projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix array bound checking
[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:
5b0e95bf
JB		 1795 if (root != root->fs_info->tree_root)
1796 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1797 if (trans) {
1798 if (nolock)
0cb59c99 1799 btrfs_end_transaction_nolock(trans, root);
5b0e95bf 1800 else
0cb59c99
JB		 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;
8f6d7f4f 2233 u64 last_objectid = 0;
7b128766
JB		 2234 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2235
d68fc57b 2236 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2237 return 0;
c71bf099
YZ		 2238
2239 path = btrfs_alloc_path();
66b4ffd1
JB		 2240 if (!path) {
2241 ret = -ENOMEM;
2242 goto out;
2243 }
7b128766
JB		 2244 path->reada = -1;
2245
2246 key.objectid = BTRFS_ORPHAN_OBJECTID;
2247 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2248 key.offset = (u64)-1;
2249
7b128766
JB		 2250 while (1) {
2251 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2252 if (ret < 0)
2253 goto out;
7b128766
JB		 2254
2255 /*
2256 * if ret == 0 means we found what we were searching for, which
25985edc 2257 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2258 * find the key and see if we have stuff that matches
2259 */
2260 if (ret > 0) {
66b4ffd1 2261 ret = 0;
7b128766
JB		 2262 if (path->slots[0] == 0)
2263 break;
2264 path->slots[0]--;
2265 }
2266
2267 /* pull out the item */
2268 leaf = path->nodes[0];
7b128766
JB		 2269 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2270
2271 /* make sure the item matches what we want */
2272 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2273 break;
2274 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2275 break;
2276
2277 /* release the path since we're done with it */
b3b4aa74 2278 btrfs_release_path(path);
7b128766
JB		 2279
2280 /*
2281 * this is where we are basically btrfs_lookup, without the
2282 * crossing root thing. we store the inode number in the
2283 * offset of the orphan item.
2284 */
8f6d7f4f
JB		 2285
2286 if (found_key.offset == last_objectid) {
2287 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2288 "stopping orphan cleanup\n");
2289 ret = -EINVAL;
2290 goto out;
2291 }
2292
2293 last_objectid = found_key.offset;
2294
5d4f98a2
YZ		 2295 found_key.objectid = found_key.offset;
2296 found_key.type = BTRFS_INODE_ITEM_KEY;
2297 found_key.offset = 0;
73f73415 2298 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2299 ret = PTR_RET(inode);
2300 if (ret && ret != -ESTALE)
66b4ffd1 2301 goto out;
7b128766
JB		 2302
2303 /*
a8c9e576
JB		 2304 * Inode is already gone but the orphan item is still there,
2305 * kill the orphan item.
7b128766 2306 */
a8c9e576
JB		 2307 if (ret == -ESTALE) {
2308 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2309 if (IS_ERR(trans)) {
2310 ret = PTR_ERR(trans);
2311 goto out;
2312 }
a8c9e576
JB		 2313 ret = btrfs_del_orphan_item(trans, root,
2314 found_key.objectid);
2315 BUG_ON(ret);
5b21f2ed 2316 btrfs_end_transaction(trans, root);
7b128766
JB		 2317 continue;
2318 }
2319
a8c9e576
JB		 2320 /*
2321 * add this inode to the orphan list so btrfs_orphan_del does
2322 * the proper thing when we hit it
2323 */
2324 spin_lock(&root->orphan_lock);
2325 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2326 spin_unlock(&root->orphan_lock);
2327
7b128766
JB		 2328 /* if we have links, this was a truncate, lets do that */
2329 if (inode->i_nlink) {
a41ad394
JB		 2330 if (!S_ISREG(inode->i_mode)) {
2331 WARN_ON(1);
2332 iput(inode);
2333 continue;
2334 }
7b128766 2335 nr_truncate++;
66b4ffd1 2336 ret = btrfs_truncate(inode);
7b128766
JB		 2337 } else {
2338 nr_unlink++;
2339 }
2340
2341 /* this will do delete_inode and everything for us */
2342 iput(inode);
66b4ffd1
JB		 2343 if (ret)
2344 goto out;
7b128766 2345 }
d68fc57b
YZ		 2346 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2347
2348 if (root->orphan_block_rsv)
2349 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2350 (u64)-1);
2351
2352 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2353 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2354 if (!IS_ERR(trans))
2355 btrfs_end_transaction(trans, root);
d68fc57b 2356 }
7b128766
JB		 2357
2358 if (nr_unlink)
2359 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2360 if (nr_truncate)
2361 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2362
2363out:
2364 if (ret)
2365 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2366 btrfs_free_path(path);
2367 return ret;
7b128766
JB		 2368}
2369
46a53cca
CM		 2370/*
2371 * very simple check to peek ahead in the leaf looking for xattrs. If we
2372 * don't find any xattrs, we know there can't be any acls.
2373 *
2374 * slot is the slot the inode is in, objectid is the objectid of the inode
2375 */
2376static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2377 int slot, u64 objectid)
2378{
2379 u32 nritems = btrfs_header_nritems(leaf);
2380 struct btrfs_key found_key;
2381 int scanned = 0;
2382
2383 slot++;
2384 while (slot < nritems) {
2385 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2386
2387 /* we found a different objectid, there must not be acls */
2388 if (found_key.objectid != objectid)
2389 return 0;
2390
2391 /* we found an xattr, assume we've got an acl */
2392 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2393 return 1;
2394
2395 /*
2396 * we found a key greater than an xattr key, there can't
2397 * be any acls later on
2398 */
2399 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2400 return 0;
2401
2402 slot++;
2403 scanned++;
2404
2405 /*
2406 * it goes inode, inode backrefs, xattrs, extents,
2407 * so if there are a ton of hard links to an inode there can
2408 * be a lot of backrefs. Don't waste time searching too hard,
2409 * this is just an optimization
2410 */
2411 if (scanned >= 8)
2412 break;
2413 }
2414 /* we hit the end of the leaf before we found an xattr or
2415 * something larger than an xattr. We have to assume the inode
2416 * has acls
2417 */
2418 return 1;
2419}
2420
d352ac68
CM		 2421/*
2422 * read an inode from the btree into the in-memory inode
2423 */
5d4f98a2 2424static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2425{
2426 struct btrfs_path *path;
5f39d397 2427 struct extent_buffer *leaf;
39279cc3 2428 struct btrfs_inode_item *inode_item;
0b86a832 2429 struct btrfs_timespec *tspec;
39279cc3
CM		 2430 struct btrfs_root *root = BTRFS_I(inode)->root;
2431 struct btrfs_key location;
46a53cca 2432 int maybe_acls;
618e21d5 2433 u32 rdev;
39279cc3 2434 int ret;
2f7e33d4
MX		 2435 bool filled = false;
2436
2437 ret = btrfs_fill_inode(inode, &rdev);
2438 if (!ret)
2439 filled = true;
39279cc3
CM		 2440
2441 path = btrfs_alloc_path();
1748f843
MF		 2442 if (!path)
2443 goto make_bad;
2444
d90c7321 2445 path->leave_spinning = 1;
39279cc3 2446 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2447
39279cc3 2448 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2449 if (ret)
39279cc3 2450 goto make_bad;
39279cc3 2451
5f39d397 2452 leaf = path->nodes[0];
2f7e33d4
MX		 2453
2454 if (filled)
2455 goto cache_acl;
2456
5f39d397
CM		 2457 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458 struct btrfs_inode_item);
5f39d397
CM		 2459 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2460 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2461 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2462 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2463 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2464
2465 tspec = btrfs_inode_atime(inode_item);
2466 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2467 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2468
2469 tspec = btrfs_inode_mtime(inode_item);
2470 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2472
2473 tspec = btrfs_inode_ctime(inode_item);
2474 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2476
a76a3cd4 2477 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2478 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2479 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2480 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2481 inode->i_rdev = 0;
5f39d397
CM		 2482 rdev = btrfs_inode_rdev(leaf, inode_item);
2483
aec7477b 2484 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2485 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2486cache_acl:
46a53cca
CM		 2487 /*
2488 * try to precache a NULL acl entry for files that don't have
2489 * any xattrs or acls
2490 */
33345d01
LZ		 2491 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2492 btrfs_ino(inode));
72c04902
AV		 2493 if (!maybe_acls)
2494 cache_no_acl(inode);
46a53cca 2495
39279cc3 2496 btrfs_free_path(path);
39279cc3 2497
39279cc3 2498 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2499 case S_IFREG:
2500 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2501 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2502 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2503 inode->i_fop = &btrfs_file_operations;
2504 inode->i_op = &btrfs_file_inode_operations;
2505 break;
2506 case S_IFDIR:
2507 inode->i_fop = &btrfs_dir_file_operations;
2508 if (root == root->fs_info->tree_root)
2509 inode->i_op = &btrfs_dir_ro_inode_operations;
2510 else
2511 inode->i_op = &btrfs_dir_inode_operations;
2512 break;
2513 case S_IFLNK:
2514 inode->i_op = &btrfs_symlink_inode_operations;
2515 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2516 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2517 break;
618e21d5 2518 default:
0279b4cd 2519 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2520 init_special_inode(inode, inode->i_mode, rdev);
2521 break;
39279cc3 2522 }
6cbff00f
CH		 2523
2524 btrfs_update_iflags(inode);
39279cc3
CM		 2525 return;
2526
2527make_bad:
39279cc3 2528 btrfs_free_path(path);
39279cc3
CM		 2529 make_bad_inode(inode);
2530}
2531
d352ac68
CM		 2532/*
2533 * given a leaf and an inode, copy the inode fields into the leaf
2534 */
e02119d5
CM		 2535static void fill_inode_item(struct btrfs_trans_handle *trans,
2536 struct extent_buffer *leaf,
5f39d397 2537 struct btrfs_inode_item *item,
39279cc3
CM		 2538 struct inode *inode)
2539{
5f39d397
CM		 2540 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2541 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2542 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2543 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2544 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2545
2546 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2547 inode->i_atime.tv_sec);
2548 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2549 inode->i_atime.tv_nsec);
2550
2551 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2552 inode->i_mtime.tv_sec);
2553 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2554 inode->i_mtime.tv_nsec);
2555
2556 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2557 inode->i_ctime.tv_sec);
2558 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2559 inode->i_ctime.tv_nsec);
2560
a76a3cd4 2561 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2562 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2563 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2564 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2565 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2566 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2567 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2568}
2569
d352ac68
CM		 2570/*
2571 * copy everything in the in-memory inode into the btree.
2572 */
d397712b
CM		 2573noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2574 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2575{
2576 struct btrfs_inode_item *inode_item;
2577 struct btrfs_path *path;
5f39d397 2578 struct extent_buffer *leaf;
39279cc3
CM		 2579 int ret;
2580
16cdcec7 2581 /*
149e2d76
MX		 2582 * If the inode is a free space inode, we can deadlock during commit
2583 * if we put it into the delayed code.
2584 *
2585 * The data relocation inode should also be directly updated
2586 * without delay
16cdcec7 2587 */
2cf8572d 2588 if (!btrfs_is_free_space_inode(root, inode)
149e2d76 2589 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
16cdcec7
MX		 2590 ret = btrfs_delayed_update_inode(trans, root, inode);
2591 if (!ret)
2592 btrfs_set_inode_last_trans(trans, inode);
2593 return ret;
2594 }
2595
39279cc3 2596 path = btrfs_alloc_path();
16cdcec7
MX		 2597 if (!path)
2598 return -ENOMEM;
2599
b9473439 2600 path->leave_spinning = 1;
16cdcec7
MX		 2601 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2602 1);
39279cc3
CM		 2603 if (ret) {
2604 if (ret > 0)
2605 ret = -ENOENT;
2606 goto failed;
2607 }
2608
b4ce94de 2609 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2610 leaf = path->nodes[0];
2611 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2612 struct btrfs_inode_item);
39279cc3 2613
e02119d5 2614 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2615 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2616 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2617 ret = 0;
2618failed:
39279cc3
CM		 2619 btrfs_free_path(path);
2620 return ret;
2621}
2622
d352ac68
CM		 2623/*
2624 * unlink helper that gets used here in inode.c and in the tree logging
2625 * recovery code. It remove a link in a directory with a given name, and
2626 * also drops the back refs in the inode to the directory
2627 */
92986796
AV		 2628static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root,
2630 struct inode *dir, struct inode *inode,
2631 const char *name, int name_len)
39279cc3
CM		 2632{
2633 struct btrfs_path *path;
39279cc3 2634 int ret = 0;
5f39d397 2635 struct extent_buffer *leaf;
39279cc3 2636 struct btrfs_dir_item *di;
5f39d397 2637 struct btrfs_key key;
aec7477b 2638 u64 index;
33345d01
LZ		 2639 u64 ino = btrfs_ino(inode);
2640 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2641
2642 path = btrfs_alloc_path();
54aa1f4d
CM		 2643 if (!path) {
2644 ret = -ENOMEM;
554233a6 2645 goto out;
54aa1f4d
CM		 2646 }
2647
b9473439 2648 path->leave_spinning = 1;
33345d01 2649 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2650 name, name_len, -1);
2651 if (IS_ERR(di)) {
2652 ret = PTR_ERR(di);
2653 goto err;
2654 }
2655 if (!di) {
2656 ret = -ENOENT;
2657 goto err;
2658 }
5f39d397
CM		 2659 leaf = path->nodes[0];
2660 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2661 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2662 if (ret)
2663 goto err;
b3b4aa74 2664 btrfs_release_path(path);
39279cc3 2665
33345d01
LZ		 2666 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2667 dir_ino, &index);
aec7477b 2668 if (ret) {
d397712b 2669 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2670 "inode %llu parent %llu\n", name_len, name,
2671 (unsigned long long)ino, (unsigned long long)dir_ino);
aec7477b
JB		 2672 goto err;
2673 }
2674
16cdcec7
MX		 2675 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2676 if (ret)
39279cc3 2677 goto err;
39279cc3 2678
e02119d5 2679 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2680 inode, dir_ino);
49eb7e46 2681 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2682
2683 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2684 dir, index);
6418c961
CM		 2685 if (ret == -ENOENT)
2686 ret = 0;
39279cc3
CM		 2687err:
2688 btrfs_free_path(path);
e02119d5
CM		 2689 if (ret)
2690 goto out;
2691
2692 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2693 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2694 btrfs_update_inode(trans, root, dir);
e02119d5 2695out:
39279cc3
CM		 2696 return ret;
2697}
2698
92986796
AV		 2699int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2700 struct btrfs_root *root,
2701 struct inode *dir, struct inode *inode,
2702 const char *name, int name_len)
2703{
2704 int ret;
2705 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2706 if (!ret) {
2707 btrfs_drop_nlink(inode);
2708 ret = btrfs_update_inode(trans, root, inode);
2709 }
2710 return ret;
2711}
2712
2713
a22285a6
YZ		 2714/* helper to check if there is any shared block in the path */
2715static int check_path_shared(struct btrfs_root *root,
2716 struct btrfs_path *path)
39279cc3 2717{
a22285a6
YZ		 2718 struct extent_buffer *eb;
2719 int level;
0e4dcbef 2720 u64 refs = 1;
5df6a9f6 2721
a22285a6 2722 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2723 int ret;
2724
a22285a6
YZ		 2725 if (!path->nodes[level])
2726 break;
2727 eb = path->nodes[level];
2728 if (!btrfs_block_can_be_shared(root, eb))
2729 continue;
2730 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2731 &refs, NULL);
2732 if (refs > 1)
2733 return 1;
5df6a9f6 2734 }
dedefd72 2735 return 0;
39279cc3
CM		 2736}
2737
a22285a6
YZ		 2738/*
2739 * helper to start transaction for unlink and rmdir.
2740 *
2741 * unlink and rmdir are special in btrfs, they do not always free space.
2742 * so in enospc case, we should make sure they will free space before
2743 * allowing them to use the global metadata reservation.
2744 */
2745static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2746 struct dentry *dentry)
4df27c4d 2747{
39279cc3 2748 struct btrfs_trans_handle *trans;
a22285a6 2749 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2750 struct btrfs_path *path;
a22285a6 2751 struct btrfs_inode_ref *ref;
4df27c4d 2752 struct btrfs_dir_item *di;
7b128766 2753 struct inode *inode = dentry->d_inode;
4df27c4d 2754 u64 index;
a22285a6
YZ		 2755 int check_link = 1;
2756 int err = -ENOSPC;
4df27c4d 2757 int ret;
33345d01
LZ		 2758 u64 ino = btrfs_ino(inode);
2759 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2760
e70bea5f
JB		 2761 /*
2762 * 1 for the possible orphan item
2763 * 1 for the dir item
2764 * 1 for the dir index
2765 * 1 for the inode ref
2766 * 1 for the inode ref in the tree log
2767 * 2 for the dir entries in the log
2768 * 1 for the inode
2769 */
2770 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2771 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2772 return trans;
4df27c4d 2773
33345d01 2774 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2775 return ERR_PTR(-ENOSPC);
4df27c4d 2776
a22285a6
YZ		 2777 /* check if there is someone else holds reference */
2778 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2779 return ERR_PTR(-ENOSPC);
4df27c4d 2780
a22285a6
YZ		 2781 if (atomic_read(&inode->i_count) > 2)
2782 return ERR_PTR(-ENOSPC);
4df27c4d 2783
a22285a6
YZ		 2784 if (xchg(&root->fs_info->enospc_unlink, 1))
2785 return ERR_PTR(-ENOSPC);
2786
2787 path = btrfs_alloc_path();
2788 if (!path) {
2789 root->fs_info->enospc_unlink = 0;
2790 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2791 }
2792
3880a1b4
JB		 2793 /* 1 for the orphan item */
2794 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2795 if (IS_ERR(trans)) {
a22285a6
YZ		 2796 btrfs_free_path(path);
2797 root->fs_info->enospc_unlink = 0;
2798 return trans;
2799 }
4df27c4d 2800
a22285a6
YZ		 2801 path->skip_locking = 1;
2802 path->search_commit_root = 1;
4df27c4d 2803
a22285a6
YZ		 2804 ret = btrfs_lookup_inode(trans, root, path,
2805 &BTRFS_I(dir)->location, 0);
2806 if (ret < 0) {
2807 err = ret;
2808 goto out;
2809 }
2810 if (ret == 0) {
2811 if (check_path_shared(root, path))
2812 goto out;
2813 } else {
2814 check_link = 0;
5df6a9f6 2815 }
b3b4aa74 2816 btrfs_release_path(path);
a22285a6
YZ		 2817
2818 ret = btrfs_lookup_inode(trans, root, path,
2819 &BTRFS_I(inode)->location, 0);
2820 if (ret < 0) {
2821 err = ret;
2822 goto out;
2823 }
2824 if (ret == 0) {
2825 if (check_path_shared(root, path))
2826 goto out;
2827 } else {
2828 check_link = 0;
2829 }
b3b4aa74 2830 btrfs_release_path(path);
a22285a6
YZ		 2831
2832 if (ret == 0 && S_ISREG(inode->i_mode)) {
2833 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2834 ino, (u64)-1, 0);
a22285a6
YZ		 2835 if (ret < 0) {
2836 err = ret;
2837 goto out;
2838 }
2839 BUG_ON(ret == 0);
2840 if (check_path_shared(root, path))
2841 goto out;
b3b4aa74 2842 btrfs_release_path(path);
a22285a6
YZ		 2843 }
2844
2845 if (!check_link) {
2846 err = 0;
2847 goto out;
2848 }
2849
33345d01 2850 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2851 dentry->d_name.name, dentry->d_name.len, 0);
2852 if (IS_ERR(di)) {
2853 err = PTR_ERR(di);
2854 goto out;
2855 }
2856 if (di) {
2857 if (check_path_shared(root, path))
2858 goto out;
2859 } else {
2860 err = 0;
2861 goto out;
2862 }
b3b4aa74 2863 btrfs_release_path(path);
a22285a6
YZ		 2864
2865 ref = btrfs_lookup_inode_ref(trans, root, path,
2866 dentry->d_name.name, dentry->d_name.len,
33345d01 2867 ino, dir_ino, 0);
a22285a6
YZ		 2868 if (IS_ERR(ref)) {
2869 err = PTR_ERR(ref);
2870 goto out;
2871 }
2872 BUG_ON(!ref);
2873 if (check_path_shared(root, path))
2874 goto out;
2875 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 2876 btrfs_release_path(path);
a22285a6 2877
16cdcec7
MX		 2878 /*
2879 * This is a commit root search, if we can lookup inode item and other
2880 * relative items in the commit root, it means the transaction of
2881 * dir/file creation has been committed, and the dir index item that we
2882 * delay to insert has also been inserted into the commit root. So
2883 * we needn't worry about the delayed insertion of the dir index item
2884 * here.
2885 */
33345d01 2886 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 2887 dentry->d_name.name, dentry->d_name.len, 0);
2888 if (IS_ERR(di)) {
2889 err = PTR_ERR(di);
2890 goto out;
2891 }
2892 BUG_ON(ret == -ENOENT);
2893 if (check_path_shared(root, path))
2894 goto out;
2895
2896 err = 0;
2897out:
2898 btrfs_free_path(path);
3880a1b4
JB		 2899 /* Migrate the orphan reservation over */
2900 if (!err)
2901 err = btrfs_block_rsv_migrate(trans->block_rsv,
2902 &root->fs_info->global_block_rsv,
2903 btrfs_calc_trans_metadata_size(root, 1));
2904
a22285a6
YZ		 2905 if (err) {
2906 btrfs_end_transaction(trans, root);
2907 root->fs_info->enospc_unlink = 0;
2908 return ERR_PTR(err);
2909 }
2910
2911 trans->block_rsv = &root->fs_info->global_block_rsv;
2912 return trans;
2913}
2914
2915static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2916 struct btrfs_root *root)
2917{
2918 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2919 BUG_ON(!root->fs_info->enospc_unlink);
2920 root->fs_info->enospc_unlink = 0;
2921 }
2922 btrfs_end_transaction_throttle(trans, root);
2923}
2924
2925static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2926{
2927 struct btrfs_root *root = BTRFS_I(dir)->root;
2928 struct btrfs_trans_handle *trans;
2929 struct inode *inode = dentry->d_inode;
2930 int ret;
2931 unsigned long nr = 0;
2932
2933 trans = __unlink_start_trans(dir, dentry);
2934 if (IS_ERR(trans))
2935 return PTR_ERR(trans);
5f39d397 2936
12fcfd22
CM		 2937 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2938
e02119d5
CM		 2939 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2940 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 2941 if (ret)
2942 goto out;
7b128766 2943
a22285a6 2944 if (inode->i_nlink == 0) {
7b128766 2945 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 2946 if (ret)
2947 goto out;
a22285a6 2948 }
7b128766 2949
b532402e 2950out:
d3c2fdcf 2951 nr = trans->blocks_used;
a22285a6 2952 __unlink_end_trans(trans, root);
d3c2fdcf 2953 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2954 return ret;
2955}
2956
4df27c4d
YZ		 2957int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2958 struct btrfs_root *root,
2959 struct inode *dir, u64 objectid,
2960 const char *name, int name_len)
2961{
2962 struct btrfs_path *path;
2963 struct extent_buffer *leaf;
2964 struct btrfs_dir_item *di;
2965 struct btrfs_key key;
2966 u64 index;
2967 int ret;
33345d01 2968 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 2969
2970 path = btrfs_alloc_path();
2971 if (!path)
2972 return -ENOMEM;
2973
33345d01 2974 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 2975 name, name_len, -1);
c704005d 2976 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2977
2978 leaf = path->nodes[0];
2979 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2980 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2981 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2982 BUG_ON(ret);
b3b4aa74 2983 btrfs_release_path(path);
4df27c4d
YZ		 2984
2985 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2986 objectid, root->root_key.objectid,
33345d01 2987 dir_ino, &index, name, name_len);
4df27c4d
YZ		 2988 if (ret < 0) {
2989 BUG_ON(ret != -ENOENT);
33345d01 2990 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 2991 name, name_len);
c704005d 2992 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2993
2994 leaf = path->nodes[0];
2995 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2996 btrfs_release_path(path);
4df27c4d
YZ		 2997 index = key.offset;
2998 }
945d8962 2999 btrfs_release_path(path);
4df27c4d 3000
16cdcec7 3001 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4df27c4d 3002 BUG_ON(ret);
4df27c4d
YZ		 3003
3004 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3005 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3006 ret = btrfs_update_inode(trans, root, dir);
3007 BUG_ON(ret);
4df27c4d 3008
71d7aed0 3009 btrfs_free_path(path);
4df27c4d
YZ		 3010 return 0;
3011}
3012
39279cc3
CM		 3013static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3014{
3015 struct inode *inode = dentry->d_inode;
1832a6d5 3016 int err = 0;
39279cc3 3017 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3018 struct btrfs_trans_handle *trans;
1832a6d5 3019 unsigned long nr = 0;
39279cc3 3020
3394e160 3021 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 3022 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 3023 return -ENOTEMPTY;
3024
a22285a6
YZ		 3025 trans = __unlink_start_trans(dir, dentry);
3026 if (IS_ERR(trans))
5df6a9f6 3027 return PTR_ERR(trans);
5df6a9f6 3028
33345d01 3029 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3030 err = btrfs_unlink_subvol(trans, root, dir,
3031 BTRFS_I(inode)->location.objectid,
3032 dentry->d_name.name,
3033 dentry->d_name.len);
3034 goto out;
3035 }
3036
7b128766
JB		 3037 err = btrfs_orphan_add(trans, inode);
3038 if (err)
4df27c4d 3039 goto out;
7b128766 3040
39279cc3 3041 /* now the directory is empty */
e02119d5
CM		 3042 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3043 dentry->d_name.name, dentry->d_name.len);
d397712b 3044 if (!err)
dbe674a9 3045 btrfs_i_size_write(inode, 0);
4df27c4d 3046out:
d3c2fdcf 3047 nr = trans->blocks_used;
a22285a6 3048 __unlink_end_trans(trans, root);
d3c2fdcf 3049 btrfs_btree_balance_dirty(root, nr);
3954401f 3050
39279cc3
CM		 3051 return err;
3052}
3053
39279cc3
CM		 3054/*
3055 * this can truncate away extent items, csum items and directory items.
3056 * It starts at a high offset and removes keys until it can't find
d352ac68 3057 * any higher than new_size
39279cc3
CM		 3058 *
3059 * csum items that cross the new i_size are truncated to the new size
3060 * as well.
7b128766
JB		 3061 *
3062 * min_type is the minimum key type to truncate down to. If set to 0, this
3063 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3064 */
8082510e
YZ		 3065int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3066 struct btrfs_root *root,
3067 struct inode *inode,
3068 u64 new_size, u32 min_type)
39279cc3 3069{
39279cc3 3070 struct btrfs_path *path;
5f39d397 3071 struct extent_buffer *leaf;
39279cc3 3072 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3073 struct btrfs_key key;
3074 struct btrfs_key found_key;
39279cc3 3075 u64 extent_start = 0;
db94535d 3076 u64 extent_num_bytes = 0;
5d4f98a2 3077 u64 extent_offset = 0;
39279cc3 3078 u64 item_end = 0;
8082510e
YZ		 3079 u64 mask = root->sectorsize - 1;
3080 u32 found_type = (u8)-1;
39279cc3
CM		 3081 int found_extent;
3082 int del_item;
85e21bac
CM		 3083 int pending_del_nr = 0;
3084 int pending_del_slot = 0;
179e29e4 3085 int extent_type = -1;
771ed689 3086 int encoding;
8082510e
YZ		 3087 int ret;
3088 int err = 0;
33345d01 3089 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3090
3091 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3092
0eb0e19c
MF		 3093 path = btrfs_alloc_path();
3094 if (!path)
3095 return -ENOMEM;
3096 path->reada = -1;
3097
0af3d00b 3098 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 3099 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 3100
16cdcec7
MX		 3101 /*
3102 * This function is also used to drop the items in the log tree before
3103 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3104 * it is used to drop the loged items. So we shouldn't kill the delayed
3105 * items.
3106 */
3107 if (min_type == 0 && root == BTRFS_I(inode)->root)
3108 btrfs_kill_delayed_inode_items(inode);
3109
33345d01 3110 key.objectid = ino;
39279cc3 3111 key.offset = (u64)-1;
5f39d397
CM		 3112 key.type = (u8)-1;
3113
85e21bac 3114search_again:
b9473439 3115 path->leave_spinning = 1;
85e21bac 3116 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3117 if (ret < 0) {
3118 err = ret;
3119 goto out;
3120 }
d397712b 3121
85e21bac 3122 if (ret > 0) {
e02119d5
CM		 3123 /* there are no items in the tree for us to truncate, we're
3124 * done
3125 */
8082510e
YZ		 3126 if (path->slots[0] == 0)
3127 goto out;
85e21bac
CM		 3128 path->slots[0]--;
3129 }
3130
d397712b 3131 while (1) {
39279cc3 3132 fi = NULL;
5f39d397
CM		 3133 leaf = path->nodes[0];
3134 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3135 found_type = btrfs_key_type(&found_key);
771ed689 3136 encoding = 0;
39279cc3 3137
33345d01 3138 if (found_key.objectid != ino)
39279cc3 3139 break;
5f39d397 3140
85e21bac 3141 if (found_type < min_type)
39279cc3
CM		 3142 break;
3143
5f39d397 3144 item_end = found_key.offset;
39279cc3 3145 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3146 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3147 struct btrfs_file_extent_item);
179e29e4 3148 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 3149 encoding = btrfs_file_extent_compression(leaf, fi);
3150 encoding |= btrfs_file_extent_encryption(leaf, fi);
3151 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3152
179e29e4 3153 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3154 item_end +=
db94535d 3155 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3156 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3157 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3158 fi);
39279cc3 3159 }
008630c1 3160 item_end--;
39279cc3 3161 }
8082510e
YZ		 3162 if (found_type > min_type) {
3163 del_item = 1;
3164 } else {
3165 if (item_end < new_size)
b888db2b 3166 break;
8082510e
YZ		 3167 if (found_key.offset >= new_size)
3168 del_item = 1;
3169 else
3170 del_item = 0;
39279cc3 3171 }
39279cc3 3172 found_extent = 0;
39279cc3 3173 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3174 if (found_type != BTRFS_EXTENT_DATA_KEY)
3175 goto delete;
3176
3177 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3178 u64 num_dec;
db94535d 3179 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 3180 if (!del_item && !encoding) {
db94535d
CM		 3181 u64 orig_num_bytes =
3182 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3183 extent_num_bytes = new_size -
5f39d397 3184 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3185 extent_num_bytes = extent_num_bytes &
3186 ~((u64)root->sectorsize - 1);
db94535d
CM		 3187 btrfs_set_file_extent_num_bytes(leaf, fi,
3188 extent_num_bytes);
3189 num_dec = (orig_num_bytes -
9069218d 3190 extent_num_bytes);
e02119d5 3191 if (root->ref_cows && extent_start != 0)
a76a3cd4 3192 inode_sub_bytes(inode, num_dec);
5f39d397 3193 btrfs_mark_buffer_dirty(leaf);
39279cc3 3194 } else {
db94535d
CM		 3195 extent_num_bytes =
3196 btrfs_file_extent_disk_num_bytes(leaf,
3197 fi);
5d4f98a2
YZ		 3198 extent_offset = found_key.offset -
3199 btrfs_file_extent_offset(leaf, fi);
3200
39279cc3 3201 /* FIXME blocksize != 4096 */
9069218d 3202 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3203 if (extent_start != 0) {
3204 found_extent = 1;
e02119d5 3205 if (root->ref_cows)
a76a3cd4 3206 inode_sub_bytes(inode, num_dec);
e02119d5 3207 }
39279cc3 3208 }
9069218d 3209 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3210 /*
3211 * we can't truncate inline items that have had
3212 * special encodings
3213 */
3214 if (!del_item &&
3215 btrfs_file_extent_compression(leaf, fi) == 0 &&
3216 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3217 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3218 u32 size = new_size - found_key.offset;
3219
3220 if (root->ref_cows) {
a76a3cd4
YZ		 3221 inode_sub_bytes(inode, item_end + 1 -
3222 new_size);
e02119d5
CM		 3223 }
3224 size =
3225 btrfs_file_extent_calc_inline_size(size);
9069218d 3226 ret = btrfs_truncate_item(trans, root, path,
e02119d5 3227 size, 1);
e02119d5 3228 } else if (root->ref_cows) {
a76a3cd4
YZ		 3229 inode_sub_bytes(inode, item_end + 1 -
3230 found_key.offset);
9069218d 3231 }
39279cc3 3232 }
179e29e4 3233delete:
39279cc3 3234 if (del_item) {
85e21bac
CM		 3235 if (!pending_del_nr) {
3236 /* no pending yet, add ourselves */
3237 pending_del_slot = path->slots[0];
3238 pending_del_nr = 1;
3239 } else if (pending_del_nr &&
3240 path->slots[0] + 1 == pending_del_slot) {
3241 /* hop on the pending chunk */
3242 pending_del_nr++;
3243 pending_del_slot = path->slots[0];
3244 } else {
d397712b 3245 BUG();
85e21bac 3246 }
39279cc3
CM		 3247 } else {
3248 break;
3249 }
0af3d00b
JB		 3250 if (found_extent && (root->ref_cows ||
3251 root == root->fs_info->tree_root)) {
b9473439 3252 btrfs_set_path_blocking(path);
39279cc3 3253 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3254 extent_num_bytes, 0,
3255 btrfs_header_owner(leaf),
33345d01 3256 ino, extent_offset);
39279cc3
CM		 3257 BUG_ON(ret);
3258 }
85e21bac 3259
8082510e
YZ		 3260 if (found_type == BTRFS_INODE_ITEM_KEY)
3261 break;
3262
3263 if (path->slots[0] == 0 ||
3264 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3265 if (root->ref_cows &&
3266 BTRFS_I(inode)->location.objectid !=
3267 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3268 err = -EAGAIN;
3269 goto out;
3270 }
3271 if (pending_del_nr) {
3272 ret = btrfs_del_items(trans, root, path,
3273 pending_del_slot,
3274 pending_del_nr);
3275 BUG_ON(ret);
3276 pending_del_nr = 0;
3277 }
b3b4aa74 3278 btrfs_release_path(path);
85e21bac 3279 goto search_again;
8082510e
YZ		 3280 } else {
3281 path->slots[0]--;
85e21bac 3282 }
39279cc3 3283 }
8082510e 3284out:
85e21bac
CM		 3285 if (pending_del_nr) {
3286 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3287 pending_del_nr);
d68fc57b 3288 BUG_ON(ret);
85e21bac 3289 }
39279cc3 3290 btrfs_free_path(path);
8082510e 3291 return err;
39279cc3
CM		 3292}
3293
3294/*
3295 * taken from block_truncate_page, but does cow as it zeros out
3296 * any bytes left in the last page in the file.
3297 */
3298static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3299{
3300 struct inode *inode = mapping->host;
db94535d 3301 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3302 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3303 struct btrfs_ordered_extent *ordered;
2ac55d41 3304 struct extent_state *cached_state = NULL;
e6dcd2dc 3305 char *kaddr;
db94535d 3306 u32 blocksize = root->sectorsize;
39279cc3
CM		 3307 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3308 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3309 struct page *page;
3b16a4e3 3310 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3311 int ret = 0;
a52d9a80 3312 u64 page_start;
e6dcd2dc 3313 u64 page_end;
39279cc3
CM		 3314
3315 if ((offset & (blocksize - 1)) == 0)
3316 goto out;
0ca1f7ce 3317 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3318 if (ret)
3319 goto out;
39279cc3
CM		 3320
3321 ret = -ENOMEM;
211c17f5 3322again:
3b16a4e3 3323 page = find_or_create_page(mapping, index, mask);
5d5e103a 3324 if (!page) {
0ca1f7ce 3325 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3326 goto out;
5d5e103a 3327 }
e6dcd2dc
CM		 3328
3329 page_start = page_offset(page);
3330 page_end = page_start + PAGE_CACHE_SIZE - 1;
3331
39279cc3 3332 if (!PageUptodate(page)) {
9ebefb18 3333 ret = btrfs_readpage(NULL, page);
39279cc3 3334 lock_page(page);
211c17f5
CM		 3335 if (page->mapping != mapping) {
3336 unlock_page(page);
3337 page_cache_release(page);
3338 goto again;
3339 }
39279cc3
CM		 3340 if (!PageUptodate(page)) {
3341 ret = -EIO;
89642229 3342 goto out_unlock;
39279cc3
CM		 3343 }
3344 }
211c17f5 3345 wait_on_page_writeback(page);
e6dcd2dc 3346
2ac55d41
JB		 3347 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3348 GFP_NOFS);
e6dcd2dc
CM		 3349 set_page_extent_mapped(page);
3350
3351 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3352 if (ordered) {
2ac55d41
JB		 3353 unlock_extent_cached(io_tree, page_start, page_end,
3354 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3355 unlock_page(page);
3356 page_cache_release(page);
eb84ae03 3357 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3358 btrfs_put_ordered_extent(ordered);
3359 goto again;
3360 }
3361
2ac55d41 3362 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3363 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3364 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3365
2ac55d41
JB		 3366 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3367 &cached_state);
9ed74f2d 3368 if (ret) {
2ac55d41
JB		 3369 unlock_extent_cached(io_tree, page_start, page_end,
3370 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3371 goto out_unlock;
3372 }
3373
e6dcd2dc
CM		 3374 ret = 0;
3375 if (offset != PAGE_CACHE_SIZE) {
3376 kaddr = kmap(page);
3377 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3378 flush_dcache_page(page);
3379 kunmap(page);
3380 }
247e743c 3381 ClearPageChecked(page);
e6dcd2dc 3382 set_page_dirty(page);
2ac55d41
JB		 3383 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3384 GFP_NOFS);
39279cc3 3385
89642229 3386out_unlock:
5d5e103a 3387 if (ret)
0ca1f7ce 3388 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3389 unlock_page(page);
3390 page_cache_release(page);
3391out:
3392 return ret;
3393}
3394
695a0d0d
JB		 3395/*
3396 * This function puts in dummy file extents for the area we're creating a hole
3397 * for. So if we are truncating this file to a larger size we need to insert
3398 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3399 * the range between oldsize and size
3400 */
a41ad394 3401int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3402{
9036c102
YZ		 3403 struct btrfs_trans_handle *trans;
3404 struct btrfs_root *root = BTRFS_I(inode)->root;
3405 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3406 struct extent_map *em = NULL;
2ac55d41 3407 struct extent_state *cached_state = NULL;
9036c102 3408 u64 mask = root->sectorsize - 1;
a41ad394 3409 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3410 u64 block_end = (size + mask) & ~mask;
3411 u64 last_byte;
3412 u64 cur_offset;
3413 u64 hole_size;
9ed74f2d 3414 int err = 0;
39279cc3 3415
9036c102
YZ		 3416 if (size <= hole_start)
3417 return 0;
3418
9036c102
YZ		 3419 while (1) {
3420 struct btrfs_ordered_extent *ordered;
3421 btrfs_wait_ordered_range(inode, hole_start,
3422 block_end - hole_start);
2ac55d41
JB		 3423 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3424 &cached_state, GFP_NOFS);
9036c102
YZ		 3425 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3426 if (!ordered)
3427 break;
2ac55d41
JB		 3428 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3429 &cached_state, GFP_NOFS);
9036c102
YZ		 3430 btrfs_put_ordered_extent(ordered);
3431 }
39279cc3 3432
9036c102
YZ		 3433 cur_offset = hole_start;
3434 while (1) {
3435 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3436 block_end - cur_offset, 0);
c704005d 3437 BUG_ON(IS_ERR_OR_NULL(em));
9036c102
YZ		 3438 last_byte = min(extent_map_end(em), block_end);
3439 last_byte = (last_byte + mask) & ~mask;
8082510e 3440 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3441 u64 hint_byte = 0;
9036c102 3442 hole_size = last_byte - cur_offset;
9ed74f2d 3443
a22285a6
YZ		 3444 trans = btrfs_start_transaction(root, 2);
3445 if (IS_ERR(trans)) {
3446 err = PTR_ERR(trans);
9ed74f2d 3447 break;
a22285a6 3448 }
8082510e
YZ		 3449
3450 err = btrfs_drop_extents(trans, inode, cur_offset,
3451 cur_offset + hole_size,
3452 &hint_byte, 1);
5b397377
MX		 3453 if (err) {
3454 btrfs_end_transaction(trans, root);
3893e33b 3455 break;
5b397377 3456 }
8082510e 3457
9036c102 3458 err = btrfs_insert_file_extent(trans, root,
33345d01 3459 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3460 0, hole_size, 0, hole_size,
3461 0, 0, 0);
5b397377
MX		 3462 if (err) {
3463 btrfs_end_transaction(trans, root);
3893e33b 3464 break;
5b397377 3465 }
8082510e 3466
9036c102
YZ		 3467 btrfs_drop_extent_cache(inode, hole_start,
3468 last_byte - 1, 0);
8082510e
YZ		 3469
3470 btrfs_end_transaction(trans, root);
9036c102
YZ		 3471 }
3472 free_extent_map(em);
a22285a6 3473 em = NULL;
9036c102 3474 cur_offset = last_byte;
8082510e 3475 if (cur_offset >= block_end)
9036c102
YZ		 3476 break;
3477 }
1832a6d5 3478
a22285a6 3479 free_extent_map(em);
2ac55d41
JB		 3480 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3481 GFP_NOFS);
9036c102
YZ		 3482 return err;
3483}
39279cc3 3484
a41ad394 3485static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3486{
a41ad394 3487 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3488 int ret;
3489
a41ad394 3490 if (newsize == oldsize)
8082510e
YZ		 3491 return 0;
3492
a41ad394
JB		 3493 if (newsize > oldsize) {
3494 i_size_write(inode, newsize);
3495 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3496 truncate_pagecache(inode, oldsize, newsize);
3497 ret = btrfs_cont_expand(inode, oldsize, newsize);
8082510e 3498 if (ret) {
a41ad394 3499 btrfs_setsize(inode, oldsize);
8082510e
YZ		 3500 return ret;
3501 }
3502
930f028a 3503 mark_inode_dirty(inode);
a41ad394 3504 } else {
8082510e 3505
a41ad394
JB		 3506 /*
3507 * We're truncating a file that used to have good data down to
3508 * zero. Make sure it gets into the ordered flush list so that
3509 * any new writes get down to disk quickly.
3510 */
3511 if (newsize == 0)
3512 BTRFS_I(inode)->ordered_data_close = 1;
8082510e 3513
a41ad394
JB		 3514 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3515 truncate_setsize(inode, newsize);
3516 ret = btrfs_truncate(inode);
8082510e
YZ		 3517 }
3518
a41ad394 3519 return ret;
8082510e
YZ		 3520}
3521
9036c102
YZ		 3522static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3523{
3524 struct inode *inode = dentry->d_inode;
b83cc969 3525 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3526 int err;
39279cc3 3527
b83cc969
LZ		 3528 if (btrfs_root_readonly(root))
3529 return -EROFS;
3530
9036c102
YZ		 3531 err = inode_change_ok(inode, attr);
3532 if (err)
3533 return err;
2bf5a725 3534
5a3f23d5 3535 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3536 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3537 if (err)
3538 return err;
39279cc3 3539 }
9036c102 3540
1025774c
CH		 3541 if (attr->ia_valid) {
3542 setattr_copy(inode, attr);
3543 mark_inode_dirty(inode);
3544
3545 if (attr->ia_valid & ATTR_MODE)
3546 err = btrfs_acl_chmod(inode);
3547 }
33268eaf 3548
39279cc3
CM		 3549 return err;
3550}
61295eb8 3551
bd555975 3552void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3553{
3554 struct btrfs_trans_handle *trans;
3555 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3556 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3557 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3558 unsigned long nr;
39279cc3
CM		 3559 int ret;
3560
1abe9b8a 3561 trace_btrfs_inode_evict(inode);
3562
39279cc3 3563 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3564 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
2cf8572d 3565 btrfs_is_free_space_inode(root, inode)))
bd555975
AV		 3566 goto no_delete;
3567
39279cc3 3568 if (is_bad_inode(inode)) {
7b128766 3569 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3570 goto no_delete;
3571 }
bd555975 3572 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3573 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3574
c71bf099
YZ		 3575 if (root->fs_info->log_root_recovering) {
3576 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3577 goto no_delete;
3578 }
3579
76dda93c
YZ		 3580 if (inode->i_nlink > 0) {
3581 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3582 goto no_delete;
3583 }
3584
4289a667
JB		 3585 rsv = btrfs_alloc_block_rsv(root);
3586 if (!rsv) {
3587 btrfs_orphan_del(NULL, inode);
3588 goto no_delete;
3589 }
4a338542 3590 rsv->size = min_size;
726c35fa 3591 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3592
dbe674a9 3593 btrfs_i_size_write(inode, 0);
5f39d397 3594
4289a667
JB		 3595 /*
3596 * This is a bit simpler than btrfs_truncate since
3597 *
3598 * 1) We've already reserved our space for our orphan item in the
3599 * unlink.
3600 * 2) We're going to delete the inode item, so we don't need to update
3601 * it at all.
3602 *
3603 * So we just need to reserve some slack space in case we add bytes when
3604 * doing the truncate.
3605 */
8082510e 3606 while (1) {
36ba022a 3607 ret = btrfs_block_rsv_refill(root, rsv, min_size);
726c35fa
JB		 3608
3609 /*
3610 * Try and steal from the global reserve since we will
3611 * likely not use this space anyway, we want to try as
3612 * hard as possible to get this to work.
3613 */
3614 if (ret)
3615 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3616
d68fc57b 3617 if (ret) {
4289a667 3618 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3619 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3620 btrfs_orphan_del(NULL, inode);
3621 btrfs_free_block_rsv(root, rsv);
3622 goto no_delete;
3623 }
3624
3625 trans = btrfs_start_transaction(root, 0);
3626 if (IS_ERR(trans)) {
3627 btrfs_orphan_del(NULL, inode);
3628 btrfs_free_block_rsv(root, rsv);
3629 goto no_delete;
d68fc57b 3630 }
7b128766 3631
4289a667
JB		 3632 trans->block_rsv = rsv;
3633
d68fc57b 3634 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3635 if (ret != -EAGAIN)
3636 break;
85e21bac 3637
8082510e
YZ		 3638 nr = trans->blocks_used;
3639 btrfs_end_transaction(trans, root);
3640 trans = NULL;
3641 btrfs_btree_balance_dirty(root, nr);
3642 }
5f39d397 3643
4289a667
JB		 3644 btrfs_free_block_rsv(root, rsv);
3645
8082510e 3646 if (ret == 0) {
4289a667 3647 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3648 ret = btrfs_orphan_del(trans, inode);
3649 BUG_ON(ret);
3650 }
54aa1f4d 3651
4289a667 3652 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3653 if (!(root == root->fs_info->tree_root ||
3654 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3655 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3656
d3c2fdcf 3657 nr = trans->blocks_used;
54aa1f4d 3658 btrfs_end_transaction(trans, root);
d3c2fdcf 3659 btrfs_btree_balance_dirty(root, nr);
39279cc3 3660no_delete:
bd555975 3661 end_writeback(inode);
8082510e 3662 return;
39279cc3
CM		 3663}
3664
3665/*
3666 * this returns the key found in the dir entry in the location pointer.
3667 * If no dir entries were found, location->objectid is 0.
3668 */
3669static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3670 struct btrfs_key *location)
3671{
3672 const char *name = dentry->d_name.name;
3673 int namelen = dentry->d_name.len;
3674 struct btrfs_dir_item *di;
3675 struct btrfs_path *path;
3676 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3677 int ret = 0;
39279cc3
CM		 3678
3679 path = btrfs_alloc_path();
d8926bb3
MF		 3680 if (!path)
3681 return -ENOMEM;
3954401f 3682
33345d01 3683 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3684 namelen, 0);
0d9f7f3e
Y		 3685 if (IS_ERR(di))
3686 ret = PTR_ERR(di);
d397712b 3687
c704005d 3688 if (IS_ERR_OR_NULL(di))
3954401f 3689 goto out_err;
d397712b 3690
5f39d397 3691 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3692out:
39279cc3
CM		 3693 btrfs_free_path(path);
3694 return ret;
3954401f
CM		 3695out_err:
3696 location->objectid = 0;
3697 goto out;
39279cc3
CM		 3698}
3699
3700/*
3701 * when we hit a tree root in a directory, the btrfs part of the inode
3702 * needs to be changed to reflect the root directory of the tree root. This
3703 * is kind of like crossing a mount point.
3704 */
3705static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3706 struct inode *dir,
3707 struct dentry *dentry,
3708 struct btrfs_key *location,
3709 struct btrfs_root **sub_root)
39279cc3 3710{
4df27c4d
YZ		 3711 struct btrfs_path *path;
3712 struct btrfs_root *new_root;
3713 struct btrfs_root_ref *ref;
3714 struct extent_buffer *leaf;
3715 int ret;
3716 int err = 0;
39279cc3 3717
4df27c4d
YZ		 3718 path = btrfs_alloc_path();
3719 if (!path) {
3720 err = -ENOMEM;
3721 goto out;
3722 }
39279cc3 3723
4df27c4d
YZ		 3724 err = -ENOENT;
3725 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3726 BTRFS_I(dir)->root->root_key.objectid,
3727 location->objectid);
3728 if (ret) {
3729 if (ret < 0)
3730 err = ret;
3731 goto out;
3732 }
39279cc3 3733
4df27c4d
YZ		 3734 leaf = path->nodes[0];
3735 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3736 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3737 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3738 goto out;
39279cc3 3739
4df27c4d
YZ		 3740 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3741 (unsigned long)(ref + 1),
3742 dentry->d_name.len);
3743 if (ret)
3744 goto out;
3745
b3b4aa74 3746 btrfs_release_path(path);
4df27c4d
YZ		 3747
3748 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3749 if (IS_ERR(new_root)) {
3750 err = PTR_ERR(new_root);
3751 goto out;
3752 }
3753
3754 if (btrfs_root_refs(&new_root->root_item) == 0) {
3755 err = -ENOENT;
3756 goto out;
3757 }
3758
3759 *sub_root = new_root;
3760 location->objectid = btrfs_root_dirid(&new_root->root_item);
3761 location->type = BTRFS_INODE_ITEM_KEY;
3762 location->offset = 0;
3763 err = 0;
3764out:
3765 btrfs_free_path(path);
3766 return err;
39279cc3
CM		 3767}
3768
5d4f98a2
YZ		 3769static void inode_tree_add(struct inode *inode)
3770{
3771 struct btrfs_root *root = BTRFS_I(inode)->root;
3772 struct btrfs_inode *entry;
03e860bd
FNP		 3773 struct rb_node **p;
3774 struct rb_node *parent;
33345d01 3775 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3776again:
3777 p = &root->inode_tree.rb_node;
3778 parent = NULL;
5d4f98a2 3779
1d3382cb 3780 if (inode_unhashed(inode))
76dda93c
YZ		 3781 return;
3782
5d4f98a2
YZ		 3783 spin_lock(&root->inode_lock);
3784 while (*p) {
3785 parent = *p;
3786 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3787
33345d01 3788 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3789 p = &parent->rb_left;
33345d01 3790 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3791 p = &parent->rb_right;
5d4f98a2
YZ		 3792 else {
3793 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3794 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3795 rb_erase(parent, &root->inode_tree);
3796 RB_CLEAR_NODE(parent);
3797 spin_unlock(&root->inode_lock);
3798 goto again;
5d4f98a2
YZ		 3799 }
3800 }
3801 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3802 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3803 spin_unlock(&root->inode_lock);
3804}
3805
3806static void inode_tree_del(struct inode *inode)
3807{
3808 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3809 int empty = 0;
5d4f98a2 3810
03e860bd 3811 spin_lock(&root->inode_lock);
5d4f98a2 3812 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3813 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3814 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3815 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3816 }
03e860bd 3817 spin_unlock(&root->inode_lock);
76dda93c 3818
0af3d00b
JB		 3819 /*
3820 * Free space cache has inodes in the tree root, but the tree root has a
3821 * root_refs of 0, so this could end up dropping the tree root as a
3822 * snapshot, so we need the extra !root->fs_info->tree_root check to
3823 * make sure we don't drop it.
3824 */
3825 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3826 root != root->fs_info->tree_root) {
76dda93c
YZ		 3827 synchronize_srcu(&root->fs_info->subvol_srcu);
3828 spin_lock(&root->inode_lock);
3829 empty = RB_EMPTY_ROOT(&root->inode_tree);
3830 spin_unlock(&root->inode_lock);
3831 if (empty)
3832 btrfs_add_dead_root(root);
3833 }
3834}
3835
3836int btrfs_invalidate_inodes(struct btrfs_root *root)
3837{
3838 struct rb_node *node;
3839 struct rb_node *prev;
3840 struct btrfs_inode *entry;
3841 struct inode *inode;
3842 u64 objectid = 0;
3843
3844 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3845
3846 spin_lock(&root->inode_lock);
3847again:
3848 node = root->inode_tree.rb_node;
3849 prev = NULL;
3850 while (node) {
3851 prev = node;
3852 entry = rb_entry(node, struct btrfs_inode, rb_node);
3853
33345d01 3854 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 3855 node = node->rb_left;
33345d01 3856 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 3857 node = node->rb_right;
3858 else
3859 break;
3860 }
3861 if (!node) {
3862 while (prev) {
3863 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 3864 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 3865 node = prev;
3866 break;
3867 }
3868 prev = rb_next(prev);
3869 }
3870 }
3871 while (node) {
3872 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 3873 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 3874 inode = igrab(&entry->vfs_inode);
3875 if (inode) {
3876 spin_unlock(&root->inode_lock);
3877 if (atomic_read(&inode->i_count) > 1)
3878 d_prune_aliases(inode);
3879 /*
45321ac5 3880 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 3881 * the inode cache when its usage count
3882 * hits zero.
3883 */
3884 iput(inode);
3885 cond_resched();
3886 spin_lock(&root->inode_lock);
3887 goto again;
3888 }
3889
3890 if (cond_resched_lock(&root->inode_lock))
3891 goto again;
3892
3893 node = rb_next(node);
3894 }
3895 spin_unlock(&root->inode_lock);
3896 return 0;
5d4f98a2
YZ		 3897}
3898
e02119d5
CM		 3899static int btrfs_init_locked_inode(struct inode *inode, void *p)
3900{
3901 struct btrfs_iget_args *args = p;
3902 inode->i_ino = args->ino;
e02119d5 3903 BTRFS_I(inode)->root = args->root;
6a63209f 3904 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3905 return 0;
3906}
3907
3908static int btrfs_find_actor(struct inode *inode, void *opaque)
3909{
3910 struct btrfs_iget_args *args = opaque;
33345d01 3911 return args->ino == btrfs_ino(inode) &&
d397712b 3912 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3913}
3914
5d4f98a2
YZ		 3915static struct inode *btrfs_iget_locked(struct super_block *s,
3916 u64 objectid,
3917 struct btrfs_root *root)
39279cc3
CM		 3918{
3919 struct inode *inode;
3920 struct btrfs_iget_args args;
3921 args.ino = objectid;
3922 args.root = root;
3923
3924 inode = iget5_locked(s, objectid, btrfs_find_actor,
3925 btrfs_init_locked_inode,
3926 (void *)&args);
3927 return inode;
3928}
3929
1a54ef8c
BR		 3930/* Get an inode object given its location and corresponding root.
3931 * Returns in *is_new if the inode was read from disk
3932 */
3933struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 3934 struct btrfs_root *root, int *new)
1a54ef8c
BR		 3935{
3936 struct inode *inode;
3937
3938 inode = btrfs_iget_locked(s, location->objectid, root);
3939 if (!inode)
5d4f98a2 3940 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3941
3942 if (inode->i_state & I_NEW) {
3943 BTRFS_I(inode)->root = root;
3944 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3945 btrfs_read_locked_inode(inode);
1748f843
MF		 3946 if (!is_bad_inode(inode)) {
3947 inode_tree_add(inode);
3948 unlock_new_inode(inode);
3949 if (new)
3950 *new = 1;
3951 } else {
e0b6d65b
ST		 3952 unlock_new_inode(inode);
3953 iput(inode);
3954 inode = ERR_PTR(-ESTALE);
1748f843
MF		 3955 }
3956 }
3957
1a54ef8c
BR		 3958 return inode;
3959}
3960
4df27c4d
YZ		 3961static struct inode *new_simple_dir(struct super_block *s,
3962 struct btrfs_key *key,
3963 struct btrfs_root *root)
3964{
3965 struct inode *inode = new_inode(s);
3966
3967 if (!inode)
3968 return ERR_PTR(-ENOMEM);
3969
4df27c4d
YZ		 3970 BTRFS_I(inode)->root = root;
3971 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3972 BTRFS_I(inode)->dummy_inode = 1;
3973
3974 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3975 inode->i_op = &simple_dir_inode_operations;
3976 inode->i_fop = &simple_dir_operations;
3977 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3978 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3979
3980 return inode;
3981}
3982
3de4586c 3983struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3984{
d397712b 3985 struct inode *inode;
4df27c4d 3986 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 3987 struct btrfs_root *sub_root = root;
3988 struct btrfs_key location;
76dda93c 3989 int index;
b4aff1f8 3990 int ret = 0;
39279cc3
CM		 3991
3992 if (dentry->d_name.len > BTRFS_NAME_LEN)
3993 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3994
b4aff1f8
JB		 3995 if (unlikely(d_need_lookup(dentry))) {
3996 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3997 kfree(dentry->d_fsdata);
3998 dentry->d_fsdata = NULL;
a66e7cc6
JB		 3999 /* This thing is hashed, drop it for now */
4000 d_drop(dentry);
b4aff1f8
JB		 4001 } else {
4002 ret = btrfs_inode_by_name(dir, dentry, &location);
4003 }
5f39d397 4004
39279cc3
CM		 4005 if (ret < 0)
4006 return ERR_PTR(ret);
5f39d397 4007
4df27c4d
YZ		 4008 if (location.objectid == 0)
4009 return NULL;
4010
4011 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4012 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4013 return inode;
4014 }
4015
4016 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4017
76dda93c 4018 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4019 ret = fixup_tree_root_location(root, dir, dentry,
4020 &location, &sub_root);
4021 if (ret < 0) {
4022 if (ret != -ENOENT)
4023 inode = ERR_PTR(ret);
4024 else
4025 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4026 } else {
73f73415 4027 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4028 }
76dda93c
YZ		 4029 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4030
34d19bad 4031 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4032 down_read(&root->fs_info->cleanup_work_sem);
4033 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4034 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4035 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 4036 if (ret)
4037 inode = ERR_PTR(ret);
c71bf099
YZ		 4038 }
4039
3de4586c
CM		 4040 return inode;
4041}
4042
fe15ce44 4043static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4044{
4045 struct btrfs_root *root;
4046
efefb143
YZ		 4047 if (!dentry->d_inode && !IS_ROOT(dentry))
4048 dentry = dentry->d_parent;
76dda93c 4049
efefb143
YZ		 4050 if (dentry->d_inode) {
4051 root = BTRFS_I(dentry->d_inode)->root;
4052 if (btrfs_root_refs(&root->root_item) == 0)
4053 return 1;
4054 }
76dda93c
YZ		 4055 return 0;
4056}
4057
b4aff1f8
JB		 4058static void btrfs_dentry_release(struct dentry *dentry)
4059{
4060 if (dentry->d_fsdata)
4061 kfree(dentry->d_fsdata);
4062}
4063
3de4586c
CM		 4064static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4065 struct nameidata *nd)
4066{
a66e7cc6
JB		 4067 struct dentry *ret;
4068
4069 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4070 if (unlikely(d_need_lookup(dentry))) {
4071 spin_lock(&dentry->d_lock);
4072 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4073 spin_unlock(&dentry->d_lock);
4074 }
4075 return ret;
39279cc3
CM		 4076}
4077
16cdcec7 4078unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4079 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4080};
4081
cbdf5a24
DW		 4082static int btrfs_real_readdir(struct file *filp, void *dirent,
4083 filldir_t filldir)
39279cc3 4084{
6da6abae 4085 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4086 struct btrfs_root *root = BTRFS_I(inode)->root;
4087 struct btrfs_item *item;
4088 struct btrfs_dir_item *di;
4089 struct btrfs_key key;
5f39d397 4090 struct btrfs_key found_key;
39279cc3 4091 struct btrfs_path *path;
16cdcec7
MX		 4092 struct list_head ins_list;
4093 struct list_head del_list;
b4aff1f8 4094 struct qstr q;
39279cc3 4095 int ret;
5f39d397 4096 struct extent_buffer *leaf;
39279cc3 4097 int slot;
39279cc3
CM		 4098 unsigned char d_type;
4099 int over = 0;
4100 u32 di_cur;
4101 u32 di_total;
4102 u32 di_len;
4103 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4104 char tmp_name[32];
4105 char *name_ptr;
4106 int name_len;
16cdcec7 4107 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4108
4109 /* FIXME, use a real flag for deciding about the key type */
4110 if (root->fs_info->tree_root == root)
4111 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4112
3954401f
CM		 4113 /* special case for "." */
4114 if (filp->f_pos == 0) {
3765fefa
HS		 4115 over = filldir(dirent, ".", 1,
4116 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4117 if (over)
4118 return 0;
4119 filp->f_pos = 1;
4120 }
3954401f
CM		 4121 /* special case for .., just use the back ref */
4122 if (filp->f_pos == 1) {
5ecc7e5d 4123 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4124 over = filldir(dirent, "..", 2,
3765fefa 4125 filp->f_pos, pino, DT_DIR);
3954401f 4126 if (over)
49593bfa 4127 return 0;
3954401f
CM		 4128 filp->f_pos = 2;
4129 }
49593bfa 4130 path = btrfs_alloc_path();
16cdcec7
MX		 4131 if (!path)
4132 return -ENOMEM;
ff5714cc 4133
026fd317 4134 path->reada = 1;
49593bfa 4135
16cdcec7
MX		 4136 if (key_type == BTRFS_DIR_INDEX_KEY) {
4137 INIT_LIST_HEAD(&ins_list);
4138 INIT_LIST_HEAD(&del_list);
4139 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4140 }
4141
39279cc3
CM		 4142 btrfs_set_key_type(&key, key_type);
4143 key.offset = filp->f_pos;
33345d01 4144 key.objectid = btrfs_ino(inode);
5f39d397 4145
39279cc3
CM		 4146 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4147 if (ret < 0)
4148 goto err;
49593bfa
DW		 4149
4150 while (1) {
5f39d397 4151 leaf = path->nodes[0];
39279cc3 4152 slot = path->slots[0];
b9e03af0
LZ		 4153 if (slot >= btrfs_header_nritems(leaf)) {
4154 ret = btrfs_next_leaf(root, path);
4155 if (ret < 0)
4156 goto err;
4157 else if (ret > 0)
4158 break;
4159 continue;
39279cc3 4160 }
3de4586c 4161
5f39d397
CM		 4162 item = btrfs_item_nr(leaf, slot);
4163 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4164
4165 if (found_key.objectid != key.objectid)
39279cc3 4166 break;
5f39d397 4167 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4168 break;
5f39d397 4169 if (found_key.offset < filp->f_pos)
b9e03af0 4170 goto next;
16cdcec7
MX		 4171 if (key_type == BTRFS_DIR_INDEX_KEY &&
4172 btrfs_should_delete_dir_index(&del_list,
4173 found_key.offset))
4174 goto next;
5f39d397
CM		 4175
4176 filp->f_pos = found_key.offset;
16cdcec7 4177 is_curr = 1;
49593bfa 4178
39279cc3
CM		 4179 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4180 di_cur = 0;
5f39d397 4181 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4182
4183 while (di_cur < di_total) {
5f39d397 4184 struct btrfs_key location;
b4aff1f8 4185 struct dentry *tmp;
5f39d397 4186
22a94d44
JB		 4187 if (verify_dir_item(root, leaf, di))
4188 break;
4189
5f39d397 4190 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4191 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4192 name_ptr = tmp_name;
4193 } else {
4194 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4195 if (!name_ptr) {
4196 ret = -ENOMEM;
4197 goto err;
4198 }
5f39d397
CM		 4199 }
4200 read_extent_buffer(leaf, name_ptr,
4201 (unsigned long)(di + 1), name_len);
4202
4203 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4204 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4205
b4aff1f8
JB		 4206 q.name = name_ptr;
4207 q.len = name_len;
4208 q.hash = full_name_hash(q.name, q.len);
4209 tmp = d_lookup(filp->f_dentry, &q);
4210 if (!tmp) {
4211 struct btrfs_key *newkey;
4212
4213 newkey = kzalloc(sizeof(struct btrfs_key),
4214 GFP_NOFS);
4215 if (!newkey)
4216 goto no_dentry;
4217 tmp = d_alloc(filp->f_dentry, &q);
4218 if (!tmp) {
4219 kfree(newkey);
4220 dput(tmp);
4221 goto no_dentry;
4222 }
4223 memcpy(newkey, &location,
4224 sizeof(struct btrfs_key));
4225 tmp->d_fsdata = newkey;
4226 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4227 d_rehash(tmp);
4228 dput(tmp);
4229 } else {
4230 dput(tmp);
4231 }
4232no_dentry:
3de4586c
CM		 4233 /* is this a reference to our own snapshot? If so
4234 * skip it
4235 */
4236 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4237 location.objectid == root->root_key.objectid) {
4238 over = 0;
4239 goto skip;
4240 }
5f39d397 4241 over = filldir(dirent, name_ptr, name_len,
49593bfa 4242 found_key.offset, location.objectid,
39279cc3 4243 d_type);
5f39d397 4244
3de4586c 4245skip:
5f39d397
CM		 4246 if (name_ptr != tmp_name)
4247 kfree(name_ptr);
4248
39279cc3
CM		 4249 if (over)
4250 goto nopos;
5103e947 4251 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4252 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4253 di_cur += di_len;
4254 di = (struct btrfs_dir_item *)((char *)di + di_len);
4255 }
b9e03af0
LZ		 4256next:
4257 path->slots[0]++;
39279cc3 4258 }
49593bfa 4259
16cdcec7
MX		 4260 if (key_type == BTRFS_DIR_INDEX_KEY) {
4261 if (is_curr)
4262 filp->f_pos++;
4263 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4264 &ins_list);
4265 if (ret)
4266 goto nopos;
4267 }
4268
49593bfa 4269 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4270 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4271 /*
4272 * 32-bit glibc will use getdents64, but then strtol -
4273 * so the last number we can serve is this.
4274 */
4275 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4276 else
4277 filp->f_pos++;
39279cc3
CM		 4278nopos:
4279 ret = 0;
4280err:
16cdcec7
MX		 4281 if (key_type == BTRFS_DIR_INDEX_KEY)
4282 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4283 btrfs_free_path(path);
39279cc3
CM		 4284 return ret;
4285}
4286
a9185b41 4287int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4288{
4289 struct btrfs_root *root = BTRFS_I(inode)->root;
4290 struct btrfs_trans_handle *trans;
4291 int ret = 0;
0af3d00b 4292 bool nolock = false;
39279cc3 4293
8929ecfa 4294 if (BTRFS_I(inode)->dummy_inode)
4ca8b41e
CM		 4295 return 0;
4296
2cf8572d 4297 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
82d5902d 4298 nolock = true;
0af3d00b 4299
a9185b41 4300 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4301 if (nolock)
7a7eaa40 4302 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4303 else
7a7eaa40 4304 trans = btrfs_join_transaction(root);
3612b495
TI		 4305 if (IS_ERR(trans))
4306 return PTR_ERR(trans);
0af3d00b
JB		 4307 if (nolock)
4308 ret = btrfs_end_transaction_nolock(trans, root);
4309 else
4310 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4311 }
4312 return ret;
4313}
4314
4315/*
54aa1f4d 4316 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4317 * inode changes. But, it is most likely to find the inode in cache.
4318 * FIXME, needs more benchmarking...there are no reasons other than performance
4319 * to keep or drop this code.
4320 */
aa385729 4321void btrfs_dirty_inode(struct inode *inode, int flags)
39279cc3
CM		 4322{
4323 struct btrfs_root *root = BTRFS_I(inode)->root;
4324 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4325 int ret;
4326
4327 if (BTRFS_I(inode)->dummy_inode)
4328 return;
39279cc3 4329
7a7eaa40 4330 trans = btrfs_join_transaction(root);
3612b495 4331 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 4332
4333 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4334 if (ret && ret == -ENOSPC) {
4335 /* whoops, lets try again with the full transaction */
4336 btrfs_end_transaction(trans, root);
4337 trans = btrfs_start_transaction(root, 1);
9aeead73 4338 if (IS_ERR(trans)) {
7a36ddec 4339 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4340 "dirty inode %llu error %ld\n",
4341 (unsigned long long)btrfs_ino(inode),
4342 PTR_ERR(trans));
9aeead73
CM		 4343 return;
4344 }
8929ecfa 4345
94b60442
CM		 4346 ret = btrfs_update_inode(trans, root, inode);
4347 if (ret) {
7a36ddec 4348 printk_ratelimited(KERN_ERR "btrfs: fail to "
33345d01
LZ		 4349 "dirty inode %llu error %d\n",
4350 (unsigned long long)btrfs_ino(inode),
4351 ret);
94b60442
CM		 4352 }
4353 }
39279cc3 4354 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4355 if (BTRFS_I(inode)->delayed_node)
4356 btrfs_balance_delayed_items(root);
39279cc3
CM		 4357}
4358
d352ac68
CM		 4359/*
4360 * find the highest existing sequence number in a directory
4361 * and then set the in-memory index_cnt variable to reflect
4362 * free sequence numbers
4363 */
aec7477b
JB		 4364static int btrfs_set_inode_index_count(struct inode *inode)
4365{
4366 struct btrfs_root *root = BTRFS_I(inode)->root;
4367 struct btrfs_key key, found_key;
4368 struct btrfs_path *path;
4369 struct extent_buffer *leaf;
4370 int ret;
4371
33345d01 4372 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4373 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4374 key.offset = (u64)-1;
4375
4376 path = btrfs_alloc_path();
4377 if (!path)
4378 return -ENOMEM;
4379
4380 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4381 if (ret < 0)
4382 goto out;
4383 /* FIXME: we should be able to handle this */
4384 if (ret == 0)
4385 goto out;
4386 ret = 0;
4387
4388 /*
4389 * MAGIC NUMBER EXPLANATION:
4390 * since we search a directory based on f_pos we have to start at 2
4391 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4392 * else has to start at 2
4393 */
4394 if (path->slots[0] == 0) {
4395 BTRFS_I(inode)->index_cnt = 2;
4396 goto out;
4397 }
4398
4399 path->slots[0]--;
4400
4401 leaf = path->nodes[0];
4402 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4403
33345d01 4404 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4405 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4406 BTRFS_I(inode)->index_cnt = 2;
4407 goto out;
4408 }
4409
4410 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4411out:
4412 btrfs_free_path(path);
4413 return ret;
4414}
4415
d352ac68
CM		 4416/*
4417 * helper to find a free sequence number in a given directory. This current
4418 * code is very simple, later versions will do smarter things in the btree
4419 */
3de4586c 4420int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4421{
4422 int ret = 0;
4423
4424 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4425 ret = btrfs_inode_delayed_dir_index_count(dir);
4426 if (ret) {
4427 ret = btrfs_set_inode_index_count(dir);
4428 if (ret)
4429 return ret;
4430 }
aec7477b
JB		 4431 }
4432
00e4e6b3 4433 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4434 BTRFS_I(dir)->index_cnt++;
4435
4436 return ret;
4437}
4438
39279cc3
CM		 4439static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4440 struct btrfs_root *root,
aec7477b 4441 struct inode *dir,
9c58309d 4442 const char *name, int name_len,
d82a6f1d
JB		 4443 u64 ref_objectid, u64 objectid, int mode,
4444 u64 *index)
39279cc3
CM		 4445{
4446 struct inode *inode;
5f39d397 4447 struct btrfs_inode_item *inode_item;
39279cc3 4448 struct btrfs_key *location;
5f39d397 4449 struct btrfs_path *path;
9c58309d
CM		 4450 struct btrfs_inode_ref *ref;
4451 struct btrfs_key key[2];
4452 u32 sizes[2];
4453 unsigned long ptr;
39279cc3
CM		 4454 int ret;
4455 int owner;
4456
5f39d397 4457 path = btrfs_alloc_path();
d8926bb3
MF		 4458 if (!path)
4459 return ERR_PTR(-ENOMEM);
5f39d397 4460
39279cc3 4461 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4462 if (!inode) {
4463 btrfs_free_path(path);
39279cc3 4464 return ERR_PTR(-ENOMEM);
8fb27640 4465 }
39279cc3 4466
581bb050
LZ		 4467 /*
4468 * we have to initialize this early, so we can reclaim the inode
4469 * number if we fail afterwards in this function.
4470 */
4471 inode->i_ino = objectid;
4472
aec7477b 4473 if (dir) {
1abe9b8a 4474 trace_btrfs_inode_request(dir);
4475
3de4586c 4476 ret = btrfs_set_inode_index(dir, index);
09771430 4477 if (ret) {
8fb27640 4478 btrfs_free_path(path);
09771430 4479 iput(inode);
aec7477b 4480 return ERR_PTR(ret);
09771430 4481 }
aec7477b
JB		 4482 }
4483 /*
4484 * index_cnt is ignored for everything but a dir,
4485 * btrfs_get_inode_index_count has an explanation for the magic
4486 * number
4487 */
4488 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4489 BTRFS_I(inode)->root = root;
e02119d5 4490 BTRFS_I(inode)->generation = trans->transid;
76195853 4491 inode->i_generation = BTRFS_I(inode)->generation;
6a63209f 4492 btrfs_set_inode_space_info(root, inode);
b888db2b 4493
569254b0 4494 if (S_ISDIR(mode))
39279cc3
CM		 4495 owner = 0;
4496 else
4497 owner = 1;
9c58309d
CM		 4498
4499 key[0].objectid = objectid;
4500 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4501 key[0].offset = 0;
4502
4503 key[1].objectid = objectid;
4504 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4505 key[1].offset = ref_objectid;
4506
4507 sizes[0] = sizeof(struct btrfs_inode_item);
4508 sizes[1] = name_len + sizeof(*ref);
4509
b9473439 4510 path->leave_spinning = 1;
9c58309d
CM		 4511 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4512 if (ret != 0)
5f39d397
CM		 4513 goto fail;
4514
ecc11fab 4515 inode_init_owner(inode, dir, mode);
a76a3cd4 4516 inode_set_bytes(inode, 0);
39279cc3 4517 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4518 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4519 struct btrfs_inode_item);
e02119d5 4520 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4521
4522 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4523 struct btrfs_inode_ref);
4524 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4525 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4526 ptr = (unsigned long)(ref + 1);
4527 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4528
5f39d397
CM		 4529 btrfs_mark_buffer_dirty(path->nodes[0]);
4530 btrfs_free_path(path);
4531
39279cc3
CM		 4532 location = &BTRFS_I(inode)->location;
4533 location->objectid = objectid;
39279cc3
CM		 4534 location->offset = 0;
4535 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4536
6cbff00f
CH		 4537 btrfs_inherit_iflags(inode, dir);
4538
569254b0 4539 if (S_ISREG(mode)) {
94272164
CM		 4540 if (btrfs_test_opt(root, NODATASUM))
4541 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4542 if (btrfs_test_opt(root, NODATACOW) ||
4543 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4544 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4545 }
4546
39279cc3 4547 insert_inode_hash(inode);
5d4f98a2 4548 inode_tree_add(inode);
1abe9b8a 4549
4550 trace_btrfs_inode_new(inode);
1973f0fa 4551 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4552
39279cc3 4553 return inode;
5f39d397 4554fail:
aec7477b
JB		 4555 if (dir)
4556 BTRFS_I(dir)->index_cnt--;
5f39d397 4557 btrfs_free_path(path);
09771430 4558 iput(inode);
5f39d397 4559 return ERR_PTR(ret);
39279cc3
CM		 4560}
4561
4562static inline u8 btrfs_inode_type(struct inode *inode)
4563{
4564 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4565}
4566
d352ac68
CM		 4567/*
4568 * utility function to add 'inode' into 'parent_inode' with
4569 * a give name and a given sequence number.
4570 * if 'add_backref' is true, also insert a backref from the
4571 * inode to the parent directory.
4572 */
e02119d5
CM		 4573int btrfs_add_link(struct btrfs_trans_handle *trans,
4574 struct inode *parent_inode, struct inode *inode,
4575 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4576{
4df27c4d 4577 int ret = 0;
39279cc3 4578 struct btrfs_key key;
e02119d5 4579 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4580 u64 ino = btrfs_ino(inode);
4581 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4582
33345d01 4583 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4584 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4585 } else {
33345d01 4586 key.objectid = ino;
4df27c4d
YZ		 4587 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4588 key.offset = 0;
4589 }
4590
33345d01 4591 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4592 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4593 key.objectid, root->root_key.objectid,
33345d01 4594 parent_ino, index, name, name_len);
4df27c4d 4595 } else if (add_backref) {
33345d01
LZ		 4596 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4597 parent_ino, index);
4df27c4d 4598 }
39279cc3 4599
39279cc3 4600 if (ret == 0) {
4df27c4d 4601 ret = btrfs_insert_dir_item(trans, root, name, name_len,
16cdcec7 4602 parent_inode, &key,
4df27c4d
YZ		 4603 btrfs_inode_type(inode), index);
4604 BUG_ON(ret);
4605
dbe674a9 4606 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 4607 name_len * 2);
79c44584 4608 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 4609 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 4610 }
4611 return ret;
4612}
4613
4614static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4615 struct inode *dir, struct dentry *dentry,
4616 struct inode *inode, int backref, u64 index)
39279cc3 4617{
a1b075d2
JB		 4618 int err = btrfs_add_link(trans, dir, inode,
4619 dentry->d_name.name, dentry->d_name.len,
4620 backref, index);
39279cc3
CM		 4621 if (!err) {
4622 d_instantiate(dentry, inode);
4623 return 0;
4624 }
4625 if (err > 0)
4626 err = -EEXIST;
4627 return err;
4628}
4629
618e21d5
JB		 4630static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4631 int mode, dev_t rdev)
4632{
4633 struct btrfs_trans_handle *trans;
4634 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4635 struct inode *inode = NULL;
618e21d5
JB		 4636 int err;
4637 int drop_inode = 0;
4638 u64 objectid;
1832a6d5 4639 unsigned long nr = 0;
00e4e6b3 4640 u64 index = 0;
618e21d5
JB		 4641
4642 if (!new_valid_dev(rdev))
4643 return -EINVAL;
4644
9ed74f2d
JB		 4645 /*
4646 * 2 for inode item and ref
4647 * 2 for dir items
4648 * 1 for xattr if selinux is on
4649 */
a22285a6
YZ		 4650 trans = btrfs_start_transaction(root, 5);
4651 if (IS_ERR(trans))
4652 return PTR_ERR(trans);
1832a6d5 4653
581bb050
LZ		 4654 err = btrfs_find_free_ino(root, &objectid);
4655 if (err)
4656 goto out_unlock;
4657
aec7477b 4658 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4659 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4660 mode, &index);
7cf96da3
TI		 4661 if (IS_ERR(inode)) {
4662 err = PTR_ERR(inode);
618e21d5 4663 goto out_unlock;
7cf96da3 4664 }
618e21d5 4665
2a7dba39 4666 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4667 if (err) {
4668 drop_inode = 1;
4669 goto out_unlock;
4670 }
4671
a1b075d2 4672 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4673 if (err)
4674 drop_inode = 1;
4675 else {
4676 inode->i_op = &btrfs_special_inode_operations;
4677 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4678 btrfs_update_inode(trans, root, inode);
618e21d5 4679 }
618e21d5 4680out_unlock:
d3c2fdcf 4681 nr = trans->blocks_used;
89ce8a63 4682 btrfs_end_transaction_throttle(trans, root);
a22285a6 4683 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4684 if (drop_inode) {
4685 inode_dec_link_count(inode);
4686 iput(inode);
4687 }
618e21d5
JB		 4688 return err;
4689}
4690
39279cc3
CM		 4691static int btrfs_create(struct inode *dir, struct dentry *dentry,
4692 int mode, struct nameidata *nd)
4693{
4694 struct btrfs_trans_handle *trans;
4695 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4696 struct inode *inode = NULL;
39279cc3 4697 int drop_inode = 0;
a22285a6 4698 int err;
1832a6d5 4699 unsigned long nr = 0;
39279cc3 4700 u64 objectid;
00e4e6b3 4701 u64 index = 0;
39279cc3 4702
9ed74f2d
JB		 4703 /*
4704 * 2 for inode item and ref
4705 * 2 for dir items
4706 * 1 for xattr if selinux is on
4707 */
a22285a6
YZ		 4708 trans = btrfs_start_transaction(root, 5);
4709 if (IS_ERR(trans))
4710 return PTR_ERR(trans);
9ed74f2d 4711
581bb050
LZ		 4712 err = btrfs_find_free_ino(root, &objectid);
4713 if (err)
4714 goto out_unlock;
4715
aec7477b 4716 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4717 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4718 mode, &index);
7cf96da3
TI		 4719 if (IS_ERR(inode)) {
4720 err = PTR_ERR(inode);
39279cc3 4721 goto out_unlock;
7cf96da3 4722 }
39279cc3 4723
2a7dba39 4724 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4725 if (err) {
4726 drop_inode = 1;
4727 goto out_unlock;
4728 }
4729
a1b075d2 4730 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4731 if (err)
4732 drop_inode = 1;
4733 else {
4734 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4735 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4736 inode->i_fop = &btrfs_file_operations;
4737 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4738 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 4739 }
39279cc3 4740out_unlock:
d3c2fdcf 4741 nr = trans->blocks_used;
ab78c84d 4742 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4743 if (drop_inode) {
4744 inode_dec_link_count(inode);
4745 iput(inode);
4746 }
d3c2fdcf 4747 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4748 return err;
4749}
4750
4751static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4752 struct dentry *dentry)
4753{
4754 struct btrfs_trans_handle *trans;
4755 struct btrfs_root *root = BTRFS_I(dir)->root;
4756 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4757 u64 index;
1832a6d5 4758 unsigned long nr = 0;
39279cc3
CM		 4759 int err;
4760 int drop_inode = 0;
4761
4a8be425
TH		 4762 /* do not allow sys_link's with other subvols of the same device */
4763 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4764 return -EXDEV;
4a8be425 4765
c055e99e
AV		 4766 if (inode->i_nlink == ~0U)
4767 return -EMLINK;
4a8be425 4768
3de4586c 4769 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4770 if (err)
4771 goto fail;
4772
a22285a6 4773 /*
7e6b6465 4774 * 2 items for inode and inode ref
a22285a6 4775 * 2 items for dir items
7e6b6465 4776 * 1 item for parent inode
a22285a6 4777 */
7e6b6465 4778 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4779 if (IS_ERR(trans)) {
4780 err = PTR_ERR(trans);
4781 goto fail;
4782 }
5f39d397 4783
3153495d
MX		 4784 btrfs_inc_nlink(inode);
4785 inode->i_ctime = CURRENT_TIME;
7de9c6ee 4786 ihold(inode);
aec7477b 4787
a1b075d2 4788 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 4789
a5719521 4790 if (err) {
54aa1f4d 4791 drop_inode = 1;
a5719521 4792 } else {
10d9f309 4793 struct dentry *parent = dentry->d_parent;
a5719521
YZ		 4794 err = btrfs_update_inode(trans, root, inode);
4795 BUG_ON(err);
6a912213 4796 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 4797 }
39279cc3 4798
d3c2fdcf 4799 nr = trans->blocks_used;
ab78c84d 4800 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4801fail:
39279cc3
CM		 4802 if (drop_inode) {
4803 inode_dec_link_count(inode);
4804 iput(inode);
4805 }
d3c2fdcf 4806 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4807 return err;
4808}
4809
39279cc3
CM		 4810static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4811{
b9d86667 4812 struct inode *inode = NULL;
39279cc3
CM		 4813 struct btrfs_trans_handle *trans;
4814 struct btrfs_root *root = BTRFS_I(dir)->root;
4815 int err = 0;
4816 int drop_on_err = 0;
b9d86667 4817 u64 objectid = 0;
00e4e6b3 4818 u64 index = 0;
d3c2fdcf 4819 unsigned long nr = 1;
39279cc3 4820
9ed74f2d
JB		 4821 /*
4822 * 2 items for inode and ref
4823 * 2 items for dir items
4824 * 1 for xattr if selinux is on
4825 */
a22285a6
YZ		 4826 trans = btrfs_start_transaction(root, 5);
4827 if (IS_ERR(trans))
4828 return PTR_ERR(trans);
39279cc3 4829
581bb050
LZ		 4830 err = btrfs_find_free_ino(root, &objectid);
4831 if (err)
4832 goto out_fail;
4833
aec7477b 4834 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4835 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4836 S_IFDIR | mode, &index);
39279cc3
CM		 4837 if (IS_ERR(inode)) {
4838 err = PTR_ERR(inode);
4839 goto out_fail;
4840 }
5f39d397 4841
39279cc3 4842 drop_on_err = 1;
33268eaf 4843
2a7dba39 4844 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4845 if (err)
4846 goto out_fail;
4847
39279cc3
CM		 4848 inode->i_op = &btrfs_dir_inode_operations;
4849 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 4850
dbe674a9 4851 btrfs_i_size_write(inode, 0);
39279cc3
CM		 4852 err = btrfs_update_inode(trans, root, inode);
4853 if (err)
4854 goto out_fail;
5f39d397 4855
a1b075d2
JB		 4856 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4857 dentry->d_name.len, 0, index);
39279cc3
CM		 4858 if (err)
4859 goto out_fail;
5f39d397 4860
39279cc3
CM		 4861 d_instantiate(dentry, inode);
4862 drop_on_err = 0;
39279cc3
CM		 4863
4864out_fail:
d3c2fdcf 4865 nr = trans->blocks_used;
ab78c84d 4866 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 4867 if (drop_on_err)
4868 iput(inode);
d3c2fdcf 4869 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4870 return err;
4871}
4872
d352ac68
CM		 4873/* helper for btfs_get_extent. Given an existing extent in the tree,
4874 * and an extent that you want to insert, deal with overlap and insert
4875 * the new extent into the tree.
4876 */
3b951516
CM		 4877static int merge_extent_mapping(struct extent_map_tree *em_tree,
4878 struct extent_map *existing,
e6dcd2dc
CM		 4879 struct extent_map *em,
4880 u64 map_start, u64 map_len)
3b951516
CM		 4881{
4882 u64 start_diff;
3b951516 4883
e6dcd2dc
CM		 4884 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4885 start_diff = map_start - em->start;
4886 em->start = map_start;
4887 em->len = map_len;
c8b97818
CM		 4888 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4889 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 4890 em->block_start += start_diff;
c8b97818
CM		 4891 em->block_len -= start_diff;
4892 }
e6dcd2dc 4893 return add_extent_mapping(em_tree, em);
3b951516
CM		 4894}
4895
c8b97818
CM		 4896static noinline int uncompress_inline(struct btrfs_path *path,
4897 struct inode *inode, struct page *page,
4898 size_t pg_offset, u64 extent_offset,
4899 struct btrfs_file_extent_item *item)
4900{
4901 int ret;
4902 struct extent_buffer *leaf = path->nodes[0];
4903 char *tmp;
4904 size_t max_size;
4905 unsigned long inline_size;
4906 unsigned long ptr;
261507a0 4907 int compress_type;
c8b97818
CM		 4908
4909 WARN_ON(pg_offset != 0);
261507a0 4910 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 4911 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4912 inline_size = btrfs_file_extent_inline_item_len(leaf,
4913 btrfs_item_nr(leaf, path->slots[0]));
4914 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 4915 if (!tmp)
4916 return -ENOMEM;
c8b97818
CM		 4917 ptr = btrfs_file_extent_inline_start(item);
4918
4919 read_extent_buffer(leaf, tmp, ptr, inline_size);
4920
5b050f04 4921 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 4922 ret = btrfs_decompress(compress_type, tmp, page,
4923 extent_offset, inline_size, max_size);
c8b97818
CM		 4924 if (ret) {
4925 char *kaddr = kmap_atomic(page, KM_USER0);
4926 unsigned long copy_size = min_t(u64,
4927 PAGE_CACHE_SIZE - pg_offset,
4928 max_size - extent_offset);
4929 memset(kaddr + pg_offset, 0, copy_size);
4930 kunmap_atomic(kaddr, KM_USER0);
4931 }
4932 kfree(tmp);
4933 return 0;
4934}
4935
d352ac68
CM		 4936/*
4937 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4938 * the ugly parts come from merging extents from the disk with the in-ram
4939 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4940 * where the in-ram extents might be locked pending data=ordered completion.
4941 *
4942 * This also copies inline extents directly into the page.
4943 */
d397712b 4944
a52d9a80 4945struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4946 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4947 int create)
4948{
4949 int ret;
4950 int err = 0;
db94535d 4951 u64 bytenr;
a52d9a80
CM		 4952 u64 extent_start = 0;
4953 u64 extent_end = 0;
33345d01 4954 u64 objectid = btrfs_ino(inode);
a52d9a80 4955 u32 found_type;
f421950f 4956 struct btrfs_path *path = NULL;
a52d9a80
CM		 4957 struct btrfs_root *root = BTRFS_I(inode)->root;
4958 struct btrfs_file_extent_item *item;
5f39d397
CM		 4959 struct extent_buffer *leaf;
4960 struct btrfs_key found_key;
a52d9a80
CM		 4961 struct extent_map *em = NULL;
4962 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4963 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4964 struct btrfs_trans_handle *trans = NULL;
261507a0 4965 int compress_type;
a52d9a80 4966
a52d9a80 4967again:
890871be 4968 read_lock(&em_tree->lock);
d1310b2e 4969 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4970 if (em)
4971 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 4972 read_unlock(&em_tree->lock);
d1310b2e 4973
a52d9a80 4974 if (em) {
e1c4b745
CM		 4975 if (em->start > start || em->start + em->len <= start)
4976 free_extent_map(em);
4977 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4978 free_extent_map(em);
4979 else
4980 goto out;
a52d9a80 4981 }
172ddd60 4982 em = alloc_extent_map();
a52d9a80 4983 if (!em) {
d1310b2e
CM		 4984 err = -ENOMEM;
4985 goto out;
a52d9a80 4986 }
e6dcd2dc 4987 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 4988 em->start = EXTENT_MAP_HOLE;
445a6944 4989 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 4990 em->len = (u64)-1;
c8b97818 4991 em->block_len = (u64)-1;
f421950f
CM		 4992
4993 if (!path) {
4994 path = btrfs_alloc_path();
026fd317
JB		 4995 if (!path) {
4996 err = -ENOMEM;
4997 goto out;
4998 }
4999 /*
5000 * Chances are we'll be called again, so go ahead and do
5001 * readahead
5002 */
5003 path->reada = 1;
f421950f
CM		 5004 }
5005
179e29e4
CM		 5006 ret = btrfs_lookup_file_extent(trans, root, path,
5007 objectid, start, trans != NULL);
a52d9a80
CM		 5008 if (ret < 0) {
5009 err = ret;
5010 goto out;
5011 }
5012
5013 if (ret != 0) {
5014 if (path->slots[0] == 0)
5015 goto not_found;
5016 path->slots[0]--;
5017 }
5018
5f39d397
CM		 5019 leaf = path->nodes[0];
5020 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5021 struct btrfs_file_extent_item);
a52d9a80 5022 /* are we inside the extent that was found? */
5f39d397
CM		 5023 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5024 found_type = btrfs_key_type(&found_key);
5025 if (found_key.objectid != objectid ||
a52d9a80
CM		 5026 found_type != BTRFS_EXTENT_DATA_KEY) {
5027 goto not_found;
5028 }
5029
5f39d397
CM		 5030 found_type = btrfs_file_extent_type(leaf, item);
5031 extent_start = found_key.offset;
261507a0 5032 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5033 if (found_type == BTRFS_FILE_EXTENT_REG ||
5034 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5035 extent_end = extent_start +
db94535d 5036 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5037 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5038 size_t size;
5039 size = btrfs_file_extent_inline_len(leaf, item);
5040 extent_end = (extent_start + size + root->sectorsize - 1) &
5041 ~((u64)root->sectorsize - 1);
5042 }
5043
5044 if (start >= extent_end) {
5045 path->slots[0]++;
5046 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5047 ret = btrfs_next_leaf(root, path);
5048 if (ret < 0) {
5049 err = ret;
5050 goto out;
a52d9a80 5051 }
9036c102
YZ		 5052 if (ret > 0)
5053 goto not_found;
5054 leaf = path->nodes[0];
a52d9a80 5055 }
9036c102
YZ		 5056 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5057 if (found_key.objectid != objectid ||
5058 found_key.type != BTRFS_EXTENT_DATA_KEY)
5059 goto not_found;
5060 if (start + len <= found_key.offset)
5061 goto not_found;
5062 em->start = start;
5063 em->len = found_key.offset - start;
5064 goto not_found_em;
5065 }
5066
d899e052
YZ		 5067 if (found_type == BTRFS_FILE_EXTENT_REG ||
5068 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5069 em->start = extent_start;
5070 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5071 em->orig_start = extent_start -
5072 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5073 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5074 if (bytenr == 0) {
5f39d397 5075 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5076 goto insert;
5077 }
261507a0 5078 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5079 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5080 em->compress_type = compress_type;
c8b97818
CM		 5081 em->block_start = bytenr;
5082 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5083 item);
5084 } else {
5085 bytenr += btrfs_file_extent_offset(leaf, item);
5086 em->block_start = bytenr;
5087 em->block_len = em->len;
d899e052
YZ		 5088 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5089 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5090 }
a52d9a80
CM		 5091 goto insert;
5092 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5093 unsigned long ptr;
a52d9a80 5094 char *map;
3326d1b0
CM		 5095 size_t size;
5096 size_t extent_offset;
5097 size_t copy_size;
a52d9a80 5098
689f9346 5099 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5100 if (!page || create) {
689f9346 5101 em->start = extent_start;
9036c102 5102 em->len = extent_end - extent_start;
689f9346
Y		 5103 goto out;
5104 }
5f39d397 5105
9036c102
YZ		 5106 size = btrfs_file_extent_inline_len(leaf, item);
5107 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5108 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5109 size - extent_offset);
3326d1b0 5110 em->start = extent_start + extent_offset;
70dec807
CM		 5111 em->len = (copy_size + root->sectorsize - 1) &
5112 ~((u64)root->sectorsize - 1);
ff5b7ee3 5113 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5114 if (compress_type) {
c8b97818 5115 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5116 em->compress_type = compress_type;
5117 }
689f9346 5118 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5119 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5120 if (btrfs_file_extent_compression(leaf, item) !=
5121 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5122 ret = uncompress_inline(path, inode, page,
5123 pg_offset,
5124 extent_offset, item);
5125 BUG_ON(ret);
5126 } else {
5127 map = kmap(page);
5128 read_extent_buffer(leaf, map + pg_offset, ptr,
5129 copy_size);
93c82d57
CM		 5130 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5131 memset(map + pg_offset + copy_size, 0,
5132 PAGE_CACHE_SIZE - pg_offset -
5133 copy_size);
5134 }
c8b97818
CM		 5135 kunmap(page);
5136 }
179e29e4
CM		 5137 flush_dcache_page(page);
5138 } else if (create && PageUptodate(page)) {
0ca1f7ce 5139 WARN_ON(1);
179e29e4
CM		 5140 if (!trans) {
5141 kunmap(page);
5142 free_extent_map(em);
5143 em = NULL;
ff5714cc 5144
b3b4aa74 5145 btrfs_release_path(path);
7a7eaa40 5146 trans = btrfs_join_transaction(root);
ff5714cc 5147
3612b495
TI		 5148 if (IS_ERR(trans))
5149 return ERR_CAST(trans);
179e29e4
CM		 5150 goto again;
5151 }
c8b97818 5152 map = kmap(page);
70dec807 5153 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5154 copy_size);
c8b97818 5155 kunmap(page);
179e29e4 5156 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5157 }
d1310b2e 5158 set_extent_uptodate(io_tree, em->start,
507903b8 5159 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5160 goto insert;
5161 } else {
d397712b 5162 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5163 WARN_ON(1);
5164 }
5165not_found:
5166 em->start = start;
d1310b2e 5167 em->len = len;
a52d9a80 5168not_found_em:
5f39d397 5169 em->block_start = EXTENT_MAP_HOLE;
9036c102 5170 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5171insert:
b3b4aa74 5172 btrfs_release_path(path);
d1310b2e 5173 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5174 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5175 "[%llu %llu]\n", (unsigned long long)em->start,
5176 (unsigned long long)em->len,
5177 (unsigned long long)start,
5178 (unsigned long long)len);
a52d9a80
CM		 5179 err = -EIO;
5180 goto out;
5181 }
d1310b2e
CM		 5182
5183 err = 0;
890871be 5184 write_lock(&em_tree->lock);
a52d9a80 5185 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5186 /* it is possible that someone inserted the extent into the tree
5187 * while we had the lock dropped. It is also possible that
5188 * an overlapping map exists in the tree
5189 */
a52d9a80 5190 if (ret == -EEXIST) {
3b951516 5191 struct extent_map *existing;
e6dcd2dc
CM		 5192
5193 ret = 0;
5194
3b951516 5195 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5196 if (existing && (existing->start > start ||
5197 existing->start + existing->len <= start)) {
5198 free_extent_map(existing);
5199 existing = NULL;
5200 }
3b951516
CM		 5201 if (!existing) {
5202 existing = lookup_extent_mapping(em_tree, em->start,
5203 em->len);
5204 if (existing) {
5205 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5206 em, start,
5207 root->sectorsize);
3b951516
CM		 5208 free_extent_map(existing);
5209 if (err) {
5210 free_extent_map(em);
5211 em = NULL;
5212 }
5213 } else {
5214 err = -EIO;
3b951516
CM		 5215 free_extent_map(em);
5216 em = NULL;
5217 }
5218 } else {
5219 free_extent_map(em);
5220 em = existing;
e6dcd2dc 5221 err = 0;
a52d9a80 5222 }
a52d9a80 5223 }
890871be 5224 write_unlock(&em_tree->lock);
a52d9a80 5225out:
1abe9b8a 5226
5227 trace_btrfs_get_extent(root, em);
5228
f421950f
CM		 5229 if (path)
5230 btrfs_free_path(path);
a52d9a80
CM		 5231 if (trans) {
5232 ret = btrfs_end_transaction(trans, root);
d397712b 5233 if (!err)
a52d9a80
CM		 5234 err = ret;
5235 }
a52d9a80
CM		 5236 if (err) {
5237 free_extent_map(em);
a52d9a80
CM		 5238 return ERR_PTR(err);
5239 }
5240 return em;
5241}
5242
ec29ed5b
CM		 5243struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5244 size_t pg_offset, u64 start, u64 len,
5245 int create)
5246{
5247 struct extent_map *em;
5248 struct extent_map *hole_em = NULL;
5249 u64 range_start = start;
5250 u64 end;
5251 u64 found;
5252 u64 found_end;
5253 int err = 0;
5254
5255 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5256 if (IS_ERR(em))
5257 return em;
5258 if (em) {
5259 /*
5260 * if our em maps to a hole, there might
5261 * actually be delalloc bytes behind it
5262 */
5263 if (em->block_start != EXTENT_MAP_HOLE)
5264 return em;
5265 else
5266 hole_em = em;
5267 }
5268
5269 /* check to see if we've wrapped (len == -1 or similar) */
5270 end = start + len;
5271 if (end < start)
5272 end = (u64)-1;
5273 else
5274 end -= 1;
5275
5276 em = NULL;
5277
5278 /* ok, we didn't find anything, lets look for delalloc */
5279 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5280 end, len, EXTENT_DELALLOC, 1);
5281 found_end = range_start + found;
5282 if (found_end < range_start)
5283 found_end = (u64)-1;
5284
5285 /*
5286 * we didn't find anything useful, return
5287 * the original results from get_extent()
5288 */
5289 if (range_start > end || found_end <= start) {
5290 em = hole_em;
5291 hole_em = NULL;
5292 goto out;
5293 }
5294
5295 /* adjust the range_start to make sure it doesn't
5296 * go backwards from the start they passed in
5297 */
5298 range_start = max(start,range_start);
5299 found = found_end - range_start;
5300
5301 if (found > 0) {
5302 u64 hole_start = start;
5303 u64 hole_len = len;
5304
172ddd60 5305 em = alloc_extent_map();
ec29ed5b
CM		 5306 if (!em) {
5307 err = -ENOMEM;
5308 goto out;
5309 }
5310 /*
5311 * when btrfs_get_extent can't find anything it
5312 * returns one huge hole
5313 *
5314 * make sure what it found really fits our range, and
5315 * adjust to make sure it is based on the start from
5316 * the caller
5317 */
5318 if (hole_em) {
5319 u64 calc_end = extent_map_end(hole_em);
5320
5321 if (calc_end <= start || (hole_em->start > end)) {
5322 free_extent_map(hole_em);
5323 hole_em = NULL;
5324 } else {
5325 hole_start = max(hole_em->start, start);
5326 hole_len = calc_end - hole_start;
5327 }
5328 }
5329 em->bdev = NULL;
5330 if (hole_em && range_start > hole_start) {
5331 /* our hole starts before our delalloc, so we
5332 * have to return just the parts of the hole
5333 * that go until the delalloc starts
5334 */
5335 em->len = min(hole_len,
5336 range_start - hole_start);
5337 em->start = hole_start;
5338 em->orig_start = hole_start;
5339 /*
5340 * don't adjust block start at all,
5341 * it is fixed at EXTENT_MAP_HOLE
5342 */
5343 em->block_start = hole_em->block_start;
5344 em->block_len = hole_len;
5345 } else {
5346 em->start = range_start;
5347 em->len = found;
5348 em->orig_start = range_start;
5349 em->block_start = EXTENT_MAP_DELALLOC;
5350 em->block_len = found;
5351 }
5352 } else if (hole_em) {
5353 return hole_em;
5354 }
5355out:
5356
5357 free_extent_map(hole_em);
5358 if (err) {
5359 free_extent_map(em);
5360 return ERR_PTR(err);
5361 }
5362 return em;
5363}
5364
4b46fce2 5365static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5366 struct extent_map *em,
4b46fce2
JB		 5367 u64 start, u64 len)
5368{
5369 struct btrfs_root *root = BTRFS_I(inode)->root;
5370 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5371 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5372 struct btrfs_key ins;
5373 u64 alloc_hint;
5374 int ret;
16d299ac 5375 bool insert = false;
4b46fce2 5376
16d299ac
JB		 5377 /*
5378 * Ok if the extent map we looked up is a hole and is for the exact
5379 * range we want, there is no reason to allocate a new one, however if
5380 * it is not right then we need to free this one and drop the cache for
5381 * our range.
5382 */
5383 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5384 em->len != len) {
5385 free_extent_map(em);
5386 em = NULL;
5387 insert = true;
5388 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5389 }
4b46fce2 5390
7a7eaa40 5391 trans = btrfs_join_transaction(root);
3612b495
TI		 5392 if (IS_ERR(trans))
5393 return ERR_CAST(trans);
4b46fce2 5394
4cb5300b
CM		 5395 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5396 btrfs_add_inode_defrag(trans, inode);
5397
4b46fce2
JB		 5398 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5399
5400 alloc_hint = get_extent_allocation_hint(inode, start, len);
5401 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5402 alloc_hint, (u64)-1, &ins, 1);
5403 if (ret) {
5404 em = ERR_PTR(ret);
5405 goto out;
5406 }
5407
4b46fce2 5408 if (!em) {
172ddd60 5409 em = alloc_extent_map();
16d299ac
JB		 5410 if (!em) {
5411 em = ERR_PTR(-ENOMEM);
5412 goto out;
5413 }
4b46fce2
JB		 5414 }
5415
5416 em->start = start;
5417 em->orig_start = em->start;
5418 em->len = ins.offset;
5419
5420 em->block_start = ins.objectid;
5421 em->block_len = ins.offset;
5422 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5423
5424 /*
5425 * We need to do this because if we're using the original em we searched
5426 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5427 */
5428 em->flags = 0;
4b46fce2
JB		 5429 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5430
16d299ac 5431 while (insert) {
4b46fce2
JB		 5432 write_lock(&em_tree->lock);
5433 ret = add_extent_mapping(em_tree, em);
5434 write_unlock(&em_tree->lock);
5435 if (ret != -EEXIST)
5436 break;
5437 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5438 }
5439
5440 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5441 ins.offset, ins.offset, 0);
5442 if (ret) {
5443 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5444 em = ERR_PTR(ret);
5445 }
5446out:
5447 btrfs_end_transaction(trans, root);
5448 return em;
5449}
5450
46bfbb5c
CM		 5451/*
5452 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5453 * block must be cow'd
5454 */
5455static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5456 struct inode *inode, u64 offset, u64 len)
5457{
5458 struct btrfs_path *path;
5459 int ret;
5460 struct extent_buffer *leaf;
5461 struct btrfs_root *root = BTRFS_I(inode)->root;
5462 struct btrfs_file_extent_item *fi;
5463 struct btrfs_key key;
5464 u64 disk_bytenr;
5465 u64 backref_offset;
5466 u64 extent_end;
5467 u64 num_bytes;
5468 int slot;
5469 int found_type;
5470
5471 path = btrfs_alloc_path();
5472 if (!path)
5473 return -ENOMEM;
5474
33345d01 5475 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5476 offset, 0);
5477 if (ret < 0)
5478 goto out;
5479
5480 slot = path->slots[0];
5481 if (ret == 1) {
5482 if (slot == 0) {
5483 /* can't find the item, must cow */
5484 ret = 0;
5485 goto out;
5486 }
5487 slot--;
5488 }
5489 ret = 0;
5490 leaf = path->nodes[0];
5491 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5492 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5493 key.type != BTRFS_EXTENT_DATA_KEY) {
5494 /* not our file or wrong item type, must cow */
5495 goto out;
5496 }
5497
5498 if (key.offset > offset) {
5499 /* Wrong offset, must cow */
5500 goto out;
5501 }
5502
5503 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5504 found_type = btrfs_file_extent_type(leaf, fi);
5505 if (found_type != BTRFS_FILE_EXTENT_REG &&
5506 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5507 /* not a regular extent, must cow */
5508 goto out;
5509 }
5510 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5511 backref_offset = btrfs_file_extent_offset(leaf, fi);
5512
5513 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5514 if (extent_end < offset + len) {
5515 /* extent doesn't include our full range, must cow */
5516 goto out;
5517 }
5518
5519 if (btrfs_extent_readonly(root, disk_bytenr))
5520 goto out;
5521
5522 /*
5523 * look for other files referencing this extent, if we
5524 * find any we must cow
5525 */
33345d01 5526 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5527 key.offset - backref_offset, disk_bytenr))
5528 goto out;
5529
5530 /*
5531 * adjust disk_bytenr and num_bytes to cover just the bytes
5532 * in this extent we are about to write. If there
5533 * are any csums in that range we have to cow in order
5534 * to keep the csums correct
5535 */
5536 disk_bytenr += backref_offset;
5537 disk_bytenr += offset - key.offset;
5538 num_bytes = min(offset + len, extent_end) - offset;
5539 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5540 goto out;
5541 /*
5542 * all of the above have passed, it is safe to overwrite this extent
5543 * without cow
5544 */
5545 ret = 1;
5546out:
5547 btrfs_free_path(path);
5548 return ret;
5549}
5550
4b46fce2
JB		 5551static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5552 struct buffer_head *bh_result, int create)
5553{
5554 struct extent_map *em;
5555 struct btrfs_root *root = BTRFS_I(inode)->root;
5556 u64 start = iblock << inode->i_blkbits;
5557 u64 len = bh_result->b_size;
46bfbb5c 5558 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5559
5560 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5561 if (IS_ERR(em))
5562 return PTR_ERR(em);
5563
5564 /*
5565 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5566 * io. INLINE is special, and we could probably kludge it in here, but
5567 * it's still buffered so for safety lets just fall back to the generic
5568 * buffered path.
5569 *
5570 * For COMPRESSED we _have_ to read the entire extent in so we can
5571 * decompress it, so there will be buffering required no matter what we
5572 * do, so go ahead and fallback to buffered.
5573 *
5574 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5575 * to buffered IO. Don't blame me, this is the price we pay for using
5576 * the generic code.
5577 */
5578 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5579 em->block_start == EXTENT_MAP_INLINE) {
5580 free_extent_map(em);
5581 return -ENOTBLK;
5582 }
5583
5584 /* Just a good old fashioned hole, return */
5585 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5586 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5587 free_extent_map(em);
5588 /* DIO will do one hole at a time, so just unlock a sector */
5589 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5590 start + root->sectorsize - 1, GFP_NOFS);
5591 return 0;
5592 }
5593
5594 /*
5595 * We don't allocate a new extent in the following cases
5596 *
5597 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5598 * existing extent.
5599 * 2) The extent is marked as PREALLOC. We're good to go here and can
5600 * just use the extent.
5601 *
5602 */
46bfbb5c
CM		 5603 if (!create) {
5604 len = em->len - (start - em->start);
4b46fce2 5605 goto map;
46bfbb5c 5606 }
4b46fce2
JB		 5607
5608 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5609 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5610 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5611 int type;
5612 int ret;
46bfbb5c 5613 u64 block_start;
4b46fce2
JB		 5614
5615 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5616 type = BTRFS_ORDERED_PREALLOC;
5617 else
5618 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5619 len = min(len, em->len - (start - em->start));
4b46fce2 5620 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5621
5622 /*
5623 * we're not going to log anything, but we do need
5624 * to make sure the current transaction stays open
5625 * while we look for nocow cross refs
5626 */
7a7eaa40 5627 trans = btrfs_join_transaction(root);
3612b495 5628 if (IS_ERR(trans))
46bfbb5c
CM		 5629 goto must_cow;
5630
5631 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5632 ret = btrfs_add_ordered_extent_dio(inode, start,
5633 block_start, len, len, type);
5634 btrfs_end_transaction(trans, root);
5635 if (ret) {
5636 free_extent_map(em);
5637 return ret;
5638 }
5639 goto unlock;
4b46fce2 5640 }
46bfbb5c 5641 btrfs_end_transaction(trans, root);
4b46fce2 5642 }
46bfbb5c
CM		 5643must_cow:
5644 /*
5645 * this will cow the extent, reset the len in case we changed
5646 * it above
5647 */
5648 len = bh_result->b_size;
16d299ac 5649 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5650 if (IS_ERR(em))
5651 return PTR_ERR(em);
5652 len = min(len, em->len - (start - em->start));
5653unlock:
4845e44f
CM		 5654 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5655 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5656 0, NULL, GFP_NOFS);
4b46fce2
JB		 5657map:
5658 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5659 inode->i_blkbits;
46bfbb5c 5660 bh_result->b_size = len;
4b46fce2
JB		 5661 bh_result->b_bdev = em->bdev;
5662 set_buffer_mapped(bh_result);
5663 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5664 set_buffer_new(bh_result);
5665
5666 free_extent_map(em);
5667
5668 return 0;
5669}
5670
5671struct btrfs_dio_private {
5672 struct inode *inode;
5673 u64 logical_offset;
5674 u64 disk_bytenr;
5675 u64 bytes;
5676 u32 *csums;
5677 void *private;
e65e1535
MX		 5678
5679 /* number of bios pending for this dio */
5680 atomic_t pending_bios;
5681
5682 /* IO errors */
5683 int errors;
5684
5685 struct bio *orig_bio;
4b46fce2
JB		 5686};
5687
5688static void btrfs_endio_direct_read(struct bio *bio, int err)
5689{
e65e1535 5690 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5691 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5692 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5693 struct inode *inode = dip->inode;
5694 struct btrfs_root *root = BTRFS_I(inode)->root;
5695 u64 start;
5696 u32 *private = dip->csums;
5697
5698 start = dip->logical_offset;
5699 do {
5700 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5701 struct page *page = bvec->bv_page;
5702 char *kaddr;
5703 u32 csum = ~(u32)0;
5704 unsigned long flags;
5705
5706 local_irq_save(flags);
5707 kaddr = kmap_atomic(page, KM_IRQ0);
5708 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5709 csum, bvec->bv_len);
5710 btrfs_csum_final(csum, (char *)&csum);
5711 kunmap_atomic(kaddr, KM_IRQ0);
5712 local_irq_restore(flags);
5713
5714 flush_dcache_page(bvec->bv_page);
5715 if (csum != *private) {
33345d01 5716 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5717 " %llu csum %u private %u\n",
33345d01
LZ		 5718 (unsigned long long)btrfs_ino(inode),
5719 (unsigned long long)start,
4b46fce2
JB		 5720 csum, *private);
5721 err = -EIO;
5722 }
5723 }
5724
5725 start += bvec->bv_len;
5726 private++;
5727 bvec++;
5728 } while (bvec <= bvec_end);
5729
5730 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5731 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5732 bio->bi_private = dip->private;
5733
5734 kfree(dip->csums);
5735 kfree(dip);
c0da7aa1
JB		 5736
5737 /* If we had a csum failure make sure to clear the uptodate flag */
5738 if (err)
5739 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5740 dio_end_io(bio, err);
5741}
5742
5743static void btrfs_endio_direct_write(struct bio *bio, int err)
5744{
5745 struct btrfs_dio_private *dip = bio->bi_private;
5746 struct inode *inode = dip->inode;
5747 struct btrfs_root *root = BTRFS_I(inode)->root;
5748 struct btrfs_trans_handle *trans;
5749 struct btrfs_ordered_extent *ordered = NULL;
5750 struct extent_state *cached_state = NULL;
163cf09c
CM		 5751 u64 ordered_offset = dip->logical_offset;
5752 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5753 int ret;
5754
5755 if (err)
5756 goto out_done;
163cf09c
CM		 5757again:
5758 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5759 &ordered_offset,
5760 ordered_bytes);
4b46fce2 5761 if (!ret)
163cf09c 5762 goto out_test;
4b46fce2
JB		 5763
5764 BUG_ON(!ordered);
5765
7a7eaa40 5766 trans = btrfs_join_transaction(root);
3612b495 5767 if (IS_ERR(trans)) {
4b46fce2
JB		 5768 err = -ENOMEM;
5769 goto out;
5770 }
5771 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5772
5773 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5774 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5775 if (!ret)
5776 ret = btrfs_update_inode(trans, root, inode);
5777 err = ret;
5778 goto out;
5779 }
5780
5781 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5782 ordered->file_offset + ordered->len - 1, 0,
5783 &cached_state, GFP_NOFS);
5784
5785 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5786 ret = btrfs_mark_extent_written(trans, inode,
5787 ordered->file_offset,
5788 ordered->file_offset +
5789 ordered->len);
5790 if (ret) {
5791 err = ret;
5792 goto out_unlock;
5793 }
5794 } else {
5795 ret = insert_reserved_file_extent(trans, inode,
5796 ordered->file_offset,
5797 ordered->start,
5798 ordered->disk_len,
5799 ordered->len,
5800 ordered->len,
5801 0, 0, 0,
5802 BTRFS_FILE_EXTENT_REG);
5803 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5804 ordered->file_offset, ordered->len);
5805 if (ret) {
5806 err = ret;
5807 WARN_ON(1);
5808 goto out_unlock;
5809 }
5810 }
5811
5812 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
1ef30be1 5813 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
a39f7521 5814 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
1ef30be1
JB		 5815 btrfs_update_inode(trans, root, inode);
5816 ret = 0;
4b46fce2
JB		 5817out_unlock:
5818 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5819 ordered->file_offset + ordered->len - 1,
5820 &cached_state, GFP_NOFS);
5821out:
5822 btrfs_delalloc_release_metadata(inode, ordered->len);
5823 btrfs_end_transaction(trans, root);
163cf09c 5824 ordered_offset = ordered->file_offset + ordered->len;
4b46fce2
JB		 5825 btrfs_put_ordered_extent(ordered);
5826 btrfs_put_ordered_extent(ordered);
163cf09c
CM		 5827
5828out_test:
5829 /*
5830 * our bio might span multiple ordered extents. If we haven't
5831 * completed the accounting for the whole dio, go back and try again
5832 */
5833 if (ordered_offset < dip->logical_offset + dip->bytes) {
5834 ordered_bytes = dip->logical_offset + dip->bytes -
5835 ordered_offset;
5836 goto again;
5837 }
4b46fce2
JB		 5838out_done:
5839 bio->bi_private = dip->private;
5840
5841 kfree(dip->csums);
5842 kfree(dip);
c0da7aa1
JB		 5843
5844 /* If we had an error make sure to clear the uptodate flag */
5845 if (err)
5846 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5847 dio_end_io(bio, err);
5848}
5849
eaf25d93
CM		 5850static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5851 struct bio *bio, int mirror_num,
5852 unsigned long bio_flags, u64 offset)
5853{
5854 int ret;
5855 struct btrfs_root *root = BTRFS_I(inode)->root;
5856 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5857 BUG_ON(ret);
5858 return 0;
5859}
5860
e65e1535
MX		 5861static void btrfs_end_dio_bio(struct bio *bio, int err)
5862{
5863 struct btrfs_dio_private *dip = bio->bi_private;
5864
5865 if (err) {
33345d01 5866 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 5867 "sector %#Lx len %u err no %d\n",
33345d01 5868 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 5869 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 5870 dip->errors = 1;
5871
5872 /*
5873 * before atomic variable goto zero, we must make sure
5874 * dip->errors is perceived to be set.
5875 */
5876 smp_mb__before_atomic_dec();
5877 }
5878
5879 /* if there are more bios still pending for this dio, just exit */
5880 if (!atomic_dec_and_test(&dip->pending_bios))
5881 goto out;
5882
5883 if (dip->errors)
5884 bio_io_error(dip->orig_bio);
5885 else {
5886 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5887 bio_endio(dip->orig_bio, 0);
5888 }
5889out:
5890 bio_put(bio);
5891}
5892
5893static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5894 u64 first_sector, gfp_t gfp_flags)
5895{
5896 int nr_vecs = bio_get_nr_vecs(bdev);
5897 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5898}
5899
5900static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5901 int rw, u64 file_offset, int skip_sum,
1ae39938 5902 u32 *csums, int async_submit)
e65e1535
MX		 5903{
5904 int write = rw & REQ_WRITE;
5905 struct btrfs_root *root = BTRFS_I(inode)->root;
5906 int ret;
5907
5908 bio_get(bio);
5909 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5910 if (ret)
5911 goto err;
5912
1ae39938
JB		 5913 if (skip_sum)
5914 goto map;
5915
5916 if (write && async_submit) {
e65e1535
MX		 5917 ret = btrfs_wq_submit_bio(root->fs_info,
5918 inode, rw, bio, 0, 0,
5919 file_offset,
5920 __btrfs_submit_bio_start_direct_io,
5921 __btrfs_submit_bio_done);
5922 goto err;
1ae39938
JB		 5923 } else if (write) {
5924 /*
5925 * If we aren't doing async submit, calculate the csum of the
5926 * bio now.
5927 */
5928 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5929 if (ret)
5930 goto err;
c2db1073
TI		 5931 } else if (!skip_sum) {
5932 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 5933 file_offset, csums);
c2db1073
TI		 5934 if (ret)
5935 goto err;
5936 }
e65e1535 5937
1ae39938
JB		 5938map:
5939 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 5940err:
5941 bio_put(bio);
5942 return ret;
5943}
5944
5945static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5946 int skip_sum)
5947{
5948 struct inode *inode = dip->inode;
5949 struct btrfs_root *root = BTRFS_I(inode)->root;
5950 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5951 struct bio *bio;
5952 struct bio *orig_bio = dip->orig_bio;
5953 struct bio_vec *bvec = orig_bio->bi_io_vec;
5954 u64 start_sector = orig_bio->bi_sector;
5955 u64 file_offset = dip->logical_offset;
5956 u64 submit_len = 0;
5957 u64 map_length;
5958 int nr_pages = 0;
5959 u32 *csums = dip->csums;
5960 int ret = 0;
1ae39938 5961 int async_submit = 0;
98bc3149 5962 int write = rw & REQ_WRITE;
e65e1535 5963
e65e1535
MX		 5964 map_length = orig_bio->bi_size;
5965 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5966 &map_length, NULL, 0);
5967 if (ret) {
64728bbb 5968 bio_put(orig_bio);
e65e1535
MX		 5969 return -EIO;
5970 }
5971
02f57c7a
JB		 5972 if (map_length >= orig_bio->bi_size) {
5973 bio = orig_bio;
5974 goto submit;
5975 }
5976
1ae39938 5977 async_submit = 1;
02f57c7a
JB		 5978 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5979 if (!bio)
5980 return -ENOMEM;
5981 bio->bi_private = dip;
5982 bio->bi_end_io = btrfs_end_dio_bio;
5983 atomic_inc(&dip->pending_bios);
5984
e65e1535
MX		 5985 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5986 if (unlikely(map_length < submit_len + bvec->bv_len ||
5987 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5988 bvec->bv_offset) < bvec->bv_len)) {
5989 /*
5990 * inc the count before we submit the bio so
5991 * we know the end IO handler won't happen before
5992 * we inc the count. Otherwise, the dip might get freed
5993 * before we're done setting it up
5994 */
5995 atomic_inc(&dip->pending_bios);
5996 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5997 file_offset, skip_sum,
1ae39938 5998 csums, async_submit);
e65e1535
MX		 5999 if (ret) {
6000 bio_put(bio);
6001 atomic_dec(&dip->pending_bios);
6002 goto out_err;
6003 }
6004
98bc3149
JB		 6005 /* Write's use the ordered csums */
6006 if (!write && !skip_sum)
e65e1535
MX		 6007 csums = csums + nr_pages;
6008 start_sector += submit_len >> 9;
6009 file_offset += submit_len;
6010
6011 submit_len = 0;
6012 nr_pages = 0;
6013
6014 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6015 start_sector, GFP_NOFS);
6016 if (!bio)
6017 goto out_err;
6018 bio->bi_private = dip;
6019 bio->bi_end_io = btrfs_end_dio_bio;
6020
6021 map_length = orig_bio->bi_size;
6022 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6023 &map_length, NULL, 0);
6024 if (ret) {
6025 bio_put(bio);
6026 goto out_err;
6027 }
6028 } else {
6029 submit_len += bvec->bv_len;
6030 nr_pages ++;
6031 bvec++;
6032 }
6033 }
6034
02f57c7a 6035submit:
e65e1535 6036 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 6037 csums, async_submit);
e65e1535
MX		 6038 if (!ret)
6039 return 0;
6040
6041 bio_put(bio);
6042out_err:
6043 dip->errors = 1;
6044 /*
6045 * before atomic variable goto zero, we must
6046 * make sure dip->errors is perceived to be set.
6047 */
6048 smp_mb__before_atomic_dec();
6049 if (atomic_dec_and_test(&dip->pending_bios))
6050 bio_io_error(dip->orig_bio);
6051
6052 /* bio_end_io() will handle error, so we needn't return it */
6053 return 0;
6054}
6055
4b46fce2
JB		 6056static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6057 loff_t file_offset)
6058{
6059 struct btrfs_root *root = BTRFS_I(inode)->root;
6060 struct btrfs_dio_private *dip;
6061 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6062 int skip_sum;
7b6d91da 6063 int write = rw & REQ_WRITE;
4b46fce2
JB		 6064 int ret = 0;
6065
6066 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6067
6068 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6069 if (!dip) {
6070 ret = -ENOMEM;
6071 goto free_ordered;
6072 }
6073 dip->csums = NULL;
6074
98bc3149
JB		 6075 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6076 if (!write && !skip_sum) {
4b46fce2
JB		 6077 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6078 if (!dip->csums) {
b4966b77 6079 kfree(dip);
4b46fce2
JB		 6080 ret = -ENOMEM;
6081 goto free_ordered;
6082 }
6083 }
6084
6085 dip->private = bio->bi_private;
6086 dip->inode = inode;
6087 dip->logical_offset = file_offset;
6088
4b46fce2
JB		 6089 dip->bytes = 0;
6090 do {
6091 dip->bytes += bvec->bv_len;
6092 bvec++;
6093 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6094
46bfbb5c 6095 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6096 bio->bi_private = dip;
e65e1535
MX		 6097 dip->errors = 0;
6098 dip->orig_bio = bio;
6099 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6100
6101 if (write)
6102 bio->bi_end_io = btrfs_endio_direct_write;
6103 else
6104 bio->bi_end_io = btrfs_endio_direct_read;
6105
e65e1535
MX		 6106 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6107 if (!ret)
eaf25d93 6108 return;
4b46fce2
JB		 6109free_ordered:
6110 /*
6111 * If this is a write, we need to clean up the reserved space and kill
6112 * the ordered extent.
6113 */
6114 if (write) {
6115 struct btrfs_ordered_extent *ordered;
955256f2 6116 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6117 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6118 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6119 btrfs_free_reserved_extent(root, ordered->start,
6120 ordered->disk_len);
6121 btrfs_put_ordered_extent(ordered);
6122 btrfs_put_ordered_extent(ordered);
6123 }
6124 bio_endio(bio, ret);
6125}
6126
5a5f79b5
CM		 6127static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6128 const struct iovec *iov, loff_t offset,
6129 unsigned long nr_segs)
6130{
6131 int seg;
a1b75f7d 6132 int i;
5a5f79b5
CM		 6133 size_t size;
6134 unsigned long addr;
6135 unsigned blocksize_mask = root->sectorsize - 1;
6136 ssize_t retval = -EINVAL;
6137 loff_t end = offset;
6138
6139 if (offset & blocksize_mask)
6140 goto out;
6141
6142 /* Check the memory alignment. Blocks cannot straddle pages */
6143 for (seg = 0; seg < nr_segs; seg++) {
6144 addr = (unsigned long)iov[seg].iov_base;
6145 size = iov[seg].iov_len;
6146 end += size;
a1b75f7d 6147 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6148 goto out;
a1b75f7d
JB		 6149
6150 /* If this is a write we don't need to check anymore */
6151 if (rw & WRITE)
6152 continue;
6153
6154 /*
6155 * Check to make sure we don't have duplicate iov_base's in this
6156 * iovec, if so return EINVAL, otherwise we'll get csum errors
6157 * when reading back.
6158 */
6159 for (i = seg + 1; i < nr_segs; i++) {
6160 if (iov[seg].iov_base == iov[i].iov_base)
6161 goto out;
6162 }
5a5f79b5
CM		 6163 }
6164 retval = 0;
6165out:
6166 return retval;
6167}
16432985
CM		 6168static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6169 const struct iovec *iov, loff_t offset,
6170 unsigned long nr_segs)
6171{
4b46fce2
JB		 6172 struct file *file = iocb->ki_filp;
6173 struct inode *inode = file->f_mapping->host;
6174 struct btrfs_ordered_extent *ordered;
4845e44f 6175 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6176 u64 lockstart, lockend;
6177 ssize_t ret;
4845e44f
CM		 6178 int writing = rw & WRITE;
6179 int write_bits = 0;
3f7c579c 6180 size_t count = iov_length(iov, nr_segs);
4b46fce2 6181
5a5f79b5
CM		 6182 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6183 offset, nr_segs)) {
6184 return 0;
6185 }
6186
4b46fce2 6187 lockstart = offset;
3f7c579c
CM		 6188 lockend = offset + count - 1;
6189
6190 if (writing) {
6191 ret = btrfs_delalloc_reserve_space(inode, count);
6192 if (ret)
6193 goto out;
6194 }
4845e44f 6195
4b46fce2 6196 while (1) {
4845e44f
CM		 6197 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6198 0, &cached_state, GFP_NOFS);
4b46fce2
JB		 6199 /*
6200 * We're concerned with the entire range that we're going to be
6201 * doing DIO to, so we need to make sure theres no ordered
6202 * extents in this range.
6203 */
6204 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6205 lockend - lockstart + 1);
6206 if (!ordered)
6207 break;
4845e44f
CM		 6208 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6209 &cached_state, GFP_NOFS);
4b46fce2
JB		 6210 btrfs_start_ordered_extent(inode, ordered, 1);
6211 btrfs_put_ordered_extent(ordered);
6212 cond_resched();
6213 }
6214
4845e44f
CM		 6215 /*
6216 * we don't use btrfs_set_extent_delalloc because we don't want
6217 * the dirty or uptodate bits
6218 */
6219 if (writing) {
6220 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6221 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6222 EXTENT_DELALLOC, 0, NULL, &cached_state,
6223 GFP_NOFS);
6224 if (ret) {
6225 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6226 lockend, EXTENT_LOCKED | write_bits,
6227 1, 0, &cached_state, GFP_NOFS);
6228 goto out;
6229 }
6230 }
6231
6232 free_extent_state(cached_state);
6233 cached_state = NULL;
6234
5a5f79b5
CM		 6235 ret = __blockdev_direct_IO(rw, iocb, inode,
6236 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6237 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6238 btrfs_submit_direct, 0);
4b46fce2
JB		 6239
6240 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6241 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6242 offset + iov_length(iov, nr_segs) - 1,
6243 EXTENT_LOCKED | write_bits, 1, 0,
6244 &cached_state, GFP_NOFS);
4b46fce2
JB		 6245 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6246 /*
6247 * We're falling back to buffered, unlock the section we didn't
6248 * do IO on.
6249 */
4845e44f
CM		 6250 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6251 offset + iov_length(iov, nr_segs) - 1,
6252 EXTENT_LOCKED | write_bits, 1, 0,
6253 &cached_state, GFP_NOFS);
4b46fce2 6254 }
4845e44f
CM		 6255out:
6256 free_extent_state(cached_state);
4b46fce2 6257 return ret;
16432985
CM		 6258}
6259
1506fcc8
YS		 6260static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6261 __u64 start, __u64 len)
6262{
ec29ed5b 6263 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6264}
6265
a52d9a80 6266int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6267{
d1310b2e
CM		 6268 struct extent_io_tree *tree;
6269 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6270 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 6271}
1832a6d5 6272
a52d9a80 6273static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6274{
d1310b2e 6275 struct extent_io_tree *tree;
b888db2b
CM		 6276
6277
6278 if (current->flags & PF_MEMALLOC) {
6279 redirty_page_for_writepage(wbc, page);
6280 unlock_page(page);
6281 return 0;
6282 }
d1310b2e 6283 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6284 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6285}
6286
f421950f
CM		 6287int btrfs_writepages(struct address_space *mapping,
6288 struct writeback_control *wbc)
b293f02e 6289{
d1310b2e 6290 struct extent_io_tree *tree;
771ed689 6291
d1310b2e 6292 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6293 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6294}
6295
3ab2fb5a
CM		 6296static int
6297btrfs_readpages(struct file *file, struct address_space *mapping,
6298 struct list_head *pages, unsigned nr_pages)
6299{
d1310b2e
CM		 6300 struct extent_io_tree *tree;
6301 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6302 return extent_readpages(tree, mapping, pages, nr_pages,
6303 btrfs_get_extent);
6304}
e6dcd2dc 6305static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6306{
d1310b2e
CM		 6307 struct extent_io_tree *tree;
6308 struct extent_map_tree *map;
a52d9a80 6309 int ret;
8c2383c3 6310
d1310b2e
CM		 6311 tree = &BTRFS_I(page->mapping->host)->io_tree;
6312 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6313 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6314 if (ret == 1) {
6315 ClearPagePrivate(page);
6316 set_page_private(page, 0);
6317 page_cache_release(page);
39279cc3 6318 }
a52d9a80 6319 return ret;
39279cc3
CM		 6320}
6321
e6dcd2dc
CM		 6322static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6323{
98509cfc
CM		 6324 if (PageWriteback(page) || PageDirty(page))
6325 return 0;
b335b003 6326 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6327}
6328
a52d9a80 6329static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6330{
d1310b2e 6331 struct extent_io_tree *tree;
e6dcd2dc 6332 struct btrfs_ordered_extent *ordered;
2ac55d41 6333 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6334 u64 page_start = page_offset(page);
6335 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6336
8b62b72b
CM		 6337
6338 /*
6339 * we have the page locked, so new writeback can't start,
6340 * and the dirty bit won't be cleared while we are here.
6341 *
6342 * Wait for IO on this page so that we can safely clear
6343 * the PagePrivate2 bit and do ordered accounting
6344 */
e6dcd2dc 6345 wait_on_page_writeback(page);
8b62b72b 6346
d1310b2e 6347 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 6348 if (offset) {
6349 btrfs_releasepage(page, GFP_NOFS);
6350 return;
6351 }
2ac55d41
JB		 6352 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6353 GFP_NOFS);
e6dcd2dc
CM		 6354 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6355 page_offset(page));
6356 if (ordered) {
eb84ae03
CM		 6357 /*
6358 * IO on this page will never be started, so we need
6359 * to account for any ordered extents now
6360 */
e6dcd2dc
CM		 6361 clear_extent_bit(tree, page_start, page_end,
6362 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6363 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6364 &cached_state, GFP_NOFS);
8b62b72b
CM		 6365 /*
6366 * whoever cleared the private bit is responsible
6367 * for the finish_ordered_io
6368 */
6369 if (TestClearPagePrivate2(page)) {
6370 btrfs_finish_ordered_io(page->mapping->host,
6371 page_start, page_end);
6372 }
e6dcd2dc 6373 btrfs_put_ordered_extent(ordered);
2ac55d41
JB		 6374 cached_state = NULL;
6375 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6376 GFP_NOFS);
e6dcd2dc
CM		 6377 }
6378 clear_extent_bit(tree, page_start, page_end,
32c00aff 6379 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6380 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6381 __btrfs_releasepage(page, GFP_NOFS);
6382
4a096752 6383 ClearPageChecked(page);
9ad6b7bc 6384 if (PagePrivate(page)) {
9ad6b7bc
CM		 6385 ClearPagePrivate(page);
6386 set_page_private(page, 0);
6387 page_cache_release(page);
6388 }
39279cc3
CM		 6389}
6390
9ebefb18
CM		 6391/*
6392 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6393 * called from a page fault handler when a page is first dirtied. Hence we must
6394 * be careful to check for EOF conditions here. We set the page up correctly
6395 * for a written page which means we get ENOSPC checking when writing into
6396 * holes and correct delalloc and unwritten extent mapping on filesystems that
6397 * support these features.
6398 *
6399 * We are not allowed to take the i_mutex here so we have to play games to
6400 * protect against truncate races as the page could now be beyond EOF. Because
6401 * vmtruncate() writes the inode size before removing pages, once we have the
6402 * page lock we can determine safely if the page is beyond EOF. If it is not
6403 * beyond EOF, then the page is guaranteed safe against truncation until we
6404 * unlock the page.
6405 */
c2ec175c 6406int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6407{
c2ec175c 6408 struct page *page = vmf->page;
6da6abae 6409 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6410 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6411 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6412 struct btrfs_ordered_extent *ordered;
2ac55d41 6413 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6414 char *kaddr;
6415 unsigned long zero_start;
9ebefb18 6416 loff_t size;
1832a6d5 6417 int ret;
a52d9a80 6418 u64 page_start;
e6dcd2dc 6419 u64 page_end;
9ebefb18 6420
0ca1f7ce 6421 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
56a76f82
NP		 6422 if (ret) {
6423 if (ret == -ENOMEM)
6424 ret = VM_FAULT_OOM;
6425 else /* -ENOSPC, -EIO, etc */
6426 ret = VM_FAULT_SIGBUS;
1832a6d5 6427 goto out;
56a76f82 6428 }
1832a6d5 6429
56a76f82 6430 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6431again:
9ebefb18 6432 lock_page(page);
9ebefb18 6433 size = i_size_read(inode);
e6dcd2dc
CM		 6434 page_start = page_offset(page);
6435 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6436
9ebefb18 6437 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6438 (page_start >= size)) {
9ebefb18
CM		 6439 /* page got truncated out from underneath us */
6440 goto out_unlock;
6441 }
e6dcd2dc
CM		 6442 wait_on_page_writeback(page);
6443
2ac55d41
JB		 6444 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6445 GFP_NOFS);
e6dcd2dc
CM		 6446 set_page_extent_mapped(page);
6447
eb84ae03
CM		 6448 /*
6449 * we can't set the delalloc bits if there are pending ordered
6450 * extents. Drop our locks and wait for them to finish
6451 */
e6dcd2dc
CM		 6452 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6453 if (ordered) {
2ac55d41
JB		 6454 unlock_extent_cached(io_tree, page_start, page_end,
6455 &cached_state, GFP_NOFS);
e6dcd2dc 6456 unlock_page(page);
eb84ae03 6457 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6458 btrfs_put_ordered_extent(ordered);
6459 goto again;
6460 }
6461
fbf19087
JB		 6462 /*
6463 * XXX - page_mkwrite gets called every time the page is dirtied, even
6464 * if it was already dirty, so for space accounting reasons we need to
6465 * clear any delalloc bits for the range we are fixing to save. There
6466 * is probably a better way to do this, but for now keep consistent with
6467 * prepare_pages in the normal write path.
6468 */
2ac55d41 6469 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6470 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6471 0, 0, &cached_state, GFP_NOFS);
fbf19087 6472
2ac55d41
JB		 6473 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6474 &cached_state);
9ed74f2d 6475 if (ret) {
2ac55d41
JB		 6476 unlock_extent_cached(io_tree, page_start, page_end,
6477 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6478 ret = VM_FAULT_SIGBUS;
6479 goto out_unlock;
6480 }
e6dcd2dc 6481 ret = 0;
9ebefb18
CM		 6482
6483 /* page is wholly or partially inside EOF */
a52d9a80 6484 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6485 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6486 else
e6dcd2dc 6487 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6488
e6dcd2dc
CM		 6489 if (zero_start != PAGE_CACHE_SIZE) {
6490 kaddr = kmap(page);
6491 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6492 flush_dcache_page(page);
6493 kunmap(page);
6494 }
247e743c 6495 ClearPageChecked(page);
e6dcd2dc 6496 set_page_dirty(page);
50a9b214 6497 SetPageUptodate(page);
5a3f23d5 6498
257c62e1
CM		 6499 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6500 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6501
2ac55d41 6502 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6503
6504out_unlock:
50a9b214
CM		 6505 if (!ret)
6506 return VM_FAULT_LOCKED;
9ebefb18 6507 unlock_page(page);
0ca1f7ce 6508 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
1832a6d5 6509out:
9ebefb18
CM		 6510 return ret;
6511}
6512
a41ad394 6513static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6514{
6515 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6516 struct btrfs_block_rsv *rsv;
39279cc3 6517 int ret;
3893e33b 6518 int err = 0;
39279cc3 6519 struct btrfs_trans_handle *trans;
d3c2fdcf 6520 unsigned long nr;
dbe674a9 6521 u64 mask = root->sectorsize - 1;
07127184 6522 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6523
5d5e103a
JB		 6524 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6525 if (ret)
a41ad394 6526 return ret;
8082510e 6527
4a096752 6528 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6529 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6530
fcb80c2a
JB		 6531 /*
6532 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6533 * 3 things going on here
6534 *
6535 * 1) We need to reserve space for our orphan item and the space to
6536 * delete our orphan item. Lord knows we don't want to have a dangling
6537 * orphan item because we didn't reserve space to remove it.
6538 *
6539 * 2) We need to reserve space to update our inode.
6540 *
6541 * 3) We need to have something to cache all the space that is going to
6542 * be free'd up by the truncate operation, but also have some slack
6543 * space reserved in case it uses space during the truncate (thank you
6544 * very much snapshotting).
6545 *
6546 * And we need these to all be seperate. The fact is we can use alot of
6547 * space doing the truncate, and we have no earthly idea how much space
6548 * we will use, so we need the truncate reservation to be seperate so it
6549 * doesn't end up using space reserved for updating the inode or
6550 * removing the orphan item. We also need to be able to stop the
6551 * transaction and start a new one, which means we need to be able to
6552 * update the inode several times, and we have no idea of knowing how
6553 * many times that will be, so we can't just reserve 1 item for the
6554 * entirety of the opration, so that has to be done seperately as well.
6555 * Then there is the orphan item, which does indeed need to be held on
6556 * to for the whole operation, and we need nobody to touch this reserved
6557 * space except the orphan code.
6558 *
6559 * So that leaves us with
6560 *
6561 * 1) root->orphan_block_rsv - for the orphan deletion.
6562 * 2) rsv - for the truncate reservation, which we will steal from the
6563 * transaction reservation.
6564 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6565 * updating the inode.
6566 */
6567 rsv = btrfs_alloc_block_rsv(root);
6568 if (!rsv)
6569 return -ENOMEM;
4a338542 6570 rsv->size = min_size;
f0cd846e 6571
907cbceb 6572 /*
07127184 6573 * 1 for the truncate slack space
907cbceb
JB		 6574 * 1 for the orphan item we're going to add
6575 * 1 for the orphan item deletion
6576 * 1 for updating the inode.
6577 */
07127184 6578 trans = btrfs_start_transaction(root, 4);
fcb80c2a
JB		 6579 if (IS_ERR(trans)) {
6580 err = PTR_ERR(trans);
6581 goto out;
6582 }
f0cd846e 6583
907cbceb
JB		 6584 /* Migrate the slack space for the truncate to our reserve */
6585 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6586 min_size);
fcb80c2a 6587 BUG_ON(ret);
f0cd846e
JB		 6588
6589 ret = btrfs_orphan_add(trans, inode);
6590 if (ret) {
6591 btrfs_end_transaction(trans, root);
fcb80c2a 6592 goto out;
f0cd846e
JB		 6593 }
6594
5a3f23d5
CM		 6595 /*
6596 * setattr is responsible for setting the ordered_data_close flag,
6597 * but that is only tested during the last file release. That
6598 * could happen well after the next commit, leaving a great big
6599 * window where new writes may get lost if someone chooses to write
6600 * to this file after truncating to zero
6601 *
6602 * The inode doesn't have any dirty data here, and so if we commit
6603 * this is a noop. If someone immediately starts writing to the inode
6604 * it is very likely we'll catch some of their writes in this
6605 * transaction, and the commit will find this file on the ordered
6606 * data list with good things to send down.
6607 *
6608 * This is a best effort solution, there is still a window where
6609 * using truncate to replace the contents of the file will
6610 * end up with a zero length file after a crash.
6611 */
6612 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6613 btrfs_add_ordered_operation(trans, root, inode);
6614
8082510e 6615 while (1) {
36ba022a 6616 ret = btrfs_block_rsv_refill(root, rsv, min_size);
907cbceb
JB		 6617 if (ret) {
6618 /*
6619 * This can only happen with the original transaction we
6620 * started above, every other time we shouldn't have a
6621 * transaction started yet.
6622 */
6623 if (ret == -EAGAIN)
6624 goto end_trans;
6625 err = ret;
6626 break;
6627 }
6628
d68fc57b 6629 if (!trans) {
907cbceb
JB		 6630 /* Just need the 1 for updating the inode */
6631 trans = btrfs_start_transaction(root, 1);
fcb80c2a
JB		 6632 if (IS_ERR(trans)) {
6633 err = PTR_ERR(trans);
6634 goto out;
6635 }
d68fc57b
YZ		 6636 }
6637
907cbceb
JB		 6638 trans->block_rsv = rsv;
6639
8082510e
YZ		 6640 ret = btrfs_truncate_inode_items(trans, root, inode,
6641 inode->i_size,
6642 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6643 if (ret != -EAGAIN) {
6644 err = ret;
8082510e 6645 break;
3893e33b 6646 }
39279cc3 6647
fcb80c2a 6648 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6649 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6650 if (ret) {
6651 err = ret;
6652 break;
6653 }
907cbceb 6654end_trans:
8082510e
YZ		 6655 nr = trans->blocks_used;
6656 btrfs_end_transaction(trans, root);
d68fc57b 6657 trans = NULL;
8082510e 6658 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6659 }
6660
6661 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6662 trans->block_rsv = root->orphan_block_rsv;
8082510e 6663 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6664 if (ret)
6665 err = ret;
ded5db9d
JB		 6666 } else if (ret && inode->i_nlink > 0) {
6667 /*
6668 * Failed to do the truncate, remove us from the in memory
6669 * orphan list.
6670 */
6671 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6672 }
6673
fcb80c2a 6674 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6675 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6676 if (ret && !err)
6677 err = ret;
7b128766 6678
7b128766 6679 nr = trans->blocks_used;
89ce8a63 6680 ret = btrfs_end_transaction_throttle(trans, root);
fcb80c2a
JB		 6681 btrfs_btree_balance_dirty(root, nr);
6682
6683out:
6684 btrfs_free_block_rsv(root, rsv);
6685
3893e33b
JB		 6686 if (ret && !err)
6687 err = ret;
a41ad394 6688
3893e33b 6689 return err;
39279cc3
CM		 6690}
6691
d352ac68
CM		 6692/*
6693 * create a new subvolume directory/inode (helper for the ioctl).
6694 */
d2fb3437 6695int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6696 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6697{
39279cc3 6698 struct inode *inode;
76dda93c 6699 int err;
00e4e6b3 6700 u64 index = 0;
39279cc3 6701
aec7477b 6702 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d82a6f1d 6703 new_dirid, S_IFDIR | 0700, &index);
54aa1f4d 6704 if (IS_ERR(inode))
f46b5a66 6705 return PTR_ERR(inode);
39279cc3
CM		 6706 inode->i_op = &btrfs_dir_inode_operations;
6707 inode->i_fop = &btrfs_dir_file_operations;
6708
39279cc3 6709 inode->i_nlink = 1;
dbe674a9 6710 btrfs_i_size_write(inode, 0);
3b96362c 6711
76dda93c
YZ		 6712 err = btrfs_update_inode(trans, new_root, inode);
6713 BUG_ON(err);
cb8e7090 6714
76dda93c 6715 iput(inode);
cb8e7090 6716 return 0;
39279cc3
CM		 6717}
6718
39279cc3
CM		 6719struct inode *btrfs_alloc_inode(struct super_block *sb)
6720{
6721 struct btrfs_inode *ei;
2ead6ae7 6722 struct inode *inode;
39279cc3
CM		 6723
6724 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6725 if (!ei)
6726 return NULL;
2ead6ae7
YZ		 6727
6728 ei->root = NULL;
6729 ei->space_info = NULL;
6730 ei->generation = 0;
6731 ei->sequence = 0;
15ee9bc7 6732 ei->last_trans = 0;
257c62e1 6733 ei->last_sub_trans = 0;
e02119d5 6734 ei->logged_trans = 0;
2ead6ae7 6735 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6736 ei->disk_i_size = 0;
6737 ei->flags = 0;
7709cde3 6738 ei->csum_bytes = 0;
2ead6ae7
YZ		 6739 ei->index_cnt = (u64)-1;
6740 ei->last_unlink_trans = 0;
6741
9e0baf60
JB		 6742 spin_lock_init(&ei->lock);
6743 ei->outstanding_extents = 0;
6744 ei->reserved_extents = 0;
2ead6ae7
YZ		 6745
6746 ei->ordered_data_close = 0;
d68fc57b 6747 ei->orphan_meta_reserved = 0;
2ead6ae7 6748 ei->dummy_inode = 0;
4cb5300b 6749 ei->in_defrag = 0;
261507a0 6750 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6751
16cdcec7
MX		 6752 ei->delayed_node = NULL;
6753
2ead6ae7 6754 inode = &ei->vfs_inode;
a8067e02 6755 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6756 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6757 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
2ead6ae7 6758 mutex_init(&ei->log_mutex);
e6dcd2dc 6759 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7b128766 6760 INIT_LIST_HEAD(&ei->i_orphan);
2ead6ae7 6761 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6762 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6763 RB_CLEAR_NODE(&ei->rb_node);
6764
6765 return inode;
39279cc3
CM		 6766}
6767
fa0d7e3d
NP		 6768static void btrfs_i_callback(struct rcu_head *head)
6769{
6770 struct inode *inode = container_of(head, struct inode, i_rcu);
6771 INIT_LIST_HEAD(&inode->i_dentry);
6772 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6773}
6774
39279cc3
CM		 6775void btrfs_destroy_inode(struct inode *inode)
6776{
e6dcd2dc 6777 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6778 struct btrfs_root *root = BTRFS_I(inode)->root;
6779
39279cc3
CM		 6780 WARN_ON(!list_empty(&inode->i_dentry));
6781 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6782 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6783 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6784 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6785 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6786
a6dbd429
JB		 6787 /*
6788 * This can happen where we create an inode, but somebody else also
6789 * created the same inode and we need to destroy the one we already
6790 * created.
6791 */
6792 if (!root)
6793 goto free;
6794
5a3f23d5
CM		 6795 /*
6796 * Make sure we're properly removed from the ordered operation
6797 * lists.
6798 */
6799 smp_mb();
6800 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6801 spin_lock(&root->fs_info->ordered_extent_lock);
6802 list_del_init(&BTRFS_I(inode)->ordered_operations);
6803 spin_unlock(&root->fs_info->ordered_extent_lock);
6804 }
6805
d68fc57b 6806 spin_lock(&root->orphan_lock);
7b128766 6807 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
33345d01
LZ		 6808 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6809 (unsigned long long)btrfs_ino(inode));
8082510e 6810 list_del_init(&BTRFS_I(inode)->i_orphan);
7b128766 6811 }
d68fc57b 6812 spin_unlock(&root->orphan_lock);
7b128766 6813
d397712b 6814 while (1) {
e6dcd2dc
CM		 6815 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6816 if (!ordered)
6817 break;
6818 else {
d397712b
CM		 6819 printk(KERN_ERR "btrfs found ordered "
6820 "extent %llu %llu on inode cleanup\n",
6821 (unsigned long long)ordered->file_offset,
6822 (unsigned long long)ordered->len);
e6dcd2dc
CM		 6823 btrfs_remove_ordered_extent(inode, ordered);
6824 btrfs_put_ordered_extent(ordered);
6825 btrfs_put_ordered_extent(ordered);
6826 }
6827 }
5d4f98a2 6828 inode_tree_del(inode);
5b21f2ed 6829 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 6830free:
16cdcec7 6831 btrfs_remove_delayed_node(inode);
fa0d7e3d 6832 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 6833}
6834
45321ac5 6835int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 6836{
6837 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 6838
0af3d00b 6839 if (btrfs_root_refs(&root->root_item) == 0 &&
2cf8572d 6840 !btrfs_is_free_space_inode(root, inode))
45321ac5 6841 return 1;
76dda93c 6842 else
45321ac5 6843 return generic_drop_inode(inode);
76dda93c
YZ		 6844}
6845
0ee0fda0 6846static void init_once(void *foo)
39279cc3
CM		 6847{
6848 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6849
6850 inode_init_once(&ei->vfs_inode);
6851}
6852
6853void btrfs_destroy_cachep(void)
6854{
6855 if (btrfs_inode_cachep)
6856 kmem_cache_destroy(btrfs_inode_cachep);
6857 if (btrfs_trans_handle_cachep)
6858 kmem_cache_destroy(btrfs_trans_handle_cachep);
6859 if (btrfs_transaction_cachep)
6860 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 6861 if (btrfs_path_cachep)
6862 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 6863 if (btrfs_free_space_cachep)
6864 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 6865}
6866
6867int btrfs_init_cachep(void)
6868{
9601e3f6
CH		 6869 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6870 sizeof(struct btrfs_inode), 0,
6871 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 6872 if (!btrfs_inode_cachep)
6873 goto fail;
9601e3f6
CH		 6874
6875 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6876 sizeof(struct btrfs_trans_handle), 0,
6877 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6878 if (!btrfs_trans_handle_cachep)
6879 goto fail;
9601e3f6
CH		 6880
6881 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6882 sizeof(struct btrfs_transaction), 0,
6883 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6884 if (!btrfs_transaction_cachep)
6885 goto fail;
9601e3f6
CH		 6886
6887 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6888 sizeof(struct btrfs_path), 0,
6889 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6890 if (!btrfs_path_cachep)
6891 goto fail;
9601e3f6 6892
dc89e982
JB		 6893 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6894 sizeof(struct btrfs_free_space), 0,
6895 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6896 if (!btrfs_free_space_cachep)
6897 goto fail;
6898
39279cc3
CM		 6899 return 0;
6900fail:
6901 btrfs_destroy_cachep();
6902 return -ENOMEM;
6903}
6904
6905static int btrfs_getattr(struct vfsmount *mnt,
6906 struct dentry *dentry, struct kstat *stat)
6907{
6908 struct inode *inode = dentry->d_inode;
6909 generic_fillattr(inode, stat);
0ee5dc67 6910 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 6911 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 6912 stat->blocks = (inode_get_bytes(inode) +
6913 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 6914 return 0;
6915}
6916
75e7cb7f
LB		 6917/*
6918 * If a file is moved, it will inherit the cow and compression flags of the new
6919 * directory.
6920 */
6921static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6922{
6923 struct btrfs_inode *b_dir = BTRFS_I(dir);
6924 struct btrfs_inode *b_inode = BTRFS_I(inode);
6925
6926 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6927 b_inode->flags |= BTRFS_INODE_NODATACOW;
6928 else
6929 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6930
6931 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6932 b_inode->flags |= BTRFS_INODE_COMPRESS;
6933 else
6934 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6935}
6936
d397712b
CM		 6937static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6938 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 6939{
6940 struct btrfs_trans_handle *trans;
6941 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 6942 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 6943 struct inode *new_inode = new_dentry->d_inode;
6944 struct inode *old_inode = old_dentry->d_inode;
6945 struct timespec ctime = CURRENT_TIME;
00e4e6b3 6946 u64 index = 0;
4df27c4d 6947 u64 root_objectid;
39279cc3 6948 int ret;
33345d01 6949 u64 old_ino = btrfs_ino(old_inode);
39279cc3 6950
33345d01 6951 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 6952 return -EPERM;
6953
4df27c4d 6954 /* we only allow rename subvolume link between subvolumes */
33345d01 6955 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 6956 return -EXDEV;
6957
33345d01
LZ		 6958 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6959 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 6960 return -ENOTEMPTY;
5f39d397 6961
4df27c4d
YZ		 6962 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6963 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6964 return -ENOTEMPTY;
5a3f23d5
CM		 6965 /*
6966 * we're using rename to replace one file with another.
6967 * and the replacement file is large. Start IO on it now so
6968 * we don't add too much work to the end of the transaction
6969 */
4baf8c92 6970 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 6971 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6972 filemap_flush(old_inode->i_mapping);
6973
76dda93c 6974 /* close the racy window with snapshot create/destroy ioctl */
33345d01 6975 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 6976 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 6977 /*
6978 * We want to reserve the absolute worst case amount of items. So if
6979 * both inodes are subvols and we need to unlink them then that would
6980 * require 4 item modifications, but if they are both normal inodes it
6981 * would require 5 item modifications, so we'll assume their normal
6982 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6983 * should cover the worst case number of items we'll modify.
6984 */
6985 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 6986 if (IS_ERR(trans)) {
6987 ret = PTR_ERR(trans);
6988 goto out_notrans;
6989 }
76dda93c 6990
4df27c4d
YZ		 6991 if (dest != root)
6992 btrfs_record_root_in_trans(trans, dest);
5f39d397 6993
a5719521
YZ		 6994 ret = btrfs_set_inode_index(new_dir, &index);
6995 if (ret)
6996 goto out_fail;
5a3f23d5 6997
33345d01 6998 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6999 /* force full log commit if subvolume involved. */
7000 root->fs_info->last_trans_log_full_commit = trans->transid;
7001 } else {
a5719521
YZ		 7002 ret = btrfs_insert_inode_ref(trans, dest,
7003 new_dentry->d_name.name,
7004 new_dentry->d_name.len,
33345d01
LZ		 7005 old_ino,
7006 btrfs_ino(new_dir), index);
a5719521
YZ		 7007 if (ret)
7008 goto out_fail;
4df27c4d
YZ		 7009 /*
7010 * this is an ugly little race, but the rename is required
7011 * to make sure that if we crash, the inode is either at the
7012 * old name or the new one. pinning the log transaction lets
7013 * us make sure we don't allow a log commit to come in after
7014 * we unlink the name but before we add the new name back in.
7015 */
7016 btrfs_pin_log_trans(root);
7017 }
5a3f23d5
CM		 7018 /*
7019 * make sure the inode gets flushed if it is replacing
7020 * something.
7021 */
33345d01 7022 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7023 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7024
39279cc3
CM		 7025 old_dir->i_ctime = old_dir->i_mtime = ctime;
7026 new_dir->i_ctime = new_dir->i_mtime = ctime;
7027 old_inode->i_ctime = ctime;
5f39d397 7028
12fcfd22
CM		 7029 if (old_dentry->d_parent != new_dentry->d_parent)
7030 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7031
33345d01 7032 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7033 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7034 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7035 old_dentry->d_name.name,
7036 old_dentry->d_name.len);
7037 } else {
92986796
AV		 7038 ret = __btrfs_unlink_inode(trans, root, old_dir,
7039 old_dentry->d_inode,
7040 old_dentry->d_name.name,
7041 old_dentry->d_name.len);
7042 if (!ret)
7043 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d
YZ		 7044 }
7045 BUG_ON(ret);
39279cc3
CM		 7046
7047 if (new_inode) {
7048 new_inode->i_ctime = CURRENT_TIME;
33345d01 7049 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 7050 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7051 root_objectid = BTRFS_I(new_inode)->location.objectid;
7052 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7053 root_objectid,
7054 new_dentry->d_name.name,
7055 new_dentry->d_name.len);
7056 BUG_ON(new_inode->i_nlink == 0);
7057 } else {
7058 ret = btrfs_unlink_inode(trans, dest, new_dir,
7059 new_dentry->d_inode,
7060 new_dentry->d_name.name,
7061 new_dentry->d_name.len);
7062 }
7063 BUG_ON(ret);
7b128766 7064 if (new_inode->i_nlink == 0) {
e02119d5 7065 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7066 BUG_ON(ret);
7b128766 7067 }
39279cc3 7068 }
aec7477b 7069
75e7cb7f
LB		 7070 fixup_inode_flags(new_dir, old_inode);
7071
4df27c4d
YZ		 7072 ret = btrfs_add_link(trans, new_dir, old_inode,
7073 new_dentry->d_name.name,
a5719521 7074 new_dentry->d_name.len, 0, index);
4df27c4d 7075 BUG_ON(ret);
39279cc3 7076
33345d01 7077 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7078 struct dentry *parent = new_dentry->d_parent;
6a912213 7079 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7080 btrfs_end_log_trans(root);
7081 }
39279cc3 7082out_fail:
ab78c84d 7083 btrfs_end_transaction_throttle(trans, root);
b44c59a8 7084out_notrans:
33345d01 7085 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7086 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7087
39279cc3
CM		 7088 return ret;
7089}
7090
d352ac68
CM		 7091/*
7092 * some fairly slow code that needs optimization. This walks the list
7093 * of all the inodes with pending delalloc and forces them to disk.
7094 */
24bbcf04 7095int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7096{
7097 struct list_head *head = &root->fs_info->delalloc_inodes;
7098 struct btrfs_inode *binode;
5b21f2ed 7099 struct inode *inode;
ea8c2819 7100
c146afad
YZ		 7101 if (root->fs_info->sb->s_flags & MS_RDONLY)
7102 return -EROFS;
7103
75eff68e 7104 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7105 while (!list_empty(head)) {
ea8c2819
CM		 7106 binode = list_entry(head->next, struct btrfs_inode,
7107 delalloc_inodes);
5b21f2ed
ZY		 7108 inode = igrab(&binode->vfs_inode);
7109 if (!inode)
7110 list_del_init(&binode->delalloc_inodes);
75eff68e 7111 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7112 if (inode) {
8c8bee1d 7113 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7114 if (delay_iput)
7115 btrfs_add_delayed_iput(inode);
7116 else
7117 iput(inode);
5b21f2ed
ZY		 7118 }
7119 cond_resched();
75eff68e 7120 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7121 }
75eff68e 7122 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7123
7124 /* the filemap_flush will queue IO into the worker threads, but
7125 * we have to make sure the IO is actually started and that
7126 * ordered extents get created before we return
7127 */
7128 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7129 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7130 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7131 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7132 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7133 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7134 }
7135 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7136 return 0;
7137}
7138
39279cc3
CM		 7139static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7140 const char *symname)
7141{
7142 struct btrfs_trans_handle *trans;
7143 struct btrfs_root *root = BTRFS_I(dir)->root;
7144 struct btrfs_path *path;
7145 struct btrfs_key key;
1832a6d5 7146 struct inode *inode = NULL;
39279cc3
CM		 7147 int err;
7148 int drop_inode = 0;
7149 u64 objectid;
00e4e6b3 7150 u64 index = 0 ;
39279cc3
CM		 7151 int name_len;
7152 int datasize;
5f39d397 7153 unsigned long ptr;
39279cc3 7154 struct btrfs_file_extent_item *ei;
5f39d397 7155 struct extent_buffer *leaf;
1832a6d5 7156 unsigned long nr = 0;
39279cc3
CM		 7157
7158 name_len = strlen(symname) + 1;
7159 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7160 return -ENAMETOOLONG;
1832a6d5 7161
9ed74f2d
JB		 7162 /*
7163 * 2 items for inode item and ref
7164 * 2 items for dir items
7165 * 1 item for xattr if selinux is on
7166 */
a22285a6
YZ		 7167 trans = btrfs_start_transaction(root, 5);
7168 if (IS_ERR(trans))
7169 return PTR_ERR(trans);
1832a6d5 7170
581bb050
LZ		 7171 err = btrfs_find_free_ino(root, &objectid);
7172 if (err)
7173 goto out_unlock;
7174
aec7477b 7175 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7176 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7177 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7178 if (IS_ERR(inode)) {
7179 err = PTR_ERR(inode);
39279cc3 7180 goto out_unlock;
7cf96da3 7181 }
39279cc3 7182
2a7dba39 7183 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7184 if (err) {
7185 drop_inode = 1;
7186 goto out_unlock;
7187 }
7188
a1b075d2 7189 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7190 if (err)
7191 drop_inode = 1;
7192 else {
7193 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7194 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 7195 inode->i_fop = &btrfs_file_operations;
7196 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 7197 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7198 }
39279cc3
CM		 7199 if (drop_inode)
7200 goto out_unlock;
7201
7202 path = btrfs_alloc_path();
d8926bb3
MF		 7203 if (!path) {
7204 err = -ENOMEM;
7205 drop_inode = 1;
7206 goto out_unlock;
7207 }
33345d01 7208 key.objectid = btrfs_ino(inode);
39279cc3 7209 key.offset = 0;
39279cc3
CM		 7210 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7211 datasize = btrfs_file_extent_calc_inline_size(name_len);
7212 err = btrfs_insert_empty_item(trans, root, path, &key,
7213 datasize);
54aa1f4d
CM		 7214 if (err) {
7215 drop_inode = 1;
b0839166 7216 btrfs_free_path(path);
54aa1f4d
CM		 7217 goto out_unlock;
7218 }
5f39d397
CM		 7219 leaf = path->nodes[0];
7220 ei = btrfs_item_ptr(leaf, path->slots[0],
7221 struct btrfs_file_extent_item);
7222 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7223 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7224 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7225 btrfs_set_file_extent_encryption(leaf, ei, 0);
7226 btrfs_set_file_extent_compression(leaf, ei, 0);
7227 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7228 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7229
39279cc3 7230 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7231 write_extent_buffer(leaf, symname, ptr, name_len);
7232 btrfs_mark_buffer_dirty(leaf);
39279cc3 7233 btrfs_free_path(path);
5f39d397 7234
39279cc3
CM		 7235 inode->i_op = &btrfs_symlink_inode_operations;
7236 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7237 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7238 inode_set_bytes(inode, name_len);
dbe674a9 7239 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7240 err = btrfs_update_inode(trans, root, inode);
7241 if (err)
7242 drop_inode = 1;
39279cc3
CM		 7243
7244out_unlock:
d3c2fdcf 7245 nr = trans->blocks_used;
ab78c84d 7246 btrfs_end_transaction_throttle(trans, root);
39279cc3
CM		 7247 if (drop_inode) {
7248 inode_dec_link_count(inode);
7249 iput(inode);
7250 }
d3c2fdcf 7251 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7252 return err;
7253}
16432985 7254
0af3d00b
JB		 7255static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7256 u64 start, u64 num_bytes, u64 min_size,
7257 loff_t actual_len, u64 *alloc_hint,
7258 struct btrfs_trans_handle *trans)
d899e052 7259{
d899e052
YZ		 7260 struct btrfs_root *root = BTRFS_I(inode)->root;
7261 struct btrfs_key ins;
d899e052 7262 u64 cur_offset = start;
55a61d1d 7263 u64 i_size;
d899e052 7264 int ret = 0;
0af3d00b 7265 bool own_trans = true;
d899e052 7266
0af3d00b
JB		 7267 if (trans)
7268 own_trans = false;
d899e052 7269 while (num_bytes > 0) {
0af3d00b
JB		 7270 if (own_trans) {
7271 trans = btrfs_start_transaction(root, 3);
7272 if (IS_ERR(trans)) {
7273 ret = PTR_ERR(trans);
7274 break;
7275 }
5a303d5d
YZ		 7276 }
7277
efa56464
YZ		 7278 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7279 0, *alloc_hint, (u64)-1, &ins, 1);
5a303d5d 7280 if (ret) {
0af3d00b
JB		 7281 if (own_trans)
7282 btrfs_end_transaction(trans, root);
a22285a6 7283 break;
d899e052 7284 }
5a303d5d 7285
d899e052
YZ		 7286 ret = insert_reserved_file_extent(trans, inode,
7287 cur_offset, ins.objectid,
7288 ins.offset, ins.offset,
920bbbfb 7289 ins.offset, 0, 0, 0,
d899e052
YZ		 7290 BTRFS_FILE_EXTENT_PREALLOC);
7291 BUG_ON(ret);
a1ed835e
CM		 7292 btrfs_drop_extent_cache(inode, cur_offset,
7293 cur_offset + ins.offset -1, 0);
5a303d5d 7294
d899e052
YZ		 7295 num_bytes -= ins.offset;
7296 cur_offset += ins.offset;
efa56464 7297 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7298
d899e052 7299 inode->i_ctime = CURRENT_TIME;
6cbff00f 7300 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7301 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7302 (actual_len > inode->i_size) &&
7303 (cur_offset > inode->i_size)) {
d1ea6a61 7304 if (cur_offset > actual_len)
55a61d1d 7305 i_size = actual_len;
d1ea6a61 7306 else
55a61d1d
JB		 7307 i_size = cur_offset;
7308 i_size_write(inode, i_size);
7309 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7310 }
7311
d899e052
YZ		 7312 ret = btrfs_update_inode(trans, root, inode);
7313 BUG_ON(ret);
d899e052 7314
0af3d00b
JB		 7315 if (own_trans)
7316 btrfs_end_transaction(trans, root);
5a303d5d 7317 }
d899e052
YZ		 7318 return ret;
7319}
7320
0af3d00b
JB		 7321int btrfs_prealloc_file_range(struct inode *inode, int mode,
7322 u64 start, u64 num_bytes, u64 min_size,
7323 loff_t actual_len, u64 *alloc_hint)
7324{
7325 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7326 min_size, actual_len, alloc_hint,
7327 NULL);
7328}
7329
7330int btrfs_prealloc_file_range_trans(struct inode *inode,
7331 struct btrfs_trans_handle *trans, int mode,
7332 u64 start, u64 num_bytes, u64 min_size,
7333 loff_t actual_len, u64 *alloc_hint)
7334{
7335 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7336 min_size, actual_len, alloc_hint, trans);
7337}
7338
e6dcd2dc
CM		 7339static int btrfs_set_page_dirty(struct page *page)
7340{
e6dcd2dc
CM		 7341 return __set_page_dirty_nobuffers(page);
7342}
7343
10556cb2 7344static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7345{
b83cc969 7346 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7347 umode_t mode = inode->i_mode;
b83cc969 7348
cb6db4e5
JM		 7349 if (mask & MAY_WRITE &&
7350 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7351 if (btrfs_root_readonly(root))
7352 return -EROFS;
7353 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7354 return -EACCES;
7355 }
2830ba7f 7356 return generic_permission(inode, mask);
fdebe2bd 7357}
39279cc3 7358
6e1d5dcc 7359static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7360 .getattr = btrfs_getattr,
39279cc3
CM		 7361 .lookup = btrfs_lookup,
7362 .create = btrfs_create,
7363 .unlink = btrfs_unlink,
7364 .link = btrfs_link,
7365 .mkdir = btrfs_mkdir,
7366 .rmdir = btrfs_rmdir,
7367 .rename = btrfs_rename,
7368 .symlink = btrfs_symlink,
7369 .setattr = btrfs_setattr,
618e21d5 7370 .mknod = btrfs_mknod,
95819c05
CH		 7371 .setxattr = btrfs_setxattr,
7372 .getxattr = btrfs_getxattr,
5103e947 7373 .listxattr = btrfs_listxattr,
95819c05 7374 .removexattr = btrfs_removexattr,
fdebe2bd 7375 .permission = btrfs_permission,
4e34e719 7376 .get_acl = btrfs_get_acl,
39279cc3 7377};
6e1d5dcc 7378static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7379 .lookup = btrfs_lookup,
fdebe2bd 7380 .permission = btrfs_permission,
4e34e719 7381 .get_acl = btrfs_get_acl,
39279cc3 7382};
76dda93c 7383
828c0950 7384static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7385 .llseek = generic_file_llseek,
7386 .read = generic_read_dir,
cbdf5a24 7387 .readdir = btrfs_real_readdir,
34287aa3 7388 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7389#ifdef CONFIG_COMPAT
34287aa3 7390 .compat_ioctl = btrfs_ioctl,
39279cc3 7391#endif
6bf13c0c 7392 .release = btrfs_release_file,
e02119d5 7393 .fsync = btrfs_sync_file,
39279cc3
CM		 7394};
7395
d1310b2e 7396static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7397 .fill_delalloc = run_delalloc_range,
065631f6 7398 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7399 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7400 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7401 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7402 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 7403 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 7404 .set_bit_hook = btrfs_set_bit_hook,
7405 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7406 .merge_extent_hook = btrfs_merge_extent_hook,
7407 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7408};
7409
35054394
CM		 7410/*
7411 * btrfs doesn't support the bmap operation because swapfiles
7412 * use bmap to make a mapping of extents in the file. They assume
7413 * these extents won't change over the life of the file and they
7414 * use the bmap result to do IO directly to the drive.
7415 *
7416 * the btrfs bmap call would return logical addresses that aren't
7417 * suitable for IO and they also will change frequently as COW
7418 * operations happen. So, swapfile + btrfs == corruption.
7419 *
7420 * For now we're avoiding this by dropping bmap.
7421 */
7f09410b 7422static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7423 .readpage = btrfs_readpage,
7424 .writepage = btrfs_writepage,
b293f02e 7425 .writepages = btrfs_writepages,
3ab2fb5a 7426 .readpages = btrfs_readpages,
16432985 7427 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7428 .invalidatepage = btrfs_invalidatepage,
7429 .releasepage = btrfs_releasepage,
e6dcd2dc 7430 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7431 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7432};
7433
7f09410b 7434static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7435 .readpage = btrfs_readpage,
7436 .writepage = btrfs_writepage,
2bf5a725
CM		 7437 .invalidatepage = btrfs_invalidatepage,
7438 .releasepage = btrfs_releasepage,
39279cc3
CM		 7439};
7440
6e1d5dcc 7441static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7442 .getattr = btrfs_getattr,
7443 .setattr = btrfs_setattr,
95819c05
CH		 7444 .setxattr = btrfs_setxattr,
7445 .getxattr = btrfs_getxattr,
5103e947 7446 .listxattr = btrfs_listxattr,
95819c05 7447 .removexattr = btrfs_removexattr,
fdebe2bd 7448 .permission = btrfs_permission,
1506fcc8 7449 .fiemap = btrfs_fiemap,
4e34e719 7450 .get_acl = btrfs_get_acl,
39279cc3 7451};
6e1d5dcc 7452static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7453 .getattr = btrfs_getattr,
7454 .setattr = btrfs_setattr,
fdebe2bd 7455 .permission = btrfs_permission,
95819c05
CH		 7456 .setxattr = btrfs_setxattr,
7457 .getxattr = btrfs_getxattr,
33268eaf 7458 .listxattr = btrfs_listxattr,
95819c05 7459 .removexattr = btrfs_removexattr,
4e34e719 7460 .get_acl = btrfs_get_acl,
618e21d5 7461};
6e1d5dcc 7462static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7463 .readlink = generic_readlink,
7464 .follow_link = page_follow_link_light,
7465 .put_link = page_put_link,
f209561a 7466 .getattr = btrfs_getattr,
fdebe2bd 7467 .permission = btrfs_permission,
0279b4cd
JO		 7468 .setxattr = btrfs_setxattr,
7469 .getxattr = btrfs_getxattr,
7470 .listxattr = btrfs_listxattr,
7471 .removexattr = btrfs_removexattr,
4e34e719 7472 .get_acl = btrfs_get_acl,
39279cc3 7473};
76dda93c 7474
82d339d9 7475const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7476 .d_delete = btrfs_dentry_delete,
b4aff1f8 7477 .d_release = btrfs_dentry_release,
76dda93c 7478};
Linux 6.x block layer and io_uring tree(s)