projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: __add_reloc_root error push-up
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
4b4e25f2 42#include "compat.h"
39279cc3
CM		 43#include "ctree.h"
44#include "disk-io.h"
45#include "transaction.h"
46#include "btrfs_inode.h"
47#include "ioctl.h"
48#include "print-tree.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 51#include "tree-log.h"
4a54c8c1 52#include "volumes.h"
c8b97818 53#include "compression.h"
b4ce94de 54#include "locking.h"
dc89e982 55#include "free-space-cache.h"
581bb050 56#include "inode-map.h"
39279cc3
CM		 57
58struct btrfs_iget_args {
59 u64 ino;
60 struct btrfs_root *root;
61};
62
6e1d5dcc
AD		 63static const struct inode_operations btrfs_dir_inode_operations;
64static const struct inode_operations btrfs_symlink_inode_operations;
65static const struct inode_operations btrfs_dir_ro_inode_operations;
66static const struct inode_operations btrfs_special_inode_operations;
67static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 68static const struct address_space_operations btrfs_aops;
69static const struct address_space_operations btrfs_symlink_aops;
828c0950 70static const struct file_operations btrfs_dir_file_operations;
d1310b2e 71static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 72
73static struct kmem_cache *btrfs_inode_cachep;
74struct kmem_cache *btrfs_trans_handle_cachep;
75struct kmem_cache *btrfs_transaction_cachep;
39279cc3 76struct kmem_cache *btrfs_path_cachep;
dc89e982 77struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 78
79#define S_SHIFT 12
80static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
88};
89
a41ad394
JB		 90static int btrfs_setsize(struct inode *inode, loff_t newsize);
91static int btrfs_truncate(struct inode *inode);
c8b97818 92static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 93static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
2115133f
CM		 97static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
7b128766 99
f34f57a3 100static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
0279b4cd
JO		 103{
104 int err;
105
f34f57a3 106 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 107 if (!err)
2a7dba39 108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 109 return err;
110}
111
c8b97818
CM		 112/*
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
116 */
d397712b 117static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
fe3f566c 120 int compress_type,
c8b97818
CM		 121 struct page **compressed_pages)
122{
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
127 char *kaddr;
128 unsigned long ptr;
129 struct btrfs_file_extent_item *ei;
130 int err = 0;
131 int ret;
132 size_t cur_size = size;
133 size_t datasize;
134 unsigned long offset;
c8b97818 135
fe3f566c 136 if (compressed_size && compressed_pages)
c8b97818 137 cur_size = compressed_size;
c8b97818 138
d397712b
CM		 139 path = btrfs_alloc_path();
140 if (!path)
c8b97818
CM		 141 return -ENOMEM;
142
b9473439 143 path->leave_spinning = 1;
c8b97818 144
33345d01 145 key.objectid = btrfs_ino(inode);
c8b97818
CM		 146 key.offset = start;
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 datasize);
153 BUG_ON(ret);
154 if (ret) {
155 err = ret;
c8b97818
CM		 156 goto fail;
157 }
158 leaf = path->nodes[0];
159 ei = btrfs_item_ptr(leaf, path->slots[0],
160 struct btrfs_file_extent_item);
161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
162 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
163 btrfs_set_file_extent_encryption(leaf, ei, 0);
164 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
165 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
166 ptr = btrfs_file_extent_inline_start(ei);
167
261507a0 168 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 169 struct page *cpage;
170 int i = 0;
d397712b 171 while (compressed_size > 0) {
c8b97818 172 cpage = compressed_pages[i];
5b050f04 173 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 174 PAGE_CACHE_SIZE);
175
b9473439 176 kaddr = kmap_atomic(cpage, KM_USER0);
c8b97818 177 write_extent_buffer(leaf, kaddr, ptr, cur_size);
b9473439 178 kunmap_atomic(kaddr, KM_USER0);
c8b97818
CM		 179
180 i++;
181 ptr += cur_size;
182 compressed_size -= cur_size;
183 }
184 btrfs_set_file_extent_compression(leaf, ei,
261507a0 185 compress_type);
c8b97818
CM		 186 } else {
187 page = find_get_page(inode->i_mapping,
188 start >> PAGE_CACHE_SHIFT);
189 btrfs_set_file_extent_compression(leaf, ei, 0);
190 kaddr = kmap_atomic(page, KM_USER0);
191 offset = start & (PAGE_CACHE_SIZE - 1);
192 write_extent_buffer(leaf, kaddr + offset, ptr, size);
193 kunmap_atomic(kaddr, KM_USER0);
194 page_cache_release(page);
195 }
196 btrfs_mark_buffer_dirty(leaf);
197 btrfs_free_path(path);
198
c2167754
YZ		 199 /*
200 * we're an inline extent, so nobody can
201 * extend the file past i_size without locking
202 * a page we already have locked.
203 *
204 * We must do any isize and inode updates
205 * before we unlock the pages. Otherwise we
206 * could end up racing with unlink.
207 */
c8b97818
CM		 208 BTRFS_I(inode)->disk_i_size = inode->i_size;
209 btrfs_update_inode(trans, root, inode);
c2167754 210
c8b97818
CM		 211 return 0;
212fail:
213 btrfs_free_path(path);
214 return err;
215}
216
217
218/*
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
222 */
7f366cfe 223static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 224 struct btrfs_root *root,
225 struct inode *inode, u64 start, u64 end,
fe3f566c 226 size_t compressed_size, int compress_type,
c8b97818
CM		 227 struct page **compressed_pages)
228{
229 u64 isize = i_size_read(inode);
230 u64 actual_end = min(end + 1, isize);
231 u64 inline_len = actual_end - start;
232 u64 aligned_end = (end + root->sectorsize - 1) &
233 ~((u64)root->sectorsize - 1);
234 u64 hint_byte;
235 u64 data_len = inline_len;
236 int ret;
237
238 if (compressed_size)
239 data_len = compressed_size;
240
241 if (start > 0 ||
70b99e69 242 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 243 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (!compressed_size &&
245 (actual_end & (root->sectorsize - 1)) == 0) ||
246 end + 1 < isize ||
247 data_len > root->fs_info->max_inline) {
248 return 1;
249 }
250
920bbbfb 251 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
a1ed835e 252 &hint_byte, 1);
c8b97818
CM		 253 BUG_ON(ret);
254
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
fe3f566c 259 compress_type, compressed_pages);
c8b97818 260 BUG_ON(ret);
0ca1f7ce 261 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 262 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 263 return 0;
264}
265
771ed689
CM		 266struct async_extent {
267 u64 start;
268 u64 ram_size;
269 u64 compressed_size;
270 struct page **pages;
271 unsigned long nr_pages;
261507a0 272 int compress_type;
771ed689
CM		 273 struct list_head list;
274};
275
276struct async_cow {
277 struct inode *inode;
278 struct btrfs_root *root;
279 struct page *locked_page;
280 u64 start;
281 u64 end;
282 struct list_head extents;
283 struct btrfs_work work;
284};
285
286static noinline int add_async_extent(struct async_cow *cow,
287 u64 start, u64 ram_size,
288 u64 compressed_size,
289 struct page **pages,
261507a0
LZ		 290 unsigned long nr_pages,
291 int compress_type)
771ed689
CM		 292{
293 struct async_extent *async_extent;
294
295 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
dac97e51 296 BUG_ON(!async_extent);
771ed689
CM		 297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
261507a0 302 async_extent->compress_type = compress_type;
771ed689
CM		 303 list_add_tail(&async_extent->list, &cow->extents);
304 return 0;
305}
306
d352ac68 307/*
771ed689
CM		 308 * we create compressed extents in two phases. The first
309 * phase compresses a range of pages that have already been
310 * locked (both pages and state bits are locked).
c8b97818 311 *
771ed689
CM		 312 * This is done inside an ordered work queue, and the compression
313 * is spread across many cpus. The actual IO submission is step
314 * two, and the ordered work queue takes care of making sure that
315 * happens in the same order things were put onto the queue by
316 * writepages and friends.
c8b97818 317 *
771ed689
CM		 318 * If this code finds it can't get good compression, it puts an
319 * entry onto the work queue to write the uncompressed bytes. This
320 * makes sure that both compressed inodes and uncompressed inodes
321 * are written in the same order that pdflush sent them down.
d352ac68 322 */
771ed689
CM		 323static noinline int compress_file_range(struct inode *inode,
324 struct page *locked_page,
325 u64 start, u64 end,
326 struct async_cow *async_cow,
327 int *num_added)
b888db2b
CM		 328{
329 struct btrfs_root *root = BTRFS_I(inode)->root;
330 struct btrfs_trans_handle *trans;
db94535d 331 u64 num_bytes;
db94535d 332 u64 blocksize = root->sectorsize;
c8b97818 333 u64 actual_end;
42dc7bab 334 u64 isize = i_size_read(inode);
e6dcd2dc 335 int ret = 0;
c8b97818
CM		 336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
771ed689 342 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 343 int i;
344 int will_compress;
261507a0 345 int compress_type = root->fs_info->compress_type;
b888db2b 346
4cb5300b
CM		 347 /* if this is a small write inside eof, kick off a defragbot */
348 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
349 btrfs_add_inode_defrag(NULL, inode);
350
42dc7bab 351 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 352again:
353 will_compress = 0;
354 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
355 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 356
f03d9301
CM		 357 /*
358 * we don't want to send crud past the end of i_size through
359 * compression, that's just a waste of CPU time. So, if the
360 * end of the file is before the start of our current
361 * requested range of bytes, we bail out to the uncompressed
362 * cleanup code that can deal with all of this.
363 *
364 * It isn't really the fastest way to fix things, but this is a
365 * very uncommon corner.
366 */
367 if (actual_end <= start)
368 goto cleanup_and_bail_uncompressed;
369
c8b97818
CM		 370 total_compressed = actual_end - start;
371
372 /* we want to make sure that amount of ram required to uncompress
373 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 374 * of a compressed extent to 128k. This is a crucial number
375 * because it also controls how easily we can spread reads across
376 * cpus for decompression.
377 *
378 * We also want to make sure the amount of IO required to do
379 * a random read is reasonably small, so we limit the size of
380 * a compressed extent to 128k.
c8b97818
CM		 381 */
382 total_compressed = min(total_compressed, max_uncompressed);
db94535d 383 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 384 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 385 total_in = 0;
386 ret = 0;
db94535d 387
771ed689
CM		 388 /*
389 * we do compression for mount -o compress and when the
390 * inode has not been flagged as nocompress. This flag can
391 * change at any time if we discover bad compression ratios.
c8b97818 392 */
6cbff00f 393 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 394 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 395 (BTRFS_I(inode)->force_compress) ||
396 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 397 WARN_ON(pages);
cfbc246e 398 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 399 if (!pages) {
400 /* just bail out to the uncompressed code */
401 goto cont;
402 }
c8b97818 403
261507a0
LZ		 404 if (BTRFS_I(inode)->force_compress)
405 compress_type = BTRFS_I(inode)->force_compress;
406
407 ret = btrfs_compress_pages(compress_type,
408 inode->i_mapping, start,
409 total_compressed, pages,
410 nr_pages, &nr_pages_ret,
411 &total_in,
412 &total_compressed,
413 max_compressed);
c8b97818
CM		 414
415 if (!ret) {
416 unsigned long offset = total_compressed &
417 (PAGE_CACHE_SIZE - 1);
418 struct page *page = pages[nr_pages_ret - 1];
419 char *kaddr;
420
421 /* zero the tail end of the last page, we might be
422 * sending it down to disk
423 */
424 if (offset) {
425 kaddr = kmap_atomic(page, KM_USER0);
426 memset(kaddr + offset, 0,
427 PAGE_CACHE_SIZE - offset);
428 kunmap_atomic(kaddr, KM_USER0);
429 }
430 will_compress = 1;
431 }
432 }
560f7d75 433cont:
c8b97818 434 if (start == 0) {
7a7eaa40 435 trans = btrfs_join_transaction(root);
3612b495 436 BUG_ON(IS_ERR(trans));
0ca1f7ce 437 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 438
c8b97818 439 /* lets try to make an inline extent */
771ed689 440 if (ret || total_in < (actual_end - start)) {
c8b97818 441 /* we didn't compress the entire range, try
771ed689 442 * to make an uncompressed inline extent.
c8b97818
CM		 443 */
444 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 445 start, end, 0, 0, NULL);
c8b97818 446 } else {
771ed689 447 /* try making a compressed inline extent */
c8b97818
CM		 448 ret = cow_file_range_inline(trans, root, inode,
449 start, end,
fe3f566c
LZ		 450 total_compressed,
451 compress_type, pages);
c8b97818
CM		 452 }
453 if (ret == 0) {
771ed689
CM		 454 /*
455 * inline extent creation worked, we don't need
456 * to create any more async work items. Unlock
457 * and free up our temp pages.
458 */
c8b97818 459 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 460 &BTRFS_I(inode)->io_tree,
461 start, end, NULL,
462 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 463 EXTENT_CLEAR_DELALLOC |
a791e35e 464 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 465
466 btrfs_end_transaction(trans, root);
c8b97818
CM		 467 goto free_pages_out;
468 }
c2167754 469 btrfs_end_transaction(trans, root);
c8b97818
CM		 470 }
471
472 if (will_compress) {
473 /*
474 * we aren't doing an inline extent round the compressed size
475 * up to a block size boundary so the allocator does sane
476 * things
477 */
478 total_compressed = (total_compressed + blocksize - 1) &
479 ~(blocksize - 1);
480
481 /*
482 * one last check to make sure the compression is really a
483 * win, compare the page count read with the blocks on disk
484 */
485 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
486 ~(PAGE_CACHE_SIZE - 1);
487 if (total_compressed >= total_in) {
488 will_compress = 0;
489 } else {
c8b97818
CM		 490 num_bytes = total_in;
491 }
492 }
493 if (!will_compress && pages) {
494 /*
495 * the compression code ran but failed to make things smaller,
496 * free any pages it allocated and our page pointer array
497 */
498 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 499 WARN_ON(pages[i]->mapping);
c8b97818
CM		 500 page_cache_release(pages[i]);
501 }
502 kfree(pages);
503 pages = NULL;
504 total_compressed = 0;
505 nr_pages_ret = 0;
506
507 /* flag the file so we don't compress in the future */
1e701a32
CM		 508 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
509 !(BTRFS_I(inode)->force_compress)) {
a555f810 510 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 511 }
c8b97818 512 }
771ed689
CM		 513 if (will_compress) {
514 *num_added += 1;
c8b97818 515
771ed689
CM		 516 /* the async work queues will take care of doing actual
517 * allocation on disk for these compressed pages,
518 * and will submit them to the elevator.
519 */
520 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 521 total_compressed, pages, nr_pages_ret,
522 compress_type);
179e29e4 523
24ae6365 524 if (start + num_bytes < end) {
771ed689
CM		 525 start += num_bytes;
526 pages = NULL;
527 cond_resched();
528 goto again;
529 }
530 } else {
f03d9301 531cleanup_and_bail_uncompressed:
771ed689
CM		 532 /*
533 * No compression, but we still need to write the pages in
534 * the file we've been given so far. redirty the locked
535 * page if it corresponds to our extent and set things up
536 * for the async work queue to run cow_file_range to do
537 * the normal delalloc dance
538 */
539 if (page_offset(locked_page) >= start &&
540 page_offset(locked_page) <= end) {
541 __set_page_dirty_nobuffers(locked_page);
542 /* unlocked later on in the async handlers */
543 }
261507a0
LZ		 544 add_async_extent(async_cow, start, end - start + 1,
545 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 546 *num_added += 1;
547 }
3b951516 548
771ed689
CM		 549out:
550 return 0;
551
552free_pages_out:
553 for (i = 0; i < nr_pages_ret; i++) {
554 WARN_ON(pages[i]->mapping);
555 page_cache_release(pages[i]);
556 }
d397712b 557 kfree(pages);
771ed689
CM		 558
559 goto out;
560}
561
562/*
563 * phase two of compressed writeback. This is the ordered portion
564 * of the code, which only gets called in the order the work was
565 * queued. We walk all the async extents created by compress_file_range
566 * and send them down to the disk.
567 */
568static noinline int submit_compressed_extents(struct inode *inode,
569 struct async_cow *async_cow)
570{
571 struct async_extent *async_extent;
572 u64 alloc_hint = 0;
573 struct btrfs_trans_handle *trans;
574 struct btrfs_key ins;
575 struct extent_map *em;
576 struct btrfs_root *root = BTRFS_I(inode)->root;
577 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
578 struct extent_io_tree *io_tree;
f5a84ee3 579 int ret = 0;
771ed689
CM		 580
581 if (list_empty(&async_cow->extents))
582 return 0;
583
771ed689 584
d397712b 585 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 586 async_extent = list_entry(async_cow->extents.next,
587 struct async_extent, list);
588 list_del(&async_extent->list);
c8b97818 589
771ed689
CM		 590 io_tree = &BTRFS_I(inode)->io_tree;
591
f5a84ee3 592retry:
771ed689
CM		 593 /* did the compression code fall back to uncompressed IO? */
594 if (!async_extent->pages) {
595 int page_started = 0;
596 unsigned long nr_written = 0;
597
598 lock_extent(io_tree, async_extent->start,
2ac55d41
JB		 599 async_extent->start +
600 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 601
602 /* allocate blocks */
f5a84ee3
JB		 603 ret = cow_file_range(inode, async_cow->locked_page,
604 async_extent->start,
605 async_extent->start +
606 async_extent->ram_size - 1,
607 &page_started, &nr_written, 0);
771ed689
CM		 608
609 /*
610 * if page_started, cow_file_range inserted an
611 * inline extent and took care of all the unlocking
612 * and IO for us. Otherwise, we need to submit
613 * all those pages down to the drive.
614 */
f5a84ee3 615 if (!page_started && !ret)
771ed689
CM		 616 extent_write_locked_range(io_tree,
617 inode, async_extent->start,
d397712b 618 async_extent->start +
771ed689
CM		 619 async_extent->ram_size - 1,
620 btrfs_get_extent,
621 WB_SYNC_ALL);
622 kfree(async_extent);
623 cond_resched();
624 continue;
625 }
626
627 lock_extent(io_tree, async_extent->start,
628 async_extent->start + async_extent->ram_size - 1,
629 GFP_NOFS);
771ed689 630
7a7eaa40 631 trans = btrfs_join_transaction(root);
3612b495 632 BUG_ON(IS_ERR(trans));
74b21075 633 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689
CM		 634 ret = btrfs_reserve_extent(trans, root,
635 async_extent->compressed_size,
636 async_extent->compressed_size,
637 0, alloc_hint,
638 (u64)-1, &ins, 1);
c2167754
YZ		 639 btrfs_end_transaction(trans, root);
640
f5a84ee3
JB		 641 if (ret) {
642 int i;
643 for (i = 0; i < async_extent->nr_pages; i++) {
644 WARN_ON(async_extent->pages[i]->mapping);
645 page_cache_release(async_extent->pages[i]);
646 }
647 kfree(async_extent->pages);
648 async_extent->nr_pages = 0;
649 async_extent->pages = NULL;
650 unlock_extent(io_tree, async_extent->start,
651 async_extent->start +
652 async_extent->ram_size - 1, GFP_NOFS);
653 goto retry;
654 }
655
c2167754
YZ		 656 /*
657 * here we're doing allocation and writeback of the
658 * compressed pages
659 */
660 btrfs_drop_extent_cache(inode, async_extent->start,
661 async_extent->start +
662 async_extent->ram_size - 1, 0);
663
172ddd60 664 em = alloc_extent_map();
c26a9203 665 BUG_ON(!em);
771ed689
CM		 666 em->start = async_extent->start;
667 em->len = async_extent->ram_size;
445a6944 668 em->orig_start = em->start;
c8b97818 669
771ed689
CM		 670 em->block_start = ins.objectid;
671 em->block_len = ins.offset;
672 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 673 em->compress_type = async_extent->compress_type;
771ed689
CM		 674 set_bit(EXTENT_FLAG_PINNED, &em->flags);
675 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
676
d397712b 677 while (1) {
890871be 678 write_lock(&em_tree->lock);
771ed689 679 ret = add_extent_mapping(em_tree, em);
890871be 680 write_unlock(&em_tree->lock);
771ed689
CM		 681 if (ret != -EEXIST) {
682 free_extent_map(em);
683 break;
684 }
685 btrfs_drop_extent_cache(inode, async_extent->start,
686 async_extent->start +
687 async_extent->ram_size - 1, 0);
688 }
689
261507a0
LZ		 690 ret = btrfs_add_ordered_extent_compress(inode,
691 async_extent->start,
692 ins.objectid,
693 async_extent->ram_size,
694 ins.offset,
695 BTRFS_ORDERED_COMPRESSED,
696 async_extent->compress_type);
771ed689
CM		 697 BUG_ON(ret);
698
771ed689
CM		 699 /*
700 * clear dirty, set writeback and unlock the pages.
701 */
702 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 703 &BTRFS_I(inode)->io_tree,
704 async_extent->start,
705 async_extent->start +
706 async_extent->ram_size - 1,
707 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
708 EXTENT_CLEAR_UNLOCK |
a3429ab7 709 EXTENT_CLEAR_DELALLOC |
a791e35e 710 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 711
712 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 713 async_extent->start,
714 async_extent->ram_size,
715 ins.objectid,
716 ins.offset, async_extent->pages,
717 async_extent->nr_pages);
771ed689
CM		 718
719 BUG_ON(ret);
771ed689
CM		 720 alloc_hint = ins.objectid + ins.offset;
721 kfree(async_extent);
722 cond_resched();
723 }
724
771ed689
CM		 725 return 0;
726}
727
4b46fce2
JB		 728static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
729 u64 num_bytes)
730{
731 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
732 struct extent_map *em;
733 u64 alloc_hint = 0;
734
735 read_lock(&em_tree->lock);
736 em = search_extent_mapping(em_tree, start, num_bytes);
737 if (em) {
738 /*
739 * if block start isn't an actual block number then find the
740 * first block in this inode and use that as a hint. If that
741 * block is also bogus then just don't worry about it.
742 */
743 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
744 free_extent_map(em);
745 em = search_extent_mapping(em_tree, 0, 0);
746 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
747 alloc_hint = em->block_start;
748 if (em)
749 free_extent_map(em);
750 } else {
751 alloc_hint = em->block_start;
752 free_extent_map(em);
753 }
754 }
755 read_unlock(&em_tree->lock);
756
757 return alloc_hint;
758}
759
771ed689
CM		 760/*
761 * when extent_io.c finds a delayed allocation range in the file,
762 * the call backs end up in this code. The basic idea is to
763 * allocate extents on disk for the range, and create ordered data structs
764 * in ram to track those extents.
765 *
766 * locked_page is the page that writepage had locked already. We use
767 * it to make sure we don't do extra locks or unlocks.
768 *
769 * *page_started is set to one if we unlock locked_page and do everything
770 * required to start IO on it. It may be clean and already done with
771 * IO when we return.
772 */
773static noinline int cow_file_range(struct inode *inode,
774 struct page *locked_page,
775 u64 start, u64 end, int *page_started,
776 unsigned long *nr_written,
777 int unlock)
778{
779 struct btrfs_root *root = BTRFS_I(inode)->root;
780 struct btrfs_trans_handle *trans;
781 u64 alloc_hint = 0;
782 u64 num_bytes;
783 unsigned long ram_size;
784 u64 disk_num_bytes;
785 u64 cur_alloc_size;
786 u64 blocksize = root->sectorsize;
771ed689
CM		 787 struct btrfs_key ins;
788 struct extent_map *em;
789 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
790 int ret = 0;
791
2cf8572d 792 BUG_ON(btrfs_is_free_space_inode(root, inode));
7a7eaa40 793 trans = btrfs_join_transaction(root);
3612b495 794 BUG_ON(IS_ERR(trans));
0ca1f7ce 795 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 796
771ed689
CM		 797 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
798 num_bytes = max(blocksize, num_bytes);
799 disk_num_bytes = num_bytes;
800 ret = 0;
801
4cb5300b
CM		 802 /* if this is a small write inside eof, kick off defrag */
803 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
804 btrfs_add_inode_defrag(trans, inode);
805
771ed689
CM		 806 if (start == 0) {
807 /* lets try to make an inline extent */
808 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 809 start, end, 0, 0, NULL);
771ed689
CM		 810 if (ret == 0) {
811 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 812 &BTRFS_I(inode)->io_tree,
813 start, end, NULL,
814 EXTENT_CLEAR_UNLOCK_PAGE |
815 EXTENT_CLEAR_UNLOCK |
816 EXTENT_CLEAR_DELALLOC |
817 EXTENT_CLEAR_DIRTY |
818 EXTENT_SET_WRITEBACK |
819 EXTENT_END_WRITEBACK);
c2167754 820
771ed689
CM		 821 *nr_written = *nr_written +
822 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
823 *page_started = 1;
824 ret = 0;
825 goto out;
826 }
827 }
828
829 BUG_ON(disk_num_bytes >
6c41761f 830 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 831
4b46fce2 832 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 833 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
834
d397712b 835 while (disk_num_bytes > 0) {
a791e35e
CM		 836 unsigned long op;
837
287a0ab9 838 cur_alloc_size = disk_num_bytes;
e6dcd2dc 839 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 840 root->sectorsize, 0, alloc_hint,
e6dcd2dc 841 (u64)-1, &ins, 1);
d397712b
CM		 842 BUG_ON(ret);
843
172ddd60 844 em = alloc_extent_map();
c26a9203 845 BUG_ON(!em);
e6dcd2dc 846 em->start = start;
445a6944 847 em->orig_start = em->start;
771ed689
CM		 848 ram_size = ins.offset;
849 em->len = ins.offset;
c8b97818 850
e6dcd2dc 851 em->block_start = ins.objectid;
c8b97818 852 em->block_len = ins.offset;
e6dcd2dc 853 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 854 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 855
d397712b 856 while (1) {
890871be 857 write_lock(&em_tree->lock);
e6dcd2dc 858 ret = add_extent_mapping(em_tree, em);
890871be 859 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 860 if (ret != -EEXIST) {
861 free_extent_map(em);
862 break;
863 }
864 btrfs_drop_extent_cache(inode, start,
c8b97818 865 start + ram_size - 1, 0);
e6dcd2dc
CM		 866 }
867
98d20f67 868 cur_alloc_size = ins.offset;
e6dcd2dc 869 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 870 ram_size, cur_alloc_size, 0);
e6dcd2dc 871 BUG_ON(ret);
c8b97818 872
17d217fe
YZ		 873 if (root->root_key.objectid ==
874 BTRFS_DATA_RELOC_TREE_OBJECTID) {
875 ret = btrfs_reloc_clone_csums(inode, start,
876 cur_alloc_size);
877 BUG_ON(ret);
878 }
879
d397712b 880 if (disk_num_bytes < cur_alloc_size)
3b951516 881 break;
d397712b 882
c8b97818
CM		 883 /* we're not doing compressed IO, don't unlock the first
884 * page (which the caller expects to stay locked), don't
885 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 886 *
887 * Do set the Private2 bit so we know this page was properly
888 * setup for writepage
c8b97818 889 */
a791e35e
CM		 890 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
891 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
892 EXTENT_SET_PRIVATE2;
893
c8b97818
CM		 894 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
895 start, start + ram_size - 1,
a791e35e 896 locked_page, op);
c8b97818 897 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 898 num_bytes -= cur_alloc_size;
899 alloc_hint = ins.objectid + ins.offset;
900 start += cur_alloc_size;
b888db2b 901 }
b888db2b 902out:
771ed689 903 ret = 0;
b888db2b 904 btrfs_end_transaction(trans, root);
c8b97818 905
be20aa9d 906 return ret;
771ed689 907}
c8b97818 908
771ed689
CM		 909/*
910 * work queue call back to started compression on a file and pages
911 */
912static noinline void async_cow_start(struct btrfs_work *work)
913{
914 struct async_cow *async_cow;
915 int num_added = 0;
916 async_cow = container_of(work, struct async_cow, work);
917
918 compress_file_range(async_cow->inode, async_cow->locked_page,
919 async_cow->start, async_cow->end, async_cow,
920 &num_added);
921 if (num_added == 0)
922 async_cow->inode = NULL;
923}
924
925/*
926 * work queue call back to submit previously compressed pages
927 */
928static noinline void async_cow_submit(struct btrfs_work *work)
929{
930 struct async_cow *async_cow;
931 struct btrfs_root *root;
932 unsigned long nr_pages;
933
934 async_cow = container_of(work, struct async_cow, work);
935
936 root = async_cow->root;
937 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
938 PAGE_CACHE_SHIFT;
939
940 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
941
942 if (atomic_read(&root->fs_info->async_delalloc_pages) <
943 5 * 1042 * 1024 &&
944 waitqueue_active(&root->fs_info->async_submit_wait))
945 wake_up(&root->fs_info->async_submit_wait);
946
d397712b 947 if (async_cow->inode)
771ed689 948 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 949}
c8b97818 950
771ed689
CM		 951static noinline void async_cow_free(struct btrfs_work *work)
952{
953 struct async_cow *async_cow;
954 async_cow = container_of(work, struct async_cow, work);
955 kfree(async_cow);
956}
957
958static int cow_file_range_async(struct inode *inode, struct page *locked_page,
959 u64 start, u64 end, int *page_started,
960 unsigned long *nr_written)
961{
962 struct async_cow *async_cow;
963 struct btrfs_root *root = BTRFS_I(inode)->root;
964 unsigned long nr_pages;
965 u64 cur_end;
966 int limit = 10 * 1024 * 1042;
967
a3429ab7
CM		 968 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
969 1, 0, NULL, GFP_NOFS);
d397712b 970 while (start < end) {
771ed689 971 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
8d413713 972 BUG_ON(!async_cow);
771ed689
CM		 973 async_cow->inode = inode;
974 async_cow->root = root;
975 async_cow->locked_page = locked_page;
976 async_cow->start = start;
977
6cbff00f 978 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 979 cur_end = end;
980 else
981 cur_end = min(end, start + 512 * 1024 - 1);
982
983 async_cow->end = cur_end;
984 INIT_LIST_HEAD(&async_cow->extents);
985
986 async_cow->work.func = async_cow_start;
987 async_cow->work.ordered_func = async_cow_submit;
988 async_cow->work.ordered_free = async_cow_free;
989 async_cow->work.flags = 0;
990
771ed689
CM		 991 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
992 PAGE_CACHE_SHIFT;
993 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
994
995 btrfs_queue_worker(&root->fs_info->delalloc_workers,
996 &async_cow->work);
997
998 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
999 wait_event(root->fs_info->async_submit_wait,
1000 (atomic_read(&root->fs_info->async_delalloc_pages) <
1001 limit));
1002 }
1003
d397712b 1004 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1005 atomic_read(&root->fs_info->async_delalloc_pages)) {
1006 wait_event(root->fs_info->async_submit_wait,
1007 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008 0));
1009 }
1010
1011 *nr_written += nr_pages;
1012 start = cur_end + 1;
1013 }
1014 *page_started = 1;
1015 return 0;
be20aa9d
CM		 1016}
1017
d397712b 1018static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1019 u64 bytenr, u64 num_bytes)
1020{
1021 int ret;
1022 struct btrfs_ordered_sum *sums;
1023 LIST_HEAD(list);
1024
07d400a6 1025 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1026 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1027 if (ret == 0 && list_empty(&list))
1028 return 0;
1029
1030 while (!list_empty(&list)) {
1031 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1032 list_del(&sums->list);
1033 kfree(sums);
1034 }
1035 return 1;
1036}
1037
d352ac68
CM		 1038/*
1039 * when nowcow writeback call back. This checks for snapshots or COW copies
1040 * of the extents that exist in the file, and COWs the file as required.
1041 *
1042 * If no cow copies or snapshots exist, we write directly to the existing
1043 * blocks on disk
1044 */
7f366cfe
CM		 1045static noinline int run_delalloc_nocow(struct inode *inode,
1046 struct page *locked_page,
771ed689
CM		 1047 u64 start, u64 end, int *page_started, int force,
1048 unsigned long *nr_written)
be20aa9d 1049{
be20aa9d 1050 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1051 struct btrfs_trans_handle *trans;
be20aa9d 1052 struct extent_buffer *leaf;
be20aa9d 1053 struct btrfs_path *path;
80ff3856 1054 struct btrfs_file_extent_item *fi;
be20aa9d 1055 struct btrfs_key found_key;
80ff3856
YZ		 1056 u64 cow_start;
1057 u64 cur_offset;
1058 u64 extent_end;
5d4f98a2 1059 u64 extent_offset;
80ff3856
YZ		 1060 u64 disk_bytenr;
1061 u64 num_bytes;
1062 int extent_type;
1063 int ret;
d899e052 1064 int type;
80ff3856
YZ		 1065 int nocow;
1066 int check_prev = 1;
82d5902d 1067 bool nolock;
33345d01 1068 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1069
1070 path = btrfs_alloc_path();
d8926bb3
MF		 1071 if (!path)
1072 return -ENOMEM;
82d5902d 1073
2cf8572d 1074 nolock = btrfs_is_free_space_inode(root, inode);
82d5902d
LZ		 1075
1076 if (nolock)
7a7eaa40 1077 trans = btrfs_join_transaction_nolock(root);
82d5902d 1078 else
7a7eaa40 1079 trans = btrfs_join_transaction(root);
ff5714cc 1080
3612b495 1081 BUG_ON(IS_ERR(trans));
74b21075 1082 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1083
80ff3856
YZ		 1084 cow_start = (u64)-1;
1085 cur_offset = start;
1086 while (1) {
33345d01 1087 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856
YZ		 1088 cur_offset, 0);
1089 BUG_ON(ret < 0);
1090 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1091 leaf = path->nodes[0];
1092 btrfs_item_key_to_cpu(leaf, &found_key,
1093 path->slots[0] - 1);
33345d01 1094 if (found_key.objectid == ino &&
80ff3856
YZ		 1095 found_key.type == BTRFS_EXTENT_DATA_KEY)
1096 path->slots[0]--;
1097 }
1098 check_prev = 0;
1099next_slot:
1100 leaf = path->nodes[0];
1101 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1102 ret = btrfs_next_leaf(root, path);
1103 if (ret < 0)
1104 BUG_ON(1);
1105 if (ret > 0)
1106 break;
1107 leaf = path->nodes[0];
1108 }
be20aa9d 1109
80ff3856
YZ		 1110 nocow = 0;
1111 disk_bytenr = 0;
17d217fe 1112 num_bytes = 0;
80ff3856
YZ		 1113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114
33345d01 1115 if (found_key.objectid > ino ||
80ff3856
YZ		 1116 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1117 found_key.offset > end)
1118 break;
1119
1120 if (found_key.offset > cur_offset) {
1121 extent_end = found_key.offset;
e9061e21 1122 extent_type = 0;
80ff3856
YZ		 1123 goto out_check;
1124 }
1125
1126 fi = btrfs_item_ptr(leaf, path->slots[0],
1127 struct btrfs_file_extent_item);
1128 extent_type = btrfs_file_extent_type(leaf, fi);
1129
d899e052
YZ		 1130 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1131 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1132 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1133 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1134 extent_end = found_key.offset +
1135 btrfs_file_extent_num_bytes(leaf, fi);
1136 if (extent_end <= start) {
1137 path->slots[0]++;
1138 goto next_slot;
1139 }
17d217fe
YZ		 1140 if (disk_bytenr == 0)
1141 goto out_check;
80ff3856
YZ		 1142 if (btrfs_file_extent_compression(leaf, fi) ||
1143 btrfs_file_extent_encryption(leaf, fi) ||
1144 btrfs_file_extent_other_encoding(leaf, fi))
1145 goto out_check;
d899e052
YZ		 1146 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147 goto out_check;
d2fb3437 1148 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1149 goto out_check;
33345d01 1150 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1151 found_key.offset -
1152 extent_offset, disk_bytenr))
17d217fe 1153 goto out_check;
5d4f98a2 1154 disk_bytenr += extent_offset;
17d217fe
YZ		 1155 disk_bytenr += cur_offset - found_key.offset;
1156 num_bytes = min(end + 1, extent_end) - cur_offset;
1157 /*
1158 * force cow if csum exists in the range.
1159 * this ensure that csum for a given extent are
1160 * either valid or do not exist.
1161 */
1162 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1163 goto out_check;
80ff3856
YZ		 1164 nocow = 1;
1165 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1166 extent_end = found_key.offset +
1167 btrfs_file_extent_inline_len(leaf, fi);
1168 extent_end = ALIGN(extent_end, root->sectorsize);
1169 } else {
1170 BUG_ON(1);
1171 }
1172out_check:
1173 if (extent_end <= start) {
1174 path->slots[0]++;
1175 goto next_slot;
1176 }
1177 if (!nocow) {
1178 if (cow_start == (u64)-1)
1179 cow_start = cur_offset;
1180 cur_offset = extent_end;
1181 if (cur_offset > end)
1182 break;
1183 path->slots[0]++;
1184 goto next_slot;
7ea394f1
YZ		 1185 }
1186
b3b4aa74 1187 btrfs_release_path(path);
80ff3856
YZ		 1188 if (cow_start != (u64)-1) {
1189 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1190 found_key.offset - 1, page_started,
1191 nr_written, 1);
80ff3856
YZ		 1192 BUG_ON(ret);
1193 cow_start = (u64)-1;
7ea394f1 1194 }
80ff3856 1195
d899e052
YZ		 1196 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1197 struct extent_map *em;
1198 struct extent_map_tree *em_tree;
1199 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1200 em = alloc_extent_map();
c26a9203 1201 BUG_ON(!em);
d899e052 1202 em->start = cur_offset;
445a6944 1203 em->orig_start = em->start;
d899e052
YZ		 1204 em->len = num_bytes;
1205 em->block_len = num_bytes;
1206 em->block_start = disk_bytenr;
1207 em->bdev = root->fs_info->fs_devices->latest_bdev;
1208 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209 while (1) {
890871be 1210 write_lock(&em_tree->lock);
d899e052 1211 ret = add_extent_mapping(em_tree, em);
890871be 1212 write_unlock(&em_tree->lock);
d899e052
YZ		 1213 if (ret != -EEXIST) {
1214 free_extent_map(em);
1215 break;
1216 }
1217 btrfs_drop_extent_cache(inode, em->start,
1218 em->start + em->len - 1, 0);
1219 }
1220 type = BTRFS_ORDERED_PREALLOC;
1221 } else {
1222 type = BTRFS_ORDERED_NOCOW;
1223 }
80ff3856
YZ		 1224
1225 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1226 num_bytes, num_bytes, type);
1227 BUG_ON(ret);
771ed689 1228
efa56464
YZ		 1229 if (root->root_key.objectid ==
1230 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1231 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1232 num_bytes);
1233 BUG_ON(ret);
1234 }
1235
d899e052 1236 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1237 cur_offset, cur_offset + num_bytes - 1,
1238 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1239 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1240 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1241 cur_offset = extent_end;
1242 if (cur_offset > end)
1243 break;
be20aa9d 1244 }
b3b4aa74 1245 btrfs_release_path(path);
80ff3856
YZ		 1246
1247 if (cur_offset <= end && cow_start == (u64)-1)
1248 cow_start = cur_offset;
1249 if (cow_start != (u64)-1) {
1250 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1251 page_started, nr_written, 1);
80ff3856
YZ		 1252 BUG_ON(ret);
1253 }
1254
0cb59c99
JB		 1255 if (nolock) {
1256 ret = btrfs_end_transaction_nolock(trans, root);
1257 BUG_ON(ret);
1258 } else {
1259 ret = btrfs_end_transaction(trans, root);
1260 BUG_ON(ret);
1261 }
7ea394f1 1262 btrfs_free_path(path);
80ff3856 1263 return 0;
be20aa9d
CM		 1264}
1265
d352ac68
CM		 1266/*
1267 * extent_io.c call back to do delayed allocation processing
1268 */
c8b97818 1269static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1270 u64 start, u64 end, int *page_started,
1271 unsigned long *nr_written)
be20aa9d 1272{
be20aa9d 1273 int ret;
7f366cfe 1274 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1275
6cbff00f 1276 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
c8b97818 1277 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1278 page_started, 1, nr_written);
6cbff00f 1279 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
d899e052 1280 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1281 page_started, 0, nr_written);
1e701a32 1282 else if (!btrfs_test_opt(root, COMPRESS) &&
75e7cb7f
LB		 1283 !(BTRFS_I(inode)->force_compress) &&
1284 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
7f366cfe
CM		 1285 ret = cow_file_range(inode, locked_page, start, end,
1286 page_started, nr_written, 1);
be20aa9d 1287 else
771ed689 1288 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1289 page_started, nr_written);
b888db2b
CM		 1290 return ret;
1291}
1292
1bf85046
JM		 1293static void btrfs_split_extent_hook(struct inode *inode,
1294 struct extent_state *orig, u64 split)
9ed74f2d 1295{
0ca1f7ce 1296 /* not delalloc, ignore it */
9ed74f2d 1297 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1298 return;
9ed74f2d 1299
9e0baf60
JB		 1300 spin_lock(&BTRFS_I(inode)->lock);
1301 BTRFS_I(inode)->outstanding_extents++;
1302 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1303}
1304
1305/*
1306 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307 * extents so we can keep track of new extents that are just merged onto old
1308 * extents, such as when we are doing sequential writes, so we can properly
1309 * account for the metadata space we'll need.
1310 */
1bf85046
JM		 1311static void btrfs_merge_extent_hook(struct inode *inode,
1312 struct extent_state *new,
1313 struct extent_state *other)
9ed74f2d 1314{
9ed74f2d
JB		 1315 /* not delalloc, ignore it */
1316 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1317 return;
9ed74f2d 1318
9e0baf60
JB		 1319 spin_lock(&BTRFS_I(inode)->lock);
1320 BTRFS_I(inode)->outstanding_extents--;
1321 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1322}
1323
d352ac68
CM		 1324/*
1325 * extent_io.c set_bit_hook, used to track delayed allocation
1326 * bytes in this file, and to maintain the list of inodes that
1327 * have pending delalloc work to be done.
1328 */
1bf85046
JM		 1329static void btrfs_set_bit_hook(struct inode *inode,
1330 struct extent_state *state, int *bits)
291d673e 1331{
9ed74f2d 1332
75eff68e
CM		 1333 /*
1334 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1335 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1336 * bit, which is only set or cleared with irqs on
1337 */
0ca1f7ce 1338 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1339 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1340 u64 len = state->end + 1 - state->start;
2cf8572d 1341 bool do_list = !btrfs_is_free_space_inode(root, inode);
9ed74f2d 1342
9e0baf60 1343 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1344 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1345 } else {
1346 spin_lock(&BTRFS_I(inode)->lock);
1347 BTRFS_I(inode)->outstanding_extents++;
1348 spin_unlock(&BTRFS_I(inode)->lock);
1349 }
287a0ab9 1350
75eff68e 1351 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1352 BTRFS_I(inode)->delalloc_bytes += len;
1353 root->fs_info->delalloc_bytes += len;
0cb59c99 1354 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
ea8c2819
CM		 1355 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1356 &root->fs_info->delalloc_inodes);
1357 }
75eff68e 1358 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1359 }
291d673e
CM		 1360}
1361
d352ac68
CM		 1362/*
1363 * extent_io.c clear_bit_hook, see set_bit_hook for why
1364 */
1bf85046
JM		 1365static void btrfs_clear_bit_hook(struct inode *inode,
1366 struct extent_state *state, int *bits)
291d673e 1367{
75eff68e
CM		 1368 /*
1369 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1370 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1371 * bit, which is only set or cleared with irqs on
1372 */
0ca1f7ce 1373 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1374 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1375 u64 len = state->end + 1 - state->start;
2cf8572d 1376 bool do_list = !btrfs_is_free_space_inode(root, inode);
bcbfce8a 1377
9e0baf60 1378 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1379 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1380 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1381 spin_lock(&BTRFS_I(inode)->lock);
1382 BTRFS_I(inode)->outstanding_extents--;
1383 spin_unlock(&BTRFS_I(inode)->lock);
1384 }
0ca1f7ce
YZ		 1385
1386 if (*bits & EXTENT_DO_ACCOUNTING)
1387 btrfs_delalloc_release_metadata(inode, len);
1388
0cb59c99
JB		 1389 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390 && do_list)
0ca1f7ce 1391 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1392
75eff68e 1393 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1394 root->fs_info->delalloc_bytes -= len;
1395 BTRFS_I(inode)->delalloc_bytes -= len;
1396
0cb59c99 1397 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
ea8c2819
CM		 1398 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1399 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400 }
75eff68e 1401 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1402 }
291d673e
CM		 1403}
1404
d352ac68
CM		 1405/*
1406 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407 * we don't create bios that span stripes or chunks
1408 */
239b14b3 1409int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1410 size_t size, struct bio *bio,
1411 unsigned long bio_flags)
239b14b3
CM		 1412{
1413 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1414 struct btrfs_mapping_tree *map_tree;
a62b9401 1415 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1416 u64 length = 0;
1417 u64 map_length;
239b14b3
CM		 1418 int ret;
1419
771ed689
CM		 1420 if (bio_flags & EXTENT_BIO_COMPRESSED)
1421 return 0;
1422
f2d8d74d 1423 length = bio->bi_size;
239b14b3
CM		 1424 map_tree = &root->fs_info->mapping_tree;
1425 map_length = length;
cea9e445 1426 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1427 &map_length, NULL, 0);
3444a972
JM		 1428 /* Will always return 0 or 1 with map_multi == NULL */
1429 BUG_ON(ret < 0);
d397712b 1430 if (map_length < length + size)
239b14b3 1431 return 1;
3444a972 1432 return 0;
239b14b3
CM		 1433}
1434
d352ac68
CM		 1435/*
1436 * in order to insert checksums into the metadata in large chunks,
1437 * we wait until bio submission time. All the pages in the bio are
1438 * checksummed and sums are attached onto the ordered extent record.
1439 *
1440 * At IO completion time the cums attached on the ordered extent record
1441 * are inserted into the btree
1442 */
d397712b
CM		 1443static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1444 struct bio *bio, int mirror_num,
eaf25d93
CM		 1445 unsigned long bio_flags,
1446 u64 bio_offset)
065631f6 1447{
065631f6 1448 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1449 int ret = 0;
e015640f 1450
d20f7043 1451 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1452 BUG_ON(ret);
4a69a410
CM		 1453 return 0;
1454}
e015640f 1455
4a69a410
CM		 1456/*
1457 * in order to insert checksums into the metadata in large chunks,
1458 * we wait until bio submission time. All the pages in the bio are
1459 * checksummed and sums are attached onto the ordered extent record.
1460 *
1461 * At IO completion time the cums attached on the ordered extent record
1462 * are inserted into the btree
1463 */
b2950863 1464static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1465 int mirror_num, unsigned long bio_flags,
1466 u64 bio_offset)
4a69a410
CM		 1467{
1468 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1469 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1470}
1471
d352ac68 1472/*
cad321ad
CM		 1473 * extent_io.c submission hook. This does the right thing for csum calculation
1474 * on write, or reading the csums from the tree before a read
d352ac68 1475 */
b2950863 1476static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1477 int mirror_num, unsigned long bio_flags,
1478 u64 bio_offset)
44b8bd7e
CM		 1479{
1480 struct btrfs_root *root = BTRFS_I(inode)->root;
1481 int ret = 0;
19b9bdb0 1482 int skip_sum;
0417341e 1483 int metadata = 0;
44b8bd7e 1484
6cbff00f 1485 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1486
2cf8572d 1487 if (btrfs_is_free_space_inode(root, inode))
0417341e
JM		 1488 metadata = 2;
1489
1490 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
355808c2
JM		 1491 if (ret)
1492 return ret;
065631f6 1493
7b6d91da 1494 if (!(rw & REQ_WRITE)) {
d20f7043 1495 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1496 return btrfs_submit_compressed_read(inode, bio,
1497 mirror_num, bio_flags);
c2db1073
TI		 1498 } else if (!skip_sum) {
1499 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1500 if (ret)
1501 return ret;
1502 }
4d1b5fb4 1503 goto mapit;
19b9bdb0 1504 } else if (!skip_sum) {
17d217fe
YZ		 1505 /* csum items have already been cloned */
1506 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1507 goto mapit;
19b9bdb0
CM		 1508 /* we're doing a write, do the async checksumming */
1509 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1510 inode, rw, bio, mirror_num,
eaf25d93
CM		 1511 bio_flags, bio_offset,
1512 __btrfs_submit_bio_start,
4a69a410 1513 __btrfs_submit_bio_done);
19b9bdb0
CM		 1514 }
1515
0b86a832 1516mapit:
8b712842 1517 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1518}
6885f308 1519
d352ac68
CM		 1520/*
1521 * given a list of ordered sums record them in the inode. This happens
1522 * at IO completion time based on sums calculated at bio submission time.
1523 */
ba1da2f4 1524static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1525 struct inode *inode, u64 file_offset,
1526 struct list_head *list)
1527{
e6dcd2dc
CM		 1528 struct btrfs_ordered_sum *sum;
1529
c6e30871 1530 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1531 btrfs_csum_file_blocks(trans,
1532 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1533 }
1534 return 0;
1535}
1536
2ac55d41
JB		 1537int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1538 struct extent_state **cached_state)
ea8c2819 1539{
d397712b 1540 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1541 WARN_ON(1);
ea8c2819 1542 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1543 cached_state, GFP_NOFS);
ea8c2819
CM		 1544}
1545
d352ac68 1546/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1547struct btrfs_writepage_fixup {
1548 struct page *page;
1549 struct btrfs_work work;
1550};
1551
b2950863 1552static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1553{
1554 struct btrfs_writepage_fixup *fixup;
1555 struct btrfs_ordered_extent *ordered;
2ac55d41 1556 struct extent_state *cached_state = NULL;
247e743c
CM		 1557 struct page *page;
1558 struct inode *inode;
1559 u64 page_start;
1560 u64 page_end;
87826df0 1561 int ret;
247e743c
CM		 1562
1563 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1564 page = fixup->page;
4a096752 1565again:
247e743c
CM		 1566 lock_page(page);
1567 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1568 ClearPageChecked(page);
1569 goto out_page;
1570 }
1571
1572 inode = page->mapping->host;
1573 page_start = page_offset(page);
1574 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1575
2ac55d41
JB		 1576 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1577 &cached_state, GFP_NOFS);
4a096752
CM		 1578
1579 /* already ordered? We're done */
8b62b72b 1580 if (PagePrivate2(page))
247e743c 1581 goto out;
4a096752
CM		 1582
1583 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1584 if (ordered) {
2ac55d41
JB		 1585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1586 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1587 unlock_page(page);
1588 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1589 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1590 goto again;
1591 }
247e743c 1592
87826df0
JM		 1593 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1594 if (ret) {
1595 mapping_set_error(page->mapping, ret);
1596 end_extent_writepage(page, ret, page_start, page_end);
1597 ClearPageChecked(page);
1598 goto out;
1599 }
1600
2ac55d41 1601 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1602 ClearPageChecked(page);
87826df0 1603 set_page_dirty(page);
247e743c 1604out:
2ac55d41
JB		 1605 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1606 &cached_state, GFP_NOFS);
247e743c
CM		 1607out_page:
1608 unlock_page(page);
1609 page_cache_release(page);
b897abec 1610 kfree(fixup);
247e743c
CM		 1611}
1612
1613/*
1614 * There are a few paths in the higher layers of the kernel that directly
1615 * set the page dirty bit without asking the filesystem if it is a
1616 * good idea. This causes problems because we want to make sure COW
1617 * properly happens and the data=ordered rules are followed.
1618 *
c8b97818 1619 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1620 * hasn't been properly setup for IO. We kick off an async process
1621 * to fix it up. The async helper will wait for ordered extents, set
1622 * the delalloc bit and make it safe to write the page.
1623 */
b2950863 1624static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1625{
1626 struct inode *inode = page->mapping->host;
1627 struct btrfs_writepage_fixup *fixup;
1628 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1629
8b62b72b
CM		 1630 /* this page is properly in the ordered list */
1631 if (TestClearPagePrivate2(page))
247e743c
CM		 1632 return 0;
1633
1634 if (PageChecked(page))
1635 return -EAGAIN;
1636
1637 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1638 if (!fixup)
1639 return -EAGAIN;
f421950f 1640
247e743c
CM		 1641 SetPageChecked(page);
1642 page_cache_get(page);
1643 fixup->work.func = btrfs_writepage_fixup_worker;
1644 fixup->page = page;
1645 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1646 return -EBUSY;
247e743c
CM		 1647}
1648
d899e052
YZ		 1649static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1650 struct inode *inode, u64 file_pos,
1651 u64 disk_bytenr, u64 disk_num_bytes,
1652 u64 num_bytes, u64 ram_bytes,
1653 u8 compression, u8 encryption,
1654 u16 other_encoding, int extent_type)
1655{
1656 struct btrfs_root *root = BTRFS_I(inode)->root;
1657 struct btrfs_file_extent_item *fi;
1658 struct btrfs_path *path;
1659 struct extent_buffer *leaf;
1660 struct btrfs_key ins;
1661 u64 hint;
1662 int ret;
1663
1664 path = btrfs_alloc_path();
d8926bb3
MF		 1665 if (!path)
1666 return -ENOMEM;
d899e052 1667
b9473439 1668 path->leave_spinning = 1;
a1ed835e
CM		 1669
1670 /*
1671 * we may be replacing one extent in the tree with another.
1672 * The new extent is pinned in the extent map, and we don't want
1673 * to drop it from the cache until it is completely in the btree.
1674 *
1675 * So, tell btrfs_drop_extents to leave this extent in the cache.
1676 * the caller is expected to unpin it and allow it to be merged
1677 * with the others.
1678 */
920bbbfb
YZ		 1679 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1680 &hint, 0);
d899e052
YZ		 1681 BUG_ON(ret);
1682
33345d01 1683 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1684 ins.offset = file_pos;
1685 ins.type = BTRFS_EXTENT_DATA_KEY;
1686 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1687 BUG_ON(ret);
1688 leaf = path->nodes[0];
1689 fi = btrfs_item_ptr(leaf, path->slots[0],
1690 struct btrfs_file_extent_item);
1691 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1692 btrfs_set_file_extent_type(leaf, fi, extent_type);
1693 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1694 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1695 btrfs_set_file_extent_offset(leaf, fi, 0);
1696 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1697 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1698 btrfs_set_file_extent_compression(leaf, fi, compression);
1699 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1700 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1701
1702 btrfs_unlock_up_safe(path, 1);
1703 btrfs_set_lock_blocking(leaf);
1704
d899e052
YZ		 1705 btrfs_mark_buffer_dirty(leaf);
1706
1707 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1708
1709 ins.objectid = disk_bytenr;
1710 ins.offset = disk_num_bytes;
1711 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1712 ret = btrfs_alloc_reserved_file_extent(trans, root,
1713 root->root_key.objectid,
33345d01 1714 btrfs_ino(inode), file_pos, &ins);
d899e052 1715 BUG_ON(ret);
d899e052 1716 btrfs_free_path(path);
b9473439 1717
d899e052
YZ		 1718 return 0;
1719}
1720
5d13a98f
CM		 1721/*
1722 * helper function for btrfs_finish_ordered_io, this
1723 * just reads in some of the csum leaves to prime them into ram
1724 * before we start the transaction. It limits the amount of btree
1725 * reads required while inside the transaction.
1726 */
d352ac68
CM		 1727/* as ordered data IO finishes, this gets called so we can finish
1728 * an ordered extent if the range of bytes in the file it covers are
1729 * fully written.
1730 */
211f90e6 1731static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1732{
e6dcd2dc 1733 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1734 struct btrfs_trans_handle *trans = NULL;
5d13a98f 1735 struct btrfs_ordered_extent *ordered_extent = NULL;
e6dcd2dc 1736 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1737 struct extent_state *cached_state = NULL;
261507a0 1738 int compress_type = 0;
e6dcd2dc 1739 int ret;
82d5902d 1740 bool nolock;
e6dcd2dc 1741
5a1a3df1
JB		 1742 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1743 end - start + 1);
ba1da2f4 1744 if (!ret)
e6dcd2dc 1745 return 0;
e6dcd2dc 1746 BUG_ON(!ordered_extent);
efd049fb 1747
2cf8572d 1748 nolock = btrfs_is_free_space_inode(root, inode);
0cb59c99 1749
c2167754
YZ		 1750 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1751 BUG_ON(!list_empty(&ordered_extent->list));
1752 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1753 if (!ret) {
0cb59c99 1754 if (nolock)
7a7eaa40 1755 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1756 else
7a7eaa40 1757 trans = btrfs_join_transaction(root);
3612b495 1758 BUG_ON(IS_ERR(trans));
0ca1f7ce 1759 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2115133f 1760 ret = btrfs_update_inode_fallback(trans, root, inode);
c2167754 1761 BUG_ON(ret);
c2167754
YZ		 1762 }
1763 goto out;
1764 }
e6dcd2dc 1765
2ac55d41
JB		 1766 lock_extent_bits(io_tree, ordered_extent->file_offset,
1767 ordered_extent->file_offset + ordered_extent->len - 1,
1768 0, &cached_state, GFP_NOFS);
e6dcd2dc 1769
0cb59c99 1770 if (nolock)
7a7eaa40 1771 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1772 else
7a7eaa40 1773 trans = btrfs_join_transaction(root);
3612b495 1774 BUG_ON(IS_ERR(trans));
0ca1f7ce 1775 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1776
c8b97818 1777 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1778 compress_type = ordered_extent->compress_type;
d899e052 1779 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1780 BUG_ON(compress_type);
920bbbfb 1781 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1782 ordered_extent->file_offset,
1783 ordered_extent->file_offset +
1784 ordered_extent->len);
1785 BUG_ON(ret);
1786 } else {
0af3d00b 1787 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1788 ret = insert_reserved_file_extent(trans, inode,
1789 ordered_extent->file_offset,
1790 ordered_extent->start,
1791 ordered_extent->disk_len,
1792 ordered_extent->len,
1793 ordered_extent->len,
261507a0 1794 compress_type, 0, 0,
d899e052 1795 BTRFS_FILE_EXTENT_REG);
a1ed835e
CM		 1796 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1797 ordered_extent->file_offset,
1798 ordered_extent->len);
d899e052
YZ		 1799 BUG_ON(ret);
1800 }
2ac55d41
JB		 1801 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1802 ordered_extent->file_offset +
1803 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1804
e6dcd2dc
CM		 1805 add_pending_csums(trans, inode, ordered_extent->file_offset,
1806 &ordered_extent->list);
1807
1ef30be1 1808 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1809 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2115133f 1810 ret = btrfs_update_inode_fallback(trans, root, inode);
1ef30be1
JB		 1811 BUG_ON(ret);
1812 }
1813 ret = 0;
c2167754 1814out:
5b0e95bf 1815 if (root != root->fs_info->tree_root)
0cb59c99 1816 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
5b0e95bf
JB		 1817 if (trans) {
1818 if (nolock)
0cb59c99 1819 btrfs_end_transaction_nolock(trans, root);
5b0e95bf 1820 else
0cb59c99
JB		 1821 btrfs_end_transaction(trans, root);
1822 }
1823
e6dcd2dc
CM		 1824 /* once for us */
1825 btrfs_put_ordered_extent(ordered_extent);
1826 /* once for the tree */
1827 btrfs_put_ordered_extent(ordered_extent);
1828
e6dcd2dc
CM		 1829 return 0;
1830}
1831
b2950863 1832static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1833 struct extent_state *state, int uptodate)
1834{
1abe9b8a 1835 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1836
8b62b72b 1837 ClearPagePrivate2(page);
211f90e6
CM		 1838 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1839}
1840
d352ac68
CM		 1841/*
1842 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 1843 * if there's a match, we allow the bio to finish. If not, the code in
1844 * extent_io.c will try to find good copies for us.
d352ac68 1845 */
b2950863 1846static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1847 struct extent_state *state)
07157aac 1848{
35ebb934 1849 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1850 struct inode *inode = page->mapping->host;
d1310b2e 1851 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1852 char *kaddr;
aadfeb6e 1853 u64 private = ~(u32)0;
07157aac 1854 int ret;
ff79f819
CM		 1855 struct btrfs_root *root = BTRFS_I(inode)->root;
1856 u32 csum = ~(u32)0;
d1310b2e 1857
d20f7043
CM		 1858 if (PageChecked(page)) {
1859 ClearPageChecked(page);
1860 goto good;
1861 }
6cbff00f
CH		 1862
1863 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 1864 goto good;
17d217fe
YZ		 1865
1866 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 1867 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 1868 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1869 GFP_NOFS);
b6cda9bc 1870 return 0;
17d217fe 1871 }
d20f7043 1872
c2e639f0 1873 if (state && state->start == start) {
70dec807
CM		 1874 private = state->private;
1875 ret = 0;
1876 } else {
1877 ret = get_state_private(io_tree, start, &private);
1878 }
9ab86c8e 1879 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1880 if (ret)
07157aac 1881 goto zeroit;
d397712b 1882
ff79f819
CM		 1883 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1884 btrfs_csum_final(csum, (char *)&csum);
d397712b 1885 if (csum != private)
07157aac 1886 goto zeroit;
d397712b 1887
9ab86c8e 1888 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1889good:
07157aac
CM		 1890 return 0;
1891
1892zeroit:
945d8962 1893 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 1894 "private %llu\n",
1895 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 1896 (unsigned long long)start, csum,
1897 (unsigned long long)private);
db94535d
CM		 1898 memset(kaddr + offset, 1, end - start + 1);
1899 flush_dcache_page(page);
9ab86c8e 1900 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1901 if (private == 0)
1902 return 0;
7e38326f 1903 return -EIO;
07157aac 1904}
b888db2b 1905
24bbcf04
YZ		 1906struct delayed_iput {
1907 struct list_head list;
1908 struct inode *inode;
1909};
1910
1911void btrfs_add_delayed_iput(struct inode *inode)
1912{
1913 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1914 struct delayed_iput *delayed;
1915
1916 if (atomic_add_unless(&inode->i_count, -1, 1))
1917 return;
1918
1919 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1920 delayed->inode = inode;
1921
1922 spin_lock(&fs_info->delayed_iput_lock);
1923 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1924 spin_unlock(&fs_info->delayed_iput_lock);
1925}
1926
1927void btrfs_run_delayed_iputs(struct btrfs_root *root)
1928{
1929 LIST_HEAD(list);
1930 struct btrfs_fs_info *fs_info = root->fs_info;
1931 struct delayed_iput *delayed;
1932 int empty;
1933
1934 spin_lock(&fs_info->delayed_iput_lock);
1935 empty = list_empty(&fs_info->delayed_iputs);
1936 spin_unlock(&fs_info->delayed_iput_lock);
1937 if (empty)
1938 return;
1939
1940 down_read(&root->fs_info->cleanup_work_sem);
1941 spin_lock(&fs_info->delayed_iput_lock);
1942 list_splice_init(&fs_info->delayed_iputs, &list);
1943 spin_unlock(&fs_info->delayed_iput_lock);
1944
1945 while (!list_empty(&list)) {
1946 delayed = list_entry(list.next, struct delayed_iput, list);
1947 list_del(&delayed->list);
1948 iput(delayed->inode);
1949 kfree(delayed);
1950 }
1951 up_read(&root->fs_info->cleanup_work_sem);
1952}
1953
d68fc57b
YZ		 1954enum btrfs_orphan_cleanup_state {
1955 ORPHAN_CLEANUP_STARTED = 1,
1956 ORPHAN_CLEANUP_DONE = 2,
1957};
1958
1959/*
42b2aa86 1960 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 1961 * files in the subvolume, it removes orphan item and frees block_rsv
1962 * structure.
1963 */
1964void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root)
1966{
90290e19 1967 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 1968 int ret;
1969
1970 if (!list_empty(&root->orphan_list) ||
1971 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1972 return;
1973
90290e19
JB		 1974 spin_lock(&root->orphan_lock);
1975 if (!list_empty(&root->orphan_list)) {
1976 spin_unlock(&root->orphan_lock);
1977 return;
1978 }
1979
1980 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1981 spin_unlock(&root->orphan_lock);
1982 return;
1983 }
1984
1985 block_rsv = root->orphan_block_rsv;
1986 root->orphan_block_rsv = NULL;
1987 spin_unlock(&root->orphan_lock);
1988
d68fc57b
YZ		 1989 if (root->orphan_item_inserted &&
1990 btrfs_root_refs(&root->root_item) > 0) {
1991 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1992 root->root_key.objectid);
1993 BUG_ON(ret);
1994 root->orphan_item_inserted = 0;
1995 }
1996
90290e19
JB		 1997 if (block_rsv) {
1998 WARN_ON(block_rsv->size > 0);
1999 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2000 }
2001}
2002
7b128766
JB		 2003/*
2004 * This creates an orphan entry for the given inode in case something goes
2005 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2006 *
2007 * NOTE: caller of this function should reserve 5 units of metadata for
2008 * this function.
7b128766
JB		 2009 */
2010int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2011{
2012 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2013 struct btrfs_block_rsv *block_rsv = NULL;
2014 int reserve = 0;
2015 int insert = 0;
2016 int ret;
7b128766 2017
d68fc57b
YZ		 2018 if (!root->orphan_block_rsv) {
2019 block_rsv = btrfs_alloc_block_rsv(root);
b532402e
TI		 2020 if (!block_rsv)
2021 return -ENOMEM;
d68fc57b 2022 }
7b128766 2023
d68fc57b
YZ		 2024 spin_lock(&root->orphan_lock);
2025 if (!root->orphan_block_rsv) {
2026 root->orphan_block_rsv = block_rsv;
2027 } else if (block_rsv) {
2028 btrfs_free_block_rsv(root, block_rsv);
2029 block_rsv = NULL;
7b128766 2030 }
7b128766 2031
d68fc57b
YZ		 2032 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2033 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2034#if 0
2035 /*
2036 * For proper ENOSPC handling, we should do orphan
2037 * cleanup when mounting. But this introduces backward
2038 * compatibility issue.
2039 */
2040 if (!xchg(&root->orphan_item_inserted, 1))
2041 insert = 2;
2042 else
2043 insert = 1;
2044#endif
2045 insert = 1;
7b128766
JB		 2046 }
2047
d68fc57b
YZ		 2048 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2049 BTRFS_I(inode)->orphan_meta_reserved = 1;
2050 reserve = 1;
2051 }
2052 spin_unlock(&root->orphan_lock);
7b128766 2053
d68fc57b
YZ		 2054 /* grab metadata reservation from transaction handle */
2055 if (reserve) {
2056 ret = btrfs_orphan_reserve_metadata(trans, inode);
2057 BUG_ON(ret);
2058 }
7b128766 2059
d68fc57b
YZ		 2060 /* insert an orphan item to track this unlinked/truncated file */
2061 if (insert >= 1) {
33345d01 2062 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
ee4d89f0 2063 BUG_ON(ret && ret != -EEXIST);
d68fc57b
YZ		 2064 }
2065
2066 /* insert an orphan item to track subvolume contains orphan files */
2067 if (insert >= 2) {
2068 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2069 root->root_key.objectid);
2070 BUG_ON(ret);
2071 }
2072 return 0;
7b128766
JB		 2073}
2074
2075/*
2076 * We have done the truncate/delete so we can go ahead and remove the orphan
2077 * item for this particular inode.
2078 */
2079int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2080{
2081 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2082 int delete_item = 0;
2083 int release_rsv = 0;
7b128766
JB		 2084 int ret = 0;
2085
d68fc57b
YZ		 2086 spin_lock(&root->orphan_lock);
2087 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2088 list_del_init(&BTRFS_I(inode)->i_orphan);
2089 delete_item = 1;
7b128766
JB		 2090 }
2091
d68fc57b
YZ		 2092 if (BTRFS_I(inode)->orphan_meta_reserved) {
2093 BTRFS_I(inode)->orphan_meta_reserved = 0;
2094 release_rsv = 1;
7b128766 2095 }
d68fc57b 2096 spin_unlock(&root->orphan_lock);
7b128766 2097
d68fc57b 2098 if (trans && delete_item) {
33345d01 2099 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2100 BUG_ON(ret);
2101 }
7b128766 2102
d68fc57b
YZ		 2103 if (release_rsv)
2104 btrfs_orphan_release_metadata(inode);
7b128766 2105
d68fc57b 2106 return 0;
7b128766
JB		 2107}
2108
2109/*
2110 * this cleans up any orphans that may be left on the list from the last use
2111 * of this root.
2112 */
66b4ffd1 2113int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2114{
2115 struct btrfs_path *path;
2116 struct extent_buffer *leaf;
7b128766
JB		 2117 struct btrfs_key key, found_key;
2118 struct btrfs_trans_handle *trans;
2119 struct inode *inode;
8f6d7f4f 2120 u64 last_objectid = 0;
7b128766
JB		 2121 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2122
d68fc57b 2123 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2124 return 0;
c71bf099
YZ		 2125
2126 path = btrfs_alloc_path();
66b4ffd1
JB		 2127 if (!path) {
2128 ret = -ENOMEM;
2129 goto out;
2130 }
7b128766
JB		 2131 path->reada = -1;
2132
2133 key.objectid = BTRFS_ORPHAN_OBJECTID;
2134 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2135 key.offset = (u64)-1;
2136
7b128766
JB		 2137 while (1) {
2138 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2139 if (ret < 0)
2140 goto out;
7b128766
JB		 2141
2142 /*
2143 * if ret == 0 means we found what we were searching for, which
25985edc 2144 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2145 * find the key and see if we have stuff that matches
2146 */
2147 if (ret > 0) {
66b4ffd1 2148 ret = 0;
7b128766
JB		 2149 if (path->slots[0] == 0)
2150 break;
2151 path->slots[0]--;
2152 }
2153
2154 /* pull out the item */
2155 leaf = path->nodes[0];
7b128766
JB		 2156 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2157
2158 /* make sure the item matches what we want */
2159 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2160 break;
2161 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2162 break;
2163
2164 /* release the path since we're done with it */
b3b4aa74 2165 btrfs_release_path(path);
7b128766
JB		 2166
2167 /*
2168 * this is where we are basically btrfs_lookup, without the
2169 * crossing root thing. we store the inode number in the
2170 * offset of the orphan item.
2171 */
8f6d7f4f
JB		 2172
2173 if (found_key.offset == last_objectid) {
2174 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2175 "stopping orphan cleanup\n");
2176 ret = -EINVAL;
2177 goto out;
2178 }
2179
2180 last_objectid = found_key.offset;
2181
5d4f98a2
YZ		 2182 found_key.objectid = found_key.offset;
2183 found_key.type = BTRFS_INODE_ITEM_KEY;
2184 found_key.offset = 0;
73f73415 2185 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2186 ret = PTR_RET(inode);
2187 if (ret && ret != -ESTALE)
66b4ffd1 2188 goto out;
7b128766 2189
f8e9e0b0
AJ		 2190 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2191 struct btrfs_root *dead_root;
2192 struct btrfs_fs_info *fs_info = root->fs_info;
2193 int is_dead_root = 0;
2194
2195 /*
2196 * this is an orphan in the tree root. Currently these
2197 * could come from 2 sources:
2198 * a) a snapshot deletion in progress
2199 * b) a free space cache inode
2200 * We need to distinguish those two, as the snapshot
2201 * orphan must not get deleted.
2202 * find_dead_roots already ran before us, so if this
2203 * is a snapshot deletion, we should find the root
2204 * in the dead_roots list
2205 */
2206 spin_lock(&fs_info->trans_lock);
2207 list_for_each_entry(dead_root, &fs_info->dead_roots,
2208 root_list) {
2209 if (dead_root->root_key.objectid ==
2210 found_key.objectid) {
2211 is_dead_root = 1;
2212 break;
2213 }
2214 }
2215 spin_unlock(&fs_info->trans_lock);
2216 if (is_dead_root) {
2217 /* prevent this orphan from being found again */
2218 key.offset = found_key.objectid - 1;
2219 continue;
2220 }
2221 }
7b128766 2222 /*
a8c9e576
JB		 2223 * Inode is already gone but the orphan item is still there,
2224 * kill the orphan item.
7b128766 2225 */
a8c9e576
JB		 2226 if (ret == -ESTALE) {
2227 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2228 if (IS_ERR(trans)) {
2229 ret = PTR_ERR(trans);
2230 goto out;
2231 }
a8c9e576
JB		 2232 ret = btrfs_del_orphan_item(trans, root,
2233 found_key.objectid);
2234 BUG_ON(ret);
5b21f2ed 2235 btrfs_end_transaction(trans, root);
7b128766
JB		 2236 continue;
2237 }
2238
a8c9e576
JB		 2239 /*
2240 * add this inode to the orphan list so btrfs_orphan_del does
2241 * the proper thing when we hit it
2242 */
2243 spin_lock(&root->orphan_lock);
2244 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2245 spin_unlock(&root->orphan_lock);
2246
7b128766
JB		 2247 /* if we have links, this was a truncate, lets do that */
2248 if (inode->i_nlink) {
a41ad394
JB		 2249 if (!S_ISREG(inode->i_mode)) {
2250 WARN_ON(1);
2251 iput(inode);
2252 continue;
2253 }
7b128766 2254 nr_truncate++;
66b4ffd1 2255 ret = btrfs_truncate(inode);
7b128766
JB		 2256 } else {
2257 nr_unlink++;
2258 }
2259
2260 /* this will do delete_inode and everything for us */
2261 iput(inode);
66b4ffd1
JB		 2262 if (ret)
2263 goto out;
7b128766 2264 }
3254c876
MX		 2265 /* release the path since we're done with it */
2266 btrfs_release_path(path);
2267
d68fc57b
YZ		 2268 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2269
2270 if (root->orphan_block_rsv)
2271 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2272 (u64)-1);
2273
2274 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2275 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2276 if (!IS_ERR(trans))
2277 btrfs_end_transaction(trans, root);
d68fc57b 2278 }
7b128766
JB		 2279
2280 if (nr_unlink)
2281 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2282 if (nr_truncate)
2283 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2284
2285out:
2286 if (ret)
2287 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2288 btrfs_free_path(path);
2289 return ret;
7b128766
JB		 2290}
2291
46a53cca
CM		 2292/*
2293 * very simple check to peek ahead in the leaf looking for xattrs. If we
2294 * don't find any xattrs, we know there can't be any acls.
2295 *
2296 * slot is the slot the inode is in, objectid is the objectid of the inode
2297 */
2298static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2299 int slot, u64 objectid)
2300{
2301 u32 nritems = btrfs_header_nritems(leaf);
2302 struct btrfs_key found_key;
2303 int scanned = 0;
2304
2305 slot++;
2306 while (slot < nritems) {
2307 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2308
2309 /* we found a different objectid, there must not be acls */
2310 if (found_key.objectid != objectid)
2311 return 0;
2312
2313 /* we found an xattr, assume we've got an acl */
2314 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2315 return 1;
2316
2317 /*
2318 * we found a key greater than an xattr key, there can't
2319 * be any acls later on
2320 */
2321 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2322 return 0;
2323
2324 slot++;
2325 scanned++;
2326
2327 /*
2328 * it goes inode, inode backrefs, xattrs, extents,
2329 * so if there are a ton of hard links to an inode there can
2330 * be a lot of backrefs. Don't waste time searching too hard,
2331 * this is just an optimization
2332 */
2333 if (scanned >= 8)
2334 break;
2335 }
2336 /* we hit the end of the leaf before we found an xattr or
2337 * something larger than an xattr. We have to assume the inode
2338 * has acls
2339 */
2340 return 1;
2341}
2342
d352ac68
CM		 2343/*
2344 * read an inode from the btree into the in-memory inode
2345 */
5d4f98a2 2346static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2347{
2348 struct btrfs_path *path;
5f39d397 2349 struct extent_buffer *leaf;
39279cc3 2350 struct btrfs_inode_item *inode_item;
0b86a832 2351 struct btrfs_timespec *tspec;
39279cc3
CM		 2352 struct btrfs_root *root = BTRFS_I(inode)->root;
2353 struct btrfs_key location;
46a53cca 2354 int maybe_acls;
618e21d5 2355 u32 rdev;
39279cc3 2356 int ret;
2f7e33d4
MX		 2357 bool filled = false;
2358
2359 ret = btrfs_fill_inode(inode, &rdev);
2360 if (!ret)
2361 filled = true;
39279cc3
CM		 2362
2363 path = btrfs_alloc_path();
1748f843
MF		 2364 if (!path)
2365 goto make_bad;
2366
d90c7321 2367 path->leave_spinning = 1;
39279cc3 2368 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2369
39279cc3 2370 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2371 if (ret)
39279cc3 2372 goto make_bad;
39279cc3 2373
5f39d397 2374 leaf = path->nodes[0];
2f7e33d4
MX		 2375
2376 if (filled)
2377 goto cache_acl;
2378
5f39d397
CM		 2379 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2380 struct btrfs_inode_item);
5f39d397 2381 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 2382 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
5f39d397
CM		 2383 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2384 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2385 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2386
2387 tspec = btrfs_inode_atime(inode_item);
2388 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2389 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2390
2391 tspec = btrfs_inode_mtime(inode_item);
2392 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2393 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2394
2395 tspec = btrfs_inode_ctime(inode_item);
2396 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2397 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2398
a76a3cd4 2399 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2400 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2401 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2402 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2403 inode->i_rdev = 0;
5f39d397
CM		 2404 rdev = btrfs_inode_rdev(leaf, inode_item);
2405
aec7477b 2406 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2407 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2408cache_acl:
46a53cca
CM		 2409 /*
2410 * try to precache a NULL acl entry for files that don't have
2411 * any xattrs or acls
2412 */
33345d01
LZ		 2413 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2414 btrfs_ino(inode));
72c04902
AV		 2415 if (!maybe_acls)
2416 cache_no_acl(inode);
46a53cca 2417
39279cc3 2418 btrfs_free_path(path);
39279cc3 2419
39279cc3 2420 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2421 case S_IFREG:
2422 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2423 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2424 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2425 inode->i_fop = &btrfs_file_operations;
2426 inode->i_op = &btrfs_file_inode_operations;
2427 break;
2428 case S_IFDIR:
2429 inode->i_fop = &btrfs_dir_file_operations;
2430 if (root == root->fs_info->tree_root)
2431 inode->i_op = &btrfs_dir_ro_inode_operations;
2432 else
2433 inode->i_op = &btrfs_dir_inode_operations;
2434 break;
2435 case S_IFLNK:
2436 inode->i_op = &btrfs_symlink_inode_operations;
2437 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2438 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2439 break;
618e21d5 2440 default:
0279b4cd 2441 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2442 init_special_inode(inode, inode->i_mode, rdev);
2443 break;
39279cc3 2444 }
6cbff00f
CH		 2445
2446 btrfs_update_iflags(inode);
39279cc3
CM		 2447 return;
2448
2449make_bad:
39279cc3 2450 btrfs_free_path(path);
39279cc3
CM		 2451 make_bad_inode(inode);
2452}
2453
d352ac68
CM		 2454/*
2455 * given a leaf and an inode, copy the inode fields into the leaf
2456 */
e02119d5
CM		 2457static void fill_inode_item(struct btrfs_trans_handle *trans,
2458 struct extent_buffer *leaf,
5f39d397 2459 struct btrfs_inode_item *item,
39279cc3
CM		 2460 struct inode *inode)
2461{
5f39d397
CM		 2462 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2463 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2464 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2465 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2466 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2467
2468 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2469 inode->i_atime.tv_sec);
2470 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2471 inode->i_atime.tv_nsec);
2472
2473 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2474 inode->i_mtime.tv_sec);
2475 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2476 inode->i_mtime.tv_nsec);
2477
2478 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2479 inode->i_ctime.tv_sec);
2480 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2481 inode->i_ctime.tv_nsec);
2482
a76a3cd4 2483 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2484 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2485 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2486 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2487 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2488 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2489 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2490}
2491
d352ac68
CM		 2492/*
2493 * copy everything in the in-memory inode into the btree.
2494 */
2115133f 2495static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2496 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2497{
2498 struct btrfs_inode_item *inode_item;
2499 struct btrfs_path *path;
5f39d397 2500 struct extent_buffer *leaf;
39279cc3
CM		 2501 int ret;
2502
2503 path = btrfs_alloc_path();
16cdcec7
MX		 2504 if (!path)
2505 return -ENOMEM;
2506
b9473439 2507 path->leave_spinning = 1;
16cdcec7
MX		 2508 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2509 1);
39279cc3
CM		 2510 if (ret) {
2511 if (ret > 0)
2512 ret = -ENOENT;
2513 goto failed;
2514 }
2515
b4ce94de 2516 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2517 leaf = path->nodes[0];
2518 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2519 struct btrfs_inode_item);
39279cc3 2520
e02119d5 2521 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2522 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2523 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2524 ret = 0;
2525failed:
39279cc3
CM		 2526 btrfs_free_path(path);
2527 return ret;
2528}
2529
2115133f
CM		 2530/*
2531 * copy everything in the in-memory inode into the btree.
2532 */
2533noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2534 struct btrfs_root *root, struct inode *inode)
2535{
2536 int ret;
2537
2538 /*
2539 * If the inode is a free space inode, we can deadlock during commit
2540 * if we put it into the delayed code.
2541 *
2542 * The data relocation inode should also be directly updated
2543 * without delay
2544 */
2545 if (!btrfs_is_free_space_inode(root, inode)
2546 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2547 ret = btrfs_delayed_update_inode(trans, root, inode);
2548 if (!ret)
2549 btrfs_set_inode_last_trans(trans, inode);
2550 return ret;
2551 }
2552
2553 return btrfs_update_inode_item(trans, root, inode);
2554}
2555
2556static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2557 struct btrfs_root *root, struct inode *inode)
2558{
2559 int ret;
2560
2561 ret = btrfs_update_inode(trans, root, inode);
2562 if (ret == -ENOSPC)
2563 return btrfs_update_inode_item(trans, root, inode);
2564 return ret;
2565}
2566
d352ac68
CM		 2567/*
2568 * unlink helper that gets used here in inode.c and in the tree logging
2569 * recovery code. It remove a link in a directory with a given name, and
2570 * also drops the back refs in the inode to the directory
2571 */
92986796
AV		 2572static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2573 struct btrfs_root *root,
2574 struct inode *dir, struct inode *inode,
2575 const char *name, int name_len)
39279cc3
CM		 2576{
2577 struct btrfs_path *path;
39279cc3 2578 int ret = 0;
5f39d397 2579 struct extent_buffer *leaf;
39279cc3 2580 struct btrfs_dir_item *di;
5f39d397 2581 struct btrfs_key key;
aec7477b 2582 u64 index;
33345d01
LZ		 2583 u64 ino = btrfs_ino(inode);
2584 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2585
2586 path = btrfs_alloc_path();
54aa1f4d
CM		 2587 if (!path) {
2588 ret = -ENOMEM;
554233a6 2589 goto out;
54aa1f4d
CM		 2590 }
2591
b9473439 2592 path->leave_spinning = 1;
33345d01 2593 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2594 name, name_len, -1);
2595 if (IS_ERR(di)) {
2596 ret = PTR_ERR(di);
2597 goto err;
2598 }
2599 if (!di) {
2600 ret = -ENOENT;
2601 goto err;
2602 }
5f39d397
CM		 2603 leaf = path->nodes[0];
2604 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2605 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2606 if (ret)
2607 goto err;
b3b4aa74 2608 btrfs_release_path(path);
39279cc3 2609
33345d01
LZ		 2610 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2611 dir_ino, &index);
aec7477b 2612 if (ret) {
d397712b 2613 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2614 "inode %llu parent %llu\n", name_len, name,
2615 (unsigned long long)ino, (unsigned long long)dir_ino);
aec7477b
JB		 2616 goto err;
2617 }
2618
16cdcec7
MX		 2619 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2620 if (ret)
39279cc3 2621 goto err;
39279cc3 2622
e02119d5 2623 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2624 inode, dir_ino);
49eb7e46 2625 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2626
2627 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2628 dir, index);
6418c961
CM		 2629 if (ret == -ENOENT)
2630 ret = 0;
39279cc3
CM		 2631err:
2632 btrfs_free_path(path);
e02119d5
CM		 2633 if (ret)
2634 goto out;
2635
2636 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2637 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2638 btrfs_update_inode(trans, root, dir);
e02119d5 2639out:
39279cc3
CM		 2640 return ret;
2641}
2642
92986796
AV		 2643int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2644 struct btrfs_root *root,
2645 struct inode *dir, struct inode *inode,
2646 const char *name, int name_len)
2647{
2648 int ret;
2649 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2650 if (!ret) {
2651 btrfs_drop_nlink(inode);
2652 ret = btrfs_update_inode(trans, root, inode);
2653 }
2654 return ret;
2655}
2656
2657
a22285a6
YZ		 2658/* helper to check if there is any shared block in the path */
2659static int check_path_shared(struct btrfs_root *root,
2660 struct btrfs_path *path)
39279cc3 2661{
a22285a6
YZ		 2662 struct extent_buffer *eb;
2663 int level;
0e4dcbef 2664 u64 refs = 1;
5df6a9f6 2665
a22285a6 2666 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2667 int ret;
2668
a22285a6
YZ		 2669 if (!path->nodes[level])
2670 break;
2671 eb = path->nodes[level];
2672 if (!btrfs_block_can_be_shared(root, eb))
2673 continue;
2674 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2675 &refs, NULL);
2676 if (refs > 1)
2677 return 1;
5df6a9f6 2678 }
dedefd72 2679 return 0;
39279cc3
CM		 2680}
2681
a22285a6
YZ		 2682/*
2683 * helper to start transaction for unlink and rmdir.
2684 *
2685 * unlink and rmdir are special in btrfs, they do not always free space.
2686 * so in enospc case, we should make sure they will free space before
2687 * allowing them to use the global metadata reservation.
2688 */
2689static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2690 struct dentry *dentry)
4df27c4d 2691{
39279cc3 2692 struct btrfs_trans_handle *trans;
a22285a6 2693 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2694 struct btrfs_path *path;
a22285a6 2695 struct btrfs_inode_ref *ref;
4df27c4d 2696 struct btrfs_dir_item *di;
7b128766 2697 struct inode *inode = dentry->d_inode;
4df27c4d 2698 u64 index;
a22285a6
YZ		 2699 int check_link = 1;
2700 int err = -ENOSPC;
4df27c4d 2701 int ret;
33345d01
LZ		 2702 u64 ino = btrfs_ino(inode);
2703 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2704
e70bea5f
JB		 2705 /*
2706 * 1 for the possible orphan item
2707 * 1 for the dir item
2708 * 1 for the dir index
2709 * 1 for the inode ref
2710 * 1 for the inode ref in the tree log
2711 * 2 for the dir entries in the log
2712 * 1 for the inode
2713 */
2714 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2715 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2716 return trans;
4df27c4d 2717
33345d01 2718 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2719 return ERR_PTR(-ENOSPC);
4df27c4d 2720
a22285a6
YZ		 2721 /* check if there is someone else holds reference */
2722 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2723 return ERR_PTR(-ENOSPC);
4df27c4d 2724
a22285a6
YZ		 2725 if (atomic_read(&inode->i_count) > 2)
2726 return ERR_PTR(-ENOSPC);
4df27c4d 2727
a22285a6
YZ		 2728 if (xchg(&root->fs_info->enospc_unlink, 1))
2729 return ERR_PTR(-ENOSPC);
2730
2731 path = btrfs_alloc_path();
2732 if (!path) {
2733 root->fs_info->enospc_unlink = 0;
2734 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2735 }
2736
3880a1b4
JB		 2737 /* 1 for the orphan item */
2738 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2739 if (IS_ERR(trans)) {
a22285a6
YZ		 2740 btrfs_free_path(path);
2741 root->fs_info->enospc_unlink = 0;
2742 return trans;
2743 }
4df27c4d 2744
a22285a6
YZ		 2745 path->skip_locking = 1;
2746 path->search_commit_root = 1;
4df27c4d 2747
a22285a6
YZ		 2748 ret = btrfs_lookup_inode(trans, root, path,
2749 &BTRFS_I(dir)->location, 0);
2750 if (ret < 0) {
2751 err = ret;
2752 goto out;
2753 }
2754 if (ret == 0) {
2755 if (check_path_shared(root, path))
2756 goto out;
2757 } else {
2758 check_link = 0;
5df6a9f6 2759 }
b3b4aa74 2760 btrfs_release_path(path);
a22285a6
YZ		 2761
2762 ret = btrfs_lookup_inode(trans, root, path,
2763 &BTRFS_I(inode)->location, 0);
2764 if (ret < 0) {
2765 err = ret;
2766 goto out;
2767 }
2768 if (ret == 0) {
2769 if (check_path_shared(root, path))
2770 goto out;
2771 } else {
2772 check_link = 0;
2773 }
b3b4aa74 2774 btrfs_release_path(path);
a22285a6
YZ		 2775
2776 if (ret == 0 && S_ISREG(inode->i_mode)) {
2777 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2778 ino, (u64)-1, 0);
a22285a6
YZ		 2779 if (ret < 0) {
2780 err = ret;
2781 goto out;
2782 }
2783 BUG_ON(ret == 0);
2784 if (check_path_shared(root, path))
2785 goto out;
b3b4aa74 2786 btrfs_release_path(path);
a22285a6
YZ		 2787 }
2788
2789 if (!check_link) {
2790 err = 0;
2791 goto out;
2792 }
2793
33345d01 2794 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2795 dentry->d_name.name, dentry->d_name.len, 0);
2796 if (IS_ERR(di)) {
2797 err = PTR_ERR(di);
2798 goto out;
2799 }
2800 if (di) {
2801 if (check_path_shared(root, path))
2802 goto out;
2803 } else {
2804 err = 0;
2805 goto out;
2806 }
b3b4aa74 2807 btrfs_release_path(path);
a22285a6
YZ		 2808
2809 ref = btrfs_lookup_inode_ref(trans, root, path,
2810 dentry->d_name.name, dentry->d_name.len,
33345d01 2811 ino, dir_ino, 0);
a22285a6
YZ		 2812 if (IS_ERR(ref)) {
2813 err = PTR_ERR(ref);
2814 goto out;
2815 }
2816 BUG_ON(!ref);
2817 if (check_path_shared(root, path))
2818 goto out;
2819 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 2820 btrfs_release_path(path);
a22285a6 2821
16cdcec7
MX		 2822 /*
2823 * This is a commit root search, if we can lookup inode item and other
2824 * relative items in the commit root, it means the transaction of
2825 * dir/file creation has been committed, and the dir index item that we
2826 * delay to insert has also been inserted into the commit root. So
2827 * we needn't worry about the delayed insertion of the dir index item
2828 * here.
2829 */
33345d01 2830 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 2831 dentry->d_name.name, dentry->d_name.len, 0);
2832 if (IS_ERR(di)) {
2833 err = PTR_ERR(di);
2834 goto out;
2835 }
2836 BUG_ON(ret == -ENOENT);
2837 if (check_path_shared(root, path))
2838 goto out;
2839
2840 err = 0;
2841out:
2842 btrfs_free_path(path);
3880a1b4
JB		 2843 /* Migrate the orphan reservation over */
2844 if (!err)
2845 err = btrfs_block_rsv_migrate(trans->block_rsv,
2846 &root->fs_info->global_block_rsv,
5a77d76c 2847 trans->bytes_reserved);
3880a1b4 2848
a22285a6
YZ		 2849 if (err) {
2850 btrfs_end_transaction(trans, root);
2851 root->fs_info->enospc_unlink = 0;
2852 return ERR_PTR(err);
2853 }
2854
2855 trans->block_rsv = &root->fs_info->global_block_rsv;
2856 return trans;
2857}
2858
2859static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2860 struct btrfs_root *root)
2861{
2862 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
5a77d76c
JB		 2863 btrfs_block_rsv_release(root, trans->block_rsv,
2864 trans->bytes_reserved);
2865 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 2866 BUG_ON(!root->fs_info->enospc_unlink);
2867 root->fs_info->enospc_unlink = 0;
2868 }
7ad85bb7 2869 btrfs_end_transaction(trans, root);
a22285a6
YZ		 2870}
2871
2872static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2873{
2874 struct btrfs_root *root = BTRFS_I(dir)->root;
2875 struct btrfs_trans_handle *trans;
2876 struct inode *inode = dentry->d_inode;
2877 int ret;
2878 unsigned long nr = 0;
2879
2880 trans = __unlink_start_trans(dir, dentry);
2881 if (IS_ERR(trans))
2882 return PTR_ERR(trans);
5f39d397 2883
12fcfd22
CM		 2884 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2885
e02119d5
CM		 2886 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2887 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 2888 if (ret)
2889 goto out;
7b128766 2890
a22285a6 2891 if (inode->i_nlink == 0) {
7b128766 2892 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 2893 if (ret)
2894 goto out;
a22285a6 2895 }
7b128766 2896
b532402e 2897out:
d3c2fdcf 2898 nr = trans->blocks_used;
a22285a6 2899 __unlink_end_trans(trans, root);
d3c2fdcf 2900 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2901 return ret;
2902}
2903
4df27c4d
YZ		 2904int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2905 struct btrfs_root *root,
2906 struct inode *dir, u64 objectid,
2907 const char *name, int name_len)
2908{
2909 struct btrfs_path *path;
2910 struct extent_buffer *leaf;
2911 struct btrfs_dir_item *di;
2912 struct btrfs_key key;
2913 u64 index;
2914 int ret;
33345d01 2915 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 2916
2917 path = btrfs_alloc_path();
2918 if (!path)
2919 return -ENOMEM;
2920
33345d01 2921 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 2922 name, name_len, -1);
c704005d 2923 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2924
2925 leaf = path->nodes[0];
2926 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2927 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2928 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2929 BUG_ON(ret);
b3b4aa74 2930 btrfs_release_path(path);
4df27c4d
YZ		 2931
2932 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2933 objectid, root->root_key.objectid,
33345d01 2934 dir_ino, &index, name, name_len);
4df27c4d
YZ		 2935 if (ret < 0) {
2936 BUG_ON(ret != -ENOENT);
33345d01 2937 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 2938 name, name_len);
c704005d 2939 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2940
2941 leaf = path->nodes[0];
2942 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2943 btrfs_release_path(path);
4df27c4d
YZ		 2944 index = key.offset;
2945 }
945d8962 2946 btrfs_release_path(path);
4df27c4d 2947
16cdcec7 2948 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4df27c4d 2949 BUG_ON(ret);
4df27c4d
YZ		 2950
2951 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2952 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2953 ret = btrfs_update_inode(trans, root, dir);
2954 BUG_ON(ret);
4df27c4d 2955
71d7aed0 2956 btrfs_free_path(path);
4df27c4d
YZ		 2957 return 0;
2958}
2959
39279cc3
CM		 2960static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2961{
2962 struct inode *inode = dentry->d_inode;
1832a6d5 2963 int err = 0;
39279cc3 2964 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2965 struct btrfs_trans_handle *trans;
1832a6d5 2966 unsigned long nr = 0;
39279cc3 2967
3394e160 2968 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 2969 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 2970 return -ENOTEMPTY;
2971
a22285a6
YZ		 2972 trans = __unlink_start_trans(dir, dentry);
2973 if (IS_ERR(trans))
5df6a9f6 2974 return PTR_ERR(trans);
5df6a9f6 2975
33345d01 2976 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 2977 err = btrfs_unlink_subvol(trans, root, dir,
2978 BTRFS_I(inode)->location.objectid,
2979 dentry->d_name.name,
2980 dentry->d_name.len);
2981 goto out;
2982 }
2983
7b128766
JB		 2984 err = btrfs_orphan_add(trans, inode);
2985 if (err)
4df27c4d 2986 goto out;
7b128766 2987
39279cc3 2988 /* now the directory is empty */
e02119d5
CM		 2989 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2990 dentry->d_name.name, dentry->d_name.len);
d397712b 2991 if (!err)
dbe674a9 2992 btrfs_i_size_write(inode, 0);
4df27c4d 2993out:
d3c2fdcf 2994 nr = trans->blocks_used;
a22285a6 2995 __unlink_end_trans(trans, root);
d3c2fdcf 2996 btrfs_btree_balance_dirty(root, nr);
3954401f 2997
39279cc3
CM		 2998 return err;
2999}
3000
39279cc3
CM		 3001/*
3002 * this can truncate away extent items, csum items and directory items.
3003 * It starts at a high offset and removes keys until it can't find
d352ac68 3004 * any higher than new_size
39279cc3
CM		 3005 *
3006 * csum items that cross the new i_size are truncated to the new size
3007 * as well.
7b128766
JB		 3008 *
3009 * min_type is the minimum key type to truncate down to. If set to 0, this
3010 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3011 */
8082510e
YZ		 3012int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3013 struct btrfs_root *root,
3014 struct inode *inode,
3015 u64 new_size, u32 min_type)
39279cc3 3016{
39279cc3 3017 struct btrfs_path *path;
5f39d397 3018 struct extent_buffer *leaf;
39279cc3 3019 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3020 struct btrfs_key key;
3021 struct btrfs_key found_key;
39279cc3 3022 u64 extent_start = 0;
db94535d 3023 u64 extent_num_bytes = 0;
5d4f98a2 3024 u64 extent_offset = 0;
39279cc3 3025 u64 item_end = 0;
8082510e
YZ		 3026 u64 mask = root->sectorsize - 1;
3027 u32 found_type = (u8)-1;
39279cc3
CM		 3028 int found_extent;
3029 int del_item;
85e21bac
CM		 3030 int pending_del_nr = 0;
3031 int pending_del_slot = 0;
179e29e4 3032 int extent_type = -1;
8082510e
YZ		 3033 int ret;
3034 int err = 0;
33345d01 3035 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3036
3037 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3038
0eb0e19c
MF		 3039 path = btrfs_alloc_path();
3040 if (!path)
3041 return -ENOMEM;
3042 path->reada = -1;
3043
0af3d00b 3044 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 3045 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 3046
16cdcec7
MX		 3047 /*
3048 * This function is also used to drop the items in the log tree before
3049 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3050 * it is used to drop the loged items. So we shouldn't kill the delayed
3051 * items.
3052 */
3053 if (min_type == 0 && root == BTRFS_I(inode)->root)
3054 btrfs_kill_delayed_inode_items(inode);
3055
33345d01 3056 key.objectid = ino;
39279cc3 3057 key.offset = (u64)-1;
5f39d397
CM		 3058 key.type = (u8)-1;
3059
85e21bac 3060search_again:
b9473439 3061 path->leave_spinning = 1;
85e21bac 3062 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3063 if (ret < 0) {
3064 err = ret;
3065 goto out;
3066 }
d397712b 3067
85e21bac 3068 if (ret > 0) {
e02119d5
CM		 3069 /* there are no items in the tree for us to truncate, we're
3070 * done
3071 */
8082510e
YZ		 3072 if (path->slots[0] == 0)
3073 goto out;
85e21bac
CM		 3074 path->slots[0]--;
3075 }
3076
d397712b 3077 while (1) {
39279cc3 3078 fi = NULL;
5f39d397
CM		 3079 leaf = path->nodes[0];
3080 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3081 found_type = btrfs_key_type(&found_key);
39279cc3 3082
33345d01 3083 if (found_key.objectid != ino)
39279cc3 3084 break;
5f39d397 3085
85e21bac 3086 if (found_type < min_type)
39279cc3
CM		 3087 break;
3088
5f39d397 3089 item_end = found_key.offset;
39279cc3 3090 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3091 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3092 struct btrfs_file_extent_item);
179e29e4
CM		 3093 extent_type = btrfs_file_extent_type(leaf, fi);
3094 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3095 item_end +=
db94535d 3096 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3097 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3098 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3099 fi);
39279cc3 3100 }
008630c1 3101 item_end--;
39279cc3 3102 }
8082510e
YZ		 3103 if (found_type > min_type) {
3104 del_item = 1;
3105 } else {
3106 if (item_end < new_size)
b888db2b 3107 break;
8082510e
YZ		 3108 if (found_key.offset >= new_size)
3109 del_item = 1;
3110 else
3111 del_item = 0;
39279cc3 3112 }
39279cc3 3113 found_extent = 0;
39279cc3 3114 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3115 if (found_type != BTRFS_EXTENT_DATA_KEY)
3116 goto delete;
3117
3118 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3119 u64 num_dec;
db94535d 3120 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3121 if (!del_item) {
db94535d
CM		 3122 u64 orig_num_bytes =
3123 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3124 extent_num_bytes = new_size -
5f39d397 3125 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3126 extent_num_bytes = extent_num_bytes &
3127 ~((u64)root->sectorsize - 1);
db94535d
CM		 3128 btrfs_set_file_extent_num_bytes(leaf, fi,
3129 extent_num_bytes);
3130 num_dec = (orig_num_bytes -
9069218d 3131 extent_num_bytes);
e02119d5 3132 if (root->ref_cows && extent_start != 0)
a76a3cd4 3133 inode_sub_bytes(inode, num_dec);
5f39d397 3134 btrfs_mark_buffer_dirty(leaf);
39279cc3 3135 } else {
db94535d
CM		 3136 extent_num_bytes =
3137 btrfs_file_extent_disk_num_bytes(leaf,
3138 fi);
5d4f98a2
YZ		 3139 extent_offset = found_key.offset -
3140 btrfs_file_extent_offset(leaf, fi);
3141
39279cc3 3142 /* FIXME blocksize != 4096 */
9069218d 3143 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3144 if (extent_start != 0) {
3145 found_extent = 1;
e02119d5 3146 if (root->ref_cows)
a76a3cd4 3147 inode_sub_bytes(inode, num_dec);
e02119d5 3148 }
39279cc3 3149 }
9069218d 3150 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3151 /*
3152 * we can't truncate inline items that have had
3153 * special encodings
3154 */
3155 if (!del_item &&
3156 btrfs_file_extent_compression(leaf, fi) == 0 &&
3157 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3158 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3159 u32 size = new_size - found_key.offset;
3160
3161 if (root->ref_cows) {
a76a3cd4
YZ		 3162 inode_sub_bytes(inode, item_end + 1 -
3163 new_size);
e02119d5
CM		 3164 }
3165 size =
3166 btrfs_file_extent_calc_inline_size(size);
9069218d 3167 ret = btrfs_truncate_item(trans, root, path,
e02119d5 3168 size, 1);
e02119d5 3169 } else if (root->ref_cows) {
a76a3cd4
YZ		 3170 inode_sub_bytes(inode, item_end + 1 -
3171 found_key.offset);
9069218d 3172 }
39279cc3 3173 }
179e29e4 3174delete:
39279cc3 3175 if (del_item) {
85e21bac
CM		 3176 if (!pending_del_nr) {
3177 /* no pending yet, add ourselves */
3178 pending_del_slot = path->slots[0];
3179 pending_del_nr = 1;
3180 } else if (pending_del_nr &&
3181 path->slots[0] + 1 == pending_del_slot) {
3182 /* hop on the pending chunk */
3183 pending_del_nr++;
3184 pending_del_slot = path->slots[0];
3185 } else {
d397712b 3186 BUG();
85e21bac 3187 }
39279cc3
CM		 3188 } else {
3189 break;
3190 }
0af3d00b
JB		 3191 if (found_extent && (root->ref_cows ||
3192 root == root->fs_info->tree_root)) {
b9473439 3193 btrfs_set_path_blocking(path);
39279cc3 3194 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3195 extent_num_bytes, 0,
3196 btrfs_header_owner(leaf),
66d7e7f0 3197 ino, extent_offset, 0);
39279cc3
CM		 3198 BUG_ON(ret);
3199 }
85e21bac 3200
8082510e
YZ		 3201 if (found_type == BTRFS_INODE_ITEM_KEY)
3202 break;
3203
3204 if (path->slots[0] == 0 ||
3205 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3206 if (root->ref_cows &&
3207 BTRFS_I(inode)->location.objectid !=
3208 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3209 err = -EAGAIN;
3210 goto out;
3211 }
3212 if (pending_del_nr) {
3213 ret = btrfs_del_items(trans, root, path,
3214 pending_del_slot,
3215 pending_del_nr);
3216 BUG_ON(ret);
3217 pending_del_nr = 0;
3218 }
b3b4aa74 3219 btrfs_release_path(path);
85e21bac 3220 goto search_again;
8082510e
YZ		 3221 } else {
3222 path->slots[0]--;
85e21bac 3223 }
39279cc3 3224 }
8082510e 3225out:
85e21bac
CM		 3226 if (pending_del_nr) {
3227 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3228 pending_del_nr);
d68fc57b 3229 BUG_ON(ret);
85e21bac 3230 }
39279cc3 3231 btrfs_free_path(path);
8082510e 3232 return err;
39279cc3
CM		 3233}
3234
3235/*
3236 * taken from block_truncate_page, but does cow as it zeros out
3237 * any bytes left in the last page in the file.
3238 */
3239static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3240{
3241 struct inode *inode = mapping->host;
db94535d 3242 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3243 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3244 struct btrfs_ordered_extent *ordered;
2ac55d41 3245 struct extent_state *cached_state = NULL;
e6dcd2dc 3246 char *kaddr;
db94535d 3247 u32 blocksize = root->sectorsize;
39279cc3
CM		 3248 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3249 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3250 struct page *page;
3b16a4e3 3251 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3252 int ret = 0;
a52d9a80 3253 u64 page_start;
e6dcd2dc 3254 u64 page_end;
39279cc3
CM		 3255
3256 if ((offset & (blocksize - 1)) == 0)
3257 goto out;
0ca1f7ce 3258 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3259 if (ret)
3260 goto out;
39279cc3
CM		 3261
3262 ret = -ENOMEM;
211c17f5 3263again:
3b16a4e3 3264 page = find_or_create_page(mapping, index, mask);
5d5e103a 3265 if (!page) {
0ca1f7ce 3266 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3267 goto out;
5d5e103a 3268 }
e6dcd2dc
CM		 3269
3270 page_start = page_offset(page);
3271 page_end = page_start + PAGE_CACHE_SIZE - 1;
3272
39279cc3 3273 if (!PageUptodate(page)) {
9ebefb18 3274 ret = btrfs_readpage(NULL, page);
39279cc3 3275 lock_page(page);
211c17f5
CM		 3276 if (page->mapping != mapping) {
3277 unlock_page(page);
3278 page_cache_release(page);
3279 goto again;
3280 }
39279cc3
CM		 3281 if (!PageUptodate(page)) {
3282 ret = -EIO;
89642229 3283 goto out_unlock;
39279cc3
CM		 3284 }
3285 }
211c17f5 3286 wait_on_page_writeback(page);
e6dcd2dc 3287
2ac55d41
JB		 3288 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3289 GFP_NOFS);
e6dcd2dc
CM		 3290 set_page_extent_mapped(page);
3291
3292 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3293 if (ordered) {
2ac55d41
JB		 3294 unlock_extent_cached(io_tree, page_start, page_end,
3295 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3296 unlock_page(page);
3297 page_cache_release(page);
eb84ae03 3298 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3299 btrfs_put_ordered_extent(ordered);
3300 goto again;
3301 }
3302
2ac55d41 3303 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3304 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3305 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3306
2ac55d41
JB		 3307 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3308 &cached_state);
9ed74f2d 3309 if (ret) {
2ac55d41
JB		 3310 unlock_extent_cached(io_tree, page_start, page_end,
3311 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3312 goto out_unlock;
3313 }
3314
e6dcd2dc
CM		 3315 ret = 0;
3316 if (offset != PAGE_CACHE_SIZE) {
3317 kaddr = kmap(page);
3318 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3319 flush_dcache_page(page);
3320 kunmap(page);
3321 }
247e743c 3322 ClearPageChecked(page);
e6dcd2dc 3323 set_page_dirty(page);
2ac55d41
JB		 3324 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3325 GFP_NOFS);
39279cc3 3326
89642229 3327out_unlock:
5d5e103a 3328 if (ret)
0ca1f7ce 3329 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3330 unlock_page(page);
3331 page_cache_release(page);
3332out:
3333 return ret;
3334}
3335
695a0d0d
JB		 3336/*
3337 * This function puts in dummy file extents for the area we're creating a hole
3338 * for. So if we are truncating this file to a larger size we need to insert
3339 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3340 * the range between oldsize and size
3341 */
a41ad394 3342int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3343{
9036c102
YZ		 3344 struct btrfs_trans_handle *trans;
3345 struct btrfs_root *root = BTRFS_I(inode)->root;
3346 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3347 struct extent_map *em = NULL;
2ac55d41 3348 struct extent_state *cached_state = NULL;
9036c102 3349 u64 mask = root->sectorsize - 1;
a41ad394 3350 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3351 u64 block_end = (size + mask) & ~mask;
3352 u64 last_byte;
3353 u64 cur_offset;
3354 u64 hole_size;
9ed74f2d 3355 int err = 0;
39279cc3 3356
9036c102
YZ		 3357 if (size <= hole_start)
3358 return 0;
3359
9036c102
YZ		 3360 while (1) {
3361 struct btrfs_ordered_extent *ordered;
3362 btrfs_wait_ordered_range(inode, hole_start,
3363 block_end - hole_start);
2ac55d41
JB		 3364 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3365 &cached_state, GFP_NOFS);
9036c102
YZ		 3366 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3367 if (!ordered)
3368 break;
2ac55d41
JB		 3369 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3370 &cached_state, GFP_NOFS);
9036c102
YZ		 3371 btrfs_put_ordered_extent(ordered);
3372 }
39279cc3 3373
9036c102
YZ		 3374 cur_offset = hole_start;
3375 while (1) {
3376 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3377 block_end - cur_offset, 0);
c704005d 3378 BUG_ON(IS_ERR_OR_NULL(em));
9036c102
YZ		 3379 last_byte = min(extent_map_end(em), block_end);
3380 last_byte = (last_byte + mask) & ~mask;
8082510e 3381 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3382 u64 hint_byte = 0;
9036c102 3383 hole_size = last_byte - cur_offset;
9ed74f2d 3384
3642320e 3385 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 3386 if (IS_ERR(trans)) {
3387 err = PTR_ERR(trans);
9ed74f2d 3388 break;
a22285a6 3389 }
8082510e
YZ		 3390
3391 err = btrfs_drop_extents(trans, inode, cur_offset,
3392 cur_offset + hole_size,
3393 &hint_byte, 1);
5b397377 3394 if (err) {
3642320e 3395 btrfs_update_inode(trans, root, inode);
5b397377 3396 btrfs_end_transaction(trans, root);
3893e33b 3397 break;
5b397377 3398 }
8082510e 3399
9036c102 3400 err = btrfs_insert_file_extent(trans, root,
33345d01 3401 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3402 0, hole_size, 0, hole_size,
3403 0, 0, 0);
5b397377 3404 if (err) {
3642320e 3405 btrfs_update_inode(trans, root, inode);
5b397377 3406 btrfs_end_transaction(trans, root);
3893e33b 3407 break;
5b397377 3408 }
8082510e 3409
9036c102
YZ		 3410 btrfs_drop_extent_cache(inode, hole_start,
3411 last_byte - 1, 0);
8082510e 3412
3642320e 3413 btrfs_update_inode(trans, root, inode);
8082510e 3414 btrfs_end_transaction(trans, root);
9036c102
YZ		 3415 }
3416 free_extent_map(em);
a22285a6 3417 em = NULL;
9036c102 3418 cur_offset = last_byte;
8082510e 3419 if (cur_offset >= block_end)
9036c102
YZ		 3420 break;
3421 }
1832a6d5 3422
a22285a6 3423 free_extent_map(em);
2ac55d41
JB		 3424 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3425 GFP_NOFS);
9036c102
YZ		 3426 return err;
3427}
39279cc3 3428
a41ad394 3429static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3430{
f4a2f4c5
MX		 3431 struct btrfs_root *root = BTRFS_I(inode)->root;
3432 struct btrfs_trans_handle *trans;
a41ad394 3433 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3434 int ret;
3435
a41ad394 3436 if (newsize == oldsize)
8082510e
YZ		 3437 return 0;
3438
a41ad394 3439 if (newsize > oldsize) {
a41ad394
JB		 3440 truncate_pagecache(inode, oldsize, newsize);
3441 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 3442 if (ret)
8082510e 3443 return ret;
8082510e 3444
f4a2f4c5
MX		 3445 trans = btrfs_start_transaction(root, 1);
3446 if (IS_ERR(trans))
3447 return PTR_ERR(trans);
3448
3449 i_size_write(inode, newsize);
3450 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3451 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 3452 btrfs_end_transaction(trans, root);
a41ad394 3453 } else {
8082510e 3454
a41ad394
JB		 3455 /*
3456 * We're truncating a file that used to have good data down to
3457 * zero. Make sure it gets into the ordered flush list so that
3458 * any new writes get down to disk quickly.
3459 */
3460 if (newsize == 0)
3461 BTRFS_I(inode)->ordered_data_close = 1;
8082510e 3462
a41ad394
JB		 3463 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3464 truncate_setsize(inode, newsize);
3465 ret = btrfs_truncate(inode);
8082510e
YZ		 3466 }
3467
a41ad394 3468 return ret;
8082510e
YZ		 3469}
3470
9036c102
YZ		 3471static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3472{
3473 struct inode *inode = dentry->d_inode;
b83cc969 3474 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3475 int err;
39279cc3 3476
b83cc969
LZ		 3477 if (btrfs_root_readonly(root))
3478 return -EROFS;
3479
9036c102
YZ		 3480 err = inode_change_ok(inode, attr);
3481 if (err)
3482 return err;
2bf5a725 3483
5a3f23d5 3484 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3485 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3486 if (err)
3487 return err;
39279cc3 3488 }
9036c102 3489
1025774c
CH		 3490 if (attr->ia_valid) {
3491 setattr_copy(inode, attr);
22c44fe6 3492 err = btrfs_dirty_inode(inode);
1025774c 3493
22c44fe6 3494 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 3495 err = btrfs_acl_chmod(inode);
3496 }
33268eaf 3497
39279cc3
CM		 3498 return err;
3499}
61295eb8 3500
bd555975 3501void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3502{
3503 struct btrfs_trans_handle *trans;
3504 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3505 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3506 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3507 unsigned long nr;
39279cc3
CM		 3508 int ret;
3509
1abe9b8a 3510 trace_btrfs_inode_evict(inode);
3511
39279cc3 3512 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3513 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
2cf8572d 3514 btrfs_is_free_space_inode(root, inode)))
bd555975
AV		 3515 goto no_delete;
3516
39279cc3 3517 if (is_bad_inode(inode)) {
7b128766 3518 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3519 goto no_delete;
3520 }
bd555975 3521 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3522 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3523
c71bf099
YZ		 3524 if (root->fs_info->log_root_recovering) {
3525 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3526 goto no_delete;
3527 }
3528
76dda93c
YZ		 3529 if (inode->i_nlink > 0) {
3530 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3531 goto no_delete;
3532 }
3533
4289a667
JB		 3534 rsv = btrfs_alloc_block_rsv(root);
3535 if (!rsv) {
3536 btrfs_orphan_del(NULL, inode);
3537 goto no_delete;
3538 }
4a338542 3539 rsv->size = min_size;
726c35fa 3540 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3541
dbe674a9 3542 btrfs_i_size_write(inode, 0);
5f39d397 3543
4289a667
JB		 3544 /*
3545 * This is a bit simpler than btrfs_truncate since
3546 *
3547 * 1) We've already reserved our space for our orphan item in the
3548 * unlink.
3549 * 2) We're going to delete the inode item, so we don't need to update
3550 * it at all.
3551 *
3552 * So we just need to reserve some slack space in case we add bytes when
3553 * doing the truncate.
3554 */
8082510e 3555 while (1) {
aa38a711 3556 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
726c35fa
JB		 3557
3558 /*
3559 * Try and steal from the global reserve since we will
3560 * likely not use this space anyway, we want to try as
3561 * hard as possible to get this to work.
3562 */
3563 if (ret)
3564 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 3565
d68fc57b 3566 if (ret) {
4289a667 3567 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3568 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3569 btrfs_orphan_del(NULL, inode);
3570 btrfs_free_block_rsv(root, rsv);
3571 goto no_delete;
d68fc57b 3572 }
7b128766 3573
4289a667
JB		 3574 trans = btrfs_start_transaction(root, 0);
3575 if (IS_ERR(trans)) {
3576 btrfs_orphan_del(NULL, inode);
3577 btrfs_free_block_rsv(root, rsv);
3578 goto no_delete;
d68fc57b 3579 }
7b128766 3580
4289a667
JB		 3581 trans->block_rsv = rsv;
3582
d68fc57b 3583 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3584 if (ret != -EAGAIN)
3585 break;
85e21bac 3586
8082510e
YZ		 3587 nr = trans->blocks_used;
3588 btrfs_end_transaction(trans, root);
3589 trans = NULL;
3590 btrfs_btree_balance_dirty(root, nr);
3591 }
5f39d397 3592
4289a667
JB		 3593 btrfs_free_block_rsv(root, rsv);
3594
8082510e 3595 if (ret == 0) {
4289a667 3596 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3597 ret = btrfs_orphan_del(trans, inode);
3598 BUG_ON(ret);
3599 }
54aa1f4d 3600
4289a667 3601 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3602 if (!(root == root->fs_info->tree_root ||
3603 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3604 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3605
d3c2fdcf 3606 nr = trans->blocks_used;
54aa1f4d 3607 btrfs_end_transaction(trans, root);
d3c2fdcf 3608 btrfs_btree_balance_dirty(root, nr);
39279cc3 3609no_delete:
bd555975 3610 end_writeback(inode);
8082510e 3611 return;
39279cc3
CM		 3612}
3613
3614/*
3615 * this returns the key found in the dir entry in the location pointer.
3616 * If no dir entries were found, location->objectid is 0.
3617 */
3618static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3619 struct btrfs_key *location)
3620{
3621 const char *name = dentry->d_name.name;
3622 int namelen = dentry->d_name.len;
3623 struct btrfs_dir_item *di;
3624 struct btrfs_path *path;
3625 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3626 int ret = 0;
39279cc3
CM		 3627
3628 path = btrfs_alloc_path();
d8926bb3
MF		 3629 if (!path)
3630 return -ENOMEM;
3954401f 3631
33345d01 3632 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3633 namelen, 0);
0d9f7f3e
Y		 3634 if (IS_ERR(di))
3635 ret = PTR_ERR(di);
d397712b 3636
c704005d 3637 if (IS_ERR_OR_NULL(di))
3954401f 3638 goto out_err;
d397712b 3639
5f39d397 3640 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3641out:
39279cc3
CM		 3642 btrfs_free_path(path);
3643 return ret;
3954401f
CM		 3644out_err:
3645 location->objectid = 0;
3646 goto out;
39279cc3
CM		 3647}
3648
3649/*
3650 * when we hit a tree root in a directory, the btrfs part of the inode
3651 * needs to be changed to reflect the root directory of the tree root. This
3652 * is kind of like crossing a mount point.
3653 */
3654static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3655 struct inode *dir,
3656 struct dentry *dentry,
3657 struct btrfs_key *location,
3658 struct btrfs_root **sub_root)
39279cc3 3659{
4df27c4d
YZ		 3660 struct btrfs_path *path;
3661 struct btrfs_root *new_root;
3662 struct btrfs_root_ref *ref;
3663 struct extent_buffer *leaf;
3664 int ret;
3665 int err = 0;
39279cc3 3666
4df27c4d
YZ		 3667 path = btrfs_alloc_path();
3668 if (!path) {
3669 err = -ENOMEM;
3670 goto out;
3671 }
39279cc3 3672
4df27c4d
YZ		 3673 err = -ENOENT;
3674 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3675 BTRFS_I(dir)->root->root_key.objectid,
3676 location->objectid);
3677 if (ret) {
3678 if (ret < 0)
3679 err = ret;
3680 goto out;
3681 }
39279cc3 3682
4df27c4d
YZ		 3683 leaf = path->nodes[0];
3684 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3685 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3686 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3687 goto out;
39279cc3 3688
4df27c4d
YZ		 3689 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3690 (unsigned long)(ref + 1),
3691 dentry->d_name.len);
3692 if (ret)
3693 goto out;
3694
b3b4aa74 3695 btrfs_release_path(path);
4df27c4d
YZ		 3696
3697 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3698 if (IS_ERR(new_root)) {
3699 err = PTR_ERR(new_root);
3700 goto out;
3701 }
3702
3703 if (btrfs_root_refs(&new_root->root_item) == 0) {
3704 err = -ENOENT;
3705 goto out;
3706 }
3707
3708 *sub_root = new_root;
3709 location->objectid = btrfs_root_dirid(&new_root->root_item);
3710 location->type = BTRFS_INODE_ITEM_KEY;
3711 location->offset = 0;
3712 err = 0;
3713out:
3714 btrfs_free_path(path);
3715 return err;
39279cc3
CM		 3716}
3717
5d4f98a2
YZ		 3718static void inode_tree_add(struct inode *inode)
3719{
3720 struct btrfs_root *root = BTRFS_I(inode)->root;
3721 struct btrfs_inode *entry;
03e860bd
FNP		 3722 struct rb_node **p;
3723 struct rb_node *parent;
33345d01 3724 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3725again:
3726 p = &root->inode_tree.rb_node;
3727 parent = NULL;
5d4f98a2 3728
1d3382cb 3729 if (inode_unhashed(inode))
76dda93c
YZ		 3730 return;
3731
5d4f98a2
YZ		 3732 spin_lock(&root->inode_lock);
3733 while (*p) {
3734 parent = *p;
3735 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3736
33345d01 3737 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3738 p = &parent->rb_left;
33345d01 3739 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3740 p = &parent->rb_right;
5d4f98a2
YZ		 3741 else {
3742 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3743 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3744 rb_erase(parent, &root->inode_tree);
3745 RB_CLEAR_NODE(parent);
3746 spin_unlock(&root->inode_lock);
3747 goto again;
5d4f98a2
YZ		 3748 }
3749 }
3750 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3751 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3752 spin_unlock(&root->inode_lock);
3753}
3754
3755static void inode_tree_del(struct inode *inode)
3756{
3757 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3758 int empty = 0;
5d4f98a2 3759
03e860bd 3760 spin_lock(&root->inode_lock);
5d4f98a2 3761 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3762 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3763 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3764 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3765 }
03e860bd 3766 spin_unlock(&root->inode_lock);
76dda93c 3767
0af3d00b
JB		 3768 /*
3769 * Free space cache has inodes in the tree root, but the tree root has a
3770 * root_refs of 0, so this could end up dropping the tree root as a
3771 * snapshot, so we need the extra !root->fs_info->tree_root check to
3772 * make sure we don't drop it.
3773 */
3774 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3775 root != root->fs_info->tree_root) {
76dda93c
YZ		 3776 synchronize_srcu(&root->fs_info->subvol_srcu);
3777 spin_lock(&root->inode_lock);
3778 empty = RB_EMPTY_ROOT(&root->inode_tree);
3779 spin_unlock(&root->inode_lock);
3780 if (empty)
3781 btrfs_add_dead_root(root);
3782 }
3783}
3784
3785int btrfs_invalidate_inodes(struct btrfs_root *root)
3786{
3787 struct rb_node *node;
3788 struct rb_node *prev;
3789 struct btrfs_inode *entry;
3790 struct inode *inode;
3791 u64 objectid = 0;
3792
3793 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3794
3795 spin_lock(&root->inode_lock);
3796again:
3797 node = root->inode_tree.rb_node;
3798 prev = NULL;
3799 while (node) {
3800 prev = node;
3801 entry = rb_entry(node, struct btrfs_inode, rb_node);
3802
33345d01 3803 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 3804 node = node->rb_left;
33345d01 3805 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 3806 node = node->rb_right;
3807 else
3808 break;
3809 }
3810 if (!node) {
3811 while (prev) {
3812 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 3813 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 3814 node = prev;
3815 break;
3816 }
3817 prev = rb_next(prev);
3818 }
3819 }
3820 while (node) {
3821 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 3822 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 3823 inode = igrab(&entry->vfs_inode);
3824 if (inode) {
3825 spin_unlock(&root->inode_lock);
3826 if (atomic_read(&inode->i_count) > 1)
3827 d_prune_aliases(inode);
3828 /*
45321ac5 3829 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 3830 * the inode cache when its usage count
3831 * hits zero.
3832 */
3833 iput(inode);
3834 cond_resched();
3835 spin_lock(&root->inode_lock);
3836 goto again;
3837 }
3838
3839 if (cond_resched_lock(&root->inode_lock))
3840 goto again;
3841
3842 node = rb_next(node);
3843 }
3844 spin_unlock(&root->inode_lock);
3845 return 0;
5d4f98a2
YZ		 3846}
3847
e02119d5
CM		 3848static int btrfs_init_locked_inode(struct inode *inode, void *p)
3849{
3850 struct btrfs_iget_args *args = p;
3851 inode->i_ino = args->ino;
e02119d5 3852 BTRFS_I(inode)->root = args->root;
6a63209f 3853 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3854 return 0;
3855}
3856
3857static int btrfs_find_actor(struct inode *inode, void *opaque)
3858{
3859 struct btrfs_iget_args *args = opaque;
33345d01 3860 return args->ino == btrfs_ino(inode) &&
d397712b 3861 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3862}
3863
5d4f98a2
YZ		 3864static struct inode *btrfs_iget_locked(struct super_block *s,
3865 u64 objectid,
3866 struct btrfs_root *root)
39279cc3
CM		 3867{
3868 struct inode *inode;
3869 struct btrfs_iget_args args;
3870 args.ino = objectid;
3871 args.root = root;
3872
3873 inode = iget5_locked(s, objectid, btrfs_find_actor,
3874 btrfs_init_locked_inode,
3875 (void *)&args);
3876 return inode;
3877}
3878
1a54ef8c
BR		 3879/* Get an inode object given its location and corresponding root.
3880 * Returns in *is_new if the inode was read from disk
3881 */
3882struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 3883 struct btrfs_root *root, int *new)
1a54ef8c
BR		 3884{
3885 struct inode *inode;
3886
3887 inode = btrfs_iget_locked(s, location->objectid, root);
3888 if (!inode)
5d4f98a2 3889 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3890
3891 if (inode->i_state & I_NEW) {
3892 BTRFS_I(inode)->root = root;
3893 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3894 btrfs_read_locked_inode(inode);
1748f843
MF		 3895 if (!is_bad_inode(inode)) {
3896 inode_tree_add(inode);
3897 unlock_new_inode(inode);
3898 if (new)
3899 *new = 1;
3900 } else {
e0b6d65b
ST		 3901 unlock_new_inode(inode);
3902 iput(inode);
3903 inode = ERR_PTR(-ESTALE);
1748f843
MF		 3904 }
3905 }
3906
1a54ef8c
BR		 3907 return inode;
3908}
3909
4df27c4d
YZ		 3910static struct inode *new_simple_dir(struct super_block *s,
3911 struct btrfs_key *key,
3912 struct btrfs_root *root)
3913{
3914 struct inode *inode = new_inode(s);
3915
3916 if (!inode)
3917 return ERR_PTR(-ENOMEM);
3918
4df27c4d
YZ		 3919 BTRFS_I(inode)->root = root;
3920 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3921 BTRFS_I(inode)->dummy_inode = 1;
3922
3923 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3924 inode->i_op = &simple_dir_inode_operations;
3925 inode->i_fop = &simple_dir_operations;
3926 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3927 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3928
3929 return inode;
3930}
3931
3de4586c 3932struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3933{
d397712b 3934 struct inode *inode;
4df27c4d 3935 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 3936 struct btrfs_root *sub_root = root;
3937 struct btrfs_key location;
76dda93c 3938 int index;
b4aff1f8 3939 int ret = 0;
39279cc3
CM		 3940
3941 if (dentry->d_name.len > BTRFS_NAME_LEN)
3942 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3943
b4aff1f8
JB		 3944 if (unlikely(d_need_lookup(dentry))) {
3945 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3946 kfree(dentry->d_fsdata);
3947 dentry->d_fsdata = NULL;
a66e7cc6
JB		 3948 /* This thing is hashed, drop it for now */
3949 d_drop(dentry);
b4aff1f8
JB		 3950 } else {
3951 ret = btrfs_inode_by_name(dir, dentry, &location);
3952 }
5f39d397 3953
39279cc3
CM		 3954 if (ret < 0)
3955 return ERR_PTR(ret);
5f39d397 3956
4df27c4d
YZ		 3957 if (location.objectid == 0)
3958 return NULL;
3959
3960 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 3961 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 3962 return inode;
3963 }
3964
3965 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3966
76dda93c 3967 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 3968 ret = fixup_tree_root_location(root, dir, dentry,
3969 &location, &sub_root);
3970 if (ret < 0) {
3971 if (ret != -ENOENT)
3972 inode = ERR_PTR(ret);
3973 else
3974 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3975 } else {
73f73415 3976 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 3977 }
76dda93c
YZ		 3978 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3979
34d19bad 3980 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 3981 down_read(&root->fs_info->cleanup_work_sem);
3982 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 3983 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 3984 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 3985 if (ret)
3986 inode = ERR_PTR(ret);
c71bf099
YZ		 3987 }
3988
3de4586c
CM		 3989 return inode;
3990}
3991
fe15ce44 3992static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 3993{
3994 struct btrfs_root *root;
3995
efefb143
YZ		 3996 if (!dentry->d_inode && !IS_ROOT(dentry))
3997 dentry = dentry->d_parent;
76dda93c 3998
efefb143
YZ		 3999 if (dentry->d_inode) {
4000 root = BTRFS_I(dentry->d_inode)->root;
4001 if (btrfs_root_refs(&root->root_item) == 0)
4002 return 1;
4003 }
76dda93c
YZ		 4004 return 0;
4005}
4006
b4aff1f8
JB		 4007static void btrfs_dentry_release(struct dentry *dentry)
4008{
4009 if (dentry->d_fsdata)
4010 kfree(dentry->d_fsdata);
4011}
4012
3de4586c
CM		 4013static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4014 struct nameidata *nd)
4015{
a66e7cc6
JB		 4016 struct dentry *ret;
4017
4018 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4019 if (unlikely(d_need_lookup(dentry))) {
4020 spin_lock(&dentry->d_lock);
4021 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4022 spin_unlock(&dentry->d_lock);
4023 }
4024 return ret;
39279cc3
CM		 4025}
4026
16cdcec7 4027unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4028 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4029};
4030
cbdf5a24
DW		 4031static int btrfs_real_readdir(struct file *filp, void *dirent,
4032 filldir_t filldir)
39279cc3 4033{
6da6abae 4034 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4035 struct btrfs_root *root = BTRFS_I(inode)->root;
4036 struct btrfs_item *item;
4037 struct btrfs_dir_item *di;
4038 struct btrfs_key key;
5f39d397 4039 struct btrfs_key found_key;
39279cc3 4040 struct btrfs_path *path;
16cdcec7
MX		 4041 struct list_head ins_list;
4042 struct list_head del_list;
b4aff1f8 4043 struct qstr q;
39279cc3 4044 int ret;
5f39d397 4045 struct extent_buffer *leaf;
39279cc3 4046 int slot;
39279cc3
CM		 4047 unsigned char d_type;
4048 int over = 0;
4049 u32 di_cur;
4050 u32 di_total;
4051 u32 di_len;
4052 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4053 char tmp_name[32];
4054 char *name_ptr;
4055 int name_len;
16cdcec7 4056 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4057
4058 /* FIXME, use a real flag for deciding about the key type */
4059 if (root->fs_info->tree_root == root)
4060 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4061
3954401f
CM		 4062 /* special case for "." */
4063 if (filp->f_pos == 0) {
3765fefa
HS		 4064 over = filldir(dirent, ".", 1,
4065 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4066 if (over)
4067 return 0;
4068 filp->f_pos = 1;
4069 }
3954401f
CM		 4070 /* special case for .., just use the back ref */
4071 if (filp->f_pos == 1) {
5ecc7e5d 4072 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4073 over = filldir(dirent, "..", 2,
3765fefa 4074 filp->f_pos, pino, DT_DIR);
3954401f 4075 if (over)
49593bfa 4076 return 0;
3954401f
CM		 4077 filp->f_pos = 2;
4078 }
49593bfa 4079 path = btrfs_alloc_path();
16cdcec7
MX		 4080 if (!path)
4081 return -ENOMEM;
ff5714cc 4082
026fd317 4083 path->reada = 1;
49593bfa 4084
16cdcec7
MX		 4085 if (key_type == BTRFS_DIR_INDEX_KEY) {
4086 INIT_LIST_HEAD(&ins_list);
4087 INIT_LIST_HEAD(&del_list);
4088 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4089 }
4090
39279cc3
CM		 4091 btrfs_set_key_type(&key, key_type);
4092 key.offset = filp->f_pos;
33345d01 4093 key.objectid = btrfs_ino(inode);
5f39d397 4094
39279cc3
CM		 4095 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4096 if (ret < 0)
4097 goto err;
49593bfa
DW		 4098
4099 while (1) {
5f39d397 4100 leaf = path->nodes[0];
39279cc3 4101 slot = path->slots[0];
b9e03af0
LZ		 4102 if (slot >= btrfs_header_nritems(leaf)) {
4103 ret = btrfs_next_leaf(root, path);
4104 if (ret < 0)
4105 goto err;
4106 else if (ret > 0)
4107 break;
4108 continue;
39279cc3 4109 }
3de4586c 4110
5f39d397
CM		 4111 item = btrfs_item_nr(leaf, slot);
4112 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4113
4114 if (found_key.objectid != key.objectid)
39279cc3 4115 break;
5f39d397 4116 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4117 break;
5f39d397 4118 if (found_key.offset < filp->f_pos)
b9e03af0 4119 goto next;
16cdcec7
MX		 4120 if (key_type == BTRFS_DIR_INDEX_KEY &&
4121 btrfs_should_delete_dir_index(&del_list,
4122 found_key.offset))
4123 goto next;
5f39d397
CM		 4124
4125 filp->f_pos = found_key.offset;
16cdcec7 4126 is_curr = 1;
49593bfa 4127
39279cc3
CM		 4128 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4129 di_cur = 0;
5f39d397 4130 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4131
4132 while (di_cur < di_total) {
5f39d397 4133 struct btrfs_key location;
b4aff1f8 4134 struct dentry *tmp;
5f39d397 4135
22a94d44
JB		 4136 if (verify_dir_item(root, leaf, di))
4137 break;
4138
5f39d397 4139 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4140 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4141 name_ptr = tmp_name;
4142 } else {
4143 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4144 if (!name_ptr) {
4145 ret = -ENOMEM;
4146 goto err;
4147 }
5f39d397
CM		 4148 }
4149 read_extent_buffer(leaf, name_ptr,
4150 (unsigned long)(di + 1), name_len);
4151
4152 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4153 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4154
b4aff1f8
JB		 4155 q.name = name_ptr;
4156 q.len = name_len;
4157 q.hash = full_name_hash(q.name, q.len);
4158 tmp = d_lookup(filp->f_dentry, &q);
4159 if (!tmp) {
4160 struct btrfs_key *newkey;
4161
4162 newkey = kzalloc(sizeof(struct btrfs_key),
4163 GFP_NOFS);
4164 if (!newkey)
4165 goto no_dentry;
4166 tmp = d_alloc(filp->f_dentry, &q);
4167 if (!tmp) {
4168 kfree(newkey);
4169 dput(tmp);
4170 goto no_dentry;
4171 }
4172 memcpy(newkey, &location,
4173 sizeof(struct btrfs_key));
4174 tmp->d_fsdata = newkey;
4175 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4176 d_rehash(tmp);
4177 dput(tmp);
4178 } else {
4179 dput(tmp);
4180 }
4181no_dentry:
3de4586c
CM		 4182 /* is this a reference to our own snapshot? If so
4183 * skip it
4184 */
4185 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4186 location.objectid == root->root_key.objectid) {
4187 over = 0;
4188 goto skip;
4189 }
5f39d397 4190 over = filldir(dirent, name_ptr, name_len,
49593bfa 4191 found_key.offset, location.objectid,
39279cc3 4192 d_type);
5f39d397 4193
3de4586c 4194skip:
5f39d397
CM		 4195 if (name_ptr != tmp_name)
4196 kfree(name_ptr);
4197
39279cc3
CM		 4198 if (over)
4199 goto nopos;
5103e947 4200 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4201 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4202 di_cur += di_len;
4203 di = (struct btrfs_dir_item *)((char *)di + di_len);
4204 }
b9e03af0
LZ		 4205next:
4206 path->slots[0]++;
39279cc3 4207 }
49593bfa 4208
16cdcec7
MX		 4209 if (key_type == BTRFS_DIR_INDEX_KEY) {
4210 if (is_curr)
4211 filp->f_pos++;
4212 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4213 &ins_list);
4214 if (ret)
4215 goto nopos;
4216 }
4217
49593bfa 4218 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4219 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4220 /*
4221 * 32-bit glibc will use getdents64, but then strtol -
4222 * so the last number we can serve is this.
4223 */
4224 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4225 else
4226 filp->f_pos++;
39279cc3
CM		 4227nopos:
4228 ret = 0;
4229err:
16cdcec7
MX		 4230 if (key_type == BTRFS_DIR_INDEX_KEY)
4231 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4232 btrfs_free_path(path);
39279cc3
CM		 4233 return ret;
4234}
4235
a9185b41 4236int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4237{
4238 struct btrfs_root *root = BTRFS_I(inode)->root;
4239 struct btrfs_trans_handle *trans;
4240 int ret = 0;
0af3d00b 4241 bool nolock = false;
39279cc3 4242
8929ecfa 4243 if (BTRFS_I(inode)->dummy_inode)
4ca8b41e
CM		 4244 return 0;
4245
2cf8572d 4246 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
82d5902d 4247 nolock = true;
0af3d00b 4248
a9185b41 4249 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4250 if (nolock)
7a7eaa40 4251 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4252 else
7a7eaa40 4253 trans = btrfs_join_transaction(root);
3612b495
TI		 4254 if (IS_ERR(trans))
4255 return PTR_ERR(trans);
0af3d00b
JB		 4256 if (nolock)
4257 ret = btrfs_end_transaction_nolock(trans, root);
4258 else
4259 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4260 }
4261 return ret;
4262}
4263
4264/*
54aa1f4d 4265 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4266 * inode changes. But, it is most likely to find the inode in cache.
4267 * FIXME, needs more benchmarking...there are no reasons other than performance
4268 * to keep or drop this code.
4269 */
22c44fe6 4270int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 4271{
4272 struct btrfs_root *root = BTRFS_I(inode)->root;
4273 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4274 int ret;
4275
4276 if (BTRFS_I(inode)->dummy_inode)
22c44fe6 4277 return 0;
39279cc3 4278
7a7eaa40 4279 trans = btrfs_join_transaction(root);
22c44fe6
JB		 4280 if (IS_ERR(trans))
4281 return PTR_ERR(trans);
8929ecfa
YZ		 4282
4283 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4284 if (ret && ret == -ENOSPC) {
4285 /* whoops, lets try again with the full transaction */
4286 btrfs_end_transaction(trans, root);
4287 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 4288 if (IS_ERR(trans))
4289 return PTR_ERR(trans);
8929ecfa 4290
94b60442 4291 ret = btrfs_update_inode(trans, root, inode);
94b60442 4292 }
39279cc3 4293 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4294 if (BTRFS_I(inode)->delayed_node)
4295 btrfs_balance_delayed_items(root);
22c44fe6
JB		 4296
4297 return ret;
4298}
4299
4300/*
4301 * This is a copy of file_update_time. We need this so we can return error on
4302 * ENOSPC for updating the inode in the case of file write and mmap writes.
4303 */
4304int btrfs_update_time(struct file *file)
4305{
4306 struct inode *inode = file->f_path.dentry->d_inode;
4307 struct timespec now;
4308 int ret;
4309 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4310
4311 /* First try to exhaust all avenues to not sync */
4312 if (IS_NOCMTIME(inode))
4313 return 0;
4314
4315 now = current_fs_time(inode->i_sb);
4316 if (!timespec_equal(&inode->i_mtime, &now))
4317 sync_it = S_MTIME;
4318
4319 if (!timespec_equal(&inode->i_ctime, &now))
4320 sync_it |= S_CTIME;
4321
4322 if (IS_I_VERSION(inode))
4323 sync_it |= S_VERSION;
4324
4325 if (!sync_it)
4326 return 0;
4327
4328 /* Finally allowed to write? Takes lock. */
4329 if (mnt_want_write_file(file))
4330 return 0;
4331
4332 /* Only change inode inside the lock region */
4333 if (sync_it & S_VERSION)
4334 inode_inc_iversion(inode);
4335 if (sync_it & S_CTIME)
4336 inode->i_ctime = now;
4337 if (sync_it & S_MTIME)
4338 inode->i_mtime = now;
4339 ret = btrfs_dirty_inode(inode);
4340 if (!ret)
4341 mark_inode_dirty_sync(inode);
4342 mnt_drop_write(file->f_path.mnt);
4343 return ret;
39279cc3
CM		 4344}
4345
d352ac68
CM		 4346/*
4347 * find the highest existing sequence number in a directory
4348 * and then set the in-memory index_cnt variable to reflect
4349 * free sequence numbers
4350 */
aec7477b
JB		 4351static int btrfs_set_inode_index_count(struct inode *inode)
4352{
4353 struct btrfs_root *root = BTRFS_I(inode)->root;
4354 struct btrfs_key key, found_key;
4355 struct btrfs_path *path;
4356 struct extent_buffer *leaf;
4357 int ret;
4358
33345d01 4359 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4360 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4361 key.offset = (u64)-1;
4362
4363 path = btrfs_alloc_path();
4364 if (!path)
4365 return -ENOMEM;
4366
4367 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4368 if (ret < 0)
4369 goto out;
4370 /* FIXME: we should be able to handle this */
4371 if (ret == 0)
4372 goto out;
4373 ret = 0;
4374
4375 /*
4376 * MAGIC NUMBER EXPLANATION:
4377 * since we search a directory based on f_pos we have to start at 2
4378 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4379 * else has to start at 2
4380 */
4381 if (path->slots[0] == 0) {
4382 BTRFS_I(inode)->index_cnt = 2;
4383 goto out;
4384 }
4385
4386 path->slots[0]--;
4387
4388 leaf = path->nodes[0];
4389 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4390
33345d01 4391 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4392 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4393 BTRFS_I(inode)->index_cnt = 2;
4394 goto out;
4395 }
4396
4397 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4398out:
4399 btrfs_free_path(path);
4400 return ret;
4401}
4402
d352ac68
CM		 4403/*
4404 * helper to find a free sequence number in a given directory. This current
4405 * code is very simple, later versions will do smarter things in the btree
4406 */
3de4586c 4407int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4408{
4409 int ret = 0;
4410
4411 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4412 ret = btrfs_inode_delayed_dir_index_count(dir);
4413 if (ret) {
4414 ret = btrfs_set_inode_index_count(dir);
4415 if (ret)
4416 return ret;
4417 }
aec7477b
JB		 4418 }
4419
00e4e6b3 4420 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4421 BTRFS_I(dir)->index_cnt++;
4422
4423 return ret;
4424}
4425
39279cc3
CM		 4426static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4427 struct btrfs_root *root,
aec7477b 4428 struct inode *dir,
9c58309d 4429 const char *name, int name_len,
175a4eb7
AV		 4430 u64 ref_objectid, u64 objectid,
4431 umode_t mode, u64 *index)
39279cc3
CM		 4432{
4433 struct inode *inode;
5f39d397 4434 struct btrfs_inode_item *inode_item;
39279cc3 4435 struct btrfs_key *location;
5f39d397 4436 struct btrfs_path *path;
9c58309d
CM		 4437 struct btrfs_inode_ref *ref;
4438 struct btrfs_key key[2];
4439 u32 sizes[2];
4440 unsigned long ptr;
39279cc3
CM		 4441 int ret;
4442 int owner;
4443
5f39d397 4444 path = btrfs_alloc_path();
d8926bb3
MF		 4445 if (!path)
4446 return ERR_PTR(-ENOMEM);
5f39d397 4447
39279cc3 4448 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4449 if (!inode) {
4450 btrfs_free_path(path);
39279cc3 4451 return ERR_PTR(-ENOMEM);
8fb27640 4452 }
39279cc3 4453
581bb050
LZ		 4454 /*
4455 * we have to initialize this early, so we can reclaim the inode
4456 * number if we fail afterwards in this function.
4457 */
4458 inode->i_ino = objectid;
4459
aec7477b 4460 if (dir) {
1abe9b8a 4461 trace_btrfs_inode_request(dir);
4462
3de4586c 4463 ret = btrfs_set_inode_index(dir, index);
09771430 4464 if (ret) {
8fb27640 4465 btrfs_free_path(path);
09771430 4466 iput(inode);
aec7477b 4467 return ERR_PTR(ret);
09771430 4468 }
aec7477b
JB		 4469 }
4470 /*
4471 * index_cnt is ignored for everything but a dir,
4472 * btrfs_get_inode_index_count has an explanation for the magic
4473 * number
4474 */
4475 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4476 BTRFS_I(inode)->root = root;
e02119d5 4477 BTRFS_I(inode)->generation = trans->transid;
76195853 4478 inode->i_generation = BTRFS_I(inode)->generation;
6a63209f 4479 btrfs_set_inode_space_info(root, inode);
b888db2b 4480
569254b0 4481 if (S_ISDIR(mode))
39279cc3
CM		 4482 owner = 0;
4483 else
4484 owner = 1;
9c58309d
CM		 4485
4486 key[0].objectid = objectid;
4487 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4488 key[0].offset = 0;
4489
4490 key[1].objectid = objectid;
4491 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4492 key[1].offset = ref_objectid;
4493
4494 sizes[0] = sizeof(struct btrfs_inode_item);
4495 sizes[1] = name_len + sizeof(*ref);
4496
b9473439 4497 path->leave_spinning = 1;
9c58309d
CM		 4498 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4499 if (ret != 0)
5f39d397
CM		 4500 goto fail;
4501
ecc11fab 4502 inode_init_owner(inode, dir, mode);
a76a3cd4 4503 inode_set_bytes(inode, 0);
39279cc3 4504 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4505 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4506 struct btrfs_inode_item);
e02119d5 4507 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4508
4509 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4510 struct btrfs_inode_ref);
4511 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4512 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4513 ptr = (unsigned long)(ref + 1);
4514 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4515
5f39d397
CM		 4516 btrfs_mark_buffer_dirty(path->nodes[0]);
4517 btrfs_free_path(path);
4518
39279cc3
CM		 4519 location = &BTRFS_I(inode)->location;
4520 location->objectid = objectid;
39279cc3
CM		 4521 location->offset = 0;
4522 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4523
6cbff00f
CH		 4524 btrfs_inherit_iflags(inode, dir);
4525
569254b0 4526 if (S_ISREG(mode)) {
94272164
CM		 4527 if (btrfs_test_opt(root, NODATASUM))
4528 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4529 if (btrfs_test_opt(root, NODATACOW) ||
4530 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4531 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4532 }
4533
39279cc3 4534 insert_inode_hash(inode);
5d4f98a2 4535 inode_tree_add(inode);
1abe9b8a 4536
4537 trace_btrfs_inode_new(inode);
1973f0fa 4538 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4539
39279cc3 4540 return inode;
5f39d397 4541fail:
aec7477b
JB		 4542 if (dir)
4543 BTRFS_I(dir)->index_cnt--;
5f39d397 4544 btrfs_free_path(path);
09771430 4545 iput(inode);
5f39d397 4546 return ERR_PTR(ret);
39279cc3
CM		 4547}
4548
4549static inline u8 btrfs_inode_type(struct inode *inode)
4550{
4551 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4552}
4553
d352ac68
CM		 4554/*
4555 * utility function to add 'inode' into 'parent_inode' with
4556 * a give name and a given sequence number.
4557 * if 'add_backref' is true, also insert a backref from the
4558 * inode to the parent directory.
4559 */
e02119d5
CM		 4560int btrfs_add_link(struct btrfs_trans_handle *trans,
4561 struct inode *parent_inode, struct inode *inode,
4562 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4563{
4df27c4d 4564 int ret = 0;
39279cc3 4565 struct btrfs_key key;
e02119d5 4566 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4567 u64 ino = btrfs_ino(inode);
4568 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4569
33345d01 4570 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4571 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4572 } else {
33345d01 4573 key.objectid = ino;
4df27c4d
YZ		 4574 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4575 key.offset = 0;
4576 }
4577
33345d01 4578 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4579 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4580 key.objectid, root->root_key.objectid,
33345d01 4581 parent_ino, index, name, name_len);
4df27c4d 4582 } else if (add_backref) {
33345d01
LZ		 4583 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4584 parent_ino, index);
4df27c4d 4585 }
39279cc3 4586
39279cc3 4587 if (ret == 0) {
4df27c4d 4588 ret = btrfs_insert_dir_item(trans, root, name, name_len,
16cdcec7 4589 parent_inode, &key,
4df27c4d 4590 btrfs_inode_type(inode), index);
fe66a05a
CM		 4591 if (ret)
4592 goto fail_dir_item;
4df27c4d 4593
dbe674a9 4594 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 4595 name_len * 2);
79c44584 4596 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 4597 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 4598 }
4599 return ret;
fe66a05a
CM		 4600
4601fail_dir_item:
4602 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4603 u64 local_index;
4604 int err;
4605 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4606 key.objectid, root->root_key.objectid,
4607 parent_ino, &local_index, name, name_len);
4608
4609 } else if (add_backref) {
4610 u64 local_index;
4611 int err;
4612
4613 err = btrfs_del_inode_ref(trans, root, name, name_len,
4614 ino, parent_ino, &local_index);
4615 }
4616 return ret;
39279cc3
CM		 4617}
4618
4619static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4620 struct inode *dir, struct dentry *dentry,
4621 struct inode *inode, int backref, u64 index)
39279cc3 4622{
a1b075d2
JB		 4623 int err = btrfs_add_link(trans, dir, inode,
4624 dentry->d_name.name, dentry->d_name.len,
4625 backref, index);
39279cc3
CM		 4626 if (err > 0)
4627 err = -EEXIST;
4628 return err;
4629}
4630
618e21d5 4631static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 4632 umode_t mode, dev_t rdev)
618e21d5
JB		 4633{
4634 struct btrfs_trans_handle *trans;
4635 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4636 struct inode *inode = NULL;
618e21d5
JB		 4637 int err;
4638 int drop_inode = 0;
4639 u64 objectid;
1832a6d5 4640 unsigned long nr = 0;
00e4e6b3 4641 u64 index = 0;
618e21d5
JB		 4642
4643 if (!new_valid_dev(rdev))
4644 return -EINVAL;
4645
9ed74f2d
JB		 4646 /*
4647 * 2 for inode item and ref
4648 * 2 for dir items
4649 * 1 for xattr if selinux is on
4650 */
a22285a6
YZ		 4651 trans = btrfs_start_transaction(root, 5);
4652 if (IS_ERR(trans))
4653 return PTR_ERR(trans);
1832a6d5 4654
581bb050
LZ		 4655 err = btrfs_find_free_ino(root, &objectid);
4656 if (err)
4657 goto out_unlock;
4658
aec7477b 4659 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4660 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4661 mode, &index);
7cf96da3
TI		 4662 if (IS_ERR(inode)) {
4663 err = PTR_ERR(inode);
618e21d5 4664 goto out_unlock;
7cf96da3 4665 }
618e21d5 4666
2a7dba39 4667 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4668 if (err) {
4669 drop_inode = 1;
4670 goto out_unlock;
4671 }
4672
ad19db71
CS		 4673 /*
4674 * If the active LSM wants to access the inode during
4675 * d_instantiate it needs these. Smack checks to see
4676 * if the filesystem supports xattrs by looking at the
4677 * ops vector.
4678 */
4679
4680 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 4681 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4682 if (err)
4683 drop_inode = 1;
4684 else {
618e21d5 4685 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4686 btrfs_update_inode(trans, root, inode);
08c422c2 4687 d_instantiate(dentry, inode);
618e21d5 4688 }
618e21d5 4689out_unlock:
d3c2fdcf 4690 nr = trans->blocks_used;
7ad85bb7 4691 btrfs_end_transaction(trans, root);
a22285a6 4692 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4693 if (drop_inode) {
4694 inode_dec_link_count(inode);
4695 iput(inode);
4696 }
618e21d5
JB		 4697 return err;
4698}
4699
39279cc3 4700static int btrfs_create(struct inode *dir, struct dentry *dentry,
4acdaf27 4701 umode_t mode, struct nameidata *nd)
39279cc3
CM		 4702{
4703 struct btrfs_trans_handle *trans;
4704 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4705 struct inode *inode = NULL;
39279cc3 4706 int drop_inode = 0;
a22285a6 4707 int err;
1832a6d5 4708 unsigned long nr = 0;
39279cc3 4709 u64 objectid;
00e4e6b3 4710 u64 index = 0;
39279cc3 4711
9ed74f2d
JB		 4712 /*
4713 * 2 for inode item and ref
4714 * 2 for dir items
4715 * 1 for xattr if selinux is on
4716 */
a22285a6
YZ		 4717 trans = btrfs_start_transaction(root, 5);
4718 if (IS_ERR(trans))
4719 return PTR_ERR(trans);
9ed74f2d 4720
581bb050
LZ		 4721 err = btrfs_find_free_ino(root, &objectid);
4722 if (err)
4723 goto out_unlock;
4724
aec7477b 4725 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4726 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4727 mode, &index);
7cf96da3
TI		 4728 if (IS_ERR(inode)) {
4729 err = PTR_ERR(inode);
39279cc3 4730 goto out_unlock;
7cf96da3 4731 }
39279cc3 4732
2a7dba39 4733 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4734 if (err) {
4735 drop_inode = 1;
4736 goto out_unlock;
4737 }
4738
ad19db71
CS		 4739 /*
4740 * If the active LSM wants to access the inode during
4741 * d_instantiate it needs these. Smack checks to see
4742 * if the filesystem supports xattrs by looking at the
4743 * ops vector.
4744 */
4745 inode->i_fop = &btrfs_file_operations;
4746 inode->i_op = &btrfs_file_inode_operations;
4747
a1b075d2 4748 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4749 if (err)
4750 drop_inode = 1;
4751 else {
4752 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4753 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 4754 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
08c422c2 4755 d_instantiate(dentry, inode);
39279cc3 4756 }
39279cc3 4757out_unlock:
d3c2fdcf 4758 nr = trans->blocks_used;
7ad85bb7 4759 btrfs_end_transaction(trans, root);
39279cc3
CM		 4760 if (drop_inode) {
4761 inode_dec_link_count(inode);
4762 iput(inode);
4763 }
d3c2fdcf 4764 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4765 return err;
4766}
4767
4768static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4769 struct dentry *dentry)
4770{
4771 struct btrfs_trans_handle *trans;
4772 struct btrfs_root *root = BTRFS_I(dir)->root;
4773 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4774 u64 index;
1832a6d5 4775 unsigned long nr = 0;
39279cc3
CM		 4776 int err;
4777 int drop_inode = 0;
4778
4a8be425
TH		 4779 /* do not allow sys_link's with other subvols of the same device */
4780 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4781 return -EXDEV;
4a8be425 4782
c055e99e
AV		 4783 if (inode->i_nlink == ~0U)
4784 return -EMLINK;
4a8be425 4785
3de4586c 4786 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4787 if (err)
4788 goto fail;
4789
a22285a6 4790 /*
7e6b6465 4791 * 2 items for inode and inode ref
a22285a6 4792 * 2 items for dir items
7e6b6465 4793 * 1 item for parent inode
a22285a6 4794 */
7e6b6465 4795 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4796 if (IS_ERR(trans)) {
4797 err = PTR_ERR(trans);
4798 goto fail;
4799 }
5f39d397 4800
3153495d
MX		 4801 btrfs_inc_nlink(inode);
4802 inode->i_ctime = CURRENT_TIME;
7de9c6ee 4803 ihold(inode);
aec7477b 4804
a1b075d2 4805 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 4806
a5719521 4807 if (err) {
54aa1f4d 4808 drop_inode = 1;
a5719521 4809 } else {
10d9f309 4810 struct dentry *parent = dentry->d_parent;
a5719521
YZ		 4811 err = btrfs_update_inode(trans, root, inode);
4812 BUG_ON(err);
08c422c2 4813 d_instantiate(dentry, inode);
6a912213 4814 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 4815 }
39279cc3 4816
d3c2fdcf 4817 nr = trans->blocks_used;
7ad85bb7 4818 btrfs_end_transaction(trans, root);
1832a6d5 4819fail:
39279cc3
CM		 4820 if (drop_inode) {
4821 inode_dec_link_count(inode);
4822 iput(inode);
4823 }
d3c2fdcf 4824 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4825 return err;
4826}
4827
18bb1db3 4828static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 4829{
b9d86667 4830 struct inode *inode = NULL;
39279cc3
CM		 4831 struct btrfs_trans_handle *trans;
4832 struct btrfs_root *root = BTRFS_I(dir)->root;
4833 int err = 0;
4834 int drop_on_err = 0;
b9d86667 4835 u64 objectid = 0;
00e4e6b3 4836 u64 index = 0;
d3c2fdcf 4837 unsigned long nr = 1;
39279cc3 4838
9ed74f2d
JB		 4839 /*
4840 * 2 items for inode and ref
4841 * 2 items for dir items
4842 * 1 for xattr if selinux is on
4843 */
a22285a6
YZ		 4844 trans = btrfs_start_transaction(root, 5);
4845 if (IS_ERR(trans))
4846 return PTR_ERR(trans);
39279cc3 4847
581bb050
LZ		 4848 err = btrfs_find_free_ino(root, &objectid);
4849 if (err)
4850 goto out_fail;
4851
aec7477b 4852 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4853 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4854 S_IFDIR | mode, &index);
39279cc3
CM		 4855 if (IS_ERR(inode)) {
4856 err = PTR_ERR(inode);
4857 goto out_fail;
4858 }
5f39d397 4859
39279cc3 4860 drop_on_err = 1;
33268eaf 4861
2a7dba39 4862 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4863 if (err)
4864 goto out_fail;
4865
39279cc3
CM		 4866 inode->i_op = &btrfs_dir_inode_operations;
4867 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 4868
dbe674a9 4869 btrfs_i_size_write(inode, 0);
39279cc3
CM		 4870 err = btrfs_update_inode(trans, root, inode);
4871 if (err)
4872 goto out_fail;
5f39d397 4873
a1b075d2
JB		 4874 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4875 dentry->d_name.len, 0, index);
39279cc3
CM		 4876 if (err)
4877 goto out_fail;
5f39d397 4878
39279cc3
CM		 4879 d_instantiate(dentry, inode);
4880 drop_on_err = 0;
39279cc3
CM		 4881
4882out_fail:
d3c2fdcf 4883 nr = trans->blocks_used;
7ad85bb7 4884 btrfs_end_transaction(trans, root);
39279cc3
CM		 4885 if (drop_on_err)
4886 iput(inode);
d3c2fdcf 4887 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4888 return err;
4889}
4890
d352ac68
CM		 4891/* helper for btfs_get_extent. Given an existing extent in the tree,
4892 * and an extent that you want to insert, deal with overlap and insert
4893 * the new extent into the tree.
4894 */
3b951516
CM		 4895static int merge_extent_mapping(struct extent_map_tree *em_tree,
4896 struct extent_map *existing,
e6dcd2dc
CM		 4897 struct extent_map *em,
4898 u64 map_start, u64 map_len)
3b951516
CM		 4899{
4900 u64 start_diff;
3b951516 4901
e6dcd2dc
CM		 4902 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4903 start_diff = map_start - em->start;
4904 em->start = map_start;
4905 em->len = map_len;
c8b97818
CM		 4906 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4907 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 4908 em->block_start += start_diff;
c8b97818
CM		 4909 em->block_len -= start_diff;
4910 }
e6dcd2dc 4911 return add_extent_mapping(em_tree, em);
3b951516
CM		 4912}
4913
c8b97818
CM		 4914static noinline int uncompress_inline(struct btrfs_path *path,
4915 struct inode *inode, struct page *page,
4916 size_t pg_offset, u64 extent_offset,
4917 struct btrfs_file_extent_item *item)
4918{
4919 int ret;
4920 struct extent_buffer *leaf = path->nodes[0];
4921 char *tmp;
4922 size_t max_size;
4923 unsigned long inline_size;
4924 unsigned long ptr;
261507a0 4925 int compress_type;
c8b97818
CM		 4926
4927 WARN_ON(pg_offset != 0);
261507a0 4928 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 4929 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4930 inline_size = btrfs_file_extent_inline_item_len(leaf,
4931 btrfs_item_nr(leaf, path->slots[0]));
4932 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 4933 if (!tmp)
4934 return -ENOMEM;
c8b97818
CM		 4935 ptr = btrfs_file_extent_inline_start(item);
4936
4937 read_extent_buffer(leaf, tmp, ptr, inline_size);
4938
5b050f04 4939 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 4940 ret = btrfs_decompress(compress_type, tmp, page,
4941 extent_offset, inline_size, max_size);
c8b97818
CM		 4942 if (ret) {
4943 char *kaddr = kmap_atomic(page, KM_USER0);
4944 unsigned long copy_size = min_t(u64,
4945 PAGE_CACHE_SIZE - pg_offset,
4946 max_size - extent_offset);
4947 memset(kaddr + pg_offset, 0, copy_size);
4948 kunmap_atomic(kaddr, KM_USER0);
4949 }
4950 kfree(tmp);
4951 return 0;
4952}
4953
d352ac68
CM		 4954/*
4955 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4956 * the ugly parts come from merging extents from the disk with the in-ram
4957 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4958 * where the in-ram extents might be locked pending data=ordered completion.
4959 *
4960 * This also copies inline extents directly into the page.
4961 */
d397712b 4962
a52d9a80 4963struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4964 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4965 int create)
4966{
4967 int ret;
4968 int err = 0;
db94535d 4969 u64 bytenr;
a52d9a80
CM		 4970 u64 extent_start = 0;
4971 u64 extent_end = 0;
33345d01 4972 u64 objectid = btrfs_ino(inode);
a52d9a80 4973 u32 found_type;
f421950f 4974 struct btrfs_path *path = NULL;
a52d9a80
CM		 4975 struct btrfs_root *root = BTRFS_I(inode)->root;
4976 struct btrfs_file_extent_item *item;
5f39d397
CM		 4977 struct extent_buffer *leaf;
4978 struct btrfs_key found_key;
a52d9a80
CM		 4979 struct extent_map *em = NULL;
4980 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4981 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4982 struct btrfs_trans_handle *trans = NULL;
261507a0 4983 int compress_type;
a52d9a80 4984
a52d9a80 4985again:
890871be 4986 read_lock(&em_tree->lock);
d1310b2e 4987 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4988 if (em)
4989 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 4990 read_unlock(&em_tree->lock);
d1310b2e 4991
a52d9a80 4992 if (em) {
e1c4b745
CM		 4993 if (em->start > start || em->start + em->len <= start)
4994 free_extent_map(em);
4995 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4996 free_extent_map(em);
4997 else
4998 goto out;
a52d9a80 4999 }
172ddd60 5000 em = alloc_extent_map();
a52d9a80 5001 if (!em) {
d1310b2e
CM		 5002 err = -ENOMEM;
5003 goto out;
a52d9a80 5004 }
e6dcd2dc 5005 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5006 em->start = EXTENT_MAP_HOLE;
445a6944 5007 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5008 em->len = (u64)-1;
c8b97818 5009 em->block_len = (u64)-1;
f421950f
CM		 5010
5011 if (!path) {
5012 path = btrfs_alloc_path();
026fd317
JB		 5013 if (!path) {
5014 err = -ENOMEM;
5015 goto out;
5016 }
5017 /*
5018 * Chances are we'll be called again, so go ahead and do
5019 * readahead
5020 */
5021 path->reada = 1;
f421950f
CM		 5022 }
5023
179e29e4
CM		 5024 ret = btrfs_lookup_file_extent(trans, root, path,
5025 objectid, start, trans != NULL);
a52d9a80
CM		 5026 if (ret < 0) {
5027 err = ret;
5028 goto out;
5029 }
5030
5031 if (ret != 0) {
5032 if (path->slots[0] == 0)
5033 goto not_found;
5034 path->slots[0]--;
5035 }
5036
5f39d397
CM		 5037 leaf = path->nodes[0];
5038 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5039 struct btrfs_file_extent_item);
a52d9a80 5040 /* are we inside the extent that was found? */
5f39d397
CM		 5041 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5042 found_type = btrfs_key_type(&found_key);
5043 if (found_key.objectid != objectid ||
a52d9a80
CM		 5044 found_type != BTRFS_EXTENT_DATA_KEY) {
5045 goto not_found;
5046 }
5047
5f39d397
CM		 5048 found_type = btrfs_file_extent_type(leaf, item);
5049 extent_start = found_key.offset;
261507a0 5050 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5051 if (found_type == BTRFS_FILE_EXTENT_REG ||
5052 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5053 extent_end = extent_start +
db94535d 5054 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5055 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5056 size_t size;
5057 size = btrfs_file_extent_inline_len(leaf, item);
5058 extent_end = (extent_start + size + root->sectorsize - 1) &
5059 ~((u64)root->sectorsize - 1);
5060 }
5061
5062 if (start >= extent_end) {
5063 path->slots[0]++;
5064 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5065 ret = btrfs_next_leaf(root, path);
5066 if (ret < 0) {
5067 err = ret;
5068 goto out;
a52d9a80 5069 }
9036c102
YZ		 5070 if (ret > 0)
5071 goto not_found;
5072 leaf = path->nodes[0];
a52d9a80 5073 }
9036c102
YZ		 5074 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5075 if (found_key.objectid != objectid ||
5076 found_key.type != BTRFS_EXTENT_DATA_KEY)
5077 goto not_found;
5078 if (start + len <= found_key.offset)
5079 goto not_found;
5080 em->start = start;
5081 em->len = found_key.offset - start;
5082 goto not_found_em;
5083 }
5084
d899e052
YZ		 5085 if (found_type == BTRFS_FILE_EXTENT_REG ||
5086 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5087 em->start = extent_start;
5088 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5089 em->orig_start = extent_start -
5090 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5091 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5092 if (bytenr == 0) {
5f39d397 5093 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5094 goto insert;
5095 }
261507a0 5096 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5097 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5098 em->compress_type = compress_type;
c8b97818
CM		 5099 em->block_start = bytenr;
5100 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5101 item);
5102 } else {
5103 bytenr += btrfs_file_extent_offset(leaf, item);
5104 em->block_start = bytenr;
5105 em->block_len = em->len;
d899e052
YZ		 5106 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5107 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5108 }
a52d9a80
CM		 5109 goto insert;
5110 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5111 unsigned long ptr;
a52d9a80 5112 char *map;
3326d1b0
CM		 5113 size_t size;
5114 size_t extent_offset;
5115 size_t copy_size;
a52d9a80 5116
689f9346 5117 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5118 if (!page || create) {
689f9346 5119 em->start = extent_start;
9036c102 5120 em->len = extent_end - extent_start;
689f9346
Y		 5121 goto out;
5122 }
5f39d397 5123
9036c102
YZ		 5124 size = btrfs_file_extent_inline_len(leaf, item);
5125 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5126 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5127 size - extent_offset);
3326d1b0 5128 em->start = extent_start + extent_offset;
70dec807
CM		 5129 em->len = (copy_size + root->sectorsize - 1) &
5130 ~((u64)root->sectorsize - 1);
ff5b7ee3 5131 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5132 if (compress_type) {
c8b97818 5133 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5134 em->compress_type = compress_type;
5135 }
689f9346 5136 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5137 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5138 if (btrfs_file_extent_compression(leaf, item) !=
5139 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5140 ret = uncompress_inline(path, inode, page,
5141 pg_offset,
5142 extent_offset, item);
5143 BUG_ON(ret);
5144 } else {
5145 map = kmap(page);
5146 read_extent_buffer(leaf, map + pg_offset, ptr,
5147 copy_size);
93c82d57
CM		 5148 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5149 memset(map + pg_offset + copy_size, 0,
5150 PAGE_CACHE_SIZE - pg_offset -
5151 copy_size);
5152 }
c8b97818
CM		 5153 kunmap(page);
5154 }
179e29e4
CM		 5155 flush_dcache_page(page);
5156 } else if (create && PageUptodate(page)) {
6bf7e080 5157 BUG();
179e29e4
CM		 5158 if (!trans) {
5159 kunmap(page);
5160 free_extent_map(em);
5161 em = NULL;
ff5714cc 5162
b3b4aa74 5163 btrfs_release_path(path);
7a7eaa40 5164 trans = btrfs_join_transaction(root);
ff5714cc 5165
3612b495
TI		 5166 if (IS_ERR(trans))
5167 return ERR_CAST(trans);
179e29e4
CM		 5168 goto again;
5169 }
c8b97818 5170 map = kmap(page);
70dec807 5171 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5172 copy_size);
c8b97818 5173 kunmap(page);
179e29e4 5174 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5175 }
d1310b2e 5176 set_extent_uptodate(io_tree, em->start,
507903b8 5177 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5178 goto insert;
5179 } else {
d397712b 5180 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5181 WARN_ON(1);
5182 }
5183not_found:
5184 em->start = start;
d1310b2e 5185 em->len = len;
a52d9a80 5186not_found_em:
5f39d397 5187 em->block_start = EXTENT_MAP_HOLE;
9036c102 5188 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5189insert:
b3b4aa74 5190 btrfs_release_path(path);
d1310b2e 5191 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5192 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5193 "[%llu %llu]\n", (unsigned long long)em->start,
5194 (unsigned long long)em->len,
5195 (unsigned long long)start,
5196 (unsigned long long)len);
a52d9a80
CM		 5197 err = -EIO;
5198 goto out;
5199 }
d1310b2e
CM		 5200
5201 err = 0;
890871be 5202 write_lock(&em_tree->lock);
a52d9a80 5203 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5204 /* it is possible that someone inserted the extent into the tree
5205 * while we had the lock dropped. It is also possible that
5206 * an overlapping map exists in the tree
5207 */
a52d9a80 5208 if (ret == -EEXIST) {
3b951516 5209 struct extent_map *existing;
e6dcd2dc
CM		 5210
5211 ret = 0;
5212
3b951516 5213 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5214 if (existing && (existing->start > start ||
5215 existing->start + existing->len <= start)) {
5216 free_extent_map(existing);
5217 existing = NULL;
5218 }
3b951516
CM		 5219 if (!existing) {
5220 existing = lookup_extent_mapping(em_tree, em->start,
5221 em->len);
5222 if (existing) {
5223 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5224 em, start,
5225 root->sectorsize);
3b951516
CM		 5226 free_extent_map(existing);
5227 if (err) {
5228 free_extent_map(em);
5229 em = NULL;
5230 }
5231 } else {
5232 err = -EIO;
3b951516
CM		 5233 free_extent_map(em);
5234 em = NULL;
5235 }
5236 } else {
5237 free_extent_map(em);
5238 em = existing;
e6dcd2dc 5239 err = 0;
a52d9a80 5240 }
a52d9a80 5241 }
890871be 5242 write_unlock(&em_tree->lock);
a52d9a80 5243out:
1abe9b8a 5244
5245 trace_btrfs_get_extent(root, em);
5246
f421950f
CM		 5247 if (path)
5248 btrfs_free_path(path);
a52d9a80
CM		 5249 if (trans) {
5250 ret = btrfs_end_transaction(trans, root);
d397712b 5251 if (!err)
a52d9a80
CM		 5252 err = ret;
5253 }
a52d9a80
CM		 5254 if (err) {
5255 free_extent_map(em);
a52d9a80
CM		 5256 return ERR_PTR(err);
5257 }
5258 return em;
5259}
5260
ec29ed5b
CM		 5261struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5262 size_t pg_offset, u64 start, u64 len,
5263 int create)
5264{
5265 struct extent_map *em;
5266 struct extent_map *hole_em = NULL;
5267 u64 range_start = start;
5268 u64 end;
5269 u64 found;
5270 u64 found_end;
5271 int err = 0;
5272
5273 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5274 if (IS_ERR(em))
5275 return em;
5276 if (em) {
5277 /*
5278 * if our em maps to a hole, there might
5279 * actually be delalloc bytes behind it
5280 */
5281 if (em->block_start != EXTENT_MAP_HOLE)
5282 return em;
5283 else
5284 hole_em = em;
5285 }
5286
5287 /* check to see if we've wrapped (len == -1 or similar) */
5288 end = start + len;
5289 if (end < start)
5290 end = (u64)-1;
5291 else
5292 end -= 1;
5293
5294 em = NULL;
5295
5296 /* ok, we didn't find anything, lets look for delalloc */
5297 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5298 end, len, EXTENT_DELALLOC, 1);
5299 found_end = range_start + found;
5300 if (found_end < range_start)
5301 found_end = (u64)-1;
5302
5303 /*
5304 * we didn't find anything useful, return
5305 * the original results from get_extent()
5306 */
5307 if (range_start > end || found_end <= start) {
5308 em = hole_em;
5309 hole_em = NULL;
5310 goto out;
5311 }
5312
5313 /* adjust the range_start to make sure it doesn't
5314 * go backwards from the start they passed in
5315 */
5316 range_start = max(start,range_start);
5317 found = found_end - range_start;
5318
5319 if (found > 0) {
5320 u64 hole_start = start;
5321 u64 hole_len = len;
5322
172ddd60 5323 em = alloc_extent_map();
ec29ed5b
CM		 5324 if (!em) {
5325 err = -ENOMEM;
5326 goto out;
5327 }
5328 /*
5329 * when btrfs_get_extent can't find anything it
5330 * returns one huge hole
5331 *
5332 * make sure what it found really fits our range, and
5333 * adjust to make sure it is based on the start from
5334 * the caller
5335 */
5336 if (hole_em) {
5337 u64 calc_end = extent_map_end(hole_em);
5338
5339 if (calc_end <= start || (hole_em->start > end)) {
5340 free_extent_map(hole_em);
5341 hole_em = NULL;
5342 } else {
5343 hole_start = max(hole_em->start, start);
5344 hole_len = calc_end - hole_start;
5345 }
5346 }
5347 em->bdev = NULL;
5348 if (hole_em && range_start > hole_start) {
5349 /* our hole starts before our delalloc, so we
5350 * have to return just the parts of the hole
5351 * that go until the delalloc starts
5352 */
5353 em->len = min(hole_len,
5354 range_start - hole_start);
5355 em->start = hole_start;
5356 em->orig_start = hole_start;
5357 /*
5358 * don't adjust block start at all,
5359 * it is fixed at EXTENT_MAP_HOLE
5360 */
5361 em->block_start = hole_em->block_start;
5362 em->block_len = hole_len;
5363 } else {
5364 em->start = range_start;
5365 em->len = found;
5366 em->orig_start = range_start;
5367 em->block_start = EXTENT_MAP_DELALLOC;
5368 em->block_len = found;
5369 }
5370 } else if (hole_em) {
5371 return hole_em;
5372 }
5373out:
5374
5375 free_extent_map(hole_em);
5376 if (err) {
5377 free_extent_map(em);
5378 return ERR_PTR(err);
5379 }
5380 return em;
5381}
5382
4b46fce2 5383static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5384 struct extent_map *em,
4b46fce2
JB		 5385 u64 start, u64 len)
5386{
5387 struct btrfs_root *root = BTRFS_I(inode)->root;
5388 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5389 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5390 struct btrfs_key ins;
5391 u64 alloc_hint;
5392 int ret;
16d299ac 5393 bool insert = false;
4b46fce2 5394
16d299ac
JB		 5395 /*
5396 * Ok if the extent map we looked up is a hole and is for the exact
5397 * range we want, there is no reason to allocate a new one, however if
5398 * it is not right then we need to free this one and drop the cache for
5399 * our range.
5400 */
5401 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5402 em->len != len) {
5403 free_extent_map(em);
5404 em = NULL;
5405 insert = true;
5406 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5407 }
4b46fce2 5408
7a7eaa40 5409 trans = btrfs_join_transaction(root);
3612b495
TI		 5410 if (IS_ERR(trans))
5411 return ERR_CAST(trans);
4b46fce2 5412
4cb5300b
CM		 5413 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5414 btrfs_add_inode_defrag(trans, inode);
5415
4b46fce2
JB		 5416 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5417
5418 alloc_hint = get_extent_allocation_hint(inode, start, len);
5419 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5420 alloc_hint, (u64)-1, &ins, 1);
5421 if (ret) {
5422 em = ERR_PTR(ret);
5423 goto out;
5424 }
5425
4b46fce2 5426 if (!em) {
172ddd60 5427 em = alloc_extent_map();
16d299ac
JB		 5428 if (!em) {
5429 em = ERR_PTR(-ENOMEM);
5430 goto out;
5431 }
4b46fce2
JB		 5432 }
5433
5434 em->start = start;
5435 em->orig_start = em->start;
5436 em->len = ins.offset;
5437
5438 em->block_start = ins.objectid;
5439 em->block_len = ins.offset;
5440 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5441
5442 /*
5443 * We need to do this because if we're using the original em we searched
5444 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5445 */
5446 em->flags = 0;
4b46fce2
JB		 5447 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5448
16d299ac 5449 while (insert) {
4b46fce2
JB		 5450 write_lock(&em_tree->lock);
5451 ret = add_extent_mapping(em_tree, em);
5452 write_unlock(&em_tree->lock);
5453 if (ret != -EEXIST)
5454 break;
5455 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5456 }
5457
5458 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5459 ins.offset, ins.offset, 0);
5460 if (ret) {
5461 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5462 em = ERR_PTR(ret);
5463 }
5464out:
5465 btrfs_end_transaction(trans, root);
5466 return em;
5467}
5468
46bfbb5c
CM		 5469/*
5470 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5471 * block must be cow'd
5472 */
5473static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5474 struct inode *inode, u64 offset, u64 len)
5475{
5476 struct btrfs_path *path;
5477 int ret;
5478 struct extent_buffer *leaf;
5479 struct btrfs_root *root = BTRFS_I(inode)->root;
5480 struct btrfs_file_extent_item *fi;
5481 struct btrfs_key key;
5482 u64 disk_bytenr;
5483 u64 backref_offset;
5484 u64 extent_end;
5485 u64 num_bytes;
5486 int slot;
5487 int found_type;
5488
5489 path = btrfs_alloc_path();
5490 if (!path)
5491 return -ENOMEM;
5492
33345d01 5493 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5494 offset, 0);
5495 if (ret < 0)
5496 goto out;
5497
5498 slot = path->slots[0];
5499 if (ret == 1) {
5500 if (slot == 0) {
5501 /* can't find the item, must cow */
5502 ret = 0;
5503 goto out;
5504 }
5505 slot--;
5506 }
5507 ret = 0;
5508 leaf = path->nodes[0];
5509 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5510 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5511 key.type != BTRFS_EXTENT_DATA_KEY) {
5512 /* not our file or wrong item type, must cow */
5513 goto out;
5514 }
5515
5516 if (key.offset > offset) {
5517 /* Wrong offset, must cow */
5518 goto out;
5519 }
5520
5521 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5522 found_type = btrfs_file_extent_type(leaf, fi);
5523 if (found_type != BTRFS_FILE_EXTENT_REG &&
5524 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5525 /* not a regular extent, must cow */
5526 goto out;
5527 }
5528 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5529 backref_offset = btrfs_file_extent_offset(leaf, fi);
5530
5531 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5532 if (extent_end < offset + len) {
5533 /* extent doesn't include our full range, must cow */
5534 goto out;
5535 }
5536
5537 if (btrfs_extent_readonly(root, disk_bytenr))
5538 goto out;
5539
5540 /*
5541 * look for other files referencing this extent, if we
5542 * find any we must cow
5543 */
33345d01 5544 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5545 key.offset - backref_offset, disk_bytenr))
5546 goto out;
5547
5548 /*
5549 * adjust disk_bytenr and num_bytes to cover just the bytes
5550 * in this extent we are about to write. If there
5551 * are any csums in that range we have to cow in order
5552 * to keep the csums correct
5553 */
5554 disk_bytenr += backref_offset;
5555 disk_bytenr += offset - key.offset;
5556 num_bytes = min(offset + len, extent_end) - offset;
5557 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5558 goto out;
5559 /*
5560 * all of the above have passed, it is safe to overwrite this extent
5561 * without cow
5562 */
5563 ret = 1;
5564out:
5565 btrfs_free_path(path);
5566 return ret;
5567}
5568
4b46fce2
JB		 5569static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5570 struct buffer_head *bh_result, int create)
5571{
5572 struct extent_map *em;
5573 struct btrfs_root *root = BTRFS_I(inode)->root;
5574 u64 start = iblock << inode->i_blkbits;
5575 u64 len = bh_result->b_size;
46bfbb5c 5576 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5577
5578 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5579 if (IS_ERR(em))
5580 return PTR_ERR(em);
5581
5582 /*
5583 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5584 * io. INLINE is special, and we could probably kludge it in here, but
5585 * it's still buffered so for safety lets just fall back to the generic
5586 * buffered path.
5587 *
5588 * For COMPRESSED we _have_ to read the entire extent in so we can
5589 * decompress it, so there will be buffering required no matter what we
5590 * do, so go ahead and fallback to buffered.
5591 *
5592 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5593 * to buffered IO. Don't blame me, this is the price we pay for using
5594 * the generic code.
5595 */
5596 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5597 em->block_start == EXTENT_MAP_INLINE) {
5598 free_extent_map(em);
5599 return -ENOTBLK;
5600 }
5601
5602 /* Just a good old fashioned hole, return */
5603 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5604 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5605 free_extent_map(em);
5606 /* DIO will do one hole at a time, so just unlock a sector */
5607 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5608 start + root->sectorsize - 1, GFP_NOFS);
5609 return 0;
5610 }
5611
5612 /*
5613 * We don't allocate a new extent in the following cases
5614 *
5615 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5616 * existing extent.
5617 * 2) The extent is marked as PREALLOC. We're good to go here and can
5618 * just use the extent.
5619 *
5620 */
46bfbb5c
CM		 5621 if (!create) {
5622 len = em->len - (start - em->start);
4b46fce2 5623 goto map;
46bfbb5c 5624 }
4b46fce2
JB		 5625
5626 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5627 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5628 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5629 int type;
5630 int ret;
46bfbb5c 5631 u64 block_start;
4b46fce2
JB		 5632
5633 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5634 type = BTRFS_ORDERED_PREALLOC;
5635 else
5636 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5637 len = min(len, em->len - (start - em->start));
4b46fce2 5638 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5639
5640 /*
5641 * we're not going to log anything, but we do need
5642 * to make sure the current transaction stays open
5643 * while we look for nocow cross refs
5644 */
7a7eaa40 5645 trans = btrfs_join_transaction(root);
3612b495 5646 if (IS_ERR(trans))
46bfbb5c
CM		 5647 goto must_cow;
5648
5649 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5650 ret = btrfs_add_ordered_extent_dio(inode, start,
5651 block_start, len, len, type);
5652 btrfs_end_transaction(trans, root);
5653 if (ret) {
5654 free_extent_map(em);
5655 return ret;
5656 }
5657 goto unlock;
4b46fce2 5658 }
46bfbb5c 5659 btrfs_end_transaction(trans, root);
4b46fce2 5660 }
46bfbb5c
CM		 5661must_cow:
5662 /*
5663 * this will cow the extent, reset the len in case we changed
5664 * it above
5665 */
5666 len = bh_result->b_size;
16d299ac 5667 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5668 if (IS_ERR(em))
5669 return PTR_ERR(em);
5670 len = min(len, em->len - (start - em->start));
5671unlock:
4845e44f
CM		 5672 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5673 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5674 0, NULL, GFP_NOFS);
4b46fce2
JB		 5675map:
5676 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5677 inode->i_blkbits;
46bfbb5c 5678 bh_result->b_size = len;
4b46fce2
JB		 5679 bh_result->b_bdev = em->bdev;
5680 set_buffer_mapped(bh_result);
5681 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5682 set_buffer_new(bh_result);
5683
5684 free_extent_map(em);
5685
5686 return 0;
5687}
5688
5689struct btrfs_dio_private {
5690 struct inode *inode;
5691 u64 logical_offset;
5692 u64 disk_bytenr;
5693 u64 bytes;
5694 u32 *csums;
5695 void *private;
e65e1535
MX		 5696
5697 /* number of bios pending for this dio */
5698 atomic_t pending_bios;
5699
5700 /* IO errors */
5701 int errors;
5702
5703 struct bio *orig_bio;
4b46fce2
JB		 5704};
5705
5706static void btrfs_endio_direct_read(struct bio *bio, int err)
5707{
e65e1535 5708 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5709 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5710 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5711 struct inode *inode = dip->inode;
5712 struct btrfs_root *root = BTRFS_I(inode)->root;
5713 u64 start;
5714 u32 *private = dip->csums;
5715
5716 start = dip->logical_offset;
5717 do {
5718 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5719 struct page *page = bvec->bv_page;
5720 char *kaddr;
5721 u32 csum = ~(u32)0;
5722 unsigned long flags;
5723
5724 local_irq_save(flags);
5725 kaddr = kmap_atomic(page, KM_IRQ0);
5726 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5727 csum, bvec->bv_len);
5728 btrfs_csum_final(csum, (char *)&csum);
5729 kunmap_atomic(kaddr, KM_IRQ0);
5730 local_irq_restore(flags);
5731
5732 flush_dcache_page(bvec->bv_page);
5733 if (csum != *private) {
33345d01 5734 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5735 " %llu csum %u private %u\n",
33345d01
LZ		 5736 (unsigned long long)btrfs_ino(inode),
5737 (unsigned long long)start,
4b46fce2
JB		 5738 csum, *private);
5739 err = -EIO;
5740 }
5741 }
5742
5743 start += bvec->bv_len;
5744 private++;
5745 bvec++;
5746 } while (bvec <= bvec_end);
5747
5748 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5749 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5750 bio->bi_private = dip->private;
5751
5752 kfree(dip->csums);
5753 kfree(dip);
c0da7aa1
JB		 5754
5755 /* If we had a csum failure make sure to clear the uptodate flag */
5756 if (err)
5757 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5758 dio_end_io(bio, err);
5759}
5760
5761static void btrfs_endio_direct_write(struct bio *bio, int err)
5762{
5763 struct btrfs_dio_private *dip = bio->bi_private;
5764 struct inode *inode = dip->inode;
5765 struct btrfs_root *root = BTRFS_I(inode)->root;
5766 struct btrfs_trans_handle *trans;
5767 struct btrfs_ordered_extent *ordered = NULL;
5768 struct extent_state *cached_state = NULL;
163cf09c
CM		 5769 u64 ordered_offset = dip->logical_offset;
5770 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5771 int ret;
5772
5773 if (err)
5774 goto out_done;
163cf09c
CM		 5775again:
5776 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5777 &ordered_offset,
5778 ordered_bytes);
4b46fce2 5779 if (!ret)
163cf09c 5780 goto out_test;
4b46fce2
JB		 5781
5782 BUG_ON(!ordered);
5783
7a7eaa40 5784 trans = btrfs_join_transaction(root);
3612b495 5785 if (IS_ERR(trans)) {
4b46fce2
JB		 5786 err = -ENOMEM;
5787 goto out;
5788 }
5789 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5790
5791 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5792 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5793 if (!ret)
2115133f 5794 err = btrfs_update_inode_fallback(trans, root, inode);
4b46fce2
JB		 5795 goto out;
5796 }
5797
5798 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5799 ordered->file_offset + ordered->len - 1, 0,
5800 &cached_state, GFP_NOFS);
5801
5802 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5803 ret = btrfs_mark_extent_written(trans, inode,
5804 ordered->file_offset,
5805 ordered->file_offset +
5806 ordered->len);
5807 if (ret) {
5808 err = ret;
5809 goto out_unlock;
5810 }
5811 } else {
5812 ret = insert_reserved_file_extent(trans, inode,
5813 ordered->file_offset,
5814 ordered->start,
5815 ordered->disk_len,
5816 ordered->len,
5817 ordered->len,
5818 0, 0, 0,
5819 BTRFS_FILE_EXTENT_REG);
5820 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5821 ordered->file_offset, ordered->len);
5822 if (ret) {
5823 err = ret;
5824 WARN_ON(1);
5825 goto out_unlock;
5826 }
5827 }
5828
5829 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
1ef30be1 5830 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
a39f7521 5831 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
2115133f 5832 btrfs_update_inode_fallback(trans, root, inode);
1ef30be1 5833 ret = 0;
4b46fce2
JB		 5834out_unlock:
5835 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5836 ordered->file_offset + ordered->len - 1,
5837 &cached_state, GFP_NOFS);
5838out:
5839 btrfs_delalloc_release_metadata(inode, ordered->len);
5840 btrfs_end_transaction(trans, root);
163cf09c 5841 ordered_offset = ordered->file_offset + ordered->len;
4b46fce2
JB		 5842 btrfs_put_ordered_extent(ordered);
5843 btrfs_put_ordered_extent(ordered);
163cf09c
CM		 5844
5845out_test:
5846 /*
5847 * our bio might span multiple ordered extents. If we haven't
5848 * completed the accounting for the whole dio, go back and try again
5849 */
5850 if (ordered_offset < dip->logical_offset + dip->bytes) {
5851 ordered_bytes = dip->logical_offset + dip->bytes -
5852 ordered_offset;
5853 goto again;
5854 }
4b46fce2
JB		 5855out_done:
5856 bio->bi_private = dip->private;
5857
5858 kfree(dip->csums);
5859 kfree(dip);
c0da7aa1
JB		 5860
5861 /* If we had an error make sure to clear the uptodate flag */
5862 if (err)
5863 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5864 dio_end_io(bio, err);
5865}
5866
eaf25d93
CM		 5867static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5868 struct bio *bio, int mirror_num,
5869 unsigned long bio_flags, u64 offset)
5870{
5871 int ret;
5872 struct btrfs_root *root = BTRFS_I(inode)->root;
5873 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5874 BUG_ON(ret);
5875 return 0;
5876}
5877
e65e1535
MX		 5878static void btrfs_end_dio_bio(struct bio *bio, int err)
5879{
5880 struct btrfs_dio_private *dip = bio->bi_private;
5881
5882 if (err) {
33345d01 5883 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 5884 "sector %#Lx len %u err no %d\n",
33345d01 5885 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 5886 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 5887 dip->errors = 1;
5888
5889 /*
5890 * before atomic variable goto zero, we must make sure
5891 * dip->errors is perceived to be set.
5892 */
5893 smp_mb__before_atomic_dec();
5894 }
5895
5896 /* if there are more bios still pending for this dio, just exit */
5897 if (!atomic_dec_and_test(&dip->pending_bios))
5898 goto out;
5899
5900 if (dip->errors)
5901 bio_io_error(dip->orig_bio);
5902 else {
5903 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5904 bio_endio(dip->orig_bio, 0);
5905 }
5906out:
5907 bio_put(bio);
5908}
5909
5910static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5911 u64 first_sector, gfp_t gfp_flags)
5912{
5913 int nr_vecs = bio_get_nr_vecs(bdev);
5914 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5915}
5916
5917static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5918 int rw, u64 file_offset, int skip_sum,
1ae39938 5919 u32 *csums, int async_submit)
e65e1535
MX		 5920{
5921 int write = rw & REQ_WRITE;
5922 struct btrfs_root *root = BTRFS_I(inode)->root;
5923 int ret;
5924
5925 bio_get(bio);
5926 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5927 if (ret)
5928 goto err;
5929
1ae39938
JB		 5930 if (skip_sum)
5931 goto map;
5932
5933 if (write && async_submit) {
e65e1535
MX		 5934 ret = btrfs_wq_submit_bio(root->fs_info,
5935 inode, rw, bio, 0, 0,
5936 file_offset,
5937 __btrfs_submit_bio_start_direct_io,
5938 __btrfs_submit_bio_done);
5939 goto err;
1ae39938
JB		 5940 } else if (write) {
5941 /*
5942 * If we aren't doing async submit, calculate the csum of the
5943 * bio now.
5944 */
5945 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5946 if (ret)
5947 goto err;
c2db1073
TI		 5948 } else if (!skip_sum) {
5949 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 5950 file_offset, csums);
c2db1073
TI		 5951 if (ret)
5952 goto err;
5953 }
e65e1535 5954
1ae39938
JB		 5955map:
5956 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 5957err:
5958 bio_put(bio);
5959 return ret;
5960}
5961
5962static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5963 int skip_sum)
5964{
5965 struct inode *inode = dip->inode;
5966 struct btrfs_root *root = BTRFS_I(inode)->root;
5967 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5968 struct bio *bio;
5969 struct bio *orig_bio = dip->orig_bio;
5970 struct bio_vec *bvec = orig_bio->bi_io_vec;
5971 u64 start_sector = orig_bio->bi_sector;
5972 u64 file_offset = dip->logical_offset;
5973 u64 submit_len = 0;
5974 u64 map_length;
5975 int nr_pages = 0;
5976 u32 *csums = dip->csums;
5977 int ret = 0;
1ae39938 5978 int async_submit = 0;
98bc3149 5979 int write = rw & REQ_WRITE;
e65e1535 5980
e65e1535
MX		 5981 map_length = orig_bio->bi_size;
5982 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5983 &map_length, NULL, 0);
5984 if (ret) {
64728bbb 5985 bio_put(orig_bio);
e65e1535
MX		 5986 return -EIO;
5987 }
5988
02f57c7a
JB		 5989 if (map_length >= orig_bio->bi_size) {
5990 bio = orig_bio;
5991 goto submit;
5992 }
5993
1ae39938 5994 async_submit = 1;
02f57c7a
JB		 5995 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5996 if (!bio)
5997 return -ENOMEM;
5998 bio->bi_private = dip;
5999 bio->bi_end_io = btrfs_end_dio_bio;
6000 atomic_inc(&dip->pending_bios);
6001
e65e1535
MX		 6002 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6003 if (unlikely(map_length < submit_len + bvec->bv_len ||
6004 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6005 bvec->bv_offset) < bvec->bv_len)) {
6006 /*
6007 * inc the count before we submit the bio so
6008 * we know the end IO handler won't happen before
6009 * we inc the count. Otherwise, the dip might get freed
6010 * before we're done setting it up
6011 */
6012 atomic_inc(&dip->pending_bios);
6013 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6014 file_offset, skip_sum,
1ae39938 6015 csums, async_submit);
e65e1535
MX		 6016 if (ret) {
6017 bio_put(bio);
6018 atomic_dec(&dip->pending_bios);
6019 goto out_err;
6020 }
6021
98bc3149
JB		 6022 /* Write's use the ordered csums */
6023 if (!write && !skip_sum)
e65e1535
MX		 6024 csums = csums + nr_pages;
6025 start_sector += submit_len >> 9;
6026 file_offset += submit_len;
6027
6028 submit_len = 0;
6029 nr_pages = 0;
6030
6031 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6032 start_sector, GFP_NOFS);
6033 if (!bio)
6034 goto out_err;
6035 bio->bi_private = dip;
6036 bio->bi_end_io = btrfs_end_dio_bio;
6037
6038 map_length = orig_bio->bi_size;
6039 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6040 &map_length, NULL, 0);
6041 if (ret) {
6042 bio_put(bio);
6043 goto out_err;
6044 }
6045 } else {
6046 submit_len += bvec->bv_len;
6047 nr_pages ++;
6048 bvec++;
6049 }
6050 }
6051
02f57c7a 6052submit:
e65e1535 6053 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 6054 csums, async_submit);
e65e1535
MX		 6055 if (!ret)
6056 return 0;
6057
6058 bio_put(bio);
6059out_err:
6060 dip->errors = 1;
6061 /*
6062 * before atomic variable goto zero, we must
6063 * make sure dip->errors is perceived to be set.
6064 */
6065 smp_mb__before_atomic_dec();
6066 if (atomic_dec_and_test(&dip->pending_bios))
6067 bio_io_error(dip->orig_bio);
6068
6069 /* bio_end_io() will handle error, so we needn't return it */
6070 return 0;
6071}
6072
4b46fce2
JB		 6073static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6074 loff_t file_offset)
6075{
6076 struct btrfs_root *root = BTRFS_I(inode)->root;
6077 struct btrfs_dio_private *dip;
6078 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6079 int skip_sum;
7b6d91da 6080 int write = rw & REQ_WRITE;
4b46fce2
JB		 6081 int ret = 0;
6082
6083 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6084
6085 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6086 if (!dip) {
6087 ret = -ENOMEM;
6088 goto free_ordered;
6089 }
6090 dip->csums = NULL;
6091
98bc3149
JB		 6092 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6093 if (!write && !skip_sum) {
4b46fce2
JB		 6094 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6095 if (!dip->csums) {
b4966b77 6096 kfree(dip);
4b46fce2
JB		 6097 ret = -ENOMEM;
6098 goto free_ordered;
6099 }
6100 }
6101
6102 dip->private = bio->bi_private;
6103 dip->inode = inode;
6104 dip->logical_offset = file_offset;
6105
4b46fce2
JB		 6106 dip->bytes = 0;
6107 do {
6108 dip->bytes += bvec->bv_len;
6109 bvec++;
6110 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6111
46bfbb5c 6112 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6113 bio->bi_private = dip;
e65e1535
MX		 6114 dip->errors = 0;
6115 dip->orig_bio = bio;
6116 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6117
6118 if (write)
6119 bio->bi_end_io = btrfs_endio_direct_write;
6120 else
6121 bio->bi_end_io = btrfs_endio_direct_read;
6122
e65e1535
MX		 6123 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6124 if (!ret)
eaf25d93 6125 return;
4b46fce2
JB		 6126free_ordered:
6127 /*
6128 * If this is a write, we need to clean up the reserved space and kill
6129 * the ordered extent.
6130 */
6131 if (write) {
6132 struct btrfs_ordered_extent *ordered;
955256f2 6133 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6134 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6135 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6136 btrfs_free_reserved_extent(root, ordered->start,
6137 ordered->disk_len);
6138 btrfs_put_ordered_extent(ordered);
6139 btrfs_put_ordered_extent(ordered);
6140 }
6141 bio_endio(bio, ret);
6142}
6143
5a5f79b5
CM		 6144static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6145 const struct iovec *iov, loff_t offset,
6146 unsigned long nr_segs)
6147{
6148 int seg;
a1b75f7d 6149 int i;
5a5f79b5
CM		 6150 size_t size;
6151 unsigned long addr;
6152 unsigned blocksize_mask = root->sectorsize - 1;
6153 ssize_t retval = -EINVAL;
6154 loff_t end = offset;
6155
6156 if (offset & blocksize_mask)
6157 goto out;
6158
6159 /* Check the memory alignment. Blocks cannot straddle pages */
6160 for (seg = 0; seg < nr_segs; seg++) {
6161 addr = (unsigned long)iov[seg].iov_base;
6162 size = iov[seg].iov_len;
6163 end += size;
a1b75f7d 6164 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6165 goto out;
a1b75f7d
JB		 6166
6167 /* If this is a write we don't need to check anymore */
6168 if (rw & WRITE)
6169 continue;
6170
6171 /*
6172 * Check to make sure we don't have duplicate iov_base's in this
6173 * iovec, if so return EINVAL, otherwise we'll get csum errors
6174 * when reading back.
6175 */
6176 for (i = seg + 1; i < nr_segs; i++) {
6177 if (iov[seg].iov_base == iov[i].iov_base)
6178 goto out;
6179 }
5a5f79b5
CM		 6180 }
6181 retval = 0;
6182out:
6183 return retval;
6184}
16432985
CM		 6185static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6186 const struct iovec *iov, loff_t offset,
6187 unsigned long nr_segs)
6188{
4b46fce2
JB		 6189 struct file *file = iocb->ki_filp;
6190 struct inode *inode = file->f_mapping->host;
6191 struct btrfs_ordered_extent *ordered;
4845e44f 6192 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6193 u64 lockstart, lockend;
6194 ssize_t ret;
4845e44f
CM		 6195 int writing = rw & WRITE;
6196 int write_bits = 0;
3f7c579c 6197 size_t count = iov_length(iov, nr_segs);
4b46fce2 6198
5a5f79b5
CM		 6199 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6200 offset, nr_segs)) {
6201 return 0;
6202 }
6203
4b46fce2 6204 lockstart = offset;
3f7c579c
CM		 6205 lockend = offset + count - 1;
6206
6207 if (writing) {
6208 ret = btrfs_delalloc_reserve_space(inode, count);
6209 if (ret)
6210 goto out;
6211 }
4845e44f 6212
4b46fce2 6213 while (1) {
4845e44f
CM		 6214 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6215 0, &cached_state, GFP_NOFS);
4b46fce2
JB		 6216 /*
6217 * We're concerned with the entire range that we're going to be
6218 * doing DIO to, so we need to make sure theres no ordered
6219 * extents in this range.
6220 */
6221 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6222 lockend - lockstart + 1);
6223 if (!ordered)
6224 break;
4845e44f
CM		 6225 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6226 &cached_state, GFP_NOFS);
4b46fce2
JB		 6227 btrfs_start_ordered_extent(inode, ordered, 1);
6228 btrfs_put_ordered_extent(ordered);
6229 cond_resched();
6230 }
6231
4845e44f
CM		 6232 /*
6233 * we don't use btrfs_set_extent_delalloc because we don't want
6234 * the dirty or uptodate bits
6235 */
6236 if (writing) {
6237 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6238 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6239 EXTENT_DELALLOC, 0, NULL, &cached_state,
6240 GFP_NOFS);
6241 if (ret) {
6242 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6243 lockend, EXTENT_LOCKED | write_bits,
6244 1, 0, &cached_state, GFP_NOFS);
6245 goto out;
6246 }
6247 }
6248
6249 free_extent_state(cached_state);
6250 cached_state = NULL;
6251
5a5f79b5
CM		 6252 ret = __blockdev_direct_IO(rw, iocb, inode,
6253 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6254 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6255 btrfs_submit_direct, 0);
4b46fce2
JB		 6256
6257 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6258 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6259 offset + iov_length(iov, nr_segs) - 1,
6260 EXTENT_LOCKED | write_bits, 1, 0,
6261 &cached_state, GFP_NOFS);
4b46fce2
JB		 6262 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6263 /*
6264 * We're falling back to buffered, unlock the section we didn't
6265 * do IO on.
6266 */
4845e44f
CM		 6267 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6268 offset + iov_length(iov, nr_segs) - 1,
6269 EXTENT_LOCKED | write_bits, 1, 0,
6270 &cached_state, GFP_NOFS);
4b46fce2 6271 }
4845e44f
CM		 6272out:
6273 free_extent_state(cached_state);
4b46fce2 6274 return ret;
16432985
CM		 6275}
6276
1506fcc8
YS		 6277static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6278 __u64 start, __u64 len)
6279{
ec29ed5b 6280 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6281}
6282
a52d9a80 6283int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6284{
d1310b2e
CM		 6285 struct extent_io_tree *tree;
6286 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6287 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6288}
1832a6d5 6289
a52d9a80 6290static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6291{
d1310b2e 6292 struct extent_io_tree *tree;
b888db2b
CM		 6293
6294
6295 if (current->flags & PF_MEMALLOC) {
6296 redirty_page_for_writepage(wbc, page);
6297 unlock_page(page);
6298 return 0;
6299 }
d1310b2e 6300 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6301 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6302}
6303
f421950f
CM		 6304int btrfs_writepages(struct address_space *mapping,
6305 struct writeback_control *wbc)
b293f02e 6306{
d1310b2e 6307 struct extent_io_tree *tree;
771ed689 6308
d1310b2e 6309 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6310 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6311}
6312
3ab2fb5a
CM		 6313static int
6314btrfs_readpages(struct file *file, struct address_space *mapping,
6315 struct list_head *pages, unsigned nr_pages)
6316{
d1310b2e
CM		 6317 struct extent_io_tree *tree;
6318 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6319 return extent_readpages(tree, mapping, pages, nr_pages,
6320 btrfs_get_extent);
6321}
e6dcd2dc 6322static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6323{
d1310b2e
CM		 6324 struct extent_io_tree *tree;
6325 struct extent_map_tree *map;
a52d9a80 6326 int ret;
8c2383c3 6327
d1310b2e
CM		 6328 tree = &BTRFS_I(page->mapping->host)->io_tree;
6329 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6330 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6331 if (ret == 1) {
6332 ClearPagePrivate(page);
6333 set_page_private(page, 0);
6334 page_cache_release(page);
39279cc3 6335 }
a52d9a80 6336 return ret;
39279cc3
CM		 6337}
6338
e6dcd2dc
CM		 6339static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6340{
98509cfc
CM		 6341 if (PageWriteback(page) || PageDirty(page))
6342 return 0;
b335b003 6343 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6344}
6345
a52d9a80 6346static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6347{
d1310b2e 6348 struct extent_io_tree *tree;
e6dcd2dc 6349 struct btrfs_ordered_extent *ordered;
2ac55d41 6350 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6351 u64 page_start = page_offset(page);
6352 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6353
8b62b72b
CM		 6354
6355 /*
6356 * we have the page locked, so new writeback can't start,
6357 * and the dirty bit won't be cleared while we are here.
6358 *
6359 * Wait for IO on this page so that we can safely clear
6360 * the PagePrivate2 bit and do ordered accounting
6361 */
e6dcd2dc 6362 wait_on_page_writeback(page);
8b62b72b 6363
d1310b2e 6364 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 6365 if (offset) {
6366 btrfs_releasepage(page, GFP_NOFS);
6367 return;
6368 }
2ac55d41
JB		 6369 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6370 GFP_NOFS);
e6dcd2dc
CM		 6371 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6372 page_offset(page));
6373 if (ordered) {
eb84ae03
CM		 6374 /*
6375 * IO on this page will never be started, so we need
6376 * to account for any ordered extents now
6377 */
e6dcd2dc
CM		 6378 clear_extent_bit(tree, page_start, page_end,
6379 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6380 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6381 &cached_state, GFP_NOFS);
8b62b72b
CM		 6382 /*
6383 * whoever cleared the private bit is responsible
6384 * for the finish_ordered_io
6385 */
6386 if (TestClearPagePrivate2(page)) {
6387 btrfs_finish_ordered_io(page->mapping->host,
6388 page_start, page_end);
6389 }
e6dcd2dc 6390 btrfs_put_ordered_extent(ordered);
2ac55d41
JB		 6391 cached_state = NULL;
6392 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6393 GFP_NOFS);
e6dcd2dc
CM		 6394 }
6395 clear_extent_bit(tree, page_start, page_end,
32c00aff 6396 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6397 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6398 __btrfs_releasepage(page, GFP_NOFS);
6399
4a096752 6400 ClearPageChecked(page);
9ad6b7bc 6401 if (PagePrivate(page)) {
9ad6b7bc
CM		 6402 ClearPagePrivate(page);
6403 set_page_private(page, 0);
6404 page_cache_release(page);
6405 }
39279cc3
CM		 6406}
6407
9ebefb18
CM		 6408/*
6409 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6410 * called from a page fault handler when a page is first dirtied. Hence we must
6411 * be careful to check for EOF conditions here. We set the page up correctly
6412 * for a written page which means we get ENOSPC checking when writing into
6413 * holes and correct delalloc and unwritten extent mapping on filesystems that
6414 * support these features.
6415 *
6416 * We are not allowed to take the i_mutex here so we have to play games to
6417 * protect against truncate races as the page could now be beyond EOF. Because
6418 * vmtruncate() writes the inode size before removing pages, once we have the
6419 * page lock we can determine safely if the page is beyond EOF. If it is not
6420 * beyond EOF, then the page is guaranteed safe against truncation until we
6421 * unlock the page.
6422 */
c2ec175c 6423int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6424{
c2ec175c 6425 struct page *page = vmf->page;
6da6abae 6426 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6427 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6428 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6429 struct btrfs_ordered_extent *ordered;
2ac55d41 6430 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6431 char *kaddr;
6432 unsigned long zero_start;
9ebefb18 6433 loff_t size;
1832a6d5 6434 int ret;
9998eb70 6435 int reserved = 0;
a52d9a80 6436 u64 page_start;
e6dcd2dc 6437 u64 page_end;
9ebefb18 6438
0ca1f7ce 6439 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 6440 if (!ret) {
22c44fe6 6441 ret = btrfs_update_time(vma->vm_file);
9998eb70
CM		 6442 reserved = 1;
6443 }
56a76f82
NP		 6444 if (ret) {
6445 if (ret == -ENOMEM)
6446 ret = VM_FAULT_OOM;
6447 else /* -ENOSPC, -EIO, etc */
6448 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 6449 if (reserved)
6450 goto out;
6451 goto out_noreserve;
56a76f82 6452 }
1832a6d5 6453
56a76f82 6454 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6455again:
9ebefb18 6456 lock_page(page);
9ebefb18 6457 size = i_size_read(inode);
e6dcd2dc
CM		 6458 page_start = page_offset(page);
6459 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6460
9ebefb18 6461 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6462 (page_start >= size)) {
9ebefb18
CM		 6463 /* page got truncated out from underneath us */
6464 goto out_unlock;
6465 }
e6dcd2dc
CM		 6466 wait_on_page_writeback(page);
6467
2ac55d41
JB		 6468 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6469 GFP_NOFS);
e6dcd2dc
CM		 6470 set_page_extent_mapped(page);
6471
eb84ae03
CM		 6472 /*
6473 * we can't set the delalloc bits if there are pending ordered
6474 * extents. Drop our locks and wait for them to finish
6475 */
e6dcd2dc
CM		 6476 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6477 if (ordered) {
2ac55d41
JB		 6478 unlock_extent_cached(io_tree, page_start, page_end,
6479 &cached_state, GFP_NOFS);
e6dcd2dc 6480 unlock_page(page);
eb84ae03 6481 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6482 btrfs_put_ordered_extent(ordered);
6483 goto again;
6484 }
6485
fbf19087
JB		 6486 /*
6487 * XXX - page_mkwrite gets called every time the page is dirtied, even
6488 * if it was already dirty, so for space accounting reasons we need to
6489 * clear any delalloc bits for the range we are fixing to save. There
6490 * is probably a better way to do this, but for now keep consistent with
6491 * prepare_pages in the normal write path.
6492 */
2ac55d41 6493 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6494 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6495 0, 0, &cached_state, GFP_NOFS);
fbf19087 6496
2ac55d41
JB		 6497 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6498 &cached_state);
9ed74f2d 6499 if (ret) {
2ac55d41
JB		 6500 unlock_extent_cached(io_tree, page_start, page_end,
6501 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6502 ret = VM_FAULT_SIGBUS;
6503 goto out_unlock;
6504 }
e6dcd2dc 6505 ret = 0;
9ebefb18
CM		 6506
6507 /* page is wholly or partially inside EOF */
a52d9a80 6508 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6509 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6510 else
e6dcd2dc 6511 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6512
e6dcd2dc
CM		 6513 if (zero_start != PAGE_CACHE_SIZE) {
6514 kaddr = kmap(page);
6515 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6516 flush_dcache_page(page);
6517 kunmap(page);
6518 }
247e743c 6519 ClearPageChecked(page);
e6dcd2dc 6520 set_page_dirty(page);
50a9b214 6521 SetPageUptodate(page);
5a3f23d5 6522
257c62e1
CM		 6523 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6524 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6525
2ac55d41 6526 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6527
6528out_unlock:
50a9b214
CM		 6529 if (!ret)
6530 return VM_FAULT_LOCKED;
9ebefb18 6531 unlock_page(page);
1832a6d5 6532out:
ec39e180 6533 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 6534out_noreserve:
9ebefb18
CM		 6535 return ret;
6536}
6537
a41ad394 6538static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6539{
6540 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6541 struct btrfs_block_rsv *rsv;
39279cc3 6542 int ret;
3893e33b 6543 int err = 0;
39279cc3 6544 struct btrfs_trans_handle *trans;
d3c2fdcf 6545 unsigned long nr;
dbe674a9 6546 u64 mask = root->sectorsize - 1;
07127184 6547 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6548
5d5e103a
JB		 6549 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6550 if (ret)
a41ad394 6551 return ret;
8082510e 6552
4a096752 6553 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6554 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6555
fcb80c2a
JB		 6556 /*
6557 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6558 * 3 things going on here
6559 *
6560 * 1) We need to reserve space for our orphan item and the space to
6561 * delete our orphan item. Lord knows we don't want to have a dangling
6562 * orphan item because we didn't reserve space to remove it.
6563 *
6564 * 2) We need to reserve space to update our inode.
6565 *
6566 * 3) We need to have something to cache all the space that is going to
6567 * be free'd up by the truncate operation, but also have some slack
6568 * space reserved in case it uses space during the truncate (thank you
6569 * very much snapshotting).
6570 *
6571 * And we need these to all be seperate. The fact is we can use alot of
6572 * space doing the truncate, and we have no earthly idea how much space
6573 * we will use, so we need the truncate reservation to be seperate so it
6574 * doesn't end up using space reserved for updating the inode or
6575 * removing the orphan item. We also need to be able to stop the
6576 * transaction and start a new one, which means we need to be able to
6577 * update the inode several times, and we have no idea of knowing how
6578 * many times that will be, so we can't just reserve 1 item for the
6579 * entirety of the opration, so that has to be done seperately as well.
6580 * Then there is the orphan item, which does indeed need to be held on
6581 * to for the whole operation, and we need nobody to touch this reserved
6582 * space except the orphan code.
6583 *
6584 * So that leaves us with
6585 *
6586 * 1) root->orphan_block_rsv - for the orphan deletion.
6587 * 2) rsv - for the truncate reservation, which we will steal from the
6588 * transaction reservation.
6589 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6590 * updating the inode.
6591 */
6592 rsv = btrfs_alloc_block_rsv(root);
6593 if (!rsv)
6594 return -ENOMEM;
4a338542 6595 rsv->size = min_size;
f0cd846e 6596
907cbceb 6597 /*
07127184 6598 * 1 for the truncate slack space
907cbceb
JB		 6599 * 1 for the orphan item we're going to add
6600 * 1 for the orphan item deletion
6601 * 1 for updating the inode.
6602 */
fcb80c2a
JB		 6603 trans = btrfs_start_transaction(root, 4);
6604 if (IS_ERR(trans)) {
6605 err = PTR_ERR(trans);
6606 goto out;
6607 }
f0cd846e 6608
907cbceb
JB		 6609 /* Migrate the slack space for the truncate to our reserve */
6610 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6611 min_size);
fcb80c2a 6612 BUG_ON(ret);
f0cd846e
JB		 6613
6614 ret = btrfs_orphan_add(trans, inode);
6615 if (ret) {
6616 btrfs_end_transaction(trans, root);
fcb80c2a 6617 goto out;
f0cd846e
JB		 6618 }
6619
5a3f23d5
CM		 6620 /*
6621 * setattr is responsible for setting the ordered_data_close flag,
6622 * but that is only tested during the last file release. That
6623 * could happen well after the next commit, leaving a great big
6624 * window where new writes may get lost if someone chooses to write
6625 * to this file after truncating to zero
6626 *
6627 * The inode doesn't have any dirty data here, and so if we commit
6628 * this is a noop. If someone immediately starts writing to the inode
6629 * it is very likely we'll catch some of their writes in this
6630 * transaction, and the commit will find this file on the ordered
6631 * data list with good things to send down.
6632 *
6633 * This is a best effort solution, there is still a window where
6634 * using truncate to replace the contents of the file will
6635 * end up with a zero length file after a crash.
6636 */
6637 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6638 btrfs_add_ordered_operation(trans, root, inode);
6639
8082510e 6640 while (1) {
36ba022a 6641 ret = btrfs_block_rsv_refill(root, rsv, min_size);
907cbceb
JB		 6642 if (ret) {
6643 /*
6644 * This can only happen with the original transaction we
6645 * started above, every other time we shouldn't have a
6646 * transaction started yet.
6647 */
6648 if (ret == -EAGAIN)
6649 goto end_trans;
6650 err = ret;
6651 break;
6652 }
6653
d68fc57b 6654 if (!trans) {
907cbceb
JB		 6655 /* Just need the 1 for updating the inode */
6656 trans = btrfs_start_transaction(root, 1);
fcb80c2a 6657 if (IS_ERR(trans)) {
7041ee97
JB		 6658 ret = err = PTR_ERR(trans);
6659 trans = NULL;
6660 break;
fcb80c2a 6661 }
d68fc57b
YZ		 6662 }
6663
907cbceb
JB		 6664 trans->block_rsv = rsv;
6665
8082510e
YZ		 6666 ret = btrfs_truncate_inode_items(trans, root, inode,
6667 inode->i_size,
6668 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6669 if (ret != -EAGAIN) {
6670 err = ret;
8082510e 6671 break;
3893e33b 6672 }
39279cc3 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) {
6677 err = ret;
6678 break;
6679 }
907cbceb 6680end_trans:
8082510e
YZ		 6681 nr = trans->blocks_used;
6682 btrfs_end_transaction(trans, root);
d68fc57b 6683 trans = NULL;
8082510e 6684 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6685 }
6686
6687 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6688 trans->block_rsv = root->orphan_block_rsv;
8082510e 6689 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6690 if (ret)
6691 err = ret;
ded5db9d
JB		 6692 } else if (ret && inode->i_nlink > 0) {
6693 /*
6694 * Failed to do the truncate, remove us from the in memory
6695 * orphan list.
6696 */
6697 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6698 }
6699
917c16b2
CM		 6700 if (trans) {
6701 trans->block_rsv = &root->fs_info->trans_block_rsv;
6702 ret = btrfs_update_inode(trans, root, inode);
6703 if (ret && !err)
6704 err = ret;
7b128766 6705
917c16b2 6706 nr = trans->blocks_used;
7ad85bb7 6707 ret = btrfs_end_transaction(trans, root);
917c16b2
CM		 6708 btrfs_btree_balance_dirty(root, nr);
6709 }
fcb80c2a
JB		 6710
6711out:
6712 btrfs_free_block_rsv(root, rsv);
6713
3893e33b
JB		 6714 if (ret && !err)
6715 err = ret;
a41ad394 6716
3893e33b 6717 return err;
39279cc3
CM		 6718}
6719
d352ac68
CM		 6720/*
6721 * create a new subvolume directory/inode (helper for the ioctl).
6722 */
d2fb3437 6723int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6724 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6725{
39279cc3 6726 struct inode *inode;
76dda93c 6727 int err;
00e4e6b3 6728 u64 index = 0;
39279cc3 6729
12fc9d09
FA		 6730 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6731 new_dirid, new_dirid,
6732 S_IFDIR | (~current_umask() & S_IRWXUGO),
6733 &index);
54aa1f4d 6734 if (IS_ERR(inode))
f46b5a66 6735 return PTR_ERR(inode);
39279cc3
CM		 6736 inode->i_op = &btrfs_dir_inode_operations;
6737 inode->i_fop = &btrfs_dir_file_operations;
6738
bfe86848 6739 set_nlink(inode, 1);
dbe674a9 6740 btrfs_i_size_write(inode, 0);
3b96362c 6741
76dda93c
YZ		 6742 err = btrfs_update_inode(trans, new_root, inode);
6743 BUG_ON(err);
cb8e7090 6744
76dda93c 6745 iput(inode);
cb8e7090 6746 return 0;
39279cc3
CM		 6747}
6748
39279cc3
CM		 6749struct inode *btrfs_alloc_inode(struct super_block *sb)
6750{
6751 struct btrfs_inode *ei;
2ead6ae7 6752 struct inode *inode;
39279cc3
CM		 6753
6754 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6755 if (!ei)
6756 return NULL;
2ead6ae7
YZ		 6757
6758 ei->root = NULL;
6759 ei->space_info = NULL;
6760 ei->generation = 0;
6761 ei->sequence = 0;
15ee9bc7 6762 ei->last_trans = 0;
257c62e1 6763 ei->last_sub_trans = 0;
e02119d5 6764 ei->logged_trans = 0;
2ead6ae7 6765 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6766 ei->disk_i_size = 0;
6767 ei->flags = 0;
7709cde3 6768 ei->csum_bytes = 0;
2ead6ae7
YZ		 6769 ei->index_cnt = (u64)-1;
6770 ei->last_unlink_trans = 0;
6771
9e0baf60
JB		 6772 spin_lock_init(&ei->lock);
6773 ei->outstanding_extents = 0;
6774 ei->reserved_extents = 0;
2ead6ae7
YZ		 6775
6776 ei->ordered_data_close = 0;
d68fc57b 6777 ei->orphan_meta_reserved = 0;
2ead6ae7 6778 ei->dummy_inode = 0;
4cb5300b 6779 ei->in_defrag = 0;
7fd2ae21 6780 ei->delalloc_meta_reserved = 0;
261507a0 6781 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6782
16cdcec7
MX		 6783 ei->delayed_node = NULL;
6784
2ead6ae7 6785 inode = &ei->vfs_inode;
a8067e02 6786 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6787 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6788 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
2ead6ae7 6789 mutex_init(&ei->log_mutex);
f248679e 6790 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 6791 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7b128766 6792 INIT_LIST_HEAD(&ei->i_orphan);
2ead6ae7 6793 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6794 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6795 RB_CLEAR_NODE(&ei->rb_node);
6796
6797 return inode;
39279cc3
CM		 6798}
6799
fa0d7e3d
NP		 6800static void btrfs_i_callback(struct rcu_head *head)
6801{
6802 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 6803 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6804}
6805
39279cc3
CM		 6806void btrfs_destroy_inode(struct inode *inode)
6807{
e6dcd2dc 6808 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6809 struct btrfs_root *root = BTRFS_I(inode)->root;
6810
39279cc3
CM		 6811 WARN_ON(!list_empty(&inode->i_dentry));
6812 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6813 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6814 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6815 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6816 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6817
a6dbd429
JB		 6818 /*
6819 * This can happen where we create an inode, but somebody else also
6820 * created the same inode and we need to destroy the one we already
6821 * created.
6822 */
6823 if (!root)
6824 goto free;
6825
5a3f23d5
CM		 6826 /*
6827 * Make sure we're properly removed from the ordered operation
6828 * lists.
6829 */
6830 smp_mb();
6831 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6832 spin_lock(&root->fs_info->ordered_extent_lock);
6833 list_del_init(&BTRFS_I(inode)->ordered_operations);
6834 spin_unlock(&root->fs_info->ordered_extent_lock);
6835 }
6836
d68fc57b 6837 spin_lock(&root->orphan_lock);
7b128766 6838 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
33345d01
LZ		 6839 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6840 (unsigned long long)btrfs_ino(inode));
8082510e 6841 list_del_init(&BTRFS_I(inode)->i_orphan);
7b128766 6842 }
d68fc57b 6843 spin_unlock(&root->orphan_lock);
7b128766 6844
d397712b 6845 while (1) {
e6dcd2dc
CM		 6846 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6847 if (!ordered)
6848 break;
6849 else {
d397712b
CM		 6850 printk(KERN_ERR "btrfs found ordered "
6851 "extent %llu %llu on inode cleanup\n",
6852 (unsigned long long)ordered->file_offset,
6853 (unsigned long long)ordered->len);
e6dcd2dc
CM		 6854 btrfs_remove_ordered_extent(inode, ordered);
6855 btrfs_put_ordered_extent(ordered);
6856 btrfs_put_ordered_extent(ordered);
6857 }
6858 }
5d4f98a2 6859 inode_tree_del(inode);
5b21f2ed 6860 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 6861free:
16cdcec7 6862 btrfs_remove_delayed_node(inode);
fa0d7e3d 6863 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 6864}
6865
45321ac5 6866int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 6867{
6868 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 6869
0af3d00b 6870 if (btrfs_root_refs(&root->root_item) == 0 &&
2cf8572d 6871 !btrfs_is_free_space_inode(root, inode))
45321ac5 6872 return 1;
76dda93c 6873 else
45321ac5 6874 return generic_drop_inode(inode);
76dda93c
YZ		 6875}
6876
0ee0fda0 6877static void init_once(void *foo)
39279cc3
CM		 6878{
6879 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6880
6881 inode_init_once(&ei->vfs_inode);
6882}
6883
6884void btrfs_destroy_cachep(void)
6885{
6886 if (btrfs_inode_cachep)
6887 kmem_cache_destroy(btrfs_inode_cachep);
6888 if (btrfs_trans_handle_cachep)
6889 kmem_cache_destroy(btrfs_trans_handle_cachep);
6890 if (btrfs_transaction_cachep)
6891 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 6892 if (btrfs_path_cachep)
6893 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 6894 if (btrfs_free_space_cachep)
6895 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 6896}
6897
6898int btrfs_init_cachep(void)
6899{
9601e3f6
CH		 6900 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6901 sizeof(struct btrfs_inode), 0,
6902 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 6903 if (!btrfs_inode_cachep)
6904 goto fail;
9601e3f6
CH		 6905
6906 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6907 sizeof(struct btrfs_trans_handle), 0,
6908 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6909 if (!btrfs_trans_handle_cachep)
6910 goto fail;
9601e3f6
CH		 6911
6912 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6913 sizeof(struct btrfs_transaction), 0,
6914 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6915 if (!btrfs_transaction_cachep)
6916 goto fail;
9601e3f6
CH		 6917
6918 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6919 sizeof(struct btrfs_path), 0,
6920 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6921 if (!btrfs_path_cachep)
6922 goto fail;
9601e3f6 6923
dc89e982
JB		 6924 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6925 sizeof(struct btrfs_free_space), 0,
6926 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6927 if (!btrfs_free_space_cachep)
6928 goto fail;
6929
39279cc3
CM		 6930 return 0;
6931fail:
6932 btrfs_destroy_cachep();
6933 return -ENOMEM;
6934}
6935
6936static int btrfs_getattr(struct vfsmount *mnt,
6937 struct dentry *dentry, struct kstat *stat)
6938{
6939 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 6940 u32 blocksize = inode->i_sb->s_blocksize;
6941
39279cc3 6942 generic_fillattr(inode, stat);
0ee5dc67 6943 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 6944 stat->blksize = PAGE_CACHE_SIZE;
fadc0d8b
DS		 6945 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6946 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 6947 return 0;
6948}
6949
75e7cb7f
LB		 6950/*
6951 * If a file is moved, it will inherit the cow and compression flags of the new
6952 * directory.
6953 */
6954static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6955{
6956 struct btrfs_inode *b_dir = BTRFS_I(dir);
6957 struct btrfs_inode *b_inode = BTRFS_I(inode);
6958
6959 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6960 b_inode->flags |= BTRFS_INODE_NODATACOW;
6961 else
6962 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6963
6964 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6965 b_inode->flags |= BTRFS_INODE_COMPRESS;
6966 else
6967 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6968}
6969
d397712b
CM		 6970static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6971 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 6972{
6973 struct btrfs_trans_handle *trans;
6974 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 6975 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 6976 struct inode *new_inode = new_dentry->d_inode;
6977 struct inode *old_inode = old_dentry->d_inode;
6978 struct timespec ctime = CURRENT_TIME;
00e4e6b3 6979 u64 index = 0;
4df27c4d 6980 u64 root_objectid;
39279cc3 6981 int ret;
33345d01 6982 u64 old_ino = btrfs_ino(old_inode);
39279cc3 6983
33345d01 6984 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 6985 return -EPERM;
6986
4df27c4d 6987 /* we only allow rename subvolume link between subvolumes */
33345d01 6988 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 6989 return -EXDEV;
6990
33345d01
LZ		 6991 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6992 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 6993 return -ENOTEMPTY;
5f39d397 6994
4df27c4d
YZ		 6995 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6996 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6997 return -ENOTEMPTY;
5a3f23d5
CM		 6998 /*
6999 * we're using rename to replace one file with another.
7000 * and the replacement file is large. Start IO on it now so
7001 * we don't add too much work to the end of the transaction
7002 */
4baf8c92 7003 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 7004 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7005 filemap_flush(old_inode->i_mapping);
7006
76dda93c 7007 /* close the racy window with snapshot create/destroy ioctl */
33345d01 7008 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7009 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 7010 /*
7011 * We want to reserve the absolute worst case amount of items. So if
7012 * both inodes are subvols and we need to unlink them then that would
7013 * require 4 item modifications, but if they are both normal inodes it
7014 * would require 5 item modifications, so we'll assume their normal
7015 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7016 * should cover the worst case number of items we'll modify.
7017 */
7018 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 7019 if (IS_ERR(trans)) {
7020 ret = PTR_ERR(trans);
7021 goto out_notrans;
7022 }
76dda93c 7023
4df27c4d
YZ		 7024 if (dest != root)
7025 btrfs_record_root_in_trans(trans, dest);
5f39d397 7026
a5719521
YZ		 7027 ret = btrfs_set_inode_index(new_dir, &index);
7028 if (ret)
7029 goto out_fail;
5a3f23d5 7030
33345d01 7031 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7032 /* force full log commit if subvolume involved. */
7033 root->fs_info->last_trans_log_full_commit = trans->transid;
7034 } else {
a5719521
YZ		 7035 ret = btrfs_insert_inode_ref(trans, dest,
7036 new_dentry->d_name.name,
7037 new_dentry->d_name.len,
33345d01
LZ		 7038 old_ino,
7039 btrfs_ino(new_dir), index);
a5719521
YZ		 7040 if (ret)
7041 goto out_fail;
4df27c4d
YZ		 7042 /*
7043 * this is an ugly little race, but the rename is required
7044 * to make sure that if we crash, the inode is either at the
7045 * old name or the new one. pinning the log transaction lets
7046 * us make sure we don't allow a log commit to come in after
7047 * we unlink the name but before we add the new name back in.
7048 */
7049 btrfs_pin_log_trans(root);
7050 }
5a3f23d5
CM		 7051 /*
7052 * make sure the inode gets flushed if it is replacing
7053 * something.
7054 */
33345d01 7055 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7056 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7057
39279cc3
CM		 7058 old_dir->i_ctime = old_dir->i_mtime = ctime;
7059 new_dir->i_ctime = new_dir->i_mtime = ctime;
7060 old_inode->i_ctime = ctime;
5f39d397 7061
12fcfd22
CM		 7062 if (old_dentry->d_parent != new_dentry->d_parent)
7063 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7064
33345d01 7065 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7066 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7067 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7068 old_dentry->d_name.name,
7069 old_dentry->d_name.len);
7070 } else {
92986796
AV		 7071 ret = __btrfs_unlink_inode(trans, root, old_dir,
7072 old_dentry->d_inode,
7073 old_dentry->d_name.name,
7074 old_dentry->d_name.len);
7075 if (!ret)
7076 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d
YZ		 7077 }
7078 BUG_ON(ret);
39279cc3
CM		 7079
7080 if (new_inode) {
7081 new_inode->i_ctime = CURRENT_TIME;
33345d01 7082 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 7083 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7084 root_objectid = BTRFS_I(new_inode)->location.objectid;
7085 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7086 root_objectid,
7087 new_dentry->d_name.name,
7088 new_dentry->d_name.len);
7089 BUG_ON(new_inode->i_nlink == 0);
7090 } else {
7091 ret = btrfs_unlink_inode(trans, dest, new_dir,
7092 new_dentry->d_inode,
7093 new_dentry->d_name.name,
7094 new_dentry->d_name.len);
7095 }
7096 BUG_ON(ret);
7b128766 7097 if (new_inode->i_nlink == 0) {
e02119d5 7098 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7099 BUG_ON(ret);
7b128766 7100 }
39279cc3 7101 }
aec7477b 7102
75e7cb7f
LB		 7103 fixup_inode_flags(new_dir, old_inode);
7104
4df27c4d
YZ		 7105 ret = btrfs_add_link(trans, new_dir, old_inode,
7106 new_dentry->d_name.name,
a5719521 7107 new_dentry->d_name.len, 0, index);
4df27c4d 7108 BUG_ON(ret);
39279cc3 7109
33345d01 7110 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7111 struct dentry *parent = new_dentry->d_parent;
6a912213 7112 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7113 btrfs_end_log_trans(root);
7114 }
39279cc3 7115out_fail:
7ad85bb7 7116 btrfs_end_transaction(trans, root);
b44c59a8 7117out_notrans:
33345d01 7118 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7119 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7120
39279cc3
CM		 7121 return ret;
7122}
7123
d352ac68
CM		 7124/*
7125 * some fairly slow code that needs optimization. This walks the list
7126 * of all the inodes with pending delalloc and forces them to disk.
7127 */
24bbcf04 7128int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7129{
7130 struct list_head *head = &root->fs_info->delalloc_inodes;
7131 struct btrfs_inode *binode;
5b21f2ed 7132 struct inode *inode;
ea8c2819 7133
c146afad
YZ		 7134 if (root->fs_info->sb->s_flags & MS_RDONLY)
7135 return -EROFS;
7136
75eff68e 7137 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7138 while (!list_empty(head)) {
ea8c2819
CM		 7139 binode = list_entry(head->next, struct btrfs_inode,
7140 delalloc_inodes);
5b21f2ed
ZY		 7141 inode = igrab(&binode->vfs_inode);
7142 if (!inode)
7143 list_del_init(&binode->delalloc_inodes);
75eff68e 7144 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7145 if (inode) {
8c8bee1d 7146 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7147 if (delay_iput)
7148 btrfs_add_delayed_iput(inode);
7149 else
7150 iput(inode);
5b21f2ed
ZY		 7151 }
7152 cond_resched();
75eff68e 7153 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7154 }
75eff68e 7155 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7156
7157 /* the filemap_flush will queue IO into the worker threads, but
7158 * we have to make sure the IO is actually started and that
7159 * ordered extents get created before we return
7160 */
7161 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7162 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7163 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7164 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7165 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7166 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7167 }
7168 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7169 return 0;
7170}
7171
39279cc3
CM		 7172static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7173 const char *symname)
7174{
7175 struct btrfs_trans_handle *trans;
7176 struct btrfs_root *root = BTRFS_I(dir)->root;
7177 struct btrfs_path *path;
7178 struct btrfs_key key;
1832a6d5 7179 struct inode *inode = NULL;
39279cc3
CM		 7180 int err;
7181 int drop_inode = 0;
7182 u64 objectid;
00e4e6b3 7183 u64 index = 0 ;
39279cc3
CM		 7184 int name_len;
7185 int datasize;
5f39d397 7186 unsigned long ptr;
39279cc3 7187 struct btrfs_file_extent_item *ei;
5f39d397 7188 struct extent_buffer *leaf;
1832a6d5 7189 unsigned long nr = 0;
39279cc3
CM		 7190
7191 name_len = strlen(symname) + 1;
7192 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7193 return -ENAMETOOLONG;
1832a6d5 7194
9ed74f2d
JB		 7195 /*
7196 * 2 items for inode item and ref
7197 * 2 items for dir items
7198 * 1 item for xattr if selinux is on
7199 */
a22285a6
YZ		 7200 trans = btrfs_start_transaction(root, 5);
7201 if (IS_ERR(trans))
7202 return PTR_ERR(trans);
1832a6d5 7203
581bb050
LZ		 7204 err = btrfs_find_free_ino(root, &objectid);
7205 if (err)
7206 goto out_unlock;
7207
aec7477b 7208 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7209 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7210 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7211 if (IS_ERR(inode)) {
7212 err = PTR_ERR(inode);
39279cc3 7213 goto out_unlock;
7cf96da3 7214 }
39279cc3 7215
2a7dba39 7216 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7217 if (err) {
7218 drop_inode = 1;
7219 goto out_unlock;
7220 }
7221
ad19db71
CS		 7222 /*
7223 * If the active LSM wants to access the inode during
7224 * d_instantiate it needs these. Smack checks to see
7225 * if the filesystem supports xattrs by looking at the
7226 * ops vector.
7227 */
7228 inode->i_fop = &btrfs_file_operations;
7229 inode->i_op = &btrfs_file_inode_operations;
7230
a1b075d2 7231 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7232 if (err)
7233 drop_inode = 1;
7234 else {
7235 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7236 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 7237 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7238 }
39279cc3
CM		 7239 if (drop_inode)
7240 goto out_unlock;
7241
7242 path = btrfs_alloc_path();
d8926bb3
MF		 7243 if (!path) {
7244 err = -ENOMEM;
7245 drop_inode = 1;
7246 goto out_unlock;
7247 }
33345d01 7248 key.objectid = btrfs_ino(inode);
39279cc3 7249 key.offset = 0;
39279cc3
CM		 7250 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7251 datasize = btrfs_file_extent_calc_inline_size(name_len);
7252 err = btrfs_insert_empty_item(trans, root, path, &key,
7253 datasize);
54aa1f4d
CM		 7254 if (err) {
7255 drop_inode = 1;
b0839166 7256 btrfs_free_path(path);
54aa1f4d
CM		 7257 goto out_unlock;
7258 }
5f39d397
CM		 7259 leaf = path->nodes[0];
7260 ei = btrfs_item_ptr(leaf, path->slots[0],
7261 struct btrfs_file_extent_item);
7262 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7263 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7264 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7265 btrfs_set_file_extent_encryption(leaf, ei, 0);
7266 btrfs_set_file_extent_compression(leaf, ei, 0);
7267 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7268 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7269
39279cc3 7270 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7271 write_extent_buffer(leaf, symname, ptr, name_len);
7272 btrfs_mark_buffer_dirty(leaf);
39279cc3 7273 btrfs_free_path(path);
5f39d397 7274
39279cc3
CM		 7275 inode->i_op = &btrfs_symlink_inode_operations;
7276 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7277 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7278 inode_set_bytes(inode, name_len);
dbe674a9 7279 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7280 err = btrfs_update_inode(trans, root, inode);
7281 if (err)
7282 drop_inode = 1;
39279cc3
CM		 7283
7284out_unlock:
08c422c2
AV		 7285 if (!err)
7286 d_instantiate(dentry, inode);
d3c2fdcf 7287 nr = trans->blocks_used;
7ad85bb7 7288 btrfs_end_transaction(trans, root);
39279cc3
CM		 7289 if (drop_inode) {
7290 inode_dec_link_count(inode);
7291 iput(inode);
7292 }
d3c2fdcf 7293 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7294 return err;
7295}
16432985 7296
0af3d00b
JB		 7297static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7298 u64 start, u64 num_bytes, u64 min_size,
7299 loff_t actual_len, u64 *alloc_hint,
7300 struct btrfs_trans_handle *trans)
d899e052 7301{
d899e052
YZ		 7302 struct btrfs_root *root = BTRFS_I(inode)->root;
7303 struct btrfs_key ins;
d899e052 7304 u64 cur_offset = start;
55a61d1d 7305 u64 i_size;
d899e052 7306 int ret = 0;
0af3d00b 7307 bool own_trans = true;
d899e052 7308
0af3d00b
JB		 7309 if (trans)
7310 own_trans = false;
d899e052 7311 while (num_bytes > 0) {
0af3d00b
JB		 7312 if (own_trans) {
7313 trans = btrfs_start_transaction(root, 3);
7314 if (IS_ERR(trans)) {
7315 ret = PTR_ERR(trans);
7316 break;
7317 }
5a303d5d
YZ		 7318 }
7319
efa56464
YZ		 7320 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7321 0, *alloc_hint, (u64)-1, &ins, 1);
5a303d5d 7322 if (ret) {
0af3d00b
JB		 7323 if (own_trans)
7324 btrfs_end_transaction(trans, root);
a22285a6 7325 break;
d899e052 7326 }
5a303d5d 7327
d899e052
YZ		 7328 ret = insert_reserved_file_extent(trans, inode,
7329 cur_offset, ins.objectid,
7330 ins.offset, ins.offset,
920bbbfb 7331 ins.offset, 0, 0, 0,
d899e052
YZ		 7332 BTRFS_FILE_EXTENT_PREALLOC);
7333 BUG_ON(ret);
a1ed835e
CM		 7334 btrfs_drop_extent_cache(inode, cur_offset,
7335 cur_offset + ins.offset -1, 0);
5a303d5d 7336
d899e052
YZ		 7337 num_bytes -= ins.offset;
7338 cur_offset += ins.offset;
efa56464 7339 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7340
d899e052 7341 inode->i_ctime = CURRENT_TIME;
6cbff00f 7342 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7343 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7344 (actual_len > inode->i_size) &&
7345 (cur_offset > inode->i_size)) {
d1ea6a61 7346 if (cur_offset > actual_len)
55a61d1d 7347 i_size = actual_len;
d1ea6a61 7348 else
55a61d1d
JB		 7349 i_size = cur_offset;
7350 i_size_write(inode, i_size);
7351 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7352 }
7353
d899e052
YZ		 7354 ret = btrfs_update_inode(trans, root, inode);
7355 BUG_ON(ret);
d899e052 7356
0af3d00b
JB		 7357 if (own_trans)
7358 btrfs_end_transaction(trans, root);
5a303d5d 7359 }
d899e052
YZ		 7360 return ret;
7361}
7362
0af3d00b
JB		 7363int btrfs_prealloc_file_range(struct inode *inode, int mode,
7364 u64 start, u64 num_bytes, u64 min_size,
7365 loff_t actual_len, u64 *alloc_hint)
7366{
7367 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7368 min_size, actual_len, alloc_hint,
7369 NULL);
7370}
7371
7372int btrfs_prealloc_file_range_trans(struct inode *inode,
7373 struct btrfs_trans_handle *trans, int mode,
7374 u64 start, u64 num_bytes, u64 min_size,
7375 loff_t actual_len, u64 *alloc_hint)
7376{
7377 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7378 min_size, actual_len, alloc_hint, trans);
7379}
7380
e6dcd2dc
CM		 7381static int btrfs_set_page_dirty(struct page *page)
7382{
e6dcd2dc
CM		 7383 return __set_page_dirty_nobuffers(page);
7384}
7385
10556cb2 7386static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7387{
b83cc969 7388 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7389 umode_t mode = inode->i_mode;
b83cc969 7390
cb6db4e5
JM		 7391 if (mask & MAY_WRITE &&
7392 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7393 if (btrfs_root_readonly(root))
7394 return -EROFS;
7395 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7396 return -EACCES;
7397 }
2830ba7f 7398 return generic_permission(inode, mask);
fdebe2bd 7399}
39279cc3 7400
6e1d5dcc 7401static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7402 .getattr = btrfs_getattr,
39279cc3
CM		 7403 .lookup = btrfs_lookup,
7404 .create = btrfs_create,
7405 .unlink = btrfs_unlink,
7406 .link = btrfs_link,
7407 .mkdir = btrfs_mkdir,
7408 .rmdir = btrfs_rmdir,
7409 .rename = btrfs_rename,
7410 .symlink = btrfs_symlink,
7411 .setattr = btrfs_setattr,
618e21d5 7412 .mknod = btrfs_mknod,
95819c05
CH		 7413 .setxattr = btrfs_setxattr,
7414 .getxattr = btrfs_getxattr,
5103e947 7415 .listxattr = btrfs_listxattr,
95819c05 7416 .removexattr = btrfs_removexattr,
fdebe2bd 7417 .permission = btrfs_permission,
4e34e719 7418 .get_acl = btrfs_get_acl,
39279cc3 7419};
6e1d5dcc 7420static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7421 .lookup = btrfs_lookup,
fdebe2bd 7422 .permission = btrfs_permission,
4e34e719 7423 .get_acl = btrfs_get_acl,
39279cc3 7424};
76dda93c 7425
828c0950 7426static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7427 .llseek = generic_file_llseek,
7428 .read = generic_read_dir,
cbdf5a24 7429 .readdir = btrfs_real_readdir,
34287aa3 7430 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7431#ifdef CONFIG_COMPAT
34287aa3 7432 .compat_ioctl = btrfs_ioctl,
39279cc3 7433#endif
6bf13c0c 7434 .release = btrfs_release_file,
e02119d5 7435 .fsync = btrfs_sync_file,
39279cc3
CM		 7436};
7437
d1310b2e 7438static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7439 .fill_delalloc = run_delalloc_range,
065631f6 7440 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7441 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7442 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7443 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7444 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 7445 .set_bit_hook = btrfs_set_bit_hook,
7446 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7447 .merge_extent_hook = btrfs_merge_extent_hook,
7448 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7449};
7450
35054394
CM		 7451/*
7452 * btrfs doesn't support the bmap operation because swapfiles
7453 * use bmap to make a mapping of extents in the file. They assume
7454 * these extents won't change over the life of the file and they
7455 * use the bmap result to do IO directly to the drive.
7456 *
7457 * the btrfs bmap call would return logical addresses that aren't
7458 * suitable for IO and they also will change frequently as COW
7459 * operations happen. So, swapfile + btrfs == corruption.
7460 *
7461 * For now we're avoiding this by dropping bmap.
7462 */
7f09410b 7463static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7464 .readpage = btrfs_readpage,
7465 .writepage = btrfs_writepage,
b293f02e 7466 .writepages = btrfs_writepages,
3ab2fb5a 7467 .readpages = btrfs_readpages,
16432985 7468 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7469 .invalidatepage = btrfs_invalidatepage,
7470 .releasepage = btrfs_releasepage,
e6dcd2dc 7471 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7472 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7473};
7474
7f09410b 7475static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7476 .readpage = btrfs_readpage,
7477 .writepage = btrfs_writepage,
2bf5a725
CM		 7478 .invalidatepage = btrfs_invalidatepage,
7479 .releasepage = btrfs_releasepage,
39279cc3
CM		 7480};
7481
6e1d5dcc 7482static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7483 .getattr = btrfs_getattr,
7484 .setattr = btrfs_setattr,
95819c05
CH		 7485 .setxattr = btrfs_setxattr,
7486 .getxattr = btrfs_getxattr,
5103e947 7487 .listxattr = btrfs_listxattr,
95819c05 7488 .removexattr = btrfs_removexattr,
fdebe2bd 7489 .permission = btrfs_permission,
1506fcc8 7490 .fiemap = btrfs_fiemap,
4e34e719 7491 .get_acl = btrfs_get_acl,
39279cc3 7492};
6e1d5dcc 7493static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7494 .getattr = btrfs_getattr,
7495 .setattr = btrfs_setattr,
fdebe2bd 7496 .permission = btrfs_permission,
95819c05
CH		 7497 .setxattr = btrfs_setxattr,
7498 .getxattr = btrfs_getxattr,
33268eaf 7499 .listxattr = btrfs_listxattr,
95819c05 7500 .removexattr = btrfs_removexattr,
4e34e719 7501 .get_acl = btrfs_get_acl,
618e21d5 7502};
6e1d5dcc 7503static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7504 .readlink = generic_readlink,
7505 .follow_link = page_follow_link_light,
7506 .put_link = page_put_link,
f209561a 7507 .getattr = btrfs_getattr,
22c44fe6 7508 .setattr = btrfs_setattr,
fdebe2bd 7509 .permission = btrfs_permission,
0279b4cd
JO		 7510 .setxattr = btrfs_setxattr,
7511 .getxattr = btrfs_getxattr,
7512 .listxattr = btrfs_listxattr,
7513 .removexattr = btrfs_removexattr,
4e34e719 7514 .get_acl = btrfs_get_acl,
39279cc3 7515};
76dda93c 7516
82d339d9 7517const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7518 .d_delete = btrfs_dentry_delete,
b4aff1f8 7519 .d_release = btrfs_dentry_release,
76dda93c 7520};
Linux 6.x block layer and io_uring tree(s)