which means we get ENOSPC checking when writing into
6332 * holes and correct delalloc and unwritten extent mapping on filesystems that
6333 * support these features.
6334 *
6335 * We are not allowed to take the i_mutex here so we have to play games to
6336 * protect against truncate races as the page could now be beyond EOF. Because
6337 * vmtruncate() writes the inode size before removing pages, once we have the
6338 * page lock we can determine safely if the page is beyond EOF. If it is not
6339 * beyond EOF, then the page is guaranteed safe against truncation until we
6340 * unlock the page.
6341 */
c2ec175c 6342int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6343{
c2ec175c 6344 struct page *page = vmf->page;
6da6abae 6345 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6346 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6347 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6348 struct btrfs_ordered_extent *ordered;
2ac55d41 6349 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6350 char *kaddr;
6351 unsigned long zero_start;
9ebefb18 6352 loff_t size;
1832a6d5 6353 int ret;
a52d9a80 6354 u64 page_start;
e6dcd2dc 6355 u64 page_end;
9ebefb18 6356
0ca1f7ce 6357 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
56a76f82
NP		 6358 if (ret) {
6359 if (ret == -ENOMEM)
6360 ret = VM_FAULT_OOM;
6361 else /* -ENOSPC, -EIO, etc */
6362 ret = VM_FAULT_SIGBUS;
1832a6d5 6363 goto out;
56a76f82 6364 }
1832a6d5 6365
56a76f82 6366 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6367again:
9ebefb18 6368 lock_page(page);
9ebefb18 6369 size = i_size_read(inode);
e6dcd2dc
CM		 6370 page_start = page_offset(page);
6371 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6372
9ebefb18 6373 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6374 (page_start >= size)) {
9ebefb18
CM		 6375 /* page got truncated out from underneath us */
6376 goto out_unlock;
6377 }
e6dcd2dc
CM		 6378 wait_on_page_writeback(page);
6379
2ac55d41
JB		 6380 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6381 GFP_NOFS);
e6dcd2dc
CM		 6382 set_page_extent_mapped(page);
6383
eb84ae03
CM		 6384 /*
6385 * we can't set the delalloc bits if there are pending ordered
6386 * extents. Drop our locks and wait for them to finish
6387 */
e6dcd2dc
CM		 6388 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6389 if (ordered) {
2ac55d41
JB		 6390 unlock_extent_cached(io_tree, page_start, page_end,
6391 &cached_state, GFP_NOFS);
e6dcd2dc 6392 unlock_page(page);
eb84ae03 6393 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6394 btrfs_put_ordered_extent(ordered);
6395 goto again;
6396 }
6397
fbf19087
JB		 6398 /*
6399 * XXX - page_mkwrite gets called every time the page is dirtied, even
6400 * if it was already dirty, so for space accounting reasons we need to
6401 * clear any delalloc bits for the range we are fixing to save. There
6402 * is probably a better way to do this, but for now keep consistent with
6403 * prepare_pages in the normal write path.
6404 */
2ac55d41 6405 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6406 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6407 0, 0, &cached_state, GFP_NOFS);
fbf19087 6408
2ac55d41
JB		 6409 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6410 &cached_state);
9ed74f2d 6411 if (ret) {
2ac55d41
JB		 6412 unlock_extent_cached(io_tree, page_start, page_end,
6413 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6414 ret = VM_FAULT_SIGBUS;
6415 goto out_unlock;
6416 }
e6dcd2dc 6417 ret = 0;
9ebefb18
CM		 6418
6419 /* page is wholly or partially inside EOF */
a52d9a80 6420 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6421 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6422 else
e6dcd2dc 6423 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6424
e6dcd2dc
CM		 6425 if (zero_start != PAGE_CACHE_SIZE) {
6426 kaddr = kmap(page);
6427 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6428 flush_dcache_page(page);
6429 kunmap(page);
6430 }
247e743c 6431 ClearPageChecked(page);
e6dcd2dc 6432 set_page_dirty(page);
50a9b214 6433 SetPageUptodate(page);
5a3f23d5 6434
257c62e1
CM		 6435 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6436 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6437
2ac55d41 6438 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6439
6440out_unlock:
50a9b214
CM		 6441 if (!ret)
6442 return VM_FAULT_LOCKED;
9ebefb18 6443 unlock_page(page);
0ca1f7ce 6444 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
1832a6d5 6445out:
9ebefb18
CM		 6446 return ret;
6447}
6448
a41ad394 6449static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6450{
6451 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6452 struct btrfs_block_rsv *rsv;
39279cc3 6453 int ret;
3893e33b 6454 int err = 0;
39279cc3 6455 struct btrfs_trans_handle *trans;
d3c2fdcf 6456 unsigned long nr;
dbe674a9 6457 u64 mask = root->sectorsize - 1;
39279cc3 6458
5d5e103a
JB		 6459 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6460 if (ret)
a41ad394 6461 return ret;
8082510e 6462
4a096752 6463 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6464 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6465
fcb80c2a
JB		 6466 /*
6467 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6468 * 3 things going on here
6469 *
6470 * 1) We need to reserve space for our orphan item and the space to
6471 * delete our orphan item. Lord knows we don't want to have a dangling
6472 * orphan item because we didn't reserve space to remove it.
6473 *
6474 * 2) We need to reserve space to update our inode.
6475 *
6476 * 3) We need to have something to cache all the space that is going to
6477 * be free'd up by the truncate operation, but also have some slack
6478 * space reserved in case it uses space during the truncate (thank you
6479 * very much snapshotting).
6480 *
6481 * And we need these to all be seperate. The fact is we can use alot of
6482 * space doing the truncate, and we have no earthly idea how much space
6483 * we will use, so we need the truncate reservation to be seperate so it
6484 * doesn't end up using space reserved for updating the inode or
6485 * removing the orphan item. We also need to be able to stop the
6486 * transaction and start a new one, which means we need to be able to
6487 * update the inode several times, and we have no idea of knowing how
6488 * many times that will be, so we can't just reserve 1 item for the
6489 * entirety of the opration, so that has to be done seperately as well.
6490 * Then there is the orphan item, which does indeed need to be held on
6491 * to for the whole operation, and we need nobody to touch this reserved
6492 * space except the orphan code.
6493 *
6494 * So that leaves us with
6495 *
6496 * 1) root->orphan_block_rsv - for the orphan deletion.
6497 * 2) rsv - for the truncate reservation, which we will steal from the
6498 * transaction reservation.
6499 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6500 * updating the inode.
6501 */
6502 rsv = btrfs_alloc_block_rsv(root);
6503 if (!rsv)
6504 return -ENOMEM;
f0cd846e 6505
fcb80c2a
JB		 6506 trans = btrfs_start_transaction(root, 4);
6507 if (IS_ERR(trans)) {
6508 err = PTR_ERR(trans);
6509 goto out;
6510 }
f0cd846e 6511
fcb80c2a
JB		 6512 /*
6513 * Reserve space for the truncate process. Truncate should be adding
6514 * space, but if there are snapshots it may end up using space.
6515 */
6516 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6517 BUG_ON(ret);
f0cd846e
JB		 6518
6519 ret = btrfs_orphan_add(trans, inode);
6520 if (ret) {
6521 btrfs_end_transaction(trans, root);
fcb80c2a 6522 goto out;
f0cd846e
JB		 6523 }
6524
6525 nr = trans->blocks_used;
6526 btrfs_end_transaction(trans, root);
6527 btrfs_btree_balance_dirty(root, nr);
6528
fcb80c2a
JB		 6529 /*
6530 * Ok so we've already migrated our bytes over for the truncate, so here
6531 * just reserve the one slot we need for updating the inode.
6532 */
6533 trans = btrfs_start_transaction(root, 1);
6534 if (IS_ERR(trans)) {
6535 err = PTR_ERR(trans);
6536 goto out;
6537 }
fcb80c2a 6538 trans->block_rsv = rsv;
5a3f23d5
CM		 6539
6540 /*
6541 * setattr is responsible for setting the ordered_data_close flag,
6542 * but that is only tested during the last file release. That
6543 * could happen well after the next commit, leaving a great big
6544 * window where new writes may get lost if someone chooses to write
6545 * to this file after truncating to zero
6546 *
6547 * The inode doesn't have any dirty data here, and so if we commit
6548 * this is a noop. If someone immediately starts writing to the inode
6549 * it is very likely we'll catch some of their writes in this
6550 * transaction, and the commit will find this file on the ordered
6551 * data list with good things to send down.
6552 *
6553 * This is a best effort solution, there is still a window where
6554 * using truncate to replace the contents of the file will
6555 * end up with a zero length file after a crash.
6556 */
6557 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6558 btrfs_add_ordered_operation(trans, root, inode);
6559
8082510e 6560 while (1) {
d68fc57b 6561 if (!trans) {
fcb80c2a
JB		 6562 trans = btrfs_start_transaction(root, 3);
6563 if (IS_ERR(trans)) {
6564 err = PTR_ERR(trans);
6565 goto out;
6566 }
d68fc57b 6567
fcb80c2a
JB		 6568 ret = btrfs_truncate_reserve_metadata(trans, root,
6569 rsv);
6570 BUG_ON(ret);
6571
fcb80c2a 6572 trans->block_rsv = rsv;
d68fc57b
YZ		 6573 }
6574
8082510e
YZ		 6575 ret = btrfs_truncate_inode_items(trans, root, inode,
6576 inode->i_size,
6577 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6578 if (ret != -EAGAIN) {
6579 err = ret;
8082510e 6580 break;
3893e33b 6581 }
39279cc3 6582
fcb80c2a 6583 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6584 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6585 if (ret) {
6586 err = ret;
6587 break;
6588 }
5f39d397 6589
8082510e
YZ		 6590 nr = trans->blocks_used;
6591 btrfs_end_transaction(trans, root);
d68fc57b 6592 trans = NULL;
8082510e 6593 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6594 }
6595
6596 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6597 trans->block_rsv = root->orphan_block_rsv;
8082510e 6598 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6599 if (ret)
6600 err = ret;
ded5db9d
JB		 6601 } else if (ret && inode->i_nlink > 0) {
6602 /*
6603 * Failed to do the truncate, remove us from the in memory
6604 * orphan list.
6605 */
6606 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6607 }
6608
fcb80c2a 6609 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6610 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6611 if (ret && !err)
6612 err = ret;
7b128766 6613
7b128766 6614 nr = trans->blocks_used;
89ce8a63 6615 ret = btrfs_end_transaction_throttle(trans, root);
fcb80c2a
JB		 6616 btrfs_btree_balance_dirty(root, nr);
6617
6618out:
6619 btrfs_free_block_rsv(root, rsv);
6620
3893e33b
JB		 6621 if (ret && !err)
6622 err = ret;
a41ad394 6623
3893e33b 6624 return err;
39279cc3
CM		 6625}
6626
d352ac68
CM		 6627/*
6628 * create a new subvolume directory/inode (helper for the ioctl).
6629 */
d2fb3437 6630int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6631 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6632{
39279cc3 6633 struct inode *inode;
76dda93c 6634 int err;
00e4e6b3 6635 u64 index = 0;
39279cc3 6636
aec7477b 6637 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d82a6f1d 6638 new_dirid, S_IFDIR | 0700, &index);
54aa1f4d 6639 if (IS_ERR(inode))
f46b5a66 6640 return PTR_ERR(inode);
39279cc3
CM		 6641 inode->i_op = &btrfs_dir_inode_operations;
6642 inode->i_fop = &btrfs_dir_file_operations;
6643
39279cc3 6644 inode->i_nlink = 1;
dbe674a9 6645 btrfs_i_size_write(inode, 0);
3b96362c 6646
76dda93c
YZ		 6647 err = btrfs_update_inode(trans, new_root, inode);
6648 BUG_ON(err);
cb8e7090 6649
76dda93c 6650 iput(inode);
cb8e7090 6651 return 0;
39279cc3
CM		 6652}
6653
39279cc3
CM		 6654struct inode *btrfs_alloc_inode(struct super_block *sb)
6655{
6656 struct btrfs_inode *ei;
2ead6ae7 6657 struct inode *inode;
39279cc3
CM		 6658
6659 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6660 if (!ei)
6661 return NULL;
2ead6ae7
YZ		 6662
6663 ei->root = NULL;
6664 ei->space_info = NULL;
6665 ei->generation = 0;
6666 ei->sequence = 0;
15ee9bc7 6667 ei->last_trans = 0;
257c62e1 6668 ei->last_sub_trans = 0;
e02119d5 6669 ei->logged_trans = 0;
2ead6ae7 6670 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6671 ei->disk_i_size = 0;
6672 ei->flags = 0;
7709cde3 6673 ei->csum_bytes = 0;
2ead6ae7
YZ		 6674 ei->index_cnt = (u64)-1;
6675 ei->last_unlink_trans = 0;
6676
9e0baf60
JB		 6677 spin_lock_init(&ei->lock);
6678 ei->outstanding_extents = 0;
6679 ei->reserved_extents = 0;
2ead6ae7
YZ		 6680
6681 ei->ordered_data_close = 0;
d68fc57b 6682 ei->orphan_meta_reserved = 0;
2ead6ae7 6683 ei->dummy_inode = 0;
4cb5300b 6684 ei->in_defrag = 0;
261507a0 6685 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6686
16cdcec7
MX		 6687 ei->delayed_node = NULL;
6688
2ead6ae7 6689 inode = &ei->vfs_inode;
a8067e02 6690 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6691 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6692 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
2ead6ae7 6693 mutex_init(&ei->log_mutex);
e6dcd2dc 6694 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7b128766 6695 INIT_LIST_HEAD(&ei->i_orphan);
2ead6ae7 6696 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6697 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6698 RB_CLEAR_NODE(&ei->rb_node);
6699
6700 return inode;
39279cc3
CM		 6701}
6702
fa0d7e3d
NP		 6703static void btrfs_i_callback(struct rcu_head *head)
6704{
6705 struct inode *inode = container_of(head, struct inode, i_rcu);
6706 INIT_LIST_HEAD(&inode->i_dentry);
6707 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6708}
6709
39279cc3
CM		 6710void btrfs_destroy_inode(struct inode *inode)
6711{
e6dcd2dc 6712 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6713 struct btrfs_root *root = BTRFS_I(inode)->root;
6714
39279cc3
CM		 6715 WARN_ON(!list_empty(&inode->i_dentry));
6716 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6717 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6718 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6719 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6720 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6721
a6dbd429
JB		 6722 /*
6723 * This can happen where we create an inode, but somebody else also
6724 * created the same inode and we need to destroy the one we already
6725 * created.
6726 */
6727 if (!root)
6728 goto free;
6729
5a3f23d5
CM		 6730 /*
6731 * Make sure we're properly removed from the ordered operation
6732 * lists.
6733 */
6734 smp_mb();
6735 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6736 spin_lock(&root->fs_info->ordered_extent_lock);
6737 list_del_init(&BTRFS_I(inode)->ordered_operations);
6738 spin_unlock(&root->fs_info->ordered_extent_lock);
6739 }
6740
d68fc57b 6741 spin_lock(&root->orphan_lock);
7b128766 6742 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
33345d01
LZ		 6743 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6744 (unsigned long long)btrfs_ino(inode));
8082510e 6745 list_del_init(&BTRFS_I(inode)->i_orphan);
7b128766 6746 }
d68fc57b 6747 spin_unlock(&root->orphan_lock);
7b128766 6748
d397712b 6749 while (1) {
e6dcd2dc
CM		 6750 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6751 if (!ordered)
6752 break;
6753 else {
d397712b
CM		 6754 printk(KERN_ERR "btrfs found ordered "
6755 "extent %llu %llu on inode cleanup\n",
6756 (unsigned long long)ordered->file_offset,
6757 (unsigned long long)ordered->len);
e6dcd2dc
CM		 6758 btrfs_remove_ordered_extent(inode, ordered);
6759 btrfs_put_ordered_extent(ordered);
6760 btrfs_put_ordered_extent(ordered);
6761 }
6762 }
5d4f98a2 6763 inode_tree_del(inode);
5b21f2ed 6764 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 6765free:
16cdcec7 6766 btrfs_remove_delayed_node(inode);
fa0d7e3d 6767 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 6768}
6769
45321ac5 6770int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 6771{
6772 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 6773
0af3d00b 6774 if (btrfs_root_refs(&root->root_item) == 0 &&
2cf8572d 6775 !btrfs_is_free_space_inode(root, inode))
45321ac5 6776 return 1;
76dda93c 6777 else
45321ac5 6778 return generic_drop_inode(inode);
76dda93c
YZ		 6779}
6780
0ee0fda0 6781static void init_once(void *foo)
39279cc3
CM		 6782{
6783 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6784
6785 inode_init_once(&ei->vfs_inode);
6786}
6787
6788void btrfs_destroy_cachep(void)
6789{
6790 if (btrfs_inode_cachep)
6791 kmem_cache_destroy(btrfs_inode_cachep);
6792 if (btrfs_trans_handle_cachep)
6793 kmem_cache_destroy(btrfs_trans_handle_cachep);
6794 if (btrfs_transaction_cachep)
6795 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 6796 if (btrfs_path_cachep)
6797 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 6798 if (btrfs_free_space_cachep)
6799 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 6800}
6801
6802int btrfs_init_cachep(void)
6803{
9601e3f6
CH		 6804 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6805 sizeof(struct btrfs_inode), 0,
6806 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 6807 if (!btrfs_inode_cachep)
6808 goto fail;
9601e3f6
CH		 6809
6810 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6811 sizeof(struct btrfs_trans_handle), 0,
6812 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6813 if (!btrfs_trans_handle_cachep)
6814 goto fail;
9601e3f6
CH		 6815
6816 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6817 sizeof(struct btrfs_transaction), 0,
6818 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6819 if (!btrfs_transaction_cachep)
6820 goto fail;
9601e3f6
CH		 6821
6822 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6823 sizeof(struct btrfs_path), 0,
6824 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6825 if (!btrfs_path_cachep)
6826 goto fail;
9601e3f6 6827
dc89e982
JB		 6828 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6829 sizeof(struct btrfs_free_space), 0,
6830 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6831 if (!btrfs_free_space_cachep)
6832 goto fail;
6833
39279cc3
CM		 6834 return 0;
6835fail:
6836 btrfs_destroy_cachep();
6837 return -ENOMEM;
6838}
6839
6840static int btrfs_getattr(struct vfsmount *mnt,
6841 struct dentry *dentry, struct kstat *stat)
6842{
6843 struct inode *inode = dentry->d_inode;
6844 generic_fillattr(inode, stat);
0ee5dc67 6845 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 6846 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 6847 stat->blocks = (inode_get_bytes(inode) +
6848 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 6849 return 0;
6850}
6851
75e7cb7f
LB		 6852/*
6853 * If a file is moved, it will inherit the cow and compression flags of the new
6854 * directory.
6855 */
6856static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6857{
6858 struct btrfs_inode *b_dir = BTRFS_I(dir);
6859 struct btrfs_inode *b_inode = BTRFS_I(inode);
6860
6861 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6862 b_inode->flags |= BTRFS_INODE_NODATACOW;
6863 else
6864 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6865
6866 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6867 b_inode->flags |= BTRFS_INODE_COMPRESS;
6868 else
6869 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6870}
6871
d397712b
CM		 6872static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6873 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 6874{
6875 struct btrfs_trans_handle *trans;
6876 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 6877 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 6878 struct inode *new_inode = new_dentry->d_inode;
6879 struct inode *old_inode = old_dentry->d_inode;
6880 struct timespec ctime = CURRENT_TIME;
00e4e6b3 6881 u64 index = 0;
4df27c4d 6882 u64 root_objectid;
39279cc3 6883 int ret;
33345d01 6884 u64 old_ino = btrfs_ino(old_inode);
39279cc3 6885
33345d01 6886 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 6887 return -EPERM;
6888
4df27c4d 6889 /* we only allow rename subvolume link between subvolumes */
33345d01 6890 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 6891 return -EXDEV;
6892
33345d01
LZ		 6893 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6894 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 6895 return -ENOTEMPTY;
5f39d397 6896
4df27c4d
YZ		 6897 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6898 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6899 return -ENOTEMPTY;
5a3f23d5
CM		 6900 /*
6901 * we're using rename to replace one file with another.
6902 * and the replacement file is large. Start IO on it now so
6903 * we don't add too much work to the end of the transaction
6904 */
4baf8c92 6905 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 6906 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6907 filemap_flush(old_inode->i_mapping);
6908
76dda93c 6909 /* close the racy window with snapshot create/destroy ioctl */
33345d01 6910 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 6911 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 6912 /*
6913 * We want to reserve the absolute worst case amount of items. So if
6914 * both inodes are subvols and we need to unlink them then that would
6915 * require 4 item modifications, but if they are both normal inodes it
6916 * would require 5 item modifications, so we'll assume their normal
6917 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6918 * should cover the worst case number of items we'll modify.
6919 */
6920 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 6921 if (IS_ERR(trans)) {
6922 ret = PTR_ERR(trans);
6923 goto out_notrans;
6924 }
76dda93c 6925
4df27c4d
YZ		 6926 if (dest != root)
6927 btrfs_record_root_in_trans(trans, dest);
5f39d397 6928
a5719521
YZ		 6929 ret = btrfs_set_inode_index(new_dir, &index);
6930 if (ret)
6931 goto out_fail;
5a3f23d5 6932
33345d01 6933 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6934 /* force full log commit if subvolume involved. */
6935 root->fs_info->last_trans_log_full_commit = trans->transid;
6936 } else {
a5719521
YZ		 6937 ret = btrfs_insert_inode_ref(trans, dest,
6938 new_dentry->d_name.name,
6939 new_dentry->d_name.len,
33345d01
LZ		 6940 old_ino,
6941 btrfs_ino(new_dir), index);
a5719521
YZ		 6942 if (ret)
6943 goto out_fail;
4df27c4d
YZ		 6944 /*
6945 * this is an ugly little race, but the rename is required
6946 * to make sure that if we crash, the inode is either at the
6947 * old name or the new one. pinning the log transaction lets
6948 * us make sure we don't allow a log commit to come in after
6949 * we unlink the name but before we add the new name back in.
6950 */
6951 btrfs_pin_log_trans(root);
6952 }
5a3f23d5
CM		 6953 /*
6954 * make sure the inode gets flushed if it is replacing
6955 * something.
6956 */
33345d01 6957 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 6958 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 6959
39279cc3
CM		 6960 old_dir->i_ctime = old_dir->i_mtime = ctime;
6961 new_dir->i_ctime = new_dir->i_mtime = ctime;
6962 old_inode->i_ctime = ctime;
5f39d397 6963
12fcfd22
CM		 6964 if (old_dentry->d_parent != new_dentry->d_parent)
6965 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6966
33345d01 6967 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6968 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6969 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6970 old_dentry->d_name.name,
6971 old_dentry->d_name.len);
6972 } else {
92986796
AV		 6973 ret = __btrfs_unlink_inode(trans, root, old_dir,
6974 old_dentry->d_inode,
6975 old_dentry->d_name.name,
6976 old_dentry->d_name.len);
6977 if (!ret)
6978 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d
YZ		 6979 }
6980 BUG_ON(ret);
39279cc3
CM		 6981
6982 if (new_inode) {
6983 new_inode->i_ctime = CURRENT_TIME;
33345d01 6984 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 6985 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6986 root_objectid = BTRFS_I(new_inode)->location.objectid;
6987 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6988 root_objectid,
6989 new_dentry->d_name.name,
6990 new_dentry->d_name.len);
6991 BUG_ON(new_inode->i_nlink == 0);
6992 } else {
6993 ret = btrfs_unlink_inode(trans, dest, new_dir,
6994 new_dentry->d_inode,
6995 new_dentry->d_name.name,
6996 new_dentry->d_name.len);
6997 }
6998 BUG_ON(ret);
7b128766 6999 if (new_inode->i_nlink == 0) {
e02119d5 7000 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7001 BUG_ON(ret);
7b128766 7002 }
39279cc3 7003 }
aec7477b 7004
75e7cb7f
LB		 7005 fixup_inode_flags(new_dir, old_inode);
7006
4df27c4d
YZ		 7007 ret = btrfs_add_link(trans, new_dir, old_inode,
7008 new_dentry->d_name.name,
a5719521 7009 new_dentry->d_name.len, 0, index);
4df27c4d 7010 BUG_ON(ret);
39279cc3 7011
33345d01 7012 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7013 struct dentry *parent = new_dentry->d_parent;
6a912213 7014 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7015 btrfs_end_log_trans(root);
7016 }
39279cc3 7017out_fail:
ab78c84d 7018 btrfs_end_transaction_throttle(trans, root);
b44c59a8 7019out_notrans:
33345d01 7020 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7021 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7022
39279cc3
CM		 7023 return ret;
7024}
7025
d352ac68
CM		 7026/*
7027 * some fairly slow code that needs optimization. This walks the list
7028 * of all the inodes with pending delalloc and forces them to disk.
7029 */
24bbcf04 7030int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7031{
7032 struct list_head *head = &root->fs_info->delalloc_inodes;
7033 struct btrfs_inode *binode;
5b21f2ed 7034 struct inode *inode;
ea8c2819 7035
c146afad
YZ		 7036 if (root->fs_info->sb->s_flags & MS_RDONLY)
7037 return -EROFS;
7038
75eff68e 7039 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7040 while (!list_empty(head)) {
ea8c2819
CM		 7041 binode = list_entry(head->next, struct btrfs_inode,
7042 delalloc_inodes);
5b21f2ed
ZY		 7043 inode = igrab(&binode->vfs_inode);
7044 if (!inode)
7045 list_del_init(&binode->delalloc_inodes);
75eff68e 7046 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7047 if (inode) {
8c8bee1d 7048 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7049 if (delay_iput)
7050 btrfs_add_delayed_iput(inode);
7051 else
7052 iput(inode);
5b21f2ed
ZY		 7053 }
7054 cond_resched();
75eff68e 7055 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7056 }
75eff68e 7057 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7058
7059 /* the filemap_flush will queue IO into the worker threads, but
7060 * we have to make sure the IO is actually started and that
7061 * ordered extents get created before we return
7062 */
7063 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7064 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7065 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7066 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7067 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7068 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7069 }
7070 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7071 return 0;
7072}
7073
39279cc3
CM		 7074static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7075 const char *symname)
7076{
7077 struct btrfs_trans_handle *trans;
7078 struct btrfs_root *root = BTRFS_I(dir)->root;
7079 struct btrfs_path *path;
7080 struct btrfs_key key;
1832a6d5 7081 struct inode *inode = NULL;
39279cc3
CM		 7082 int err;
7083 int drop_inode = 0;
7084 u64 objectid;
00e4e6b3 7085 u64 index = 0 ;
39279cc3
CM		 7086 int name_len;
7087 int datasize;
5f39d397 7088 unsigned long ptr;
39279cc3 7089 struct btrfs_file_extent_item *ei;
5f39d397 7090 struct extent_buffer *leaf;
1832a6d5 7091 unsigned long nr = 0;
39279cc3
CM		 7092
7093 name_len = strlen(symname) + 1;
7094 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7095 return -ENAMETOOLONG;
1832a6d5 7096
9ed74f2d
JB		 7097 /*
7098 * 2 items for inode item and ref
7099 * 2 items for dir items
7100 * 1 item for xattr if selinux is on
7101 */
a22285a6
YZ		 7102 trans = btrfs_start_transaction(root, 5);
7103 if (IS_ERR(trans))
7104 return PTR_ERR(trans);
1832a6d5 7105
581bb050
LZ		 7106 err = btrfs_find_free_ino(root, &objectid);
7107 if (err)
7108 goto out_unlock;
7109
aec7477b 7110 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7111 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7112 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7113 if (IS_ERR(inode)) {
7114 err = PTR_ERR(inode);
39279cc3 7115 goto out_unlock;
7cf96da3 7116 }
39279cc3 7117
2a7dba39 7118 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7119 if (err) {
7120 drop_inode = 1;
7121 goto out_unlock;
7122 }
7123
a1b075d2 7124 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7125 if (err)
7126 drop_inode = 1;
7127 else {
7128 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7129 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 7130 inode->i_fop = &btrfs_file_operations;
7131 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 7132 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7133 }
39279cc3
CM		 7134 if (drop_inode)
7135 goto out_unlock;
7136
7137 path = btrfs_alloc_path();
d8926bb3
MF		 7138 if (!path) {
7139 err = -ENOMEM;
7140 drop_inode = 1;
7141 goto out_unlock;
7142 }
33345d01 7143 key.objectid = btrfs_ino(inode);
39279cc3 7144 key.offset = 0;
39279cc3
CM		 7145 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7146 datasize = btrfs_file_extent_calc_inline_size(name_len);
7147 err = btrfs_insert_empty_item(trans, root, path, &key,
7148 datasize);
54aa1f4d
CM		 7149 if (err) {
7150 drop_inode = 1;
b0839166 7151 btrfs_free_path(path);
54aa1f4d
CM		 7152 goto out_unlock;
7153 }
5f39d397
CM		 7154 leaf = path->nodes[0];
7155 ei = btrfs_item_ptr(leaf, path->slots[0],
7156 struct btrfs_file_extent_item);
7157 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7158 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7159 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7160 btrfs_set_file_extent_encryption(leaf, ei, 0);
7161 btrfs_set_file_extent_compression(leaf, ei, 0);
7162 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7163 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7164
39279cc3 7165 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7166 write_extent_buffer(leaf, symname, ptr, name_len);
7167 btrfs_mark_buffer_dirty(leaf);
39279cc3 7168 btrfs_free_path(path);
5f39d397 7169
39279cc3
CM		 7170 inode->i_op = &btrfs_symlink_inode_operations;
7171 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7172 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7173 inode_set_bytes(inode, name_len);
dbe674a9 7174 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7175 err = btrfs_update_inode(trans, root, inode);
7176 if (err)
7177 drop_inode = 1;
39279cc3
CM		 7178
7179out_unlock:
d3c2fdcf 7180 nr = trans->blocks_used;
ab78c84d 7181 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 7182 if (drop_inode) {
7183 inode_dec_link_count(inode);
7184 iput(inode);
7185 }
d3c2fdcf 7186 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7187 return err;
7188}
16432985 7189
0af3d00b
JB		 7190static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7191 u64 start, u64 num_bytes, u64 min_size,
7192 loff_t actual_len, u64 *alloc_hint,
7193 struct btrfs_trans_handle *trans)
d899e052 7194{
d899e052
YZ		 7195 struct btrfs_root *root = BTRFS_I(inode)->root;
7196 struct btrfs_key ins;
d899e052 7197 u64 cur_offset = start;
55a61d1d 7198 u64 i_size;
d899e052 7199 int ret = 0;
0af3d00b 7200 bool own_trans = true;
d899e052 7201
0af3d00b
JB		 7202 if (trans)
7203 own_trans = false;
d899e052 7204 while (num_bytes > 0) {
0af3d00b
JB		 7205 if (own_trans) {
7206 trans = btrfs_start_transaction(root, 3);
7207 if (IS_ERR(trans)) {
7208 ret = PTR_ERR(trans);
7209 break;
7210 }
5a303d5d
YZ		 7211 }
7212
efa56464
YZ		 7213 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7214 0, *alloc_hint, (u64)-1, &ins, 1);
5a303d5d 7215 if (ret) {
0af3d00b
JB		 7216 if (own_trans)
7217 btrfs_end_transaction(trans, root);
a22285a6 7218 break;
d899e052 7219 }
5a303d5d 7220
d899e052
YZ		 7221 ret = insert_reserved_file_extent(trans, inode,
7222 cur_offset, ins.objectid,
7223 ins.offset, ins.offset,
920bbbfb 7224 ins.offset, 0, 0, 0,
d899e052
YZ		 7225 BTRFS_FILE_EXTENT_PREALLOC);
7226 BUG_ON(ret);
a1ed835e
CM		 7227 btrfs_drop_extent_cache(inode, cur_offset,
7228 cur_offset + ins.offset -1, 0);
5a303d5d 7229
d899e052
YZ		 7230 num_bytes -= ins.offset;
7231 cur_offset += ins.offset;
efa56464 7232 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7233
d899e052 7234 inode->i_ctime = CURRENT_TIME;
6cbff00f 7235 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7236 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7237 (actual_len > inode->i_size) &&
7238 (cur_offset > inode->i_size)) {
d1ea6a61 7239 if (cur_offset > actual_len)
55a61d1d 7240 i_size = actual_len;
d1ea6a61 7241 else
55a61d1d
JB		 7242 i_size = cur_offset;
7243 i_size_write(inode, i_size);
7244 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7245 }
7246
d899e052
YZ		 7247 ret = btrfs_update_inode(trans, root, inode);
7248 BUG_ON(ret);
d899e052 7249
0af3d00b
JB		 7250 if (own_trans)
7251 btrfs_end_transaction(trans, root);
5a303d5d 7252 }
d899e052
YZ		 7253 return ret;
7254}
7255
0af3d00b
JB		 7256int btrfs_prealloc_file_range(struct inode *inode, int mode,
7257 u64 start, u64 num_bytes, u64 min_size,
7258 loff_t actual_len, u64 *alloc_hint)
7259{
7260 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7261 min_size, actual_len, alloc_hint,
7262 NULL);
7263}
7264
7265int btrfs_prealloc_file_range_trans(struct inode *inode,
7266 struct btrfs_trans_handle *trans, int mode,
7267 u64 start, u64 num_bytes, u64 min_size,
7268 loff_t actual_len, u64 *alloc_hint)
7269{
7270 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7271 min_size, actual_len, alloc_hint, trans);
7272}
7273
e6dcd2dc
CM		 7274static int btrfs_set_page_dirty(struct page *page)
7275{
e6dcd2dc
CM		 7276 return __set_page_dirty_nobuffers(page);
7277}
7278
10556cb2 7279static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7280{
b83cc969 7281 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7282 umode_t mode = inode->i_mode;
b83cc969 7283
cb6db4e5
JM		 7284 if (mask & MAY_WRITE &&
7285 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7286 if (btrfs_root_readonly(root))
7287 return -EROFS;
7288 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7289 return -EACCES;
7290 }
2830ba7f 7291 return generic_permission(inode, mask);
fdebe2bd 7292}
39279cc3 7293
6e1d5dcc 7294static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7295 .getattr = btrfs_getattr,
39279cc3
CM		 7296 .lookup = btrfs_lookup,
7297 .create = btrfs_create,
7298 .unlink = btrfs_unlink,
7299 .link = btrfs_link,
7300 .mkdir = btrfs_mkdir,
7301 .rmdir = btrfs_rmdir,
7302 .rename = btrfs_rename,
7303 .symlink = btrfs_symlink,
7304 .setattr = btrfs_setattr,
618e21d5 7305 .mknod = btrfs_mknod,
95819c05
CH		 7306 .setxattr = btrfs_setxattr,
7307 .getxattr = btrfs_getxattr,
5103e947 7308 .listxattr = btrfs_listxattr,
95819c05 7309 .removexattr = btrfs_removexattr,
fdebe2bd 7310 .permission = btrfs_permission,
4e34e719 7311 .get_acl = btrfs_get_acl,
39279cc3 7312};
6e1d5dcc 7313static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7314 .lookup = btrfs_lookup,
fdebe2bd 7315 .permission = btrfs_permission,
4e34e719 7316 .get_acl = btrfs_get_acl,
39279cc3 7317};
76dda93c 7318
828c0950 7319static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7320 .llseek = generic_file_llseek,
7321 .read = generic_read_dir,
cbdf5a24 7322 .readdir = btrfs_real_readdir,
34287aa3 7323 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7324#ifdef CONFIG_COMPAT
34287aa3 7325 .compat_ioctl = btrfs_ioctl,
39279cc3 7326#endif
6bf13c0c 7327 .release = btrfs_release_file,
e02119d5 7328 .fsync = btrfs_sync_file,
39279cc3
CM		 7329};
7330
d1310b2e 7331static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7332 .fill_delalloc = run_delalloc_range,
065631f6 7333 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7334 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7335 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7336 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7337 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 7338 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 7339 .set_bit_hook = btrfs_set_bit_hook,
7340 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7341 .merge_extent_hook = btrfs_merge_extent_hook,
7342 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7343};
7344
35054394
CM		 7345/*
7346 * btrfs doesn't support the bmap operation because swapfiles
7347 * use bmap to make a mapping of extents in the file. They assume
7348 * these extents won't change over the life of the file and they
7349 * use the bmap result to do IO directly to the drive.
7350 *
7351 * the btrfs bmap call would return logical addresses that aren't
7352 * suitable for IO and they also will change frequently as COW
7353 * operations happen. So, swapfile + btrfs == corruption.
7354 *
7355 * For now we're avoiding this by dropping bmap.
7356 */
7f09410b 7357static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7358 .readpage = btrfs_readpage,
7359 .writepage = btrfs_writepage,
b293f02e 7360 .writepages = btrfs_writepages,
3ab2fb5a 7361 .readpages = btrfs_readpages,
16432985 7362 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7363 .invalidatepage = btrfs_invalidatepage,
7364 .releasepage = btrfs_releasepage,
e6dcd2dc 7365 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7366 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7367};
7368
7f09410b 7369static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7370 .readpage = btrfs_readpage,
7371 .writepage = btrfs_writepage,
2bf5a725
CM		 7372 .invalidatepage = btrfs_invalidatepage,
7373 .releasepage = btrfs_releasepage,
39279cc3
CM		 7374};
7375
6e1d5dcc 7376static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7377 .getattr = btrfs_getattr,
7378 .setattr = btrfs_setattr,
95819c05
CH		 7379 .setxattr = btrfs_setxattr,
7380 .getxattr = btrfs_getxattr,
5103e947 7381 .listxattr = btrfs_listxattr,
95819c05 7382 .removexattr = btrfs_removexattr,
fdebe2bd 7383 .permission = btrfs_permission,
1506fcc8 7384 .fiemap = btrfs_fiemap,
4e34e719 7385 .get_acl = btrfs_get_acl,
39279cc3 7386};
6e1d5dcc 7387static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7388 .getattr = btrfs_getattr,
7389 .setattr = btrfs_setattr,
fdebe2bd 7390 .permission = btrfs_permission,
95819c05
CH		 7391 .setxattr = btrfs_setxattr,
7392 .getxattr = btrfs_getxattr,
33268eaf 7393 .listxattr = btrfs_listxattr,
95819c05 7394 .removexattr = btrfs_removexattr,
4e34e719 7395 .get_acl = btrfs_get_acl,
618e21d5 7396};
6e1d5dcc 7397static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7398 .readlink = generic_readlink,
7399 .follow_link = page_follow_link_light,
7400 .put_link = page_put_link,
f209561a 7401 .getattr = btrfs_getattr,
fdebe2bd 7402 .permission = btrfs_permission,
0279b4cd
JO		 7403 .setxattr = btrfs_setxattr,
7404 .getxattr = btrfs_getxattr,
7405 .listxattr = btrfs_listxattr,
7406 .removexattr = btrfs_removexattr,
4e34e719 7407 .get_acl = btrfs_get_acl,
39279cc3 7408};
76dda93c 7409
82d339d9 7410const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7411 .d_delete = btrfs_dentry_delete,
b4aff1f8 7412 .d_release = btrfs_dentry_release,
76dda93c 7413};
Linux 6.x block layer and io_uring tree(s)