projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Simplify btrfs_submit_bio_hook
[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);
cea9e445 1428
d397712b 1429 if (map_length < length + size)
239b14b3 1430 return 1;
411fc6bc 1431 return ret;
239b14b3
CM		 1432}
1433
d352ac68
CM		 1434/*
1435 * in order to insert checksums into the metadata in large chunks,
1436 * we wait until bio submission time. All the pages in the bio are
1437 * checksummed and sums are attached onto the ordered extent record.
1438 *
1439 * At IO completion time the cums attached on the ordered extent record
1440 * are inserted into the btree
1441 */
d397712b
CM		 1442static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443 struct bio *bio, int mirror_num,
eaf25d93
CM		 1444 unsigned long bio_flags,
1445 u64 bio_offset)
065631f6 1446{
065631f6 1447 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1448 int ret = 0;
e015640f 1449
d20f7043 1450 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1451 BUG_ON(ret);
4a69a410
CM		 1452 return 0;
1453}
e015640f 1454
4a69a410
CM		 1455/*
1456 * in order to insert checksums into the metadata in large chunks,
1457 * we wait until bio submission time. All the pages in the bio are
1458 * checksummed and sums are attached onto the ordered extent record.
1459 *
1460 * At IO completion time the cums attached on the ordered extent record
1461 * are inserted into the btree
1462 */
b2950863 1463static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1464 int mirror_num, unsigned long bio_flags,
1465 u64 bio_offset)
4a69a410
CM		 1466{
1467 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1468 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1469}
1470
d352ac68 1471/*
cad321ad
CM		 1472 * extent_io.c submission hook. This does the right thing for csum calculation
1473 * on write, or reading the csums from the tree before a read
d352ac68 1474 */
b2950863 1475static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1476 int mirror_num, unsigned long bio_flags,
1477 u64 bio_offset)
44b8bd7e
CM		 1478{
1479 struct btrfs_root *root = BTRFS_I(inode)->root;
1480 int ret = 0;
19b9bdb0 1481 int skip_sum;
0417341e 1482 int metadata = 0;
44b8bd7e 1483
6cbff00f 1484 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1485
2cf8572d 1486 if (btrfs_is_free_space_inode(root, inode))
0417341e
JM		 1487 metadata = 2;
1488
1489 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
e6dcd2dc 1490 BUG_ON(ret);
065631f6 1491
7b6d91da 1492 if (!(rw & REQ_WRITE)) {
d20f7043 1493 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1494 return btrfs_submit_compressed_read(inode, bio,
1495 mirror_num, bio_flags);
c2db1073
TI		 1496 } else if (!skip_sum) {
1497 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1498 if (ret)
1499 return ret;
1500 }
4d1b5fb4 1501 goto mapit;
19b9bdb0 1502 } else if (!skip_sum) {
17d217fe
YZ		 1503 /* csum items have already been cloned */
1504 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1505 goto mapit;
19b9bdb0
CM		 1506 /* we're doing a write, do the async checksumming */
1507 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1508 inode, rw, bio, mirror_num,
eaf25d93
CM		 1509 bio_flags, bio_offset,
1510 __btrfs_submit_bio_start,
4a69a410 1511 __btrfs_submit_bio_done);
19b9bdb0
CM		 1512 }
1513
0b86a832 1514mapit:
8b712842 1515 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1516}
6885f308 1517
d352ac68
CM		 1518/*
1519 * given a list of ordered sums record them in the inode. This happens
1520 * at IO completion time based on sums calculated at bio submission time.
1521 */
ba1da2f4 1522static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1523 struct inode *inode, u64 file_offset,
1524 struct list_head *list)
1525{
e6dcd2dc
CM		 1526 struct btrfs_ordered_sum *sum;
1527
c6e30871 1528 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1529 btrfs_csum_file_blocks(trans,
1530 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1531 }
1532 return 0;
1533}
1534
2ac55d41
JB		 1535int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1536 struct extent_state **cached_state)
ea8c2819 1537{
d397712b 1538 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1539 WARN_ON(1);
ea8c2819 1540 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1541 cached_state, GFP_NOFS);
ea8c2819
CM		 1542}
1543
d352ac68 1544/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1545struct btrfs_writepage_fixup {
1546 struct page *page;
1547 struct btrfs_work work;
1548};
1549
b2950863 1550static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1551{
1552 struct btrfs_writepage_fixup *fixup;
1553 struct btrfs_ordered_extent *ordered;
2ac55d41 1554 struct extent_state *cached_state = NULL;
247e743c
CM		 1555 struct page *page;
1556 struct inode *inode;
1557 u64 page_start;
1558 u64 page_end;
87826df0 1559 int ret;
247e743c
CM		 1560
1561 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1562 page = fixup->page;
4a096752 1563again:
247e743c
CM		 1564 lock_page(page);
1565 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1566 ClearPageChecked(page);
1567 goto out_page;
1568 }
1569
1570 inode = page->mapping->host;
1571 page_start = page_offset(page);
1572 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1573
2ac55d41
JB		 1574 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1575 &cached_state, GFP_NOFS);
4a096752
CM		 1576
1577 /* already ordered? We're done */
8b62b72b 1578 if (PagePrivate2(page))
247e743c 1579 goto out;
4a096752
CM		 1580
1581 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1582 if (ordered) {
2ac55d41
JB		 1583 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1584 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1585 unlock_page(page);
1586 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1587 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1588 goto again;
1589 }
247e743c 1590
87826df0
JM		 1591 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1592 if (ret) {
1593 mapping_set_error(page->mapping, ret);
1594 end_extent_writepage(page, ret, page_start, page_end);
1595 ClearPageChecked(page);
1596 goto out;
1597 }
1598
2ac55d41 1599 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1600 ClearPageChecked(page);
87826df0 1601 set_page_dirty(page);
247e743c 1602out:
2ac55d41
JB		 1603 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1604 &cached_state, GFP_NOFS);
247e743c
CM		 1605out_page:
1606 unlock_page(page);
1607 page_cache_release(page);
b897abec 1608 kfree(fixup);
247e743c
CM		 1609}
1610
1611/*
1612 * There are a few paths in the higher layers of the kernel that directly
1613 * set the page dirty bit without asking the filesystem if it is a
1614 * good idea. This causes problems because we want to make sure COW
1615 * properly happens and the data=ordered rules are followed.
1616 *
c8b97818 1617 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1618 * hasn't been properly setup for IO. We kick off an async process
1619 * to fix it up. The async helper will wait for ordered extents, set
1620 * the delalloc bit and make it safe to write the page.
1621 */
b2950863 1622static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1623{
1624 struct inode *inode = page->mapping->host;
1625 struct btrfs_writepage_fixup *fixup;
1626 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1627
8b62b72b
CM		 1628 /* this page is properly in the ordered list */
1629 if (TestClearPagePrivate2(page))
247e743c
CM		 1630 return 0;
1631
1632 if (PageChecked(page))
1633 return -EAGAIN;
1634
1635 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1636 if (!fixup)
1637 return -EAGAIN;
f421950f 1638
247e743c
CM		 1639 SetPageChecked(page);
1640 page_cache_get(page);
1641 fixup->work.func = btrfs_writepage_fixup_worker;
1642 fixup->page = page;
1643 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1644 return -EBUSY;
247e743c
CM		 1645}
1646
d899e052
YZ		 1647static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1648 struct inode *inode, u64 file_pos,
1649 u64 disk_bytenr, u64 disk_num_bytes,
1650 u64 num_bytes, u64 ram_bytes,
1651 u8 compression, u8 encryption,
1652 u16 other_encoding, int extent_type)
1653{
1654 struct btrfs_root *root = BTRFS_I(inode)->root;
1655 struct btrfs_file_extent_item *fi;
1656 struct btrfs_path *path;
1657 struct extent_buffer *leaf;
1658 struct btrfs_key ins;
1659 u64 hint;
1660 int ret;
1661
1662 path = btrfs_alloc_path();
d8926bb3
MF		 1663 if (!path)
1664 return -ENOMEM;
d899e052 1665
b9473439 1666 path->leave_spinning = 1;
a1ed835e
CM		 1667
1668 /*
1669 * we may be replacing one extent in the tree with another.
1670 * The new extent is pinned in the extent map, and we don't want
1671 * to drop it from the cache until it is completely in the btree.
1672 *
1673 * So, tell btrfs_drop_extents to leave this extent in the cache.
1674 * the caller is expected to unpin it and allow it to be merged
1675 * with the others.
1676 */
920bbbfb
YZ		 1677 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1678 &hint, 0);
d899e052
YZ		 1679 BUG_ON(ret);
1680
33345d01 1681 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1682 ins.offset = file_pos;
1683 ins.type = BTRFS_EXTENT_DATA_KEY;
1684 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1685 BUG_ON(ret);
1686 leaf = path->nodes[0];
1687 fi = btrfs_item_ptr(leaf, path->slots[0],
1688 struct btrfs_file_extent_item);
1689 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1690 btrfs_set_file_extent_type(leaf, fi, extent_type);
1691 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1692 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1693 btrfs_set_file_extent_offset(leaf, fi, 0);
1694 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1695 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1696 btrfs_set_file_extent_compression(leaf, fi, compression);
1697 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1698 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1699
1700 btrfs_unlock_up_safe(path, 1);
1701 btrfs_set_lock_blocking(leaf);
1702
d899e052
YZ		 1703 btrfs_mark_buffer_dirty(leaf);
1704
1705 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1706
1707 ins.objectid = disk_bytenr;
1708 ins.offset = disk_num_bytes;
1709 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1710 ret = btrfs_alloc_reserved_file_extent(trans, root,
1711 root->root_key.objectid,
33345d01 1712 btrfs_ino(inode), file_pos, &ins);
d899e052 1713 BUG_ON(ret);
d899e052 1714 btrfs_free_path(path);
b9473439 1715
d899e052
YZ		 1716 return 0;
1717}
1718
5d13a98f
CM		 1719/*
1720 * helper function for btrfs_finish_ordered_io, this
1721 * just reads in some of the csum leaves to prime them into ram
1722 * before we start the transaction. It limits the amount of btree
1723 * reads required while inside the transaction.
1724 */
d352ac68
CM		 1725/* as ordered data IO finishes, this gets called so we can finish
1726 * an ordered extent if the range of bytes in the file it covers are
1727 * fully written.
1728 */
211f90e6 1729static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1730{
e6dcd2dc 1731 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1732 struct btrfs_trans_handle *trans = NULL;
5d13a98f 1733 struct btrfs_ordered_extent *ordered_extent = NULL;
e6dcd2dc 1734 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1735 struct extent_state *cached_state = NULL;
261507a0 1736 int compress_type = 0;
e6dcd2dc 1737 int ret;
82d5902d 1738 bool nolock;
e6dcd2dc 1739
5a1a3df1
JB		 1740 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1741 end - start + 1);
ba1da2f4 1742 if (!ret)
e6dcd2dc 1743 return 0;
e6dcd2dc 1744 BUG_ON(!ordered_extent);
efd049fb 1745
2cf8572d 1746 nolock = btrfs_is_free_space_inode(root, inode);
0cb59c99 1747
c2167754
YZ		 1748 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1749 BUG_ON(!list_empty(&ordered_extent->list));
1750 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1751 if (!ret) {
0cb59c99 1752 if (nolock)
7a7eaa40 1753 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1754 else
7a7eaa40 1755 trans = btrfs_join_transaction(root);
3612b495 1756 BUG_ON(IS_ERR(trans));
0ca1f7ce 1757 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2115133f 1758 ret = btrfs_update_inode_fallback(trans, root, inode);
c2167754 1759 BUG_ON(ret);
c2167754
YZ		 1760 }
1761 goto out;
1762 }
e6dcd2dc 1763
2ac55d41
JB		 1764 lock_extent_bits(io_tree, ordered_extent->file_offset,
1765 ordered_extent->file_offset + ordered_extent->len - 1,
1766 0, &cached_state, GFP_NOFS);
e6dcd2dc 1767
0cb59c99 1768 if (nolock)
7a7eaa40 1769 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1770 else
7a7eaa40 1771 trans = btrfs_join_transaction(root);
3612b495 1772 BUG_ON(IS_ERR(trans));
0ca1f7ce 1773 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1774
c8b97818 1775 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1776 compress_type = ordered_extent->compress_type;
d899e052 1777 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1778 BUG_ON(compress_type);
920bbbfb 1779 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1780 ordered_extent->file_offset,
1781 ordered_extent->file_offset +
1782 ordered_extent->len);
1783 BUG_ON(ret);
1784 } else {
0af3d00b 1785 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1786 ret = insert_reserved_file_extent(trans, inode,
1787 ordered_extent->file_offset,
1788 ordered_extent->start,
1789 ordered_extent->disk_len,
1790 ordered_extent->len,
1791 ordered_extent->len,
261507a0 1792 compress_type, 0, 0,
d899e052 1793 BTRFS_FILE_EXTENT_REG);
a1ed835e
CM		 1794 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1795 ordered_extent->file_offset,
1796 ordered_extent->len);
d899e052
YZ		 1797 BUG_ON(ret);
1798 }
2ac55d41
JB		 1799 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1800 ordered_extent->file_offset +
1801 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1802
e6dcd2dc
CM		 1803 add_pending_csums(trans, inode, ordered_extent->file_offset,
1804 &ordered_extent->list);
1805
1ef30be1 1806 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1807 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2115133f 1808 ret = btrfs_update_inode_fallback(trans, root, inode);
1ef30be1
JB		 1809 BUG_ON(ret);
1810 }
1811 ret = 0;
c2167754 1812out:
5b0e95bf 1813 if (root != root->fs_info->tree_root)
0cb59c99 1814 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
5b0e95bf
JB		 1815 if (trans) {
1816 if (nolock)
0cb59c99 1817 btrfs_end_transaction_nolock(trans, root);
5b0e95bf 1818 else
0cb59c99
JB		 1819 btrfs_end_transaction(trans, root);
1820 }
1821
e6dcd2dc
CM		 1822 /* once for us */
1823 btrfs_put_ordered_extent(ordered_extent);
1824 /* once for the tree */
1825 btrfs_put_ordered_extent(ordered_extent);
1826
e6dcd2dc
CM		 1827 return 0;
1828}
1829
b2950863 1830static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1831 struct extent_state *state, int uptodate)
1832{
1abe9b8a 1833 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1834
8b62b72b 1835 ClearPagePrivate2(page);
211f90e6
CM		 1836 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1837}
1838
d352ac68
CM		 1839/*
1840 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 1841 * if there's a match, we allow the bio to finish. If not, the code in
1842 * extent_io.c will try to find good copies for us.
d352ac68 1843 */
b2950863 1844static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1845 struct extent_state *state)
07157aac 1846{
35ebb934 1847 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1848 struct inode *inode = page->mapping->host;
d1310b2e 1849 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1850 char *kaddr;
aadfeb6e 1851 u64 private = ~(u32)0;
07157aac 1852 int ret;
ff79f819
CM		 1853 struct btrfs_root *root = BTRFS_I(inode)->root;
1854 u32 csum = ~(u32)0;
d1310b2e 1855
d20f7043
CM		 1856 if (PageChecked(page)) {
1857 ClearPageChecked(page);
1858 goto good;
1859 }
6cbff00f
CH		 1860
1861 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 1862 goto good;
17d217fe
YZ		 1863
1864 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 1865 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 1866 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1867 GFP_NOFS);
b6cda9bc 1868 return 0;
17d217fe 1869 }
d20f7043 1870
c2e639f0 1871 if (state && state->start == start) {
70dec807
CM		 1872 private = state->private;
1873 ret = 0;
1874 } else {
1875 ret = get_state_private(io_tree, start, &private);
1876 }
9ab86c8e 1877 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1878 if (ret)
07157aac 1879 goto zeroit;
d397712b 1880
ff79f819
CM		 1881 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1882 btrfs_csum_final(csum, (char *)&csum);
d397712b 1883 if (csum != private)
07157aac 1884 goto zeroit;
d397712b 1885
9ab86c8e 1886 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1887good:
07157aac
CM		 1888 return 0;
1889
1890zeroit:
945d8962 1891 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 1892 "private %llu\n",
1893 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 1894 (unsigned long long)start, csum,
1895 (unsigned long long)private);
db94535d
CM		 1896 memset(kaddr + offset, 1, end - start + 1);
1897 flush_dcache_page(page);
9ab86c8e 1898 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1899 if (private == 0)
1900 return 0;
7e38326f 1901 return -EIO;
07157aac 1902}
b888db2b 1903
24bbcf04
YZ		 1904struct delayed_iput {
1905 struct list_head list;
1906 struct inode *inode;
1907};
1908
1909void btrfs_add_delayed_iput(struct inode *inode)
1910{
1911 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1912 struct delayed_iput *delayed;
1913
1914 if (atomic_add_unless(&inode->i_count, -1, 1))
1915 return;
1916
1917 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1918 delayed->inode = inode;
1919
1920 spin_lock(&fs_info->delayed_iput_lock);
1921 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1922 spin_unlock(&fs_info->delayed_iput_lock);
1923}
1924
1925void btrfs_run_delayed_iputs(struct btrfs_root *root)
1926{
1927 LIST_HEAD(list);
1928 struct btrfs_fs_info *fs_info = root->fs_info;
1929 struct delayed_iput *delayed;
1930 int empty;
1931
1932 spin_lock(&fs_info->delayed_iput_lock);
1933 empty = list_empty(&fs_info->delayed_iputs);
1934 spin_unlock(&fs_info->delayed_iput_lock);
1935 if (empty)
1936 return;
1937
1938 down_read(&root->fs_info->cleanup_work_sem);
1939 spin_lock(&fs_info->delayed_iput_lock);
1940 list_splice_init(&fs_info->delayed_iputs, &list);
1941 spin_unlock(&fs_info->delayed_iput_lock);
1942
1943 while (!list_empty(&list)) {
1944 delayed = list_entry(list.next, struct delayed_iput, list);
1945 list_del(&delayed->list);
1946 iput(delayed->inode);
1947 kfree(delayed);
1948 }
1949 up_read(&root->fs_info->cleanup_work_sem);
1950}
1951
d68fc57b
YZ		 1952enum btrfs_orphan_cleanup_state {
1953 ORPHAN_CLEANUP_STARTED = 1,
1954 ORPHAN_CLEANUP_DONE = 2,
1955};
1956
1957/*
42b2aa86 1958 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 1959 * files in the subvolume, it removes orphan item and frees block_rsv
1960 * structure.
1961 */
1962void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1963 struct btrfs_root *root)
1964{
90290e19 1965 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 1966 int ret;
1967
1968 if (!list_empty(&root->orphan_list) ||
1969 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1970 return;
1971
90290e19
JB		 1972 spin_lock(&root->orphan_lock);
1973 if (!list_empty(&root->orphan_list)) {
1974 spin_unlock(&root->orphan_lock);
1975 return;
1976 }
1977
1978 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1979 spin_unlock(&root->orphan_lock);
1980 return;
1981 }
1982
1983 block_rsv = root->orphan_block_rsv;
1984 root->orphan_block_rsv = NULL;
1985 spin_unlock(&root->orphan_lock);
1986
d68fc57b
YZ		 1987 if (root->orphan_item_inserted &&
1988 btrfs_root_refs(&root->root_item) > 0) {
1989 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1990 root->root_key.objectid);
1991 BUG_ON(ret);
1992 root->orphan_item_inserted = 0;
1993 }
1994
90290e19
JB		 1995 if (block_rsv) {
1996 WARN_ON(block_rsv->size > 0);
1997 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 1998 }
1999}
2000
7b128766
JB		 2001/*
2002 * This creates an orphan entry for the given inode in case something goes
2003 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2004 *
2005 * NOTE: caller of this function should reserve 5 units of metadata for
2006 * this function.
7b128766
JB		 2007 */
2008int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2009{
2010 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2011 struct btrfs_block_rsv *block_rsv = NULL;
2012 int reserve = 0;
2013 int insert = 0;
2014 int ret;
7b128766 2015
d68fc57b
YZ		 2016 if (!root->orphan_block_rsv) {
2017 block_rsv = btrfs_alloc_block_rsv(root);
b532402e
TI		 2018 if (!block_rsv)
2019 return -ENOMEM;
d68fc57b 2020 }
7b128766 2021
d68fc57b
YZ		 2022 spin_lock(&root->orphan_lock);
2023 if (!root->orphan_block_rsv) {
2024 root->orphan_block_rsv = block_rsv;
2025 } else if (block_rsv) {
2026 btrfs_free_block_rsv(root, block_rsv);
2027 block_rsv = NULL;
7b128766 2028 }
7b128766 2029
d68fc57b
YZ		 2030 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2031 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2032#if 0
2033 /*
2034 * For proper ENOSPC handling, we should do orphan
2035 * cleanup when mounting. But this introduces backward
2036 * compatibility issue.
2037 */
2038 if (!xchg(&root->orphan_item_inserted, 1))
2039 insert = 2;
2040 else
2041 insert = 1;
2042#endif
2043 insert = 1;
7b128766
JB		 2044 }
2045
d68fc57b
YZ		 2046 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2047 BTRFS_I(inode)->orphan_meta_reserved = 1;
2048 reserve = 1;
2049 }
2050 spin_unlock(&root->orphan_lock);
7b128766 2051
d68fc57b
YZ		 2052 /* grab metadata reservation from transaction handle */
2053 if (reserve) {
2054 ret = btrfs_orphan_reserve_metadata(trans, inode);
2055 BUG_ON(ret);
2056 }
7b128766 2057
d68fc57b
YZ		 2058 /* insert an orphan item to track this unlinked/truncated file */
2059 if (insert >= 1) {
33345d01 2060 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
ee4d89f0 2061 BUG_ON(ret && ret != -EEXIST);
d68fc57b
YZ		 2062 }
2063
2064 /* insert an orphan item to track subvolume contains orphan files */
2065 if (insert >= 2) {
2066 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2067 root->root_key.objectid);
2068 BUG_ON(ret);
2069 }
2070 return 0;
7b128766
JB		 2071}
2072
2073/*
2074 * We have done the truncate/delete so we can go ahead and remove the orphan
2075 * item for this particular inode.
2076 */
2077int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2078{
2079 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2080 int delete_item = 0;
2081 int release_rsv = 0;
7b128766
JB		 2082 int ret = 0;
2083
d68fc57b
YZ		 2084 spin_lock(&root->orphan_lock);
2085 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2086 list_del_init(&BTRFS_I(inode)->i_orphan);
2087 delete_item = 1;
7b128766
JB		 2088 }
2089
d68fc57b
YZ		 2090 if (BTRFS_I(inode)->orphan_meta_reserved) {
2091 BTRFS_I(inode)->orphan_meta_reserved = 0;
2092 release_rsv = 1;
7b128766 2093 }
d68fc57b 2094 spin_unlock(&root->orphan_lock);
7b128766 2095
d68fc57b 2096 if (trans && delete_item) {
33345d01 2097 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
d68fc57b
YZ		 2098 BUG_ON(ret);
2099 }
7b128766 2100
d68fc57b
YZ		 2101 if (release_rsv)
2102 btrfs_orphan_release_metadata(inode);
7b128766 2103
d68fc57b 2104 return 0;
7b128766
JB		 2105}
2106
2107/*
2108 * this cleans up any orphans that may be left on the list from the last use
2109 * of this root.
2110 */
66b4ffd1 2111int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2112{
2113 struct btrfs_path *path;
2114 struct extent_buffer *leaf;
7b128766
JB		 2115 struct btrfs_key key, found_key;
2116 struct btrfs_trans_handle *trans;
2117 struct inode *inode;
8f6d7f4f 2118 u64 last_objectid = 0;
7b128766
JB		 2119 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2120
d68fc57b 2121 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2122 return 0;
c71bf099
YZ		 2123
2124 path = btrfs_alloc_path();
66b4ffd1
JB		 2125 if (!path) {
2126 ret = -ENOMEM;
2127 goto out;
2128 }
7b128766
JB		 2129 path->reada = -1;
2130
2131 key.objectid = BTRFS_ORPHAN_OBJECTID;
2132 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2133 key.offset = (u64)-1;
2134
7b128766
JB		 2135 while (1) {
2136 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2137 if (ret < 0)
2138 goto out;
7b128766
JB		 2139
2140 /*
2141 * if ret == 0 means we found what we were searching for, which
25985edc 2142 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2143 * find the key and see if we have stuff that matches
2144 */
2145 if (ret > 0) {
66b4ffd1 2146 ret = 0;
7b128766
JB		 2147 if (path->slots[0] == 0)
2148 break;
2149 path->slots[0]--;
2150 }
2151
2152 /* pull out the item */
2153 leaf = path->nodes[0];
7b128766
JB		 2154 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2155
2156 /* make sure the item matches what we want */
2157 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2158 break;
2159 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2160 break;
2161
2162 /* release the path since we're done with it */
b3b4aa74 2163 btrfs_release_path(path);
7b128766
JB		 2164
2165 /*
2166 * this is where we are basically btrfs_lookup, without the
2167 * crossing root thing. we store the inode number in the
2168 * offset of the orphan item.
2169 */
8f6d7f4f
JB		 2170
2171 if (found_key.offset == last_objectid) {
2172 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2173 "stopping orphan cleanup\n");
2174 ret = -EINVAL;
2175 goto out;
2176 }
2177
2178 last_objectid = found_key.offset;
2179
5d4f98a2
YZ		 2180 found_key.objectid = found_key.offset;
2181 found_key.type = BTRFS_INODE_ITEM_KEY;
2182 found_key.offset = 0;
73f73415 2183 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2184 ret = PTR_RET(inode);
2185 if (ret && ret != -ESTALE)
66b4ffd1 2186 goto out;
7b128766 2187
f8e9e0b0
AJ		 2188 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2189 struct btrfs_root *dead_root;
2190 struct btrfs_fs_info *fs_info = root->fs_info;
2191 int is_dead_root = 0;
2192
2193 /*
2194 * this is an orphan in the tree root. Currently these
2195 * could come from 2 sources:
2196 * a) a snapshot deletion in progress
2197 * b) a free space cache inode
2198 * We need to distinguish those two, as the snapshot
2199 * orphan must not get deleted.
2200 * find_dead_roots already ran before us, so if this
2201 * is a snapshot deletion, we should find the root
2202 * in the dead_roots list
2203 */
2204 spin_lock(&fs_info->trans_lock);
2205 list_for_each_entry(dead_root, &fs_info->dead_roots,
2206 root_list) {
2207 if (dead_root->root_key.objectid ==
2208 found_key.objectid) {
2209 is_dead_root = 1;
2210 break;
2211 }
2212 }
2213 spin_unlock(&fs_info->trans_lock);
2214 if (is_dead_root) {
2215 /* prevent this orphan from being found again */
2216 key.offset = found_key.objectid - 1;
2217 continue;
2218 }
2219 }
7b128766 2220 /*
a8c9e576
JB		 2221 * Inode is already gone but the orphan item is still there,
2222 * kill the orphan item.
7b128766 2223 */
a8c9e576
JB		 2224 if (ret == -ESTALE) {
2225 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2226 if (IS_ERR(trans)) {
2227 ret = PTR_ERR(trans);
2228 goto out;
2229 }
a8c9e576
JB		 2230 ret = btrfs_del_orphan_item(trans, root,
2231 found_key.objectid);
2232 BUG_ON(ret);
5b21f2ed 2233 btrfs_end_transaction(trans, root);
7b128766
JB		 2234 continue;
2235 }
2236
a8c9e576
JB		 2237 /*
2238 * add this inode to the orphan list so btrfs_orphan_del does
2239 * the proper thing when we hit it
2240 */
2241 spin_lock(&root->orphan_lock);
2242 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2243 spin_unlock(&root->orphan_lock);
2244
7b128766
JB		 2245 /* if we have links, this was a truncate, lets do that */
2246 if (inode->i_nlink) {
a41ad394
JB		 2247 if (!S_ISREG(inode->i_mode)) {
2248 WARN_ON(1);
2249 iput(inode);
2250 continue;
2251 }
7b128766 2252 nr_truncate++;
66b4ffd1 2253 ret = btrfs_truncate(inode);
7b128766
JB		 2254 } else {
2255 nr_unlink++;
2256 }
2257
2258 /* this will do delete_inode and everything for us */
2259 iput(inode);
66b4ffd1
JB		 2260 if (ret)
2261 goto out;
7b128766 2262 }
3254c876
MX		 2263 /* release the path since we're done with it */
2264 btrfs_release_path(path);
2265
d68fc57b
YZ		 2266 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2267
2268 if (root->orphan_block_rsv)
2269 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2270 (u64)-1);
2271
2272 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2273 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2274 if (!IS_ERR(trans))
2275 btrfs_end_transaction(trans, root);
d68fc57b 2276 }
7b128766
JB		 2277
2278 if (nr_unlink)
2279 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2280 if (nr_truncate)
2281 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2282
2283out:
2284 if (ret)
2285 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2286 btrfs_free_path(path);
2287 return ret;
7b128766
JB		 2288}
2289
46a53cca
CM		 2290/*
2291 * very simple check to peek ahead in the leaf looking for xattrs. If we
2292 * don't find any xattrs, we know there can't be any acls.
2293 *
2294 * slot is the slot the inode is in, objectid is the objectid of the inode
2295 */
2296static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2297 int slot, u64 objectid)
2298{
2299 u32 nritems = btrfs_header_nritems(leaf);
2300 struct btrfs_key found_key;
2301 int scanned = 0;
2302
2303 slot++;
2304 while (slot < nritems) {
2305 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2306
2307 /* we found a different objectid, there must not be acls */
2308 if (found_key.objectid != objectid)
2309 return 0;
2310
2311 /* we found an xattr, assume we've got an acl */
2312 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2313 return 1;
2314
2315 /*
2316 * we found a key greater than an xattr key, there can't
2317 * be any acls later on
2318 */
2319 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2320 return 0;
2321
2322 slot++;
2323 scanned++;
2324
2325 /*
2326 * it goes inode, inode backrefs, xattrs, extents,
2327 * so if there are a ton of hard links to an inode there can
2328 * be a lot of backrefs. Don't waste time searching too hard,
2329 * this is just an optimization
2330 */
2331 if (scanned >= 8)
2332 break;
2333 }
2334 /* we hit the end of the leaf before we found an xattr or
2335 * something larger than an xattr. We have to assume the inode
2336 * has acls
2337 */
2338 return 1;
2339}
2340
d352ac68
CM		 2341/*
2342 * read an inode from the btree into the in-memory inode
2343 */
5d4f98a2 2344static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2345{
2346 struct btrfs_path *path;
5f39d397 2347 struct extent_buffer *leaf;
39279cc3 2348 struct btrfs_inode_item *inode_item;
0b86a832 2349 struct btrfs_timespec *tspec;
39279cc3
CM		 2350 struct btrfs_root *root = BTRFS_I(inode)->root;
2351 struct btrfs_key location;
46a53cca 2352 int maybe_acls;
618e21d5 2353 u32 rdev;
39279cc3 2354 int ret;
2f7e33d4
MX		 2355 bool filled = false;
2356
2357 ret = btrfs_fill_inode(inode, &rdev);
2358 if (!ret)
2359 filled = true;
39279cc3
CM		 2360
2361 path = btrfs_alloc_path();
1748f843
MF		 2362 if (!path)
2363 goto make_bad;
2364
d90c7321 2365 path->leave_spinning = 1;
39279cc3 2366 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2367
39279cc3 2368 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2369 if (ret)
39279cc3 2370 goto make_bad;
39279cc3 2371
5f39d397 2372 leaf = path->nodes[0];
2f7e33d4
MX		 2373
2374 if (filled)
2375 goto cache_acl;
2376
5f39d397
CM		 2377 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2378 struct btrfs_inode_item);
5f39d397 2379 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 2380 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
5f39d397
CM		 2381 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2382 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2383 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2384
2385 tspec = btrfs_inode_atime(inode_item);
2386 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2387 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2388
2389 tspec = btrfs_inode_mtime(inode_item);
2390 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2391 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2392
2393 tspec = btrfs_inode_ctime(inode_item);
2394 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2395 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2396
a76a3cd4 2397 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2398 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2399 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2400 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2401 inode->i_rdev = 0;
5f39d397
CM		 2402 rdev = btrfs_inode_rdev(leaf, inode_item);
2403
aec7477b 2404 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2405 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2406cache_acl:
46a53cca
CM		 2407 /*
2408 * try to precache a NULL acl entry for files that don't have
2409 * any xattrs or acls
2410 */
33345d01
LZ		 2411 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2412 btrfs_ino(inode));
72c04902
AV		 2413 if (!maybe_acls)
2414 cache_no_acl(inode);
46a53cca 2415
39279cc3 2416 btrfs_free_path(path);
39279cc3 2417
39279cc3 2418 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2419 case S_IFREG:
2420 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2421 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2422 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2423 inode->i_fop = &btrfs_file_operations;
2424 inode->i_op = &btrfs_file_inode_operations;
2425 break;
2426 case S_IFDIR:
2427 inode->i_fop = &btrfs_dir_file_operations;
2428 if (root == root->fs_info->tree_root)
2429 inode->i_op = &btrfs_dir_ro_inode_operations;
2430 else
2431 inode->i_op = &btrfs_dir_inode_operations;
2432 break;
2433 case S_IFLNK:
2434 inode->i_op = &btrfs_symlink_inode_operations;
2435 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2436 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2437 break;
618e21d5 2438 default:
0279b4cd 2439 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2440 init_special_inode(inode, inode->i_mode, rdev);
2441 break;
39279cc3 2442 }
6cbff00f
CH		 2443
2444 btrfs_update_iflags(inode);
39279cc3
CM		 2445 return;
2446
2447make_bad:
39279cc3 2448 btrfs_free_path(path);
39279cc3
CM		 2449 make_bad_inode(inode);
2450}
2451
d352ac68
CM		 2452/*
2453 * given a leaf and an inode, copy the inode fields into the leaf
2454 */
e02119d5
CM		 2455static void fill_inode_item(struct btrfs_trans_handle *trans,
2456 struct extent_buffer *leaf,
5f39d397 2457 struct btrfs_inode_item *item,
39279cc3
CM		 2458 struct inode *inode)
2459{
5f39d397
CM		 2460 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2461 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2462 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2463 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2464 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2465
2466 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2467 inode->i_atime.tv_sec);
2468 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2469 inode->i_atime.tv_nsec);
2470
2471 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2472 inode->i_mtime.tv_sec);
2473 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2474 inode->i_mtime.tv_nsec);
2475
2476 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2477 inode->i_ctime.tv_sec);
2478 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2479 inode->i_ctime.tv_nsec);
2480
a76a3cd4 2481 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2482 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2483 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2484 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2485 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2486 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2487 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2488}
2489
d352ac68
CM		 2490/*
2491 * copy everything in the in-memory inode into the btree.
2492 */
2115133f 2493static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2494 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2495{
2496 struct btrfs_inode_item *inode_item;
2497 struct btrfs_path *path;
5f39d397 2498 struct extent_buffer *leaf;
39279cc3
CM		 2499 int ret;
2500
2501 path = btrfs_alloc_path();
16cdcec7
MX		 2502 if (!path)
2503 return -ENOMEM;
2504
b9473439 2505 path->leave_spinning = 1;
16cdcec7
MX		 2506 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2507 1);
39279cc3
CM		 2508 if (ret) {
2509 if (ret > 0)
2510 ret = -ENOENT;
2511 goto failed;
2512 }
2513
b4ce94de 2514 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2515 leaf = path->nodes[0];
2516 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2517 struct btrfs_inode_item);
39279cc3 2518
e02119d5 2519 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2520 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2521 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2522 ret = 0;
2523failed:
39279cc3
CM		 2524 btrfs_free_path(path);
2525 return ret;
2526}
2527
2115133f
CM		 2528/*
2529 * copy everything in the in-memory inode into the btree.
2530 */
2531noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2532 struct btrfs_root *root, struct inode *inode)
2533{
2534 int ret;
2535
2536 /*
2537 * If the inode is a free space inode, we can deadlock during commit
2538 * if we put it into the delayed code.
2539 *
2540 * The data relocation inode should also be directly updated
2541 * without delay
2542 */
2543 if (!btrfs_is_free_space_inode(root, inode)
2544 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2545 ret = btrfs_delayed_update_inode(trans, root, inode);
2546 if (!ret)
2547 btrfs_set_inode_last_trans(trans, inode);
2548 return ret;
2549 }
2550
2551 return btrfs_update_inode_item(trans, root, inode);
2552}
2553
2554static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2555 struct btrfs_root *root, struct inode *inode)
2556{
2557 int ret;
2558
2559 ret = btrfs_update_inode(trans, root, inode);
2560 if (ret == -ENOSPC)
2561 return btrfs_update_inode_item(trans, root, inode);
2562 return ret;
2563}
2564
d352ac68
CM		 2565/*
2566 * unlink helper that gets used here in inode.c and in the tree logging
2567 * recovery code. It remove a link in a directory with a given name, and
2568 * also drops the back refs in the inode to the directory
2569 */
92986796
AV		 2570static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2571 struct btrfs_root *root,
2572 struct inode *dir, struct inode *inode,
2573 const char *name, int name_len)
39279cc3
CM		 2574{
2575 struct btrfs_path *path;
39279cc3 2576 int ret = 0;
5f39d397 2577 struct extent_buffer *leaf;
39279cc3 2578 struct btrfs_dir_item *di;
5f39d397 2579 struct btrfs_key key;
aec7477b 2580 u64 index;
33345d01
LZ		 2581 u64 ino = btrfs_ino(inode);
2582 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2583
2584 path = btrfs_alloc_path();
54aa1f4d
CM		 2585 if (!path) {
2586 ret = -ENOMEM;
554233a6 2587 goto out;
54aa1f4d
CM		 2588 }
2589
b9473439 2590 path->leave_spinning = 1;
33345d01 2591 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2592 name, name_len, -1);
2593 if (IS_ERR(di)) {
2594 ret = PTR_ERR(di);
2595 goto err;
2596 }
2597 if (!di) {
2598 ret = -ENOENT;
2599 goto err;
2600 }
5f39d397
CM		 2601 leaf = path->nodes[0];
2602 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2603 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2604 if (ret)
2605 goto err;
b3b4aa74 2606 btrfs_release_path(path);
39279cc3 2607
33345d01
LZ		 2608 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2609 dir_ino, &index);
aec7477b 2610 if (ret) {
d397712b 2611 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2612 "inode %llu parent %llu\n", name_len, name,
2613 (unsigned long long)ino, (unsigned long long)dir_ino);
aec7477b
JB		 2614 goto err;
2615 }
2616
16cdcec7
MX		 2617 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2618 if (ret)
39279cc3 2619 goto err;
39279cc3 2620
e02119d5 2621 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2622 inode, dir_ino);
49eb7e46 2623 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2624
2625 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2626 dir, index);
6418c961
CM		 2627 if (ret == -ENOENT)
2628 ret = 0;
39279cc3
CM		 2629err:
2630 btrfs_free_path(path);
e02119d5
CM		 2631 if (ret)
2632 goto out;
2633
2634 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2635 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2636 btrfs_update_inode(trans, root, dir);
e02119d5 2637out:
39279cc3
CM		 2638 return ret;
2639}
2640
92986796
AV		 2641int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2642 struct btrfs_root *root,
2643 struct inode *dir, struct inode *inode,
2644 const char *name, int name_len)
2645{
2646 int ret;
2647 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2648 if (!ret) {
2649 btrfs_drop_nlink(inode);
2650 ret = btrfs_update_inode(trans, root, inode);
2651 }
2652 return ret;
2653}
2654
2655
a22285a6
YZ		 2656/* helper to check if there is any shared block in the path */
2657static int check_path_shared(struct btrfs_root *root,
2658 struct btrfs_path *path)
39279cc3 2659{
a22285a6
YZ		 2660 struct extent_buffer *eb;
2661 int level;
0e4dcbef 2662 u64 refs = 1;
5df6a9f6 2663
a22285a6 2664 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2665 int ret;
2666
a22285a6
YZ		 2667 if (!path->nodes[level])
2668 break;
2669 eb = path->nodes[level];
2670 if (!btrfs_block_can_be_shared(root, eb))
2671 continue;
2672 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2673 &refs, NULL);
2674 if (refs > 1)
2675 return 1;
5df6a9f6 2676 }
dedefd72 2677 return 0;
39279cc3
CM		 2678}
2679
a22285a6
YZ		 2680/*
2681 * helper to start transaction for unlink and rmdir.
2682 *
2683 * unlink and rmdir are special in btrfs, they do not always free space.
2684 * so in enospc case, we should make sure they will free space before
2685 * allowing them to use the global metadata reservation.
2686 */
2687static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2688 struct dentry *dentry)
4df27c4d 2689{
39279cc3 2690 struct btrfs_trans_handle *trans;
a22285a6 2691 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2692 struct btrfs_path *path;
a22285a6 2693 struct btrfs_inode_ref *ref;
4df27c4d 2694 struct btrfs_dir_item *di;
7b128766 2695 struct inode *inode = dentry->d_inode;
4df27c4d 2696 u64 index;
a22285a6
YZ		 2697 int check_link = 1;
2698 int err = -ENOSPC;
4df27c4d 2699 int ret;
33345d01
LZ		 2700 u64 ino = btrfs_ino(inode);
2701 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2702
e70bea5f
JB		 2703 /*
2704 * 1 for the possible orphan item
2705 * 1 for the dir item
2706 * 1 for the dir index
2707 * 1 for the inode ref
2708 * 1 for the inode ref in the tree log
2709 * 2 for the dir entries in the log
2710 * 1 for the inode
2711 */
2712 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2713 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2714 return trans;
4df27c4d 2715
33345d01 2716 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2717 return ERR_PTR(-ENOSPC);
4df27c4d 2718
a22285a6
YZ		 2719 /* check if there is someone else holds reference */
2720 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2721 return ERR_PTR(-ENOSPC);
4df27c4d 2722
a22285a6
YZ		 2723 if (atomic_read(&inode->i_count) > 2)
2724 return ERR_PTR(-ENOSPC);
4df27c4d 2725
a22285a6
YZ		 2726 if (xchg(&root->fs_info->enospc_unlink, 1))
2727 return ERR_PTR(-ENOSPC);
2728
2729 path = btrfs_alloc_path();
2730 if (!path) {
2731 root->fs_info->enospc_unlink = 0;
2732 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2733 }
2734
3880a1b4
JB		 2735 /* 1 for the orphan item */
2736 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2737 if (IS_ERR(trans)) {
a22285a6
YZ		 2738 btrfs_free_path(path);
2739 root->fs_info->enospc_unlink = 0;
2740 return trans;
2741 }
4df27c4d 2742
a22285a6
YZ		 2743 path->skip_locking = 1;
2744 path->search_commit_root = 1;
4df27c4d 2745
a22285a6
YZ		 2746 ret = btrfs_lookup_inode(trans, root, path,
2747 &BTRFS_I(dir)->location, 0);
2748 if (ret < 0) {
2749 err = ret;
2750 goto out;
2751 }
2752 if (ret == 0) {
2753 if (check_path_shared(root, path))
2754 goto out;
2755 } else {
2756 check_link = 0;
5df6a9f6 2757 }
b3b4aa74 2758 btrfs_release_path(path);
a22285a6
YZ		 2759
2760 ret = btrfs_lookup_inode(trans, root, path,
2761 &BTRFS_I(inode)->location, 0);
2762 if (ret < 0) {
2763 err = ret;
2764 goto out;
2765 }
2766 if (ret == 0) {
2767 if (check_path_shared(root, path))
2768 goto out;
2769 } else {
2770 check_link = 0;
2771 }
b3b4aa74 2772 btrfs_release_path(path);
a22285a6
YZ		 2773
2774 if (ret == 0 && S_ISREG(inode->i_mode)) {
2775 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2776 ino, (u64)-1, 0);
a22285a6
YZ		 2777 if (ret < 0) {
2778 err = ret;
2779 goto out;
2780 }
2781 BUG_ON(ret == 0);
2782 if (check_path_shared(root, path))
2783 goto out;
b3b4aa74 2784 btrfs_release_path(path);
a22285a6
YZ		 2785 }
2786
2787 if (!check_link) {
2788 err = 0;
2789 goto out;
2790 }
2791
33345d01 2792 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 2793 dentry->d_name.name, dentry->d_name.len, 0);
2794 if (IS_ERR(di)) {
2795 err = PTR_ERR(di);
2796 goto out;
2797 }
2798 if (di) {
2799 if (check_path_shared(root, path))
2800 goto out;
2801 } else {
2802 err = 0;
2803 goto out;
2804 }
b3b4aa74 2805 btrfs_release_path(path);
a22285a6
YZ		 2806
2807 ref = btrfs_lookup_inode_ref(trans, root, path,
2808 dentry->d_name.name, dentry->d_name.len,
33345d01 2809 ino, dir_ino, 0);
a22285a6
YZ		 2810 if (IS_ERR(ref)) {
2811 err = PTR_ERR(ref);
2812 goto out;
2813 }
2814 BUG_ON(!ref);
2815 if (check_path_shared(root, path))
2816 goto out;
2817 index = btrfs_inode_ref_index(path->nodes[0], ref);
b3b4aa74 2818 btrfs_release_path(path);
a22285a6 2819
16cdcec7
MX		 2820 /*
2821 * This is a commit root search, if we can lookup inode item and other
2822 * relative items in the commit root, it means the transaction of
2823 * dir/file creation has been committed, and the dir index item that we
2824 * delay to insert has also been inserted into the commit root. So
2825 * we needn't worry about the delayed insertion of the dir index item
2826 * here.
2827 */
33345d01 2828 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 2829 dentry->d_name.name, dentry->d_name.len, 0);
2830 if (IS_ERR(di)) {
2831 err = PTR_ERR(di);
2832 goto out;
2833 }
2834 BUG_ON(ret == -ENOENT);
2835 if (check_path_shared(root, path))
2836 goto out;
2837
2838 err = 0;
2839out:
2840 btrfs_free_path(path);
3880a1b4
JB		 2841 /* Migrate the orphan reservation over */
2842 if (!err)
2843 err = btrfs_block_rsv_migrate(trans->block_rsv,
2844 &root->fs_info->global_block_rsv,
5a77d76c 2845 trans->bytes_reserved);
3880a1b4 2846
a22285a6
YZ		 2847 if (err) {
2848 btrfs_end_transaction(trans, root);
2849 root->fs_info->enospc_unlink = 0;
2850 return ERR_PTR(err);
2851 }
2852
2853 trans->block_rsv = &root->fs_info->global_block_rsv;
2854 return trans;
2855}
2856
2857static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2858 struct btrfs_root *root)
2859{
2860 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
5a77d76c
JB		 2861 btrfs_block_rsv_release(root, trans->block_rsv,
2862 trans->bytes_reserved);
2863 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 2864 BUG_ON(!root->fs_info->enospc_unlink);
2865 root->fs_info->enospc_unlink = 0;
2866 }
7ad85bb7 2867 btrfs_end_transaction(trans, root);
a22285a6
YZ		 2868}
2869
2870static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2871{
2872 struct btrfs_root *root = BTRFS_I(dir)->root;
2873 struct btrfs_trans_handle *trans;
2874 struct inode *inode = dentry->d_inode;
2875 int ret;
2876 unsigned long nr = 0;
2877
2878 trans = __unlink_start_trans(dir, dentry);
2879 if (IS_ERR(trans))
2880 return PTR_ERR(trans);
5f39d397 2881
12fcfd22
CM		 2882 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2883
e02119d5
CM		 2884 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2885 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 2886 if (ret)
2887 goto out;
7b128766 2888
a22285a6 2889 if (inode->i_nlink == 0) {
7b128766 2890 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 2891 if (ret)
2892 goto out;
a22285a6 2893 }
7b128766 2894
b532402e 2895out:
d3c2fdcf 2896 nr = trans->blocks_used;
a22285a6 2897 __unlink_end_trans(trans, root);
d3c2fdcf 2898 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2899 return ret;
2900}
2901
4df27c4d
YZ		 2902int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2903 struct btrfs_root *root,
2904 struct inode *dir, u64 objectid,
2905 const char *name, int name_len)
2906{
2907 struct btrfs_path *path;
2908 struct extent_buffer *leaf;
2909 struct btrfs_dir_item *di;
2910 struct btrfs_key key;
2911 u64 index;
2912 int ret;
33345d01 2913 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 2914
2915 path = btrfs_alloc_path();
2916 if (!path)
2917 return -ENOMEM;
2918
33345d01 2919 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 2920 name, name_len, -1);
c704005d 2921 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2922
2923 leaf = path->nodes[0];
2924 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2925 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2926 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2927 BUG_ON(ret);
b3b4aa74 2928 btrfs_release_path(path);
4df27c4d
YZ		 2929
2930 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2931 objectid, root->root_key.objectid,
33345d01 2932 dir_ino, &index, name, name_len);
4df27c4d
YZ		 2933 if (ret < 0) {
2934 BUG_ON(ret != -ENOENT);
33345d01 2935 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 2936 name, name_len);
c704005d 2937 BUG_ON(IS_ERR_OR_NULL(di));
4df27c4d
YZ		 2938
2939 leaf = path->nodes[0];
2940 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2941 btrfs_release_path(path);
4df27c4d
YZ		 2942 index = key.offset;
2943 }
945d8962 2944 btrfs_release_path(path);
4df27c4d 2945
16cdcec7 2946 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4df27c4d 2947 BUG_ON(ret);
4df27c4d
YZ		 2948
2949 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2950 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2951 ret = btrfs_update_inode(trans, root, dir);
2952 BUG_ON(ret);
4df27c4d 2953
71d7aed0 2954 btrfs_free_path(path);
4df27c4d
YZ		 2955 return 0;
2956}
2957
39279cc3
CM		 2958static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2959{
2960 struct inode *inode = dentry->d_inode;
1832a6d5 2961 int err = 0;
39279cc3 2962 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2963 struct btrfs_trans_handle *trans;
1832a6d5 2964 unsigned long nr = 0;
39279cc3 2965
3394e160 2966 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
33345d01 2967 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
134d4512
Y		 2968 return -ENOTEMPTY;
2969
a22285a6
YZ		 2970 trans = __unlink_start_trans(dir, dentry);
2971 if (IS_ERR(trans))
5df6a9f6 2972 return PTR_ERR(trans);
5df6a9f6 2973
33345d01 2974 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 2975 err = btrfs_unlink_subvol(trans, root, dir,
2976 BTRFS_I(inode)->location.objectid,
2977 dentry->d_name.name,
2978 dentry->d_name.len);
2979 goto out;
2980 }
2981
7b128766
JB		 2982 err = btrfs_orphan_add(trans, inode);
2983 if (err)
4df27c4d 2984 goto out;
7b128766 2985
39279cc3 2986 /* now the directory is empty */
e02119d5
CM		 2987 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2988 dentry->d_name.name, dentry->d_name.len);
d397712b 2989 if (!err)
dbe674a9 2990 btrfs_i_size_write(inode, 0);
4df27c4d 2991out:
d3c2fdcf 2992 nr = trans->blocks_used;
a22285a6 2993 __unlink_end_trans(trans, root);
d3c2fdcf 2994 btrfs_btree_balance_dirty(root, nr);
3954401f 2995
39279cc3
CM		 2996 return err;
2997}
2998
39279cc3
CM		 2999/*
3000 * this can truncate away extent items, csum items and directory items.
3001 * It starts at a high offset and removes keys until it can't find
d352ac68 3002 * any higher than new_size
39279cc3
CM		 3003 *
3004 * csum items that cross the new i_size are truncated to the new size
3005 * as well.
7b128766
JB		 3006 *
3007 * min_type is the minimum key type to truncate down to. If set to 0, this
3008 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3009 */
8082510e
YZ		 3010int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3011 struct btrfs_root *root,
3012 struct inode *inode,
3013 u64 new_size, u32 min_type)
39279cc3 3014{
39279cc3 3015 struct btrfs_path *path;
5f39d397 3016 struct extent_buffer *leaf;
39279cc3 3017 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3018 struct btrfs_key key;
3019 struct btrfs_key found_key;
39279cc3 3020 u64 extent_start = 0;
db94535d 3021 u64 extent_num_bytes = 0;
5d4f98a2 3022 u64 extent_offset = 0;
39279cc3 3023 u64 item_end = 0;
8082510e
YZ		 3024 u64 mask = root->sectorsize - 1;
3025 u32 found_type = (u8)-1;
39279cc3
CM		 3026 int found_extent;
3027 int del_item;
85e21bac
CM		 3028 int pending_del_nr = 0;
3029 int pending_del_slot = 0;
179e29e4 3030 int extent_type = -1;
8082510e
YZ		 3031 int ret;
3032 int err = 0;
33345d01 3033 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3034
3035 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3036
0eb0e19c
MF		 3037 path = btrfs_alloc_path();
3038 if (!path)
3039 return -ENOMEM;
3040 path->reada = -1;
3041
0af3d00b 3042 if (root->ref_cows || root == root->fs_info->tree_root)
5b21f2ed 3043 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
8082510e 3044
16cdcec7
MX		 3045 /*
3046 * This function is also used to drop the items in the log tree before
3047 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3048 * it is used to drop the loged items. So we shouldn't kill the delayed
3049 * items.
3050 */
3051 if (min_type == 0 && root == BTRFS_I(inode)->root)
3052 btrfs_kill_delayed_inode_items(inode);
3053
33345d01 3054 key.objectid = ino;
39279cc3 3055 key.offset = (u64)-1;
5f39d397
CM		 3056 key.type = (u8)-1;
3057
85e21bac 3058search_again:
b9473439 3059 path->leave_spinning = 1;
85e21bac 3060 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3061 if (ret < 0) {
3062 err = ret;
3063 goto out;
3064 }
d397712b 3065
85e21bac 3066 if (ret > 0) {
e02119d5
CM		 3067 /* there are no items in the tree for us to truncate, we're
3068 * done
3069 */
8082510e
YZ		 3070 if (path->slots[0] == 0)
3071 goto out;
85e21bac
CM		 3072 path->slots[0]--;
3073 }
3074
d397712b 3075 while (1) {
39279cc3 3076 fi = NULL;
5f39d397
CM		 3077 leaf = path->nodes[0];
3078 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3079 found_type = btrfs_key_type(&found_key);
39279cc3 3080
33345d01 3081 if (found_key.objectid != ino)
39279cc3 3082 break;
5f39d397 3083
85e21bac 3084 if (found_type < min_type)
39279cc3
CM		 3085 break;
3086
5f39d397 3087 item_end = found_key.offset;
39279cc3 3088 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3089 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3090 struct btrfs_file_extent_item);
179e29e4
CM		 3091 extent_type = btrfs_file_extent_type(leaf, fi);
3092 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3093 item_end +=
db94535d 3094 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3095 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3096 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3097 fi);
39279cc3 3098 }
008630c1 3099 item_end--;
39279cc3 3100 }
8082510e
YZ		 3101 if (found_type > min_type) {
3102 del_item = 1;
3103 } else {
3104 if (item_end < new_size)
b888db2b 3105 break;
8082510e
YZ		 3106 if (found_key.offset >= new_size)
3107 del_item = 1;
3108 else
3109 del_item = 0;
39279cc3 3110 }
39279cc3 3111 found_extent = 0;
39279cc3 3112 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3113 if (found_type != BTRFS_EXTENT_DATA_KEY)
3114 goto delete;
3115
3116 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3117 u64 num_dec;
db94535d 3118 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3119 if (!del_item) {
db94535d
CM		 3120 u64 orig_num_bytes =
3121 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3122 extent_num_bytes = new_size -
5f39d397 3123 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3124 extent_num_bytes = extent_num_bytes &
3125 ~((u64)root->sectorsize - 1);
db94535d
CM		 3126 btrfs_set_file_extent_num_bytes(leaf, fi,
3127 extent_num_bytes);
3128 num_dec = (orig_num_bytes -
9069218d 3129 extent_num_bytes);
e02119d5 3130 if (root->ref_cows && extent_start != 0)
a76a3cd4 3131 inode_sub_bytes(inode, num_dec);
5f39d397 3132 btrfs_mark_buffer_dirty(leaf);
39279cc3 3133 } else {
db94535d
CM		 3134 extent_num_bytes =
3135 btrfs_file_extent_disk_num_bytes(leaf,
3136 fi);
5d4f98a2
YZ		 3137 extent_offset = found_key.offset -
3138 btrfs_file_extent_offset(leaf, fi);
3139
39279cc3 3140 /* FIXME blocksize != 4096 */
9069218d 3141 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3142 if (extent_start != 0) {
3143 found_extent = 1;
e02119d5 3144 if (root->ref_cows)
a76a3cd4 3145 inode_sub_bytes(inode, num_dec);
e02119d5 3146 }
39279cc3 3147 }
9069218d 3148 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3149 /*
3150 * we can't truncate inline items that have had
3151 * special encodings
3152 */
3153 if (!del_item &&
3154 btrfs_file_extent_compression(leaf, fi) == 0 &&
3155 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3156 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3157 u32 size = new_size - found_key.offset;
3158
3159 if (root->ref_cows) {
a76a3cd4
YZ		 3160 inode_sub_bytes(inode, item_end + 1 -
3161 new_size);
e02119d5
CM		 3162 }
3163 size =
3164 btrfs_file_extent_calc_inline_size(size);
9069218d 3165 ret = btrfs_truncate_item(trans, root, path,
e02119d5 3166 size, 1);
e02119d5 3167 } else if (root->ref_cows) {
a76a3cd4
YZ		 3168 inode_sub_bytes(inode, item_end + 1 -
3169 found_key.offset);
9069218d 3170 }
39279cc3 3171 }
179e29e4 3172delete:
39279cc3 3173 if (del_item) {
85e21bac
CM		 3174 if (!pending_del_nr) {
3175 /* no pending yet, add ourselves */
3176 pending_del_slot = path->slots[0];
3177 pending_del_nr = 1;
3178 } else if (pending_del_nr &&
3179 path->slots[0] + 1 == pending_del_slot) {
3180 /* hop on the pending chunk */
3181 pending_del_nr++;
3182 pending_del_slot = path->slots[0];
3183 } else {
d397712b 3184 BUG();
85e21bac 3185 }
39279cc3
CM		 3186 } else {
3187 break;
3188 }
0af3d00b
JB		 3189 if (found_extent && (root->ref_cows ||
3190 root == root->fs_info->tree_root)) {
b9473439 3191 btrfs_set_path_blocking(path);
39279cc3 3192 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3193 extent_num_bytes, 0,
3194 btrfs_header_owner(leaf),
66d7e7f0 3195 ino, extent_offset, 0);
39279cc3
CM		 3196 BUG_ON(ret);
3197 }
85e21bac 3198
8082510e
YZ		 3199 if (found_type == BTRFS_INODE_ITEM_KEY)
3200 break;
3201
3202 if (path->slots[0] == 0 ||
3203 path->slots[0] != pending_del_slot) {
82d5902d
LZ		 3204 if (root->ref_cows &&
3205 BTRFS_I(inode)->location.objectid !=
3206 BTRFS_FREE_INO_OBJECTID) {
8082510e
YZ		 3207 err = -EAGAIN;
3208 goto out;
3209 }
3210 if (pending_del_nr) {
3211 ret = btrfs_del_items(trans, root, path,
3212 pending_del_slot,
3213 pending_del_nr);
3214 BUG_ON(ret);
3215 pending_del_nr = 0;
3216 }
b3b4aa74 3217 btrfs_release_path(path);
85e21bac 3218 goto search_again;
8082510e
YZ		 3219 } else {
3220 path->slots[0]--;
85e21bac 3221 }
39279cc3 3222 }
8082510e 3223out:
85e21bac
CM		 3224 if (pending_del_nr) {
3225 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3226 pending_del_nr);
d68fc57b 3227 BUG_ON(ret);
85e21bac 3228 }
39279cc3 3229 btrfs_free_path(path);
8082510e 3230 return err;
39279cc3
CM		 3231}
3232
3233/*
3234 * taken from block_truncate_page, but does cow as it zeros out
3235 * any bytes left in the last page in the file.
3236 */
3237static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3238{
3239 struct inode *inode = mapping->host;
db94535d 3240 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3241 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3242 struct btrfs_ordered_extent *ordered;
2ac55d41 3243 struct extent_state *cached_state = NULL;
e6dcd2dc 3244 char *kaddr;
db94535d 3245 u32 blocksize = root->sectorsize;
39279cc3
CM		 3246 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3247 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3248 struct page *page;
3b16a4e3 3249 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3250 int ret = 0;
a52d9a80 3251 u64 page_start;
e6dcd2dc 3252 u64 page_end;
39279cc3
CM		 3253
3254 if ((offset & (blocksize - 1)) == 0)
3255 goto out;
0ca1f7ce 3256 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3257 if (ret)
3258 goto out;
39279cc3
CM		 3259
3260 ret = -ENOMEM;
211c17f5 3261again:
3b16a4e3 3262 page = find_or_create_page(mapping, index, mask);
5d5e103a 3263 if (!page) {
0ca1f7ce 3264 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3265 goto out;
5d5e103a 3266 }
e6dcd2dc
CM		 3267
3268 page_start = page_offset(page);
3269 page_end = page_start + PAGE_CACHE_SIZE - 1;
3270
39279cc3 3271 if (!PageUptodate(page)) {
9ebefb18 3272 ret = btrfs_readpage(NULL, page);
39279cc3 3273 lock_page(page);
211c17f5
CM		 3274 if (page->mapping != mapping) {
3275 unlock_page(page);
3276 page_cache_release(page);
3277 goto again;
3278 }
39279cc3
CM		 3279 if (!PageUptodate(page)) {
3280 ret = -EIO;
89642229 3281 goto out_unlock;
39279cc3
CM		 3282 }
3283 }
211c17f5 3284 wait_on_page_writeback(page);
e6dcd2dc 3285
2ac55d41
JB		 3286 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3287 GFP_NOFS);
e6dcd2dc
CM		 3288 set_page_extent_mapped(page);
3289
3290 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3291 if (ordered) {
2ac55d41
JB		 3292 unlock_extent_cached(io_tree, page_start, page_end,
3293 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3294 unlock_page(page);
3295 page_cache_release(page);
eb84ae03 3296 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3297 btrfs_put_ordered_extent(ordered);
3298 goto again;
3299 }
3300
2ac55d41 3301 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5d5e103a 3302 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 3303 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3304
2ac55d41
JB		 3305 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3306 &cached_state);
9ed74f2d 3307 if (ret) {
2ac55d41
JB		 3308 unlock_extent_cached(io_tree, page_start, page_end,
3309 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3310 goto out_unlock;
3311 }
3312
e6dcd2dc
CM		 3313 ret = 0;
3314 if (offset != PAGE_CACHE_SIZE) {
3315 kaddr = kmap(page);
3316 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3317 flush_dcache_page(page);
3318 kunmap(page);
3319 }
247e743c 3320 ClearPageChecked(page);
e6dcd2dc 3321 set_page_dirty(page);
2ac55d41
JB		 3322 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3323 GFP_NOFS);
39279cc3 3324
89642229 3325out_unlock:
5d5e103a 3326 if (ret)
0ca1f7ce 3327 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3328 unlock_page(page);
3329 page_cache_release(page);
3330out:
3331 return ret;
3332}
3333
695a0d0d
JB		 3334/*
3335 * This function puts in dummy file extents for the area we're creating a hole
3336 * for. So if we are truncating this file to a larger size we need to insert
3337 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3338 * the range between oldsize and size
3339 */
a41ad394 3340int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3341{
9036c102
YZ		 3342 struct btrfs_trans_handle *trans;
3343 struct btrfs_root *root = BTRFS_I(inode)->root;
3344 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3345 struct extent_map *em = NULL;
2ac55d41 3346 struct extent_state *cached_state = NULL;
9036c102 3347 u64 mask = root->sectorsize - 1;
a41ad394 3348 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3349 u64 block_end = (size + mask) & ~mask;
3350 u64 last_byte;
3351 u64 cur_offset;
3352 u64 hole_size;
9ed74f2d 3353 int err = 0;
39279cc3 3354
9036c102
YZ		 3355 if (size <= hole_start)
3356 return 0;
3357
9036c102
YZ		 3358 while (1) {
3359 struct btrfs_ordered_extent *ordered;
3360 btrfs_wait_ordered_range(inode, hole_start,
3361 block_end - hole_start);
2ac55d41
JB		 3362 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3363 &cached_state, GFP_NOFS);
9036c102
YZ		 3364 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3365 if (!ordered)
3366 break;
2ac55d41
JB		 3367 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3368 &cached_state, GFP_NOFS);
9036c102
YZ		 3369 btrfs_put_ordered_extent(ordered);
3370 }
39279cc3 3371
9036c102
YZ		 3372 cur_offset = hole_start;
3373 while (1) {
3374 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3375 block_end - cur_offset, 0);
c704005d 3376 BUG_ON(IS_ERR_OR_NULL(em));
9036c102
YZ		 3377 last_byte = min(extent_map_end(em), block_end);
3378 last_byte = (last_byte + mask) & ~mask;
8082510e 3379 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
771ed689 3380 u64 hint_byte = 0;
9036c102 3381 hole_size = last_byte - cur_offset;
9ed74f2d 3382
3642320e 3383 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 3384 if (IS_ERR(trans)) {
3385 err = PTR_ERR(trans);
9ed74f2d 3386 break;
a22285a6 3387 }
8082510e
YZ		 3388
3389 err = btrfs_drop_extents(trans, inode, cur_offset,
3390 cur_offset + hole_size,
3391 &hint_byte, 1);
5b397377 3392 if (err) {
3642320e 3393 btrfs_update_inode(trans, root, inode);
5b397377 3394 btrfs_end_transaction(trans, root);
3893e33b 3395 break;
5b397377 3396 }
8082510e 3397
9036c102 3398 err = btrfs_insert_file_extent(trans, root,
33345d01 3399 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3400 0, hole_size, 0, hole_size,
3401 0, 0, 0);
5b397377 3402 if (err) {
3642320e 3403 btrfs_update_inode(trans, root, inode);
5b397377 3404 btrfs_end_transaction(trans, root);
3893e33b 3405 break;
5b397377 3406 }
8082510e 3407
9036c102
YZ		 3408 btrfs_drop_extent_cache(inode, hole_start,
3409 last_byte - 1, 0);
8082510e 3410
3642320e 3411 btrfs_update_inode(trans, root, inode);
8082510e 3412 btrfs_end_transaction(trans, root);
9036c102
YZ		 3413 }
3414 free_extent_map(em);
a22285a6 3415 em = NULL;
9036c102 3416 cur_offset = last_byte;
8082510e 3417 if (cur_offset >= block_end)
9036c102
YZ		 3418 break;
3419 }
1832a6d5 3420
a22285a6 3421 free_extent_map(em);
2ac55d41
JB		 3422 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3423 GFP_NOFS);
9036c102
YZ		 3424 return err;
3425}
39279cc3 3426
a41ad394 3427static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3428{
f4a2f4c5
MX		 3429 struct btrfs_root *root = BTRFS_I(inode)->root;
3430 struct btrfs_trans_handle *trans;
a41ad394 3431 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3432 int ret;
3433
a41ad394 3434 if (newsize == oldsize)
8082510e
YZ		 3435 return 0;
3436
a41ad394 3437 if (newsize > oldsize) {
a41ad394
JB		 3438 truncate_pagecache(inode, oldsize, newsize);
3439 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 3440 if (ret)
8082510e 3441 return ret;
8082510e 3442
f4a2f4c5
MX		 3443 trans = btrfs_start_transaction(root, 1);
3444 if (IS_ERR(trans))
3445 return PTR_ERR(trans);
3446
3447 i_size_write(inode, newsize);
3448 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3449 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 3450 btrfs_end_transaction(trans, root);
a41ad394 3451 } else {
8082510e 3452
a41ad394
JB		 3453 /*
3454 * We're truncating a file that used to have good data down to
3455 * zero. Make sure it gets into the ordered flush list so that
3456 * any new writes get down to disk quickly.
3457 */
3458 if (newsize == 0)
3459 BTRFS_I(inode)->ordered_data_close = 1;
8082510e 3460
a41ad394
JB		 3461 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3462 truncate_setsize(inode, newsize);
3463 ret = btrfs_truncate(inode);
8082510e
YZ		 3464 }
3465
a41ad394 3466 return ret;
8082510e
YZ		 3467}
3468
9036c102
YZ		 3469static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3470{
3471 struct inode *inode = dentry->d_inode;
b83cc969 3472 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3473 int err;
39279cc3 3474
b83cc969
LZ		 3475 if (btrfs_root_readonly(root))
3476 return -EROFS;
3477
9036c102
YZ		 3478 err = inode_change_ok(inode, attr);
3479 if (err)
3480 return err;
2bf5a725 3481
5a3f23d5 3482 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3483 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3484 if (err)
3485 return err;
39279cc3 3486 }
9036c102 3487
1025774c
CH		 3488 if (attr->ia_valid) {
3489 setattr_copy(inode, attr);
22c44fe6 3490 err = btrfs_dirty_inode(inode);
1025774c 3491
22c44fe6 3492 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 3493 err = btrfs_acl_chmod(inode);
3494 }
33268eaf 3495
39279cc3
CM		 3496 return err;
3497}
61295eb8 3498
bd555975 3499void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3500{
3501 struct btrfs_trans_handle *trans;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3503 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3504 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3505 unsigned long nr;
39279cc3
CM		 3506 int ret;
3507
1abe9b8a 3508 trace_btrfs_inode_evict(inode);
3509
39279cc3 3510 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3511 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
2cf8572d 3512 btrfs_is_free_space_inode(root, inode)))
bd555975
AV		 3513 goto no_delete;
3514
39279cc3 3515 if (is_bad_inode(inode)) {
7b128766 3516 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3517 goto no_delete;
3518 }
bd555975 3519 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3520 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3521
c71bf099
YZ		 3522 if (root->fs_info->log_root_recovering) {
3523 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3524 goto no_delete;
3525 }
3526
76dda93c
YZ		 3527 if (inode->i_nlink > 0) {
3528 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3529 goto no_delete;
3530 }
3531
4289a667
JB		 3532 rsv = btrfs_alloc_block_rsv(root);
3533 if (!rsv) {
3534 btrfs_orphan_del(NULL, inode);
3535 goto no_delete;
3536 }
4a338542 3537 rsv->size = min_size;
726c35fa 3538 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3539
dbe674a9 3540 btrfs_i_size_write(inode, 0);
5f39d397 3541
4289a667
JB		 3542 /*
3543 * This is a bit simpler than btrfs_truncate since
3544 *
3545 * 1) We've already reserved our space for our orphan item in the
3546 * unlink.
3547 * 2) We're going to delete the inode item, so we don't need to update
3548 * it at all.
3549 *
3550 * So we just need to reserve some slack space in case we add bytes when
3551 * doing the truncate.
3552 */
8082510e 3553 while (1) {
aa38a711 3554 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
726c35fa
JB		 3555
3556 /*
3557 * Try and steal from the global reserve since we will
3558 * likely not use this space anyway, we want to try as
3559 * hard as possible to get this to work.
3560 */
3561 if (ret)
3562 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 3563
d68fc57b 3564 if (ret) {
4289a667 3565 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3566 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3567 btrfs_orphan_del(NULL, inode);
3568 btrfs_free_block_rsv(root, rsv);
3569 goto no_delete;
d68fc57b 3570 }
7b128766 3571
4289a667
JB		 3572 trans = btrfs_start_transaction(root, 0);
3573 if (IS_ERR(trans)) {
3574 btrfs_orphan_del(NULL, inode);
3575 btrfs_free_block_rsv(root, rsv);
3576 goto no_delete;
d68fc57b 3577 }
7b128766 3578
4289a667
JB		 3579 trans->block_rsv = rsv;
3580
d68fc57b 3581 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
8082510e
YZ		 3582 if (ret != -EAGAIN)
3583 break;
85e21bac 3584
8082510e
YZ		 3585 nr = trans->blocks_used;
3586 btrfs_end_transaction(trans, root);
3587 trans = NULL;
3588 btrfs_btree_balance_dirty(root, nr);
3589 }
5f39d397 3590
4289a667
JB		 3591 btrfs_free_block_rsv(root, rsv);
3592
8082510e 3593 if (ret == 0) {
4289a667 3594 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3595 ret = btrfs_orphan_del(trans, inode);
3596 BUG_ON(ret);
3597 }
54aa1f4d 3598
4289a667 3599 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3600 if (!(root == root->fs_info->tree_root ||
3601 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3602 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3603
d3c2fdcf 3604 nr = trans->blocks_used;
54aa1f4d 3605 btrfs_end_transaction(trans, root);
d3c2fdcf 3606 btrfs_btree_balance_dirty(root, nr);
39279cc3 3607no_delete:
bd555975 3608 end_writeback(inode);
8082510e 3609 return;
39279cc3
CM		 3610}
3611
3612/*
3613 * this returns the key found in the dir entry in the location pointer.
3614 * If no dir entries were found, location->objectid is 0.
3615 */
3616static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3617 struct btrfs_key *location)
3618{
3619 const char *name = dentry->d_name.name;
3620 int namelen = dentry->d_name.len;
3621 struct btrfs_dir_item *di;
3622 struct btrfs_path *path;
3623 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3624 int ret = 0;
39279cc3
CM		 3625
3626 path = btrfs_alloc_path();
d8926bb3
MF		 3627 if (!path)
3628 return -ENOMEM;
3954401f 3629
33345d01 3630 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3631 namelen, 0);
0d9f7f3e
Y		 3632 if (IS_ERR(di))
3633 ret = PTR_ERR(di);
d397712b 3634
c704005d 3635 if (IS_ERR_OR_NULL(di))
3954401f 3636 goto out_err;
d397712b 3637
5f39d397 3638 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3639out:
39279cc3
CM		 3640 btrfs_free_path(path);
3641 return ret;
3954401f
CM		 3642out_err:
3643 location->objectid = 0;
3644 goto out;
39279cc3
CM		 3645}
3646
3647/*
3648 * when we hit a tree root in a directory, the btrfs part of the inode
3649 * needs to be changed to reflect the root directory of the tree root. This
3650 * is kind of like crossing a mount point.
3651 */
3652static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3653 struct inode *dir,
3654 struct dentry *dentry,
3655 struct btrfs_key *location,
3656 struct btrfs_root **sub_root)
39279cc3 3657{
4df27c4d
YZ		 3658 struct btrfs_path *path;
3659 struct btrfs_root *new_root;
3660 struct btrfs_root_ref *ref;
3661 struct extent_buffer *leaf;
3662 int ret;
3663 int err = 0;
39279cc3 3664
4df27c4d
YZ		 3665 path = btrfs_alloc_path();
3666 if (!path) {
3667 err = -ENOMEM;
3668 goto out;
3669 }
39279cc3 3670
4df27c4d
YZ		 3671 err = -ENOENT;
3672 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3673 BTRFS_I(dir)->root->root_key.objectid,
3674 location->objectid);
3675 if (ret) {
3676 if (ret < 0)
3677 err = ret;
3678 goto out;
3679 }
39279cc3 3680
4df27c4d
YZ		 3681 leaf = path->nodes[0];
3682 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3683 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3684 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3685 goto out;
39279cc3 3686
4df27c4d
YZ		 3687 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3688 (unsigned long)(ref + 1),
3689 dentry->d_name.len);
3690 if (ret)
3691 goto out;
3692
b3b4aa74 3693 btrfs_release_path(path);
4df27c4d
YZ		 3694
3695 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3696 if (IS_ERR(new_root)) {
3697 err = PTR_ERR(new_root);
3698 goto out;
3699 }
3700
3701 if (btrfs_root_refs(&new_root->root_item) == 0) {
3702 err = -ENOENT;
3703 goto out;
3704 }
3705
3706 *sub_root = new_root;
3707 location->objectid = btrfs_root_dirid(&new_root->root_item);
3708 location->type = BTRFS_INODE_ITEM_KEY;
3709 location->offset = 0;
3710 err = 0;
3711out:
3712 btrfs_free_path(path);
3713 return err;
39279cc3
CM		 3714}
3715
5d4f98a2
YZ		 3716static void inode_tree_add(struct inode *inode)
3717{
3718 struct btrfs_root *root = BTRFS_I(inode)->root;
3719 struct btrfs_inode *entry;
03e860bd
FNP		 3720 struct rb_node **p;
3721 struct rb_node *parent;
33345d01 3722 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 3723again:
3724 p = &root->inode_tree.rb_node;
3725 parent = NULL;
5d4f98a2 3726
1d3382cb 3727 if (inode_unhashed(inode))
76dda93c
YZ		 3728 return;
3729
5d4f98a2
YZ		 3730 spin_lock(&root->inode_lock);
3731 while (*p) {
3732 parent = *p;
3733 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3734
33345d01 3735 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 3736 p = &parent->rb_left;
33345d01 3737 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 3738 p = &parent->rb_right;
5d4f98a2
YZ		 3739 else {
3740 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 3741 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 3742 rb_erase(parent, &root->inode_tree);
3743 RB_CLEAR_NODE(parent);
3744 spin_unlock(&root->inode_lock);
3745 goto again;
5d4f98a2
YZ		 3746 }
3747 }
3748 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3749 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3750 spin_unlock(&root->inode_lock);
3751}
3752
3753static void inode_tree_del(struct inode *inode)
3754{
3755 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 3756 int empty = 0;
5d4f98a2 3757
03e860bd 3758 spin_lock(&root->inode_lock);
5d4f98a2 3759 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 3760 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 3761 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 3762 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 3763 }
03e860bd 3764 spin_unlock(&root->inode_lock);
76dda93c 3765
0af3d00b
JB		 3766 /*
3767 * Free space cache has inodes in the tree root, but the tree root has a
3768 * root_refs of 0, so this could end up dropping the tree root as a
3769 * snapshot, so we need the extra !root->fs_info->tree_root check to
3770 * make sure we don't drop it.
3771 */
3772 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3773 root != root->fs_info->tree_root) {
76dda93c
YZ		 3774 synchronize_srcu(&root->fs_info->subvol_srcu);
3775 spin_lock(&root->inode_lock);
3776 empty = RB_EMPTY_ROOT(&root->inode_tree);
3777 spin_unlock(&root->inode_lock);
3778 if (empty)
3779 btrfs_add_dead_root(root);
3780 }
3781}
3782
3783int btrfs_invalidate_inodes(struct btrfs_root *root)
3784{
3785 struct rb_node *node;
3786 struct rb_node *prev;
3787 struct btrfs_inode *entry;
3788 struct inode *inode;
3789 u64 objectid = 0;
3790
3791 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3792
3793 spin_lock(&root->inode_lock);
3794again:
3795 node = root->inode_tree.rb_node;
3796 prev = NULL;
3797 while (node) {
3798 prev = node;
3799 entry = rb_entry(node, struct btrfs_inode, rb_node);
3800
33345d01 3801 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 3802 node = node->rb_left;
33345d01 3803 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 3804 node = node->rb_right;
3805 else
3806 break;
3807 }
3808 if (!node) {
3809 while (prev) {
3810 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 3811 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 3812 node = prev;
3813 break;
3814 }
3815 prev = rb_next(prev);
3816 }
3817 }
3818 while (node) {
3819 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 3820 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 3821 inode = igrab(&entry->vfs_inode);
3822 if (inode) {
3823 spin_unlock(&root->inode_lock);
3824 if (atomic_read(&inode->i_count) > 1)
3825 d_prune_aliases(inode);
3826 /*
45321ac5 3827 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 3828 * the inode cache when its usage count
3829 * hits zero.
3830 */
3831 iput(inode);
3832 cond_resched();
3833 spin_lock(&root->inode_lock);
3834 goto again;
3835 }
3836
3837 if (cond_resched_lock(&root->inode_lock))
3838 goto again;
3839
3840 node = rb_next(node);
3841 }
3842 spin_unlock(&root->inode_lock);
3843 return 0;
5d4f98a2
YZ		 3844}
3845
e02119d5
CM		 3846static int btrfs_init_locked_inode(struct inode *inode, void *p)
3847{
3848 struct btrfs_iget_args *args = p;
3849 inode->i_ino = args->ino;
e02119d5 3850 BTRFS_I(inode)->root = args->root;
6a63209f 3851 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3852 return 0;
3853}
3854
3855static int btrfs_find_actor(struct inode *inode, void *opaque)
3856{
3857 struct btrfs_iget_args *args = opaque;
33345d01 3858 return args->ino == btrfs_ino(inode) &&
d397712b 3859 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3860}
3861
5d4f98a2
YZ		 3862static struct inode *btrfs_iget_locked(struct super_block *s,
3863 u64 objectid,
3864 struct btrfs_root *root)
39279cc3
CM		 3865{
3866 struct inode *inode;
3867 struct btrfs_iget_args args;
3868 args.ino = objectid;
3869 args.root = root;
3870
3871 inode = iget5_locked(s, objectid, btrfs_find_actor,
3872 btrfs_init_locked_inode,
3873 (void *)&args);
3874 return inode;
3875}
3876
1a54ef8c
BR		 3877/* Get an inode object given its location and corresponding root.
3878 * Returns in *is_new if the inode was read from disk
3879 */
3880struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 3881 struct btrfs_root *root, int *new)
1a54ef8c
BR		 3882{
3883 struct inode *inode;
3884
3885 inode = btrfs_iget_locked(s, location->objectid, root);
3886 if (!inode)
5d4f98a2 3887 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3888
3889 if (inode->i_state & I_NEW) {
3890 BTRFS_I(inode)->root = root;
3891 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3892 btrfs_read_locked_inode(inode);
1748f843
MF		 3893 if (!is_bad_inode(inode)) {
3894 inode_tree_add(inode);
3895 unlock_new_inode(inode);
3896 if (new)
3897 *new = 1;
3898 } else {
e0b6d65b
ST		 3899 unlock_new_inode(inode);
3900 iput(inode);
3901 inode = ERR_PTR(-ESTALE);
1748f843
MF		 3902 }
3903 }
3904
1a54ef8c
BR		 3905 return inode;
3906}
3907
4df27c4d
YZ		 3908static struct inode *new_simple_dir(struct super_block *s,
3909 struct btrfs_key *key,
3910 struct btrfs_root *root)
3911{
3912 struct inode *inode = new_inode(s);
3913
3914 if (!inode)
3915 return ERR_PTR(-ENOMEM);
3916
4df27c4d
YZ		 3917 BTRFS_I(inode)->root = root;
3918 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3919 BTRFS_I(inode)->dummy_inode = 1;
3920
3921 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3922 inode->i_op = &simple_dir_inode_operations;
3923 inode->i_fop = &simple_dir_operations;
3924 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3925 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3926
3927 return inode;
3928}
3929
3de4586c 3930struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3931{
d397712b 3932 struct inode *inode;
4df27c4d 3933 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 3934 struct btrfs_root *sub_root = root;
3935 struct btrfs_key location;
76dda93c 3936 int index;
b4aff1f8 3937 int ret = 0;
39279cc3
CM		 3938
3939 if (dentry->d_name.len > BTRFS_NAME_LEN)
3940 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3941
b4aff1f8
JB		 3942 if (unlikely(d_need_lookup(dentry))) {
3943 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3944 kfree(dentry->d_fsdata);
3945 dentry->d_fsdata = NULL;
a66e7cc6
JB		 3946 /* This thing is hashed, drop it for now */
3947 d_drop(dentry);
b4aff1f8
JB		 3948 } else {
3949 ret = btrfs_inode_by_name(dir, dentry, &location);
3950 }
5f39d397 3951
39279cc3
CM		 3952 if (ret < 0)
3953 return ERR_PTR(ret);
5f39d397 3954
4df27c4d
YZ		 3955 if (location.objectid == 0)
3956 return NULL;
3957
3958 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 3959 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 3960 return inode;
3961 }
3962
3963 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3964
76dda93c 3965 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 3966 ret = fixup_tree_root_location(root, dir, dentry,
3967 &location, &sub_root);
3968 if (ret < 0) {
3969 if (ret != -ENOENT)
3970 inode = ERR_PTR(ret);
3971 else
3972 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3973 } else {
73f73415 3974 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 3975 }
76dda93c
YZ		 3976 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3977
34d19bad 3978 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 3979 down_read(&root->fs_info->cleanup_work_sem);
3980 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 3981 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 3982 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 3983 if (ret)
3984 inode = ERR_PTR(ret);
c71bf099
YZ		 3985 }
3986
3de4586c
CM		 3987 return inode;
3988}
3989
fe15ce44 3990static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 3991{
3992 struct btrfs_root *root;
3993
efefb143
YZ		 3994 if (!dentry->d_inode && !IS_ROOT(dentry))
3995 dentry = dentry->d_parent;
76dda93c 3996
efefb143
YZ		 3997 if (dentry->d_inode) {
3998 root = BTRFS_I(dentry->d_inode)->root;
3999 if (btrfs_root_refs(&root->root_item) == 0)
4000 return 1;
4001 }
76dda93c
YZ		 4002 return 0;
4003}
4004
b4aff1f8
JB		 4005static void btrfs_dentry_release(struct dentry *dentry)
4006{
4007 if (dentry->d_fsdata)
4008 kfree(dentry->d_fsdata);
4009}
4010
3de4586c
CM		 4011static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4012 struct nameidata *nd)
4013{
a66e7cc6
JB		 4014 struct dentry *ret;
4015
4016 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4017 if (unlikely(d_need_lookup(dentry))) {
4018 spin_lock(&dentry->d_lock);
4019 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4020 spin_unlock(&dentry->d_lock);
4021 }
4022 return ret;
39279cc3
CM		 4023}
4024
16cdcec7 4025unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4026 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4027};
4028
cbdf5a24
DW		 4029static int btrfs_real_readdir(struct file *filp, void *dirent,
4030 filldir_t filldir)
39279cc3 4031{
6da6abae 4032 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4033 struct btrfs_root *root = BTRFS_I(inode)->root;
4034 struct btrfs_item *item;
4035 struct btrfs_dir_item *di;
4036 struct btrfs_key key;
5f39d397 4037 struct btrfs_key found_key;
39279cc3 4038 struct btrfs_path *path;
16cdcec7
MX		 4039 struct list_head ins_list;
4040 struct list_head del_list;
b4aff1f8 4041 struct qstr q;
39279cc3 4042 int ret;
5f39d397 4043 struct extent_buffer *leaf;
39279cc3 4044 int slot;
39279cc3
CM		 4045 unsigned char d_type;
4046 int over = 0;
4047 u32 di_cur;
4048 u32 di_total;
4049 u32 di_len;
4050 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4051 char tmp_name[32];
4052 char *name_ptr;
4053 int name_len;
16cdcec7 4054 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4055
4056 /* FIXME, use a real flag for deciding about the key type */
4057 if (root->fs_info->tree_root == root)
4058 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4059
3954401f
CM		 4060 /* special case for "." */
4061 if (filp->f_pos == 0) {
3765fefa
HS		 4062 over = filldir(dirent, ".", 1,
4063 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4064 if (over)
4065 return 0;
4066 filp->f_pos = 1;
4067 }
3954401f
CM		 4068 /* special case for .., just use the back ref */
4069 if (filp->f_pos == 1) {
5ecc7e5d 4070 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4071 over = filldir(dirent, "..", 2,
3765fefa 4072 filp->f_pos, pino, DT_DIR);
3954401f 4073 if (over)
49593bfa 4074 return 0;
3954401f
CM		 4075 filp->f_pos = 2;
4076 }
49593bfa 4077 path = btrfs_alloc_path();
16cdcec7
MX		 4078 if (!path)
4079 return -ENOMEM;
ff5714cc 4080
026fd317 4081 path->reada = 1;
49593bfa 4082
16cdcec7
MX		 4083 if (key_type == BTRFS_DIR_INDEX_KEY) {
4084 INIT_LIST_HEAD(&ins_list);
4085 INIT_LIST_HEAD(&del_list);
4086 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4087 }
4088
39279cc3
CM		 4089 btrfs_set_key_type(&key, key_type);
4090 key.offset = filp->f_pos;
33345d01 4091 key.objectid = btrfs_ino(inode);
5f39d397 4092
39279cc3
CM		 4093 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4094 if (ret < 0)
4095 goto err;
49593bfa
DW		 4096
4097 while (1) {
5f39d397 4098 leaf = path->nodes[0];
39279cc3 4099 slot = path->slots[0];
b9e03af0
LZ		 4100 if (slot >= btrfs_header_nritems(leaf)) {
4101 ret = btrfs_next_leaf(root, path);
4102 if (ret < 0)
4103 goto err;
4104 else if (ret > 0)
4105 break;
4106 continue;
39279cc3 4107 }
3de4586c 4108
5f39d397
CM		 4109 item = btrfs_item_nr(leaf, slot);
4110 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4111
4112 if (found_key.objectid != key.objectid)
39279cc3 4113 break;
5f39d397 4114 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4115 break;
5f39d397 4116 if (found_key.offset < filp->f_pos)
b9e03af0 4117 goto next;
16cdcec7
MX		 4118 if (key_type == BTRFS_DIR_INDEX_KEY &&
4119 btrfs_should_delete_dir_index(&del_list,
4120 found_key.offset))
4121 goto next;
5f39d397
CM		 4122
4123 filp->f_pos = found_key.offset;
16cdcec7 4124 is_curr = 1;
49593bfa 4125
39279cc3
CM		 4126 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4127 di_cur = 0;
5f39d397 4128 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4129
4130 while (di_cur < di_total) {
5f39d397 4131 struct btrfs_key location;
b4aff1f8 4132 struct dentry *tmp;
5f39d397 4133
22a94d44
JB		 4134 if (verify_dir_item(root, leaf, di))
4135 break;
4136
5f39d397 4137 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4138 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4139 name_ptr = tmp_name;
4140 } else {
4141 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4142 if (!name_ptr) {
4143 ret = -ENOMEM;
4144 goto err;
4145 }
5f39d397
CM		 4146 }
4147 read_extent_buffer(leaf, name_ptr,
4148 (unsigned long)(di + 1), name_len);
4149
4150 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4151 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4152
b4aff1f8
JB		 4153 q.name = name_ptr;
4154 q.len = name_len;
4155 q.hash = full_name_hash(q.name, q.len);
4156 tmp = d_lookup(filp->f_dentry, &q);
4157 if (!tmp) {
4158 struct btrfs_key *newkey;
4159
4160 newkey = kzalloc(sizeof(struct btrfs_key),
4161 GFP_NOFS);
4162 if (!newkey)
4163 goto no_dentry;
4164 tmp = d_alloc(filp->f_dentry, &q);
4165 if (!tmp) {
4166 kfree(newkey);
4167 dput(tmp);
4168 goto no_dentry;
4169 }
4170 memcpy(newkey, &location,
4171 sizeof(struct btrfs_key));
4172 tmp->d_fsdata = newkey;
4173 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4174 d_rehash(tmp);
4175 dput(tmp);
4176 } else {
4177 dput(tmp);
4178 }
4179no_dentry:
3de4586c
CM		 4180 /* is this a reference to our own snapshot? If so
4181 * skip it
4182 */
4183 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4184 location.objectid == root->root_key.objectid) {
4185 over = 0;
4186 goto skip;
4187 }
5f39d397 4188 over = filldir(dirent, name_ptr, name_len,
49593bfa 4189 found_key.offset, location.objectid,
39279cc3 4190 d_type);
5f39d397 4191
3de4586c 4192skip:
5f39d397
CM		 4193 if (name_ptr != tmp_name)
4194 kfree(name_ptr);
4195
39279cc3
CM		 4196 if (over)
4197 goto nopos;
5103e947 4198 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4199 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4200 di_cur += di_len;
4201 di = (struct btrfs_dir_item *)((char *)di + di_len);
4202 }
b9e03af0
LZ		 4203next:
4204 path->slots[0]++;
39279cc3 4205 }
49593bfa 4206
16cdcec7
MX		 4207 if (key_type == BTRFS_DIR_INDEX_KEY) {
4208 if (is_curr)
4209 filp->f_pos++;
4210 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4211 &ins_list);
4212 if (ret)
4213 goto nopos;
4214 }
4215
49593bfa 4216 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4217 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4218 /*
4219 * 32-bit glibc will use getdents64, but then strtol -
4220 * so the last number we can serve is this.
4221 */
4222 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4223 else
4224 filp->f_pos++;
39279cc3
CM		 4225nopos:
4226 ret = 0;
4227err:
16cdcec7
MX		 4228 if (key_type == BTRFS_DIR_INDEX_KEY)
4229 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4230 btrfs_free_path(path);
39279cc3
CM		 4231 return ret;
4232}
4233
a9185b41 4234int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4235{
4236 struct btrfs_root *root = BTRFS_I(inode)->root;
4237 struct btrfs_trans_handle *trans;
4238 int ret = 0;
0af3d00b 4239 bool nolock = false;
39279cc3 4240
8929ecfa 4241 if (BTRFS_I(inode)->dummy_inode)
4ca8b41e
CM		 4242 return 0;
4243
2cf8572d 4244 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
82d5902d 4245 nolock = true;
0af3d00b 4246
a9185b41 4247 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4248 if (nolock)
7a7eaa40 4249 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4250 else
7a7eaa40 4251 trans = btrfs_join_transaction(root);
3612b495
TI		 4252 if (IS_ERR(trans))
4253 return PTR_ERR(trans);
0af3d00b
JB		 4254 if (nolock)
4255 ret = btrfs_end_transaction_nolock(trans, root);
4256 else
4257 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4258 }
4259 return ret;
4260}
4261
4262/*
54aa1f4d 4263 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4264 * inode changes. But, it is most likely to find the inode in cache.
4265 * FIXME, needs more benchmarking...there are no reasons other than performance
4266 * to keep or drop this code.
4267 */
22c44fe6 4268int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 4269{
4270 struct btrfs_root *root = BTRFS_I(inode)->root;
4271 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4272 int ret;
4273
4274 if (BTRFS_I(inode)->dummy_inode)
22c44fe6 4275 return 0;
39279cc3 4276
7a7eaa40 4277 trans = btrfs_join_transaction(root);
22c44fe6
JB		 4278 if (IS_ERR(trans))
4279 return PTR_ERR(trans);
8929ecfa
YZ		 4280
4281 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4282 if (ret && ret == -ENOSPC) {
4283 /* whoops, lets try again with the full transaction */
4284 btrfs_end_transaction(trans, root);
4285 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 4286 if (IS_ERR(trans))
4287 return PTR_ERR(trans);
8929ecfa 4288
94b60442 4289 ret = btrfs_update_inode(trans, root, inode);
94b60442 4290 }
39279cc3 4291 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4292 if (BTRFS_I(inode)->delayed_node)
4293 btrfs_balance_delayed_items(root);
22c44fe6
JB		 4294
4295 return ret;
4296}
4297
4298/*
4299 * This is a copy of file_update_time. We need this so we can return error on
4300 * ENOSPC for updating the inode in the case of file write and mmap writes.
4301 */
4302int btrfs_update_time(struct file *file)
4303{
4304 struct inode *inode = file->f_path.dentry->d_inode;
4305 struct timespec now;
4306 int ret;
4307 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4308
4309 /* First try to exhaust all avenues to not sync */
4310 if (IS_NOCMTIME(inode))
4311 return 0;
4312
4313 now = current_fs_time(inode->i_sb);
4314 if (!timespec_equal(&inode->i_mtime, &now))
4315 sync_it = S_MTIME;
4316
4317 if (!timespec_equal(&inode->i_ctime, &now))
4318 sync_it |= S_CTIME;
4319
4320 if (IS_I_VERSION(inode))
4321 sync_it |= S_VERSION;
4322
4323 if (!sync_it)
4324 return 0;
4325
4326 /* Finally allowed to write? Takes lock. */
4327 if (mnt_want_write_file(file))
4328 return 0;
4329
4330 /* Only change inode inside the lock region */
4331 if (sync_it & S_VERSION)
4332 inode_inc_iversion(inode);
4333 if (sync_it & S_CTIME)
4334 inode->i_ctime = now;
4335 if (sync_it & S_MTIME)
4336 inode->i_mtime = now;
4337 ret = btrfs_dirty_inode(inode);
4338 if (!ret)
4339 mark_inode_dirty_sync(inode);
4340 mnt_drop_write(file->f_path.mnt);
4341 return ret;
39279cc3
CM		 4342}
4343
d352ac68
CM		 4344/*
4345 * find the highest existing sequence number in a directory
4346 * and then set the in-memory index_cnt variable to reflect
4347 * free sequence numbers
4348 */
aec7477b
JB		 4349static int btrfs_set_inode_index_count(struct inode *inode)
4350{
4351 struct btrfs_root *root = BTRFS_I(inode)->root;
4352 struct btrfs_key key, found_key;
4353 struct btrfs_path *path;
4354 struct extent_buffer *leaf;
4355 int ret;
4356
33345d01 4357 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4358 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4359 key.offset = (u64)-1;
4360
4361 path = btrfs_alloc_path();
4362 if (!path)
4363 return -ENOMEM;
4364
4365 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4366 if (ret < 0)
4367 goto out;
4368 /* FIXME: we should be able to handle this */
4369 if (ret == 0)
4370 goto out;
4371 ret = 0;
4372
4373 /*
4374 * MAGIC NUMBER EXPLANATION:
4375 * since we search a directory based on f_pos we have to start at 2
4376 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4377 * else has to start at 2
4378 */
4379 if (path->slots[0] == 0) {
4380 BTRFS_I(inode)->index_cnt = 2;
4381 goto out;
4382 }
4383
4384 path->slots[0]--;
4385
4386 leaf = path->nodes[0];
4387 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4388
33345d01 4389 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4390 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4391 BTRFS_I(inode)->index_cnt = 2;
4392 goto out;
4393 }
4394
4395 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4396out:
4397 btrfs_free_path(path);
4398 return ret;
4399}
4400
d352ac68
CM		 4401/*
4402 * helper to find a free sequence number in a given directory. This current
4403 * code is very simple, later versions will do smarter things in the btree
4404 */
3de4586c 4405int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4406{
4407 int ret = 0;
4408
4409 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4410 ret = btrfs_inode_delayed_dir_index_count(dir);
4411 if (ret) {
4412 ret = btrfs_set_inode_index_count(dir);
4413 if (ret)
4414 return ret;
4415 }
aec7477b
JB		 4416 }
4417
00e4e6b3 4418 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4419 BTRFS_I(dir)->index_cnt++;
4420
4421 return ret;
4422}
4423
39279cc3
CM		 4424static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4425 struct btrfs_root *root,
aec7477b 4426 struct inode *dir,
9c58309d 4427 const char *name, int name_len,
175a4eb7
AV		 4428 u64 ref_objectid, u64 objectid,
4429 umode_t mode, u64 *index)
39279cc3
CM		 4430{
4431 struct inode *inode;
5f39d397 4432 struct btrfs_inode_item *inode_item;
39279cc3 4433 struct btrfs_key *location;
5f39d397 4434 struct btrfs_path *path;
9c58309d
CM		 4435 struct btrfs_inode_ref *ref;
4436 struct btrfs_key key[2];
4437 u32 sizes[2];
4438 unsigned long ptr;
39279cc3
CM		 4439 int ret;
4440 int owner;
4441
5f39d397 4442 path = btrfs_alloc_path();
d8926bb3
MF		 4443 if (!path)
4444 return ERR_PTR(-ENOMEM);
5f39d397 4445
39279cc3 4446 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4447 if (!inode) {
4448 btrfs_free_path(path);
39279cc3 4449 return ERR_PTR(-ENOMEM);
8fb27640 4450 }
39279cc3 4451
581bb050
LZ		 4452 /*
4453 * we have to initialize this early, so we can reclaim the inode
4454 * number if we fail afterwards in this function.
4455 */
4456 inode->i_ino = objectid;
4457
aec7477b 4458 if (dir) {
1abe9b8a 4459 trace_btrfs_inode_request(dir);
4460
3de4586c 4461 ret = btrfs_set_inode_index(dir, index);
09771430 4462 if (ret) {
8fb27640 4463 btrfs_free_path(path);
09771430 4464 iput(inode);
aec7477b 4465 return ERR_PTR(ret);
09771430 4466 }
aec7477b
JB		 4467 }
4468 /*
4469 * index_cnt is ignored for everything but a dir,
4470 * btrfs_get_inode_index_count has an explanation for the magic
4471 * number
4472 */
4473 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4474 BTRFS_I(inode)->root = root;
e02119d5 4475 BTRFS_I(inode)->generation = trans->transid;
76195853 4476 inode->i_generation = BTRFS_I(inode)->generation;
6a63209f 4477 btrfs_set_inode_space_info(root, inode);
b888db2b 4478
569254b0 4479 if (S_ISDIR(mode))
39279cc3
CM		 4480 owner = 0;
4481 else
4482 owner = 1;
9c58309d
CM		 4483
4484 key[0].objectid = objectid;
4485 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4486 key[0].offset = 0;
4487
4488 key[1].objectid = objectid;
4489 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4490 key[1].offset = ref_objectid;
4491
4492 sizes[0] = sizeof(struct btrfs_inode_item);
4493 sizes[1] = name_len + sizeof(*ref);
4494
b9473439 4495 path->leave_spinning = 1;
9c58309d
CM		 4496 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4497 if (ret != 0)
5f39d397
CM		 4498 goto fail;
4499
ecc11fab 4500 inode_init_owner(inode, dir, mode);
a76a3cd4 4501 inode_set_bytes(inode, 0);
39279cc3 4502 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4503 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4504 struct btrfs_inode_item);
e02119d5 4505 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4506
4507 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4508 struct btrfs_inode_ref);
4509 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4510 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4511 ptr = (unsigned long)(ref + 1);
4512 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4513
5f39d397
CM		 4514 btrfs_mark_buffer_dirty(path->nodes[0]);
4515 btrfs_free_path(path);
4516
39279cc3
CM		 4517 location = &BTRFS_I(inode)->location;
4518 location->objectid = objectid;
39279cc3
CM		 4519 location->offset = 0;
4520 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4521
6cbff00f
CH		 4522 btrfs_inherit_iflags(inode, dir);
4523
569254b0 4524 if (S_ISREG(mode)) {
94272164
CM		 4525 if (btrfs_test_opt(root, NODATASUM))
4526 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4527 if (btrfs_test_opt(root, NODATACOW) ||
4528 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4529 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4530 }
4531
39279cc3 4532 insert_inode_hash(inode);
5d4f98a2 4533 inode_tree_add(inode);
1abe9b8a 4534
4535 trace_btrfs_inode_new(inode);
1973f0fa 4536 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4537
39279cc3 4538 return inode;
5f39d397 4539fail:
aec7477b
JB		 4540 if (dir)
4541 BTRFS_I(dir)->index_cnt--;
5f39d397 4542 btrfs_free_path(path);
09771430 4543 iput(inode);
5f39d397 4544 return ERR_PTR(ret);
39279cc3
CM		 4545}
4546
4547static inline u8 btrfs_inode_type(struct inode *inode)
4548{
4549 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4550}
4551
d352ac68
CM		 4552/*
4553 * utility function to add 'inode' into 'parent_inode' with
4554 * a give name and a given sequence number.
4555 * if 'add_backref' is true, also insert a backref from the
4556 * inode to the parent directory.
4557 */
e02119d5
CM		 4558int btrfs_add_link(struct btrfs_trans_handle *trans,
4559 struct inode *parent_inode, struct inode *inode,
4560 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4561{
4df27c4d 4562 int ret = 0;
39279cc3 4563 struct btrfs_key key;
e02119d5 4564 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4565 u64 ino = btrfs_ino(inode);
4566 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4567
33345d01 4568 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4569 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4570 } else {
33345d01 4571 key.objectid = ino;
4df27c4d
YZ		 4572 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4573 key.offset = 0;
4574 }
4575
33345d01 4576 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4577 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4578 key.objectid, root->root_key.objectid,
33345d01 4579 parent_ino, index, name, name_len);
4df27c4d 4580 } else if (add_backref) {
33345d01
LZ		 4581 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4582 parent_ino, index);
4df27c4d 4583 }
39279cc3 4584
39279cc3 4585 if (ret == 0) {
4df27c4d 4586 ret = btrfs_insert_dir_item(trans, root, name, name_len,
16cdcec7 4587 parent_inode, &key,
4df27c4d 4588 btrfs_inode_type(inode), index);
fe66a05a
CM		 4589 if (ret)
4590 goto fail_dir_item;
4df27c4d 4591
dbe674a9 4592 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 4593 name_len * 2);
79c44584 4594 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 4595 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 4596 }
4597 return ret;
fe66a05a
CM		 4598
4599fail_dir_item:
4600 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4601 u64 local_index;
4602 int err;
4603 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4604 key.objectid, root->root_key.objectid,
4605 parent_ino, &local_index, name, name_len);
4606
4607 } else if (add_backref) {
4608 u64 local_index;
4609 int err;
4610
4611 err = btrfs_del_inode_ref(trans, root, name, name_len,
4612 ino, parent_ino, &local_index);
4613 }
4614 return ret;
39279cc3
CM		 4615}
4616
4617static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4618 struct inode *dir, struct dentry *dentry,
4619 struct inode *inode, int backref, u64 index)
39279cc3 4620{
a1b075d2
JB		 4621 int err = btrfs_add_link(trans, dir, inode,
4622 dentry->d_name.name, dentry->d_name.len,
4623 backref, index);
39279cc3
CM		 4624 if (err > 0)
4625 err = -EEXIST;
4626 return err;
4627}
4628
618e21d5 4629static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 4630 umode_t mode, dev_t rdev)
618e21d5
JB		 4631{
4632 struct btrfs_trans_handle *trans;
4633 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4634 struct inode *inode = NULL;
618e21d5
JB		 4635 int err;
4636 int drop_inode = 0;
4637 u64 objectid;
1832a6d5 4638 unsigned long nr = 0;
00e4e6b3 4639 u64 index = 0;
618e21d5
JB		 4640
4641 if (!new_valid_dev(rdev))
4642 return -EINVAL;
4643
9ed74f2d
JB		 4644 /*
4645 * 2 for inode item and ref
4646 * 2 for dir items
4647 * 1 for xattr if selinux is on
4648 */
a22285a6
YZ		 4649 trans = btrfs_start_transaction(root, 5);
4650 if (IS_ERR(trans))
4651 return PTR_ERR(trans);
1832a6d5 4652
581bb050
LZ		 4653 err = btrfs_find_free_ino(root, &objectid);
4654 if (err)
4655 goto out_unlock;
4656
aec7477b 4657 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4658 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4659 mode, &index);
7cf96da3
TI		 4660 if (IS_ERR(inode)) {
4661 err = PTR_ERR(inode);
618e21d5 4662 goto out_unlock;
7cf96da3 4663 }
618e21d5 4664
2a7dba39 4665 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4666 if (err) {
4667 drop_inode = 1;
4668 goto out_unlock;
4669 }
4670
ad19db71
CS		 4671 /*
4672 * If the active LSM wants to access the inode during
4673 * d_instantiate it needs these. Smack checks to see
4674 * if the filesystem supports xattrs by looking at the
4675 * ops vector.
4676 */
4677
4678 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 4679 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4680 if (err)
4681 drop_inode = 1;
4682 else {
618e21d5 4683 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4684 btrfs_update_inode(trans, root, inode);
08c422c2 4685 d_instantiate(dentry, inode);
618e21d5 4686 }
618e21d5 4687out_unlock:
d3c2fdcf 4688 nr = trans->blocks_used;
7ad85bb7 4689 btrfs_end_transaction(trans, root);
a22285a6 4690 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4691 if (drop_inode) {
4692 inode_dec_link_count(inode);
4693 iput(inode);
4694 }
618e21d5
JB		 4695 return err;
4696}
4697
39279cc3 4698static int btrfs_create(struct inode *dir, struct dentry *dentry,
4acdaf27 4699 umode_t mode, struct nameidata *nd)
39279cc3
CM		 4700{
4701 struct btrfs_trans_handle *trans;
4702 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4703 struct inode *inode = NULL;
39279cc3 4704 int drop_inode = 0;
a22285a6 4705 int err;
1832a6d5 4706 unsigned long nr = 0;
39279cc3 4707 u64 objectid;
00e4e6b3 4708 u64 index = 0;
39279cc3 4709
9ed74f2d
JB		 4710 /*
4711 * 2 for inode item and ref
4712 * 2 for dir items
4713 * 1 for xattr if selinux is on
4714 */
a22285a6
YZ		 4715 trans = btrfs_start_transaction(root, 5);
4716 if (IS_ERR(trans))
4717 return PTR_ERR(trans);
9ed74f2d 4718
581bb050
LZ		 4719 err = btrfs_find_free_ino(root, &objectid);
4720 if (err)
4721 goto out_unlock;
4722
aec7477b 4723 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4724 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4725 mode, &index);
7cf96da3
TI		 4726 if (IS_ERR(inode)) {
4727 err = PTR_ERR(inode);
39279cc3 4728 goto out_unlock;
7cf96da3 4729 }
39279cc3 4730
2a7dba39 4731 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4732 if (err) {
4733 drop_inode = 1;
4734 goto out_unlock;
4735 }
4736
ad19db71
CS		 4737 /*
4738 * If the active LSM wants to access the inode during
4739 * d_instantiate it needs these. Smack checks to see
4740 * if the filesystem supports xattrs by looking at the
4741 * ops vector.
4742 */
4743 inode->i_fop = &btrfs_file_operations;
4744 inode->i_op = &btrfs_file_inode_operations;
4745
a1b075d2 4746 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 4747 if (err)
4748 drop_inode = 1;
4749 else {
4750 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4751 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 4752 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
08c422c2 4753 d_instantiate(dentry, inode);
39279cc3 4754 }
39279cc3 4755out_unlock:
d3c2fdcf 4756 nr = trans->blocks_used;
7ad85bb7 4757 btrfs_end_transaction(trans, root);
39279cc3
CM		 4758 if (drop_inode) {
4759 inode_dec_link_count(inode);
4760 iput(inode);
4761 }
d3c2fdcf 4762 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4763 return err;
4764}
4765
4766static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4767 struct dentry *dentry)
4768{
4769 struct btrfs_trans_handle *trans;
4770 struct btrfs_root *root = BTRFS_I(dir)->root;
4771 struct inode *inode = old_dentry->d_inode;
00e4e6b3 4772 u64 index;
1832a6d5 4773 unsigned long nr = 0;
39279cc3
CM		 4774 int err;
4775 int drop_inode = 0;
4776
4a8be425
TH		 4777 /* do not allow sys_link's with other subvols of the same device */
4778 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 4779 return -EXDEV;
4a8be425 4780
c055e99e
AV		 4781 if (inode->i_nlink == ~0U)
4782 return -EMLINK;
4a8be425 4783
3de4586c 4784 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 4785 if (err)
4786 goto fail;
4787
a22285a6 4788 /*
7e6b6465 4789 * 2 items for inode and inode ref
a22285a6 4790 * 2 items for dir items
7e6b6465 4791 * 1 item for parent inode
a22285a6 4792 */
7e6b6465 4793 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4794 if (IS_ERR(trans)) {
4795 err = PTR_ERR(trans);
4796 goto fail;
4797 }
5f39d397 4798
3153495d
MX		 4799 btrfs_inc_nlink(inode);
4800 inode->i_ctime = CURRENT_TIME;
7de9c6ee 4801 ihold(inode);
aec7477b 4802
a1b075d2 4803 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 4804
a5719521 4805 if (err) {
54aa1f4d 4806 drop_inode = 1;
a5719521 4807 } else {
10d9f309 4808 struct dentry *parent = dentry->d_parent;
a5719521
YZ		 4809 err = btrfs_update_inode(trans, root, inode);
4810 BUG_ON(err);
08c422c2 4811 d_instantiate(dentry, inode);
6a912213 4812 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 4813 }
39279cc3 4814
d3c2fdcf 4815 nr = trans->blocks_used;
7ad85bb7 4816 btrfs_end_transaction(trans, root);
1832a6d5 4817fail:
39279cc3
CM		 4818 if (drop_inode) {
4819 inode_dec_link_count(inode);
4820 iput(inode);
4821 }
d3c2fdcf 4822 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4823 return err;
4824}
4825
18bb1db3 4826static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 4827{
b9d86667 4828 struct inode *inode = NULL;
39279cc3
CM		 4829 struct btrfs_trans_handle *trans;
4830 struct btrfs_root *root = BTRFS_I(dir)->root;
4831 int err = 0;
4832 int drop_on_err = 0;
b9d86667 4833 u64 objectid = 0;
00e4e6b3 4834 u64 index = 0;
d3c2fdcf 4835 unsigned long nr = 1;
39279cc3 4836
9ed74f2d
JB		 4837 /*
4838 * 2 items for inode and ref
4839 * 2 items for dir items
4840 * 1 for xattr if selinux is on
4841 */
a22285a6
YZ		 4842 trans = btrfs_start_transaction(root, 5);
4843 if (IS_ERR(trans))
4844 return PTR_ERR(trans);
39279cc3 4845
581bb050
LZ		 4846 err = btrfs_find_free_ino(root, &objectid);
4847 if (err)
4848 goto out_fail;
4849
aec7477b 4850 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4851 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4852 S_IFDIR | mode, &index);
39279cc3
CM		 4853 if (IS_ERR(inode)) {
4854 err = PTR_ERR(inode);
4855 goto out_fail;
4856 }
5f39d397 4857
39279cc3 4858 drop_on_err = 1;
33268eaf 4859
2a7dba39 4860 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4861 if (err)
4862 goto out_fail;
4863
39279cc3
CM		 4864 inode->i_op = &btrfs_dir_inode_operations;
4865 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 4866
dbe674a9 4867 btrfs_i_size_write(inode, 0);
39279cc3
CM		 4868 err = btrfs_update_inode(trans, root, inode);
4869 if (err)
4870 goto out_fail;
5f39d397 4871
a1b075d2
JB		 4872 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4873 dentry->d_name.len, 0, index);
39279cc3
CM		 4874 if (err)
4875 goto out_fail;
5f39d397 4876
39279cc3
CM		 4877 d_instantiate(dentry, inode);
4878 drop_on_err = 0;
39279cc3
CM		 4879
4880out_fail:
d3c2fdcf 4881 nr = trans->blocks_used;
7ad85bb7 4882 btrfs_end_transaction(trans, root);
39279cc3
CM		 4883 if (drop_on_err)
4884 iput(inode);
d3c2fdcf 4885 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4886 return err;
4887}
4888
d352ac68
CM		 4889/* helper for btfs_get_extent. Given an existing extent in the tree,
4890 * and an extent that you want to insert, deal with overlap and insert
4891 * the new extent into the tree.
4892 */
3b951516
CM		 4893static int merge_extent_mapping(struct extent_map_tree *em_tree,
4894 struct extent_map *existing,
e6dcd2dc
CM		 4895 struct extent_map *em,
4896 u64 map_start, u64 map_len)
3b951516
CM		 4897{
4898 u64 start_diff;
3b951516 4899
e6dcd2dc
CM		 4900 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4901 start_diff = map_start - em->start;
4902 em->start = map_start;
4903 em->len = map_len;
c8b97818
CM		 4904 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4905 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 4906 em->block_start += start_diff;
c8b97818
CM		 4907 em->block_len -= start_diff;
4908 }
e6dcd2dc 4909 return add_extent_mapping(em_tree, em);
3b951516
CM		 4910}
4911
c8b97818
CM		 4912static noinline int uncompress_inline(struct btrfs_path *path,
4913 struct inode *inode, struct page *page,
4914 size_t pg_offset, u64 extent_offset,
4915 struct btrfs_file_extent_item *item)
4916{
4917 int ret;
4918 struct extent_buffer *leaf = path->nodes[0];
4919 char *tmp;
4920 size_t max_size;
4921 unsigned long inline_size;
4922 unsigned long ptr;
261507a0 4923 int compress_type;
c8b97818
CM		 4924
4925 WARN_ON(pg_offset != 0);
261507a0 4926 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 4927 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4928 inline_size = btrfs_file_extent_inline_item_len(leaf,
4929 btrfs_item_nr(leaf, path->slots[0]));
4930 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 4931 if (!tmp)
4932 return -ENOMEM;
c8b97818
CM		 4933 ptr = btrfs_file_extent_inline_start(item);
4934
4935 read_extent_buffer(leaf, tmp, ptr, inline_size);
4936
5b050f04 4937 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 4938 ret = btrfs_decompress(compress_type, tmp, page,
4939 extent_offset, inline_size, max_size);
c8b97818
CM		 4940 if (ret) {
4941 char *kaddr = kmap_atomic(page, KM_USER0);
4942 unsigned long copy_size = min_t(u64,
4943 PAGE_CACHE_SIZE - pg_offset,
4944 max_size - extent_offset);
4945 memset(kaddr + pg_offset, 0, copy_size);
4946 kunmap_atomic(kaddr, KM_USER0);
4947 }
4948 kfree(tmp);
4949 return 0;
4950}
4951
d352ac68
CM		 4952/*
4953 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4954 * the ugly parts come from merging extents from the disk with the in-ram
4955 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4956 * where the in-ram extents might be locked pending data=ordered completion.
4957 *
4958 * This also copies inline extents directly into the page.
4959 */
d397712b 4960
a52d9a80 4961struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4962 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4963 int create)
4964{
4965 int ret;
4966 int err = 0;
db94535d 4967 u64 bytenr;
a52d9a80
CM		 4968 u64 extent_start = 0;
4969 u64 extent_end = 0;
33345d01 4970 u64 objectid = btrfs_ino(inode);
a52d9a80 4971 u32 found_type;
f421950f 4972 struct btrfs_path *path = NULL;
a52d9a80
CM		 4973 struct btrfs_root *root = BTRFS_I(inode)->root;
4974 struct btrfs_file_extent_item *item;
5f39d397
CM		 4975 struct extent_buffer *leaf;
4976 struct btrfs_key found_key;
a52d9a80
CM		 4977 struct extent_map *em = NULL;
4978 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4979 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4980 struct btrfs_trans_handle *trans = NULL;
261507a0 4981 int compress_type;
a52d9a80 4982
a52d9a80 4983again:
890871be 4984 read_lock(&em_tree->lock);
d1310b2e 4985 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4986 if (em)
4987 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 4988 read_unlock(&em_tree->lock);
d1310b2e 4989
a52d9a80 4990 if (em) {
e1c4b745
CM		 4991 if (em->start > start || em->start + em->len <= start)
4992 free_extent_map(em);
4993 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4994 free_extent_map(em);
4995 else
4996 goto out;
a52d9a80 4997 }
172ddd60 4998 em = alloc_extent_map();
a52d9a80 4999 if (!em) {
d1310b2e
CM		 5000 err = -ENOMEM;
5001 goto out;
a52d9a80 5002 }
e6dcd2dc 5003 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5004 em->start = EXTENT_MAP_HOLE;
445a6944 5005 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5006 em->len = (u64)-1;
c8b97818 5007 em->block_len = (u64)-1;
f421950f
CM		 5008
5009 if (!path) {
5010 path = btrfs_alloc_path();
026fd317
JB		 5011 if (!path) {
5012 err = -ENOMEM;
5013 goto out;
5014 }
5015 /*
5016 * Chances are we'll be called again, so go ahead and do
5017 * readahead
5018 */
5019 path->reada = 1;
f421950f
CM		 5020 }
5021
179e29e4
CM		 5022 ret = btrfs_lookup_file_extent(trans, root, path,
5023 objectid, start, trans != NULL);
a52d9a80
CM		 5024 if (ret < 0) {
5025 err = ret;
5026 goto out;
5027 }
5028
5029 if (ret != 0) {
5030 if (path->slots[0] == 0)
5031 goto not_found;
5032 path->slots[0]--;
5033 }
5034
5f39d397
CM		 5035 leaf = path->nodes[0];
5036 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5037 struct btrfs_file_extent_item);
a52d9a80 5038 /* are we inside the extent that was found? */
5f39d397
CM		 5039 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5040 found_type = btrfs_key_type(&found_key);
5041 if (found_key.objectid != objectid ||
a52d9a80
CM		 5042 found_type != BTRFS_EXTENT_DATA_KEY) {
5043 goto not_found;
5044 }
5045
5f39d397
CM		 5046 found_type = btrfs_file_extent_type(leaf, item);
5047 extent_start = found_key.offset;
261507a0 5048 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5049 if (found_type == BTRFS_FILE_EXTENT_REG ||
5050 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5051 extent_end = extent_start +
db94535d 5052 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5053 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5054 size_t size;
5055 size = btrfs_file_extent_inline_len(leaf, item);
5056 extent_end = (extent_start + size + root->sectorsize - 1) &
5057 ~((u64)root->sectorsize - 1);
5058 }
5059
5060 if (start >= extent_end) {
5061 path->slots[0]++;
5062 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5063 ret = btrfs_next_leaf(root, path);
5064 if (ret < 0) {
5065 err = ret;
5066 goto out;
a52d9a80 5067 }
9036c102
YZ		 5068 if (ret > 0)
5069 goto not_found;
5070 leaf = path->nodes[0];
a52d9a80 5071 }
9036c102
YZ		 5072 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5073 if (found_key.objectid != objectid ||
5074 found_key.type != BTRFS_EXTENT_DATA_KEY)
5075 goto not_found;
5076 if (start + len <= found_key.offset)
5077 goto not_found;
5078 em->start = start;
5079 em->len = found_key.offset - start;
5080 goto not_found_em;
5081 }
5082
d899e052
YZ		 5083 if (found_type == BTRFS_FILE_EXTENT_REG ||
5084 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5085 em->start = extent_start;
5086 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5087 em->orig_start = extent_start -
5088 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5089 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5090 if (bytenr == 0) {
5f39d397 5091 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5092 goto insert;
5093 }
261507a0 5094 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5095 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5096 em->compress_type = compress_type;
c8b97818
CM		 5097 em->block_start = bytenr;
5098 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5099 item);
5100 } else {
5101 bytenr += btrfs_file_extent_offset(leaf, item);
5102 em->block_start = bytenr;
5103 em->block_len = em->len;
d899e052
YZ		 5104 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5105 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5106 }
a52d9a80
CM		 5107 goto insert;
5108 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5109 unsigned long ptr;
a52d9a80 5110 char *map;
3326d1b0
CM		 5111 size_t size;
5112 size_t extent_offset;
5113 size_t copy_size;
a52d9a80 5114
689f9346 5115 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5116 if (!page || create) {
689f9346 5117 em->start = extent_start;
9036c102 5118 em->len = extent_end - extent_start;
689f9346
Y		 5119 goto out;
5120 }
5f39d397 5121
9036c102
YZ		 5122 size = btrfs_file_extent_inline_len(leaf, item);
5123 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5124 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5125 size - extent_offset);
3326d1b0 5126 em->start = extent_start + extent_offset;
70dec807
CM		 5127 em->len = (copy_size + root->sectorsize - 1) &
5128 ~((u64)root->sectorsize - 1);
ff5b7ee3 5129 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5130 if (compress_type) {
c8b97818 5131 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5132 em->compress_type = compress_type;
5133 }
689f9346 5134 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5135 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5136 if (btrfs_file_extent_compression(leaf, item) !=
5137 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5138 ret = uncompress_inline(path, inode, page,
5139 pg_offset,
5140 extent_offset, item);
5141 BUG_ON(ret);
5142 } else {
5143 map = kmap(page);
5144 read_extent_buffer(leaf, map + pg_offset, ptr,
5145 copy_size);
93c82d57
CM		 5146 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5147 memset(map + pg_offset + copy_size, 0,
5148 PAGE_CACHE_SIZE - pg_offset -
5149 copy_size);
5150 }
c8b97818
CM		 5151 kunmap(page);
5152 }
179e29e4
CM		 5153 flush_dcache_page(page);
5154 } else if (create && PageUptodate(page)) {
6bf7e080 5155 BUG();
179e29e4
CM		 5156 if (!trans) {
5157 kunmap(page);
5158 free_extent_map(em);
5159 em = NULL;
ff5714cc 5160
b3b4aa74 5161 btrfs_release_path(path);
7a7eaa40 5162 trans = btrfs_join_transaction(root);
ff5714cc 5163
3612b495
TI		 5164 if (IS_ERR(trans))
5165 return ERR_CAST(trans);
179e29e4
CM		 5166 goto again;
5167 }
c8b97818 5168 map = kmap(page);
70dec807 5169 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5170 copy_size);
c8b97818 5171 kunmap(page);
179e29e4 5172 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5173 }
d1310b2e 5174 set_extent_uptodate(io_tree, em->start,
507903b8 5175 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5176 goto insert;
5177 } else {
d397712b 5178 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5179 WARN_ON(1);
5180 }
5181not_found:
5182 em->start = start;
d1310b2e 5183 em->len = len;
a52d9a80 5184not_found_em:
5f39d397 5185 em->block_start = EXTENT_MAP_HOLE;
9036c102 5186 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5187insert:
b3b4aa74 5188 btrfs_release_path(path);
d1310b2e 5189 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5190 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5191 "[%llu %llu]\n", (unsigned long long)em->start,
5192 (unsigned long long)em->len,
5193 (unsigned long long)start,
5194 (unsigned long long)len);
a52d9a80
CM		 5195 err = -EIO;
5196 goto out;
5197 }
d1310b2e
CM		 5198
5199 err = 0;
890871be 5200 write_lock(&em_tree->lock);
a52d9a80 5201 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5202 /* it is possible that someone inserted the extent into the tree
5203 * while we had the lock dropped. It is also possible that
5204 * an overlapping map exists in the tree
5205 */
a52d9a80 5206 if (ret == -EEXIST) {
3b951516 5207 struct extent_map *existing;
e6dcd2dc
CM		 5208
5209 ret = 0;
5210
3b951516 5211 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5212 if (existing && (existing->start > start ||
5213 existing->start + existing->len <= start)) {
5214 free_extent_map(existing);
5215 existing = NULL;
5216 }
3b951516
CM		 5217 if (!existing) {
5218 existing = lookup_extent_mapping(em_tree, em->start,
5219 em->len);
5220 if (existing) {
5221 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5222 em, start,
5223 root->sectorsize);
3b951516
CM		 5224 free_extent_map(existing);
5225 if (err) {
5226 free_extent_map(em);
5227 em = NULL;
5228 }
5229 } else {
5230 err = -EIO;
3b951516
CM		 5231 free_extent_map(em);
5232 em = NULL;
5233 }
5234 } else {
5235 free_extent_map(em);
5236 em = existing;
e6dcd2dc 5237 err = 0;
a52d9a80 5238 }
a52d9a80 5239 }
890871be 5240 write_unlock(&em_tree->lock);
a52d9a80 5241out:
1abe9b8a 5242
5243 trace_btrfs_get_extent(root, em);
5244
f421950f
CM		 5245 if (path)
5246 btrfs_free_path(path);
a52d9a80
CM		 5247 if (trans) {
5248 ret = btrfs_end_transaction(trans, root);
d397712b 5249 if (!err)
a52d9a80
CM		 5250 err = ret;
5251 }
a52d9a80
CM		 5252 if (err) {
5253 free_extent_map(em);
a52d9a80
CM		 5254 return ERR_PTR(err);
5255 }
5256 return em;
5257}
5258
ec29ed5b
CM		 5259struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5260 size_t pg_offset, u64 start, u64 len,
5261 int create)
5262{
5263 struct extent_map *em;
5264 struct extent_map *hole_em = NULL;
5265 u64 range_start = start;
5266 u64 end;
5267 u64 found;
5268 u64 found_end;
5269 int err = 0;
5270
5271 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5272 if (IS_ERR(em))
5273 return em;
5274 if (em) {
5275 /*
5276 * if our em maps to a hole, there might
5277 * actually be delalloc bytes behind it
5278 */
5279 if (em->block_start != EXTENT_MAP_HOLE)
5280 return em;
5281 else
5282 hole_em = em;
5283 }
5284
5285 /* check to see if we've wrapped (len == -1 or similar) */
5286 end = start + len;
5287 if (end < start)
5288 end = (u64)-1;
5289 else
5290 end -= 1;
5291
5292 em = NULL;
5293
5294 /* ok, we didn't find anything, lets look for delalloc */
5295 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5296 end, len, EXTENT_DELALLOC, 1);
5297 found_end = range_start + found;
5298 if (found_end < range_start)
5299 found_end = (u64)-1;
5300
5301 /*
5302 * we didn't find anything useful, return
5303 * the original results from get_extent()
5304 */
5305 if (range_start > end || found_end <= start) {
5306 em = hole_em;
5307 hole_em = NULL;
5308 goto out;
5309 }
5310
5311 /* adjust the range_start to make sure it doesn't
5312 * go backwards from the start they passed in
5313 */
5314 range_start = max(start,range_start);
5315 found = found_end - range_start;
5316
5317 if (found > 0) {
5318 u64 hole_start = start;
5319 u64 hole_len = len;
5320
172ddd60 5321 em = alloc_extent_map();
ec29ed5b
CM		 5322 if (!em) {
5323 err = -ENOMEM;
5324 goto out;
5325 }
5326 /*
5327 * when btrfs_get_extent can't find anything it
5328 * returns one huge hole
5329 *
5330 * make sure what it found really fits our range, and
5331 * adjust to make sure it is based on the start from
5332 * the caller
5333 */
5334 if (hole_em) {
5335 u64 calc_end = extent_map_end(hole_em);
5336
5337 if (calc_end <= start || (hole_em->start > end)) {
5338 free_extent_map(hole_em);
5339 hole_em = NULL;
5340 } else {
5341 hole_start = max(hole_em->start, start);
5342 hole_len = calc_end - hole_start;
5343 }
5344 }
5345 em->bdev = NULL;
5346 if (hole_em && range_start > hole_start) {
5347 /* our hole starts before our delalloc, so we
5348 * have to return just the parts of the hole
5349 * that go until the delalloc starts
5350 */
5351 em->len = min(hole_len,
5352 range_start - hole_start);
5353 em->start = hole_start;
5354 em->orig_start = hole_start;
5355 /*
5356 * don't adjust block start at all,
5357 * it is fixed at EXTENT_MAP_HOLE
5358 */
5359 em->block_start = hole_em->block_start;
5360 em->block_len = hole_len;
5361 } else {
5362 em->start = range_start;
5363 em->len = found;
5364 em->orig_start = range_start;
5365 em->block_start = EXTENT_MAP_DELALLOC;
5366 em->block_len = found;
5367 }
5368 } else if (hole_em) {
5369 return hole_em;
5370 }
5371out:
5372
5373 free_extent_map(hole_em);
5374 if (err) {
5375 free_extent_map(em);
5376 return ERR_PTR(err);
5377 }
5378 return em;
5379}
5380
4b46fce2 5381static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5382 struct extent_map *em,
4b46fce2
JB		 5383 u64 start, u64 len)
5384{
5385 struct btrfs_root *root = BTRFS_I(inode)->root;
5386 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5387 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5388 struct btrfs_key ins;
5389 u64 alloc_hint;
5390 int ret;
16d299ac 5391 bool insert = false;
4b46fce2 5392
16d299ac
JB		 5393 /*
5394 * Ok if the extent map we looked up is a hole and is for the exact
5395 * range we want, there is no reason to allocate a new one, however if
5396 * it is not right then we need to free this one and drop the cache for
5397 * our range.
5398 */
5399 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5400 em->len != len) {
5401 free_extent_map(em);
5402 em = NULL;
5403 insert = true;
5404 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5405 }
4b46fce2 5406
7a7eaa40 5407 trans = btrfs_join_transaction(root);
3612b495
TI		 5408 if (IS_ERR(trans))
5409 return ERR_CAST(trans);
4b46fce2 5410
4cb5300b
CM		 5411 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5412 btrfs_add_inode_defrag(trans, inode);
5413
4b46fce2
JB		 5414 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5415
5416 alloc_hint = get_extent_allocation_hint(inode, start, len);
5417 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5418 alloc_hint, (u64)-1, &ins, 1);
5419 if (ret) {
5420 em = ERR_PTR(ret);
5421 goto out;
5422 }
5423
4b46fce2 5424 if (!em) {
172ddd60 5425 em = alloc_extent_map();
16d299ac
JB		 5426 if (!em) {
5427 em = ERR_PTR(-ENOMEM);
5428 goto out;
5429 }
4b46fce2
JB		 5430 }
5431
5432 em->start = start;
5433 em->orig_start = em->start;
5434 em->len = ins.offset;
5435
5436 em->block_start = ins.objectid;
5437 em->block_len = ins.offset;
5438 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5439
5440 /*
5441 * We need to do this because if we're using the original em we searched
5442 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5443 */
5444 em->flags = 0;
4b46fce2
JB		 5445 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5446
16d299ac 5447 while (insert) {
4b46fce2
JB		 5448 write_lock(&em_tree->lock);
5449 ret = add_extent_mapping(em_tree, em);
5450 write_unlock(&em_tree->lock);
5451 if (ret != -EEXIST)
5452 break;
5453 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5454 }
5455
5456 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5457 ins.offset, ins.offset, 0);
5458 if (ret) {
5459 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5460 em = ERR_PTR(ret);
5461 }
5462out:
5463 btrfs_end_transaction(trans, root);
5464 return em;
5465}
5466
46bfbb5c
CM		 5467/*
5468 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5469 * block must be cow'd
5470 */
5471static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5472 struct inode *inode, u64 offset, u64 len)
5473{
5474 struct btrfs_path *path;
5475 int ret;
5476 struct extent_buffer *leaf;
5477 struct btrfs_root *root = BTRFS_I(inode)->root;
5478 struct btrfs_file_extent_item *fi;
5479 struct btrfs_key key;
5480 u64 disk_bytenr;
5481 u64 backref_offset;
5482 u64 extent_end;
5483 u64 num_bytes;
5484 int slot;
5485 int found_type;
5486
5487 path = btrfs_alloc_path();
5488 if (!path)
5489 return -ENOMEM;
5490
33345d01 5491 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5492 offset, 0);
5493 if (ret < 0)
5494 goto out;
5495
5496 slot = path->slots[0];
5497 if (ret == 1) {
5498 if (slot == 0) {
5499 /* can't find the item, must cow */
5500 ret = 0;
5501 goto out;
5502 }
5503 slot--;
5504 }
5505 ret = 0;
5506 leaf = path->nodes[0];
5507 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5508 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5509 key.type != BTRFS_EXTENT_DATA_KEY) {
5510 /* not our file or wrong item type, must cow */
5511 goto out;
5512 }
5513
5514 if (key.offset > offset) {
5515 /* Wrong offset, must cow */
5516 goto out;
5517 }
5518
5519 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5520 found_type = btrfs_file_extent_type(leaf, fi);
5521 if (found_type != BTRFS_FILE_EXTENT_REG &&
5522 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5523 /* not a regular extent, must cow */
5524 goto out;
5525 }
5526 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5527 backref_offset = btrfs_file_extent_offset(leaf, fi);
5528
5529 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5530 if (extent_end < offset + len) {
5531 /* extent doesn't include our full range, must cow */
5532 goto out;
5533 }
5534
5535 if (btrfs_extent_readonly(root, disk_bytenr))
5536 goto out;
5537
5538 /*
5539 * look for other files referencing this extent, if we
5540 * find any we must cow
5541 */
33345d01 5542 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5543 key.offset - backref_offset, disk_bytenr))
5544 goto out;
5545
5546 /*
5547 * adjust disk_bytenr and num_bytes to cover just the bytes
5548 * in this extent we are about to write. If there
5549 * are any csums in that range we have to cow in order
5550 * to keep the csums correct
5551 */
5552 disk_bytenr += backref_offset;
5553 disk_bytenr += offset - key.offset;
5554 num_bytes = min(offset + len, extent_end) - offset;
5555 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5556 goto out;
5557 /*
5558 * all of the above have passed, it is safe to overwrite this extent
5559 * without cow
5560 */
5561 ret = 1;
5562out:
5563 btrfs_free_path(path);
5564 return ret;
5565}
5566
4b46fce2
JB		 5567static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5568 struct buffer_head *bh_result, int create)
5569{
5570 struct extent_map *em;
5571 struct btrfs_root *root = BTRFS_I(inode)->root;
5572 u64 start = iblock << inode->i_blkbits;
5573 u64 len = bh_result->b_size;
46bfbb5c 5574 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5575
5576 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5577 if (IS_ERR(em))
5578 return PTR_ERR(em);
5579
5580 /*
5581 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5582 * io. INLINE is special, and we could probably kludge it in here, but
5583 * it's still buffered so for safety lets just fall back to the generic
5584 * buffered path.
5585 *
5586 * For COMPRESSED we _have_ to read the entire extent in so we can
5587 * decompress it, so there will be buffering required no matter what we
5588 * do, so go ahead and fallback to buffered.
5589 *
5590 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5591 * to buffered IO. Don't blame me, this is the price we pay for using
5592 * the generic code.
5593 */
5594 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5595 em->block_start == EXTENT_MAP_INLINE) {
5596 free_extent_map(em);
5597 return -ENOTBLK;
5598 }
5599
5600 /* Just a good old fashioned hole, return */
5601 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5602 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5603 free_extent_map(em);
5604 /* DIO will do one hole at a time, so just unlock a sector */
5605 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5606 start + root->sectorsize - 1, GFP_NOFS);
5607 return 0;
5608 }
5609
5610 /*
5611 * We don't allocate a new extent in the following cases
5612 *
5613 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5614 * existing extent.
5615 * 2) The extent is marked as PREALLOC. We're good to go here and can
5616 * just use the extent.
5617 *
5618 */
46bfbb5c
CM		 5619 if (!create) {
5620 len = em->len - (start - em->start);
4b46fce2 5621 goto map;
46bfbb5c 5622 }
4b46fce2
JB		 5623
5624 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5625 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5626 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 5627 int type;
5628 int ret;
46bfbb5c 5629 u64 block_start;
4b46fce2
JB		 5630
5631 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5632 type = BTRFS_ORDERED_PREALLOC;
5633 else
5634 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 5635 len = min(len, em->len - (start - em->start));
4b46fce2 5636 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 5637
5638 /*
5639 * we're not going to log anything, but we do need
5640 * to make sure the current transaction stays open
5641 * while we look for nocow cross refs
5642 */
7a7eaa40 5643 trans = btrfs_join_transaction(root);
3612b495 5644 if (IS_ERR(trans))
46bfbb5c
CM		 5645 goto must_cow;
5646
5647 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5648 ret = btrfs_add_ordered_extent_dio(inode, start,
5649 block_start, len, len, type);
5650 btrfs_end_transaction(trans, root);
5651 if (ret) {
5652 free_extent_map(em);
5653 return ret;
5654 }
5655 goto unlock;
4b46fce2 5656 }
46bfbb5c 5657 btrfs_end_transaction(trans, root);
4b46fce2 5658 }
46bfbb5c
CM		 5659must_cow:
5660 /*
5661 * this will cow the extent, reset the len in case we changed
5662 * it above
5663 */
5664 len = bh_result->b_size;
16d299ac 5665 em = btrfs_new_extent_direct(inode, em, start, len);
46bfbb5c
CM		 5666 if (IS_ERR(em))
5667 return PTR_ERR(em);
5668 len = min(len, em->len - (start - em->start));
5669unlock:
4845e44f
CM		 5670 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5671 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5672 0, NULL, GFP_NOFS);
4b46fce2
JB		 5673map:
5674 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5675 inode->i_blkbits;
46bfbb5c 5676 bh_result->b_size = len;
4b46fce2
JB		 5677 bh_result->b_bdev = em->bdev;
5678 set_buffer_mapped(bh_result);
5679 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5680 set_buffer_new(bh_result);
5681
5682 free_extent_map(em);
5683
5684 return 0;
5685}
5686
5687struct btrfs_dio_private {
5688 struct inode *inode;
5689 u64 logical_offset;
5690 u64 disk_bytenr;
5691 u64 bytes;
5692 u32 *csums;
5693 void *private;
e65e1535
MX		 5694
5695 /* number of bios pending for this dio */
5696 atomic_t pending_bios;
5697
5698 /* IO errors */
5699 int errors;
5700
5701 struct bio *orig_bio;
4b46fce2
JB		 5702};
5703
5704static void btrfs_endio_direct_read(struct bio *bio, int err)
5705{
e65e1535 5706 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 5707 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5708 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 5709 struct inode *inode = dip->inode;
5710 struct btrfs_root *root = BTRFS_I(inode)->root;
5711 u64 start;
5712 u32 *private = dip->csums;
5713
5714 start = dip->logical_offset;
5715 do {
5716 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5717 struct page *page = bvec->bv_page;
5718 char *kaddr;
5719 u32 csum = ~(u32)0;
5720 unsigned long flags;
5721
5722 local_irq_save(flags);
5723 kaddr = kmap_atomic(page, KM_IRQ0);
5724 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5725 csum, bvec->bv_len);
5726 btrfs_csum_final(csum, (char *)&csum);
5727 kunmap_atomic(kaddr, KM_IRQ0);
5728 local_irq_restore(flags);
5729
5730 flush_dcache_page(bvec->bv_page);
5731 if (csum != *private) {
33345d01 5732 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 5733 " %llu csum %u private %u\n",
33345d01
LZ		 5734 (unsigned long long)btrfs_ino(inode),
5735 (unsigned long long)start,
4b46fce2
JB		 5736 csum, *private);
5737 err = -EIO;
5738 }
5739 }
5740
5741 start += bvec->bv_len;
5742 private++;
5743 bvec++;
5744 } while (bvec <= bvec_end);
5745
5746 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5747 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5748 bio->bi_private = dip->private;
5749
5750 kfree(dip->csums);
5751 kfree(dip);
c0da7aa1
JB		 5752
5753 /* If we had a csum failure make sure to clear the uptodate flag */
5754 if (err)
5755 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5756 dio_end_io(bio, err);
5757}
5758
5759static void btrfs_endio_direct_write(struct bio *bio, int err)
5760{
5761 struct btrfs_dio_private *dip = bio->bi_private;
5762 struct inode *inode = dip->inode;
5763 struct btrfs_root *root = BTRFS_I(inode)->root;
5764 struct btrfs_trans_handle *trans;
5765 struct btrfs_ordered_extent *ordered = NULL;
5766 struct extent_state *cached_state = NULL;
163cf09c
CM		 5767 u64 ordered_offset = dip->logical_offset;
5768 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 5769 int ret;
5770
5771 if (err)
5772 goto out_done;
163cf09c
CM		 5773again:
5774 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5775 &ordered_offset,
5776 ordered_bytes);
4b46fce2 5777 if (!ret)
163cf09c 5778 goto out_test;
4b46fce2
JB		 5779
5780 BUG_ON(!ordered);
5781
7a7eaa40 5782 trans = btrfs_join_transaction(root);
3612b495 5783 if (IS_ERR(trans)) {
4b46fce2
JB		 5784 err = -ENOMEM;
5785 goto out;
5786 }
5787 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5788
5789 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5790 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5791 if (!ret)
2115133f 5792 err = btrfs_update_inode_fallback(trans, root, inode);
4b46fce2
JB		 5793 goto out;
5794 }
5795
5796 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5797 ordered->file_offset + ordered->len - 1, 0,
5798 &cached_state, GFP_NOFS);
5799
5800 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5801 ret = btrfs_mark_extent_written(trans, inode,
5802 ordered->file_offset,
5803 ordered->file_offset +
5804 ordered->len);
5805 if (ret) {
5806 err = ret;
5807 goto out_unlock;
5808 }
5809 } else {
5810 ret = insert_reserved_file_extent(trans, inode,
5811 ordered->file_offset,
5812 ordered->start,
5813 ordered->disk_len,
5814 ordered->len,
5815 ordered->len,
5816 0, 0, 0,
5817 BTRFS_FILE_EXTENT_REG);
5818 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5819 ordered->file_offset, ordered->len);
5820 if (ret) {
5821 err = ret;
5822 WARN_ON(1);
5823 goto out_unlock;
5824 }
5825 }
5826
5827 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
1ef30be1 5828 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
a39f7521 5829 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
2115133f 5830 btrfs_update_inode_fallback(trans, root, inode);
1ef30be1 5831 ret = 0;
4b46fce2
JB		 5832out_unlock:
5833 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5834 ordered->file_offset + ordered->len - 1,
5835 &cached_state, GFP_NOFS);
5836out:
5837 btrfs_delalloc_release_metadata(inode, ordered->len);
5838 btrfs_end_transaction(trans, root);
163cf09c 5839 ordered_offset = ordered->file_offset + ordered->len;
4b46fce2
JB		 5840 btrfs_put_ordered_extent(ordered);
5841 btrfs_put_ordered_extent(ordered);
163cf09c
CM		 5842
5843out_test:
5844 /*
5845 * our bio might span multiple ordered extents. If we haven't
5846 * completed the accounting for the whole dio, go back and try again
5847 */
5848 if (ordered_offset < dip->logical_offset + dip->bytes) {
5849 ordered_bytes = dip->logical_offset + dip->bytes -
5850 ordered_offset;
5851 goto again;
5852 }
4b46fce2
JB		 5853out_done:
5854 bio->bi_private = dip->private;
5855
5856 kfree(dip->csums);
5857 kfree(dip);
c0da7aa1
JB		 5858
5859 /* If we had an error make sure to clear the uptodate flag */
5860 if (err)
5861 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 5862 dio_end_io(bio, err);
5863}
5864
eaf25d93
CM		 5865static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5866 struct bio *bio, int mirror_num,
5867 unsigned long bio_flags, u64 offset)
5868{
5869 int ret;
5870 struct btrfs_root *root = BTRFS_I(inode)->root;
5871 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5872 BUG_ON(ret);
5873 return 0;
5874}
5875
e65e1535
MX		 5876static void btrfs_end_dio_bio(struct bio *bio, int err)
5877{
5878 struct btrfs_dio_private *dip = bio->bi_private;
5879
5880 if (err) {
33345d01 5881 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 5882 "sector %#Lx len %u err no %d\n",
33345d01 5883 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 5884 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 5885 dip->errors = 1;
5886
5887 /*
5888 * before atomic variable goto zero, we must make sure
5889 * dip->errors is perceived to be set.
5890 */
5891 smp_mb__before_atomic_dec();
5892 }
5893
5894 /* if there are more bios still pending for this dio, just exit */
5895 if (!atomic_dec_and_test(&dip->pending_bios))
5896 goto out;
5897
5898 if (dip->errors)
5899 bio_io_error(dip->orig_bio);
5900 else {
5901 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5902 bio_endio(dip->orig_bio, 0);
5903 }
5904out:
5905 bio_put(bio);
5906}
5907
5908static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5909 u64 first_sector, gfp_t gfp_flags)
5910{
5911 int nr_vecs = bio_get_nr_vecs(bdev);
5912 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5913}
5914
5915static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5916 int rw, u64 file_offset, int skip_sum,
1ae39938 5917 u32 *csums, int async_submit)
e65e1535
MX		 5918{
5919 int write = rw & REQ_WRITE;
5920 struct btrfs_root *root = BTRFS_I(inode)->root;
5921 int ret;
5922
5923 bio_get(bio);
5924 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5925 if (ret)
5926 goto err;
5927
1ae39938
JB		 5928 if (skip_sum)
5929 goto map;
5930
5931 if (write && async_submit) {
e65e1535
MX		 5932 ret = btrfs_wq_submit_bio(root->fs_info,
5933 inode, rw, bio, 0, 0,
5934 file_offset,
5935 __btrfs_submit_bio_start_direct_io,
5936 __btrfs_submit_bio_done);
5937 goto err;
1ae39938
JB		 5938 } else if (write) {
5939 /*
5940 * If we aren't doing async submit, calculate the csum of the
5941 * bio now.
5942 */
5943 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5944 if (ret)
5945 goto err;
c2db1073
TI		 5946 } else if (!skip_sum) {
5947 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
e65e1535 5948 file_offset, csums);
c2db1073
TI		 5949 if (ret)
5950 goto err;
5951 }
e65e1535 5952
1ae39938
JB		 5953map:
5954 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 5955err:
5956 bio_put(bio);
5957 return ret;
5958}
5959
5960static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5961 int skip_sum)
5962{
5963 struct inode *inode = dip->inode;
5964 struct btrfs_root *root = BTRFS_I(inode)->root;
5965 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5966 struct bio *bio;
5967 struct bio *orig_bio = dip->orig_bio;
5968 struct bio_vec *bvec = orig_bio->bi_io_vec;
5969 u64 start_sector = orig_bio->bi_sector;
5970 u64 file_offset = dip->logical_offset;
5971 u64 submit_len = 0;
5972 u64 map_length;
5973 int nr_pages = 0;
5974 u32 *csums = dip->csums;
5975 int ret = 0;
1ae39938 5976 int async_submit = 0;
98bc3149 5977 int write = rw & REQ_WRITE;
e65e1535 5978
e65e1535
MX		 5979 map_length = orig_bio->bi_size;
5980 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5981 &map_length, NULL, 0);
5982 if (ret) {
64728bbb 5983 bio_put(orig_bio);
e65e1535
MX		 5984 return -EIO;
5985 }
5986
02f57c7a
JB		 5987 if (map_length >= orig_bio->bi_size) {
5988 bio = orig_bio;
5989 goto submit;
5990 }
5991
1ae39938 5992 async_submit = 1;
02f57c7a
JB		 5993 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5994 if (!bio)
5995 return -ENOMEM;
5996 bio->bi_private = dip;
5997 bio->bi_end_io = btrfs_end_dio_bio;
5998 atomic_inc(&dip->pending_bios);
5999
e65e1535
MX		 6000 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6001 if (unlikely(map_length < submit_len + bvec->bv_len ||
6002 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6003 bvec->bv_offset) < bvec->bv_len)) {
6004 /*
6005 * inc the count before we submit the bio so
6006 * we know the end IO handler won't happen before
6007 * we inc the count. Otherwise, the dip might get freed
6008 * before we're done setting it up
6009 */
6010 atomic_inc(&dip->pending_bios);
6011 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6012 file_offset, skip_sum,
1ae39938 6013 csums, async_submit);
e65e1535
MX		 6014 if (ret) {
6015 bio_put(bio);
6016 atomic_dec(&dip->pending_bios);
6017 goto out_err;
6018 }
6019
98bc3149
JB		 6020 /* Write's use the ordered csums */
6021 if (!write && !skip_sum)
e65e1535
MX		 6022 csums = csums + nr_pages;
6023 start_sector += submit_len >> 9;
6024 file_offset += submit_len;
6025
6026 submit_len = 0;
6027 nr_pages = 0;
6028
6029 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6030 start_sector, GFP_NOFS);
6031 if (!bio)
6032 goto out_err;
6033 bio->bi_private = dip;
6034 bio->bi_end_io = btrfs_end_dio_bio;
6035
6036 map_length = orig_bio->bi_size;
6037 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6038 &map_length, NULL, 0);
6039 if (ret) {
6040 bio_put(bio);
6041 goto out_err;
6042 }
6043 } else {
6044 submit_len += bvec->bv_len;
6045 nr_pages ++;
6046 bvec++;
6047 }
6048 }
6049
02f57c7a 6050submit:
e65e1535 6051 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
1ae39938 6052 csums, async_submit);
e65e1535
MX		 6053 if (!ret)
6054 return 0;
6055
6056 bio_put(bio);
6057out_err:
6058 dip->errors = 1;
6059 /*
6060 * before atomic variable goto zero, we must
6061 * make sure dip->errors is perceived to be set.
6062 */
6063 smp_mb__before_atomic_dec();
6064 if (atomic_dec_and_test(&dip->pending_bios))
6065 bio_io_error(dip->orig_bio);
6066
6067 /* bio_end_io() will handle error, so we needn't return it */
6068 return 0;
6069}
6070
4b46fce2
JB		 6071static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6072 loff_t file_offset)
6073{
6074 struct btrfs_root *root = BTRFS_I(inode)->root;
6075 struct btrfs_dio_private *dip;
6076 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6077 int skip_sum;
7b6d91da 6078 int write = rw & REQ_WRITE;
4b46fce2
JB		 6079 int ret = 0;
6080
6081 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6082
6083 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6084 if (!dip) {
6085 ret = -ENOMEM;
6086 goto free_ordered;
6087 }
6088 dip->csums = NULL;
6089
98bc3149
JB		 6090 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6091 if (!write && !skip_sum) {
4b46fce2
JB		 6092 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6093 if (!dip->csums) {
b4966b77 6094 kfree(dip);
4b46fce2
JB		 6095 ret = -ENOMEM;
6096 goto free_ordered;
6097 }
6098 }
6099
6100 dip->private = bio->bi_private;
6101 dip->inode = inode;
6102 dip->logical_offset = file_offset;
6103
4b46fce2
JB		 6104 dip->bytes = 0;
6105 do {
6106 dip->bytes += bvec->bv_len;
6107 bvec++;
6108 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6109
46bfbb5c 6110 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6111 bio->bi_private = dip;
e65e1535
MX		 6112 dip->errors = 0;
6113 dip->orig_bio = bio;
6114 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6115
6116 if (write)
6117 bio->bi_end_io = btrfs_endio_direct_write;
6118 else
6119 bio->bi_end_io = btrfs_endio_direct_read;
6120
e65e1535
MX		 6121 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6122 if (!ret)
eaf25d93 6123 return;
4b46fce2
JB		 6124free_ordered:
6125 /*
6126 * If this is a write, we need to clean up the reserved space and kill
6127 * the ordered extent.
6128 */
6129 if (write) {
6130 struct btrfs_ordered_extent *ordered;
955256f2 6131 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6132 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6133 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6134 btrfs_free_reserved_extent(root, ordered->start,
6135 ordered->disk_len);
6136 btrfs_put_ordered_extent(ordered);
6137 btrfs_put_ordered_extent(ordered);
6138 }
6139 bio_endio(bio, ret);
6140}
6141
5a5f79b5
CM		 6142static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6143 const struct iovec *iov, loff_t offset,
6144 unsigned long nr_segs)
6145{
6146 int seg;
a1b75f7d 6147 int i;
5a5f79b5
CM		 6148 size_t size;
6149 unsigned long addr;
6150 unsigned blocksize_mask = root->sectorsize - 1;
6151 ssize_t retval = -EINVAL;
6152 loff_t end = offset;
6153
6154 if (offset & blocksize_mask)
6155 goto out;
6156
6157 /* Check the memory alignment. Blocks cannot straddle pages */
6158 for (seg = 0; seg < nr_segs; seg++) {
6159 addr = (unsigned long)iov[seg].iov_base;
6160 size = iov[seg].iov_len;
6161 end += size;
a1b75f7d 6162 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6163 goto out;
a1b75f7d
JB		 6164
6165 /* If this is a write we don't need to check anymore */
6166 if (rw & WRITE)
6167 continue;
6168
6169 /*
6170 * Check to make sure we don't have duplicate iov_base's in this
6171 * iovec, if so return EINVAL, otherwise we'll get csum errors
6172 * when reading back.
6173 */
6174 for (i = seg + 1; i < nr_segs; i++) {
6175 if (iov[seg].iov_base == iov[i].iov_base)
6176 goto out;
6177 }
5a5f79b5
CM		 6178 }
6179 retval = 0;
6180out:
6181 return retval;
6182}
16432985
CM		 6183static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6184 const struct iovec *iov, loff_t offset,
6185 unsigned long nr_segs)
6186{
4b46fce2
JB		 6187 struct file *file = iocb->ki_filp;
6188 struct inode *inode = file->f_mapping->host;
6189 struct btrfs_ordered_extent *ordered;
4845e44f 6190 struct extent_state *cached_state = NULL;
4b46fce2
JB		 6191 u64 lockstart, lockend;
6192 ssize_t ret;
4845e44f
CM		 6193 int writing = rw & WRITE;
6194 int write_bits = 0;
3f7c579c 6195 size_t count = iov_length(iov, nr_segs);
4b46fce2 6196
5a5f79b5
CM		 6197 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6198 offset, nr_segs)) {
6199 return 0;
6200 }
6201
4b46fce2 6202 lockstart = offset;
3f7c579c
CM		 6203 lockend = offset + count - 1;
6204
6205 if (writing) {
6206 ret = btrfs_delalloc_reserve_space(inode, count);
6207 if (ret)
6208 goto out;
6209 }
4845e44f 6210
4b46fce2 6211 while (1) {
4845e44f
CM		 6212 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6213 0, &cached_state, GFP_NOFS);
4b46fce2
JB		 6214 /*
6215 * We're concerned with the entire range that we're going to be
6216 * doing DIO to, so we need to make sure theres no ordered
6217 * extents in this range.
6218 */
6219 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6220 lockend - lockstart + 1);
6221 if (!ordered)
6222 break;
4845e44f
CM		 6223 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6224 &cached_state, GFP_NOFS);
4b46fce2
JB		 6225 btrfs_start_ordered_extent(inode, ordered, 1);
6226 btrfs_put_ordered_extent(ordered);
6227 cond_resched();
6228 }
6229
4845e44f
CM		 6230 /*
6231 * we don't use btrfs_set_extent_delalloc because we don't want
6232 * the dirty or uptodate bits
6233 */
6234 if (writing) {
6235 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6236 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6237 EXTENT_DELALLOC, 0, NULL, &cached_state,
6238 GFP_NOFS);
6239 if (ret) {
6240 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6241 lockend, EXTENT_LOCKED | write_bits,
6242 1, 0, &cached_state, GFP_NOFS);
6243 goto out;
6244 }
6245 }
6246
6247 free_extent_state(cached_state);
6248 cached_state = NULL;
6249
5a5f79b5
CM		 6250 ret = __blockdev_direct_IO(rw, iocb, inode,
6251 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6252 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6253 btrfs_submit_direct, 0);
4b46fce2
JB		 6254
6255 if (ret < 0 && ret != -EIOCBQUEUED) {
4845e44f
CM		 6256 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6257 offset + iov_length(iov, nr_segs) - 1,
6258 EXTENT_LOCKED | write_bits, 1, 0,
6259 &cached_state, GFP_NOFS);
4b46fce2
JB		 6260 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6261 /*
6262 * We're falling back to buffered, unlock the section we didn't
6263 * do IO on.
6264 */
4845e44f
CM		 6265 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6266 offset + iov_length(iov, nr_segs) - 1,
6267 EXTENT_LOCKED | write_bits, 1, 0,
6268 &cached_state, GFP_NOFS);
4b46fce2 6269 }
4845e44f
CM		 6270out:
6271 free_extent_state(cached_state);
4b46fce2 6272 return ret;
16432985
CM		 6273}
6274
1506fcc8
YS		 6275static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6276 __u64 start, __u64 len)
6277{
ec29ed5b 6278 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6279}
6280
a52d9a80 6281int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6282{
d1310b2e
CM		 6283 struct extent_io_tree *tree;
6284 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6285 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6286}
1832a6d5 6287
a52d9a80 6288static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6289{
d1310b2e 6290 struct extent_io_tree *tree;
b888db2b
CM		 6291
6292
6293 if (current->flags & PF_MEMALLOC) {
6294 redirty_page_for_writepage(wbc, page);
6295 unlock_page(page);
6296 return 0;
6297 }
d1310b2e 6298 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6299 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6300}
6301
f421950f
CM		 6302int btrfs_writepages(struct address_space *mapping,
6303 struct writeback_control *wbc)
b293f02e 6304{
d1310b2e 6305 struct extent_io_tree *tree;
771ed689 6306
d1310b2e 6307 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6308 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6309}
6310
3ab2fb5a
CM		 6311static int
6312btrfs_readpages(struct file *file, struct address_space *mapping,
6313 struct list_head *pages, unsigned nr_pages)
6314{
d1310b2e
CM		 6315 struct extent_io_tree *tree;
6316 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6317 return extent_readpages(tree, mapping, pages, nr_pages,
6318 btrfs_get_extent);
6319}
e6dcd2dc 6320static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6321{
d1310b2e
CM		 6322 struct extent_io_tree *tree;
6323 struct extent_map_tree *map;
a52d9a80 6324 int ret;
8c2383c3 6325
d1310b2e
CM		 6326 tree = &BTRFS_I(page->mapping->host)->io_tree;
6327 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6328 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6329 if (ret == 1) {
6330 ClearPagePrivate(page);
6331 set_page_private(page, 0);
6332 page_cache_release(page);
39279cc3 6333 }
a52d9a80 6334 return ret;
39279cc3
CM		 6335}
6336
e6dcd2dc
CM		 6337static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6338{
98509cfc
CM		 6339 if (PageWriteback(page) || PageDirty(page))
6340 return 0;
b335b003 6341 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6342}
6343
a52d9a80 6344static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6345{
d1310b2e 6346 struct extent_io_tree *tree;
e6dcd2dc 6347 struct btrfs_ordered_extent *ordered;
2ac55d41 6348 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6349 u64 page_start = page_offset(page);
6350 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6351
8b62b72b
CM		 6352
6353 /*
6354 * we have the page locked, so new writeback can't start,
6355 * and the dirty bit won't be cleared while we are here.
6356 *
6357 * Wait for IO on this page so that we can safely clear
6358 * the PagePrivate2 bit and do ordered accounting
6359 */
e6dcd2dc 6360 wait_on_page_writeback(page);
8b62b72b 6361
d1310b2e 6362 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 6363 if (offset) {
6364 btrfs_releasepage(page, GFP_NOFS);
6365 return;
6366 }
2ac55d41
JB		 6367 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6368 GFP_NOFS);
e6dcd2dc
CM		 6369 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6370 page_offset(page));
6371 if (ordered) {
eb84ae03
CM		 6372 /*
6373 * IO on this page will never be started, so we need
6374 * to account for any ordered extents now
6375 */
e6dcd2dc
CM		 6376 clear_extent_bit(tree, page_start, page_end,
6377 EXTENT_DIRTY | EXTENT_DELALLOC |
32c00aff 6378 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
2ac55d41 6379 &cached_state, GFP_NOFS);
8b62b72b
CM		 6380 /*
6381 * whoever cleared the private bit is responsible
6382 * for the finish_ordered_io
6383 */
6384 if (TestClearPagePrivate2(page)) {
6385 btrfs_finish_ordered_io(page->mapping->host,
6386 page_start, page_end);
6387 }
e6dcd2dc 6388 btrfs_put_ordered_extent(ordered);
2ac55d41
JB		 6389 cached_state = NULL;
6390 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6391 GFP_NOFS);
e6dcd2dc
CM		 6392 }
6393 clear_extent_bit(tree, page_start, page_end,
32c00aff 6394 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2ac55d41 6395 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6396 __btrfs_releasepage(page, GFP_NOFS);
6397
4a096752 6398 ClearPageChecked(page);
9ad6b7bc 6399 if (PagePrivate(page)) {
9ad6b7bc
CM		 6400 ClearPagePrivate(page);
6401 set_page_private(page, 0);
6402 page_cache_release(page);
6403 }
39279cc3
CM		 6404}
6405
9ebefb18
CM		 6406/*
6407 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6408 * called from a page fault handler when a page is first dirtied. Hence we must
6409 * be careful to check for EOF conditions here. We set the page up correctly
6410 * for a written page which means we get ENOSPC checking when writing into
6411 * holes and correct delalloc and unwritten extent mapping on filesystems that
6412 * support these features.
6413 *
6414 * We are not allowed to take the i_mutex here so we have to play games to
6415 * protect against truncate races as the page could now be beyond EOF. Because
6416 * vmtruncate() writes the inode size before removing pages, once we have the
6417 * page lock we can determine safely if the page is beyond EOF. If it is not
6418 * beyond EOF, then the page is guaranteed safe against truncation until we
6419 * unlock the page.
6420 */
c2ec175c 6421int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6422{
c2ec175c 6423 struct page *page = vmf->page;
6da6abae 6424 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6425 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6426 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6427 struct btrfs_ordered_extent *ordered;
2ac55d41 6428 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6429 char *kaddr;
6430 unsigned long zero_start;
9ebefb18 6431 loff_t size;
1832a6d5 6432 int ret;
9998eb70 6433 int reserved = 0;
a52d9a80 6434 u64 page_start;
e6dcd2dc 6435 u64 page_end;
9ebefb18 6436
0ca1f7ce 6437 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 6438 if (!ret) {
22c44fe6 6439 ret = btrfs_update_time(vma->vm_file);
9998eb70
CM		 6440 reserved = 1;
6441 }
56a76f82
NP		 6442 if (ret) {
6443 if (ret == -ENOMEM)
6444 ret = VM_FAULT_OOM;
6445 else /* -ENOSPC, -EIO, etc */
6446 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 6447 if (reserved)
6448 goto out;
6449 goto out_noreserve;
56a76f82 6450 }
1832a6d5 6451
56a76f82 6452 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6453again:
9ebefb18 6454 lock_page(page);
9ebefb18 6455 size = i_size_read(inode);
e6dcd2dc
CM		 6456 page_start = page_offset(page);
6457 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6458
9ebefb18 6459 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6460 (page_start >= size)) {
9ebefb18
CM		 6461 /* page got truncated out from underneath us */
6462 goto out_unlock;
6463 }
e6dcd2dc
CM		 6464 wait_on_page_writeback(page);
6465
2ac55d41
JB		 6466 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6467 GFP_NOFS);
e6dcd2dc
CM		 6468 set_page_extent_mapped(page);
6469
eb84ae03
CM		 6470 /*
6471 * we can't set the delalloc bits if there are pending ordered
6472 * extents. Drop our locks and wait for them to finish
6473 */
e6dcd2dc
CM		 6474 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6475 if (ordered) {
2ac55d41
JB		 6476 unlock_extent_cached(io_tree, page_start, page_end,
6477 &cached_state, GFP_NOFS);
e6dcd2dc 6478 unlock_page(page);
eb84ae03 6479 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6480 btrfs_put_ordered_extent(ordered);
6481 goto again;
6482 }
6483
fbf19087
JB		 6484 /*
6485 * XXX - page_mkwrite gets called every time the page is dirtied, even
6486 * if it was already dirty, so for space accounting reasons we need to
6487 * clear any delalloc bits for the range we are fixing to save. There
6488 * is probably a better way to do this, but for now keep consistent with
6489 * prepare_pages in the normal write path.
6490 */
2ac55d41 6491 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
32c00aff 6492 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
2ac55d41 6493 0, 0, &cached_state, GFP_NOFS);
fbf19087 6494
2ac55d41
JB		 6495 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6496 &cached_state);
9ed74f2d 6497 if (ret) {
2ac55d41
JB		 6498 unlock_extent_cached(io_tree, page_start, page_end,
6499 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6500 ret = VM_FAULT_SIGBUS;
6501 goto out_unlock;
6502 }
e6dcd2dc 6503 ret = 0;
9ebefb18
CM		 6504
6505 /* page is wholly or partially inside EOF */
a52d9a80 6506 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6507 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6508 else
e6dcd2dc 6509 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6510
e6dcd2dc
CM		 6511 if (zero_start != PAGE_CACHE_SIZE) {
6512 kaddr = kmap(page);
6513 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6514 flush_dcache_page(page);
6515 kunmap(page);
6516 }
247e743c 6517 ClearPageChecked(page);
e6dcd2dc 6518 set_page_dirty(page);
50a9b214 6519 SetPageUptodate(page);
5a3f23d5 6520
257c62e1
CM		 6521 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6522 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6523
2ac55d41 6524 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6525
6526out_unlock:
50a9b214
CM		 6527 if (!ret)
6528 return VM_FAULT_LOCKED;
9ebefb18 6529 unlock_page(page);
1832a6d5 6530out:
ec39e180 6531 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 6532out_noreserve:
9ebefb18
CM		 6533 return ret;
6534}
6535
a41ad394 6536static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6537{
6538 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6539 struct btrfs_block_rsv *rsv;
39279cc3 6540 int ret;
3893e33b 6541 int err = 0;
39279cc3 6542 struct btrfs_trans_handle *trans;
d3c2fdcf 6543 unsigned long nr;
dbe674a9 6544 u64 mask = root->sectorsize - 1;
07127184 6545 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6546
5d5e103a
JB		 6547 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6548 if (ret)
a41ad394 6549 return ret;
8082510e 6550
4a096752 6551 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6552 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6553
fcb80c2a
JB		 6554 /*
6555 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6556 * 3 things going on here
6557 *
6558 * 1) We need to reserve space for our orphan item and the space to
6559 * delete our orphan item. Lord knows we don't want to have a dangling
6560 * orphan item because we didn't reserve space to remove it.
6561 *
6562 * 2) We need to reserve space to update our inode.
6563 *
6564 * 3) We need to have something to cache all the space that is going to
6565 * be free'd up by the truncate operation, but also have some slack
6566 * space reserved in case it uses space during the truncate (thank you
6567 * very much snapshotting).
6568 *
6569 * And we need these to all be seperate. The fact is we can use alot of
6570 * space doing the truncate, and we have no earthly idea how much space
6571 * we will use, so we need the truncate reservation to be seperate so it
6572 * doesn't end up using space reserved for updating the inode or
6573 * removing the orphan item. We also need to be able to stop the
6574 * transaction and start a new one, which means we need to be able to
6575 * update the inode several times, and we have no idea of knowing how
6576 * many times that will be, so we can't just reserve 1 item for the
6577 * entirety of the opration, so that has to be done seperately as well.
6578 * Then there is the orphan item, which does indeed need to be held on
6579 * to for the whole operation, and we need nobody to touch this reserved
6580 * space except the orphan code.
6581 *
6582 * So that leaves us with
6583 *
6584 * 1) root->orphan_block_rsv - for the orphan deletion.
6585 * 2) rsv - for the truncate reservation, which we will steal from the
6586 * transaction reservation.
6587 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6588 * updating the inode.
6589 */
6590 rsv = btrfs_alloc_block_rsv(root);
6591 if (!rsv)
6592 return -ENOMEM;
4a338542 6593 rsv->size = min_size;
f0cd846e 6594
907cbceb 6595 /*
07127184 6596 * 1 for the truncate slack space
907cbceb
JB		 6597 * 1 for the orphan item we're going to add
6598 * 1 for the orphan item deletion
6599 * 1 for updating the inode.
6600 */
fcb80c2a
JB		 6601 trans = btrfs_start_transaction(root, 4);
6602 if (IS_ERR(trans)) {
6603 err = PTR_ERR(trans);
6604 goto out;
6605 }
f0cd846e 6606
907cbceb
JB		 6607 /* Migrate the slack space for the truncate to our reserve */
6608 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6609 min_size);
fcb80c2a 6610 BUG_ON(ret);
f0cd846e
JB		 6611
6612 ret = btrfs_orphan_add(trans, inode);
6613 if (ret) {
6614 btrfs_end_transaction(trans, root);
fcb80c2a 6615 goto out;
f0cd846e
JB		 6616 }
6617
5a3f23d5
CM		 6618 /*
6619 * setattr is responsible for setting the ordered_data_close flag,
6620 * but that is only tested during the last file release. That
6621 * could happen well after the next commit, leaving a great big
6622 * window where new writes may get lost if someone chooses to write
6623 * to this file after truncating to zero
6624 *
6625 * The inode doesn't have any dirty data here, and so if we commit
6626 * this is a noop. If someone immediately starts writing to the inode
6627 * it is very likely we'll catch some of their writes in this
6628 * transaction, and the commit will find this file on the ordered
6629 * data list with good things to send down.
6630 *
6631 * This is a best effort solution, there is still a window where
6632 * using truncate to replace the contents of the file will
6633 * end up with a zero length file after a crash.
6634 */
6635 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6636 btrfs_add_ordered_operation(trans, root, inode);
6637
8082510e 6638 while (1) {
36ba022a 6639 ret = btrfs_block_rsv_refill(root, rsv, min_size);
907cbceb
JB		 6640 if (ret) {
6641 /*
6642 * This can only happen with the original transaction we
6643 * started above, every other time we shouldn't have a
6644 * transaction started yet.
6645 */
6646 if (ret == -EAGAIN)
6647 goto end_trans;
6648 err = ret;
6649 break;
6650 }
6651
d68fc57b 6652 if (!trans) {
907cbceb
JB		 6653 /* Just need the 1 for updating the inode */
6654 trans = btrfs_start_transaction(root, 1);
fcb80c2a 6655 if (IS_ERR(trans)) {
7041ee97
JB		 6656 ret = err = PTR_ERR(trans);
6657 trans = NULL;
6658 break;
fcb80c2a 6659 }
d68fc57b
YZ		 6660 }
6661
907cbceb
JB		 6662 trans->block_rsv = rsv;
6663
8082510e
YZ		 6664 ret = btrfs_truncate_inode_items(trans, root, inode,
6665 inode->i_size,
6666 BTRFS_EXTENT_DATA_KEY);
3893e33b
JB		 6667 if (ret != -EAGAIN) {
6668 err = ret;
8082510e 6669 break;
3893e33b 6670 }
39279cc3 6671
fcb80c2a 6672 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6673 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6674 if (ret) {
6675 err = ret;
6676 break;
6677 }
907cbceb 6678end_trans:
8082510e
YZ		 6679 nr = trans->blocks_used;
6680 btrfs_end_transaction(trans, root);
d68fc57b 6681 trans = NULL;
8082510e 6682 btrfs_btree_balance_dirty(root, nr);
8082510e
YZ		 6683 }
6684
6685 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 6686 trans->block_rsv = root->orphan_block_rsv;
8082510e 6687 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 6688 if (ret)
6689 err = ret;
ded5db9d
JB		 6690 } else if (ret && inode->i_nlink > 0) {
6691 /*
6692 * Failed to do the truncate, remove us from the in memory
6693 * orphan list.
6694 */
6695 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 6696 }
6697
917c16b2
CM		 6698 if (trans) {
6699 trans->block_rsv = &root->fs_info->trans_block_rsv;
6700 ret = btrfs_update_inode(trans, root, inode);
6701 if (ret && !err)
6702 err = ret;
7b128766 6703
917c16b2 6704 nr = trans->blocks_used;
7ad85bb7 6705 ret = btrfs_end_transaction(trans, root);
917c16b2
CM		 6706 btrfs_btree_balance_dirty(root, nr);
6707 }
fcb80c2a
JB		 6708
6709out:
6710 btrfs_free_block_rsv(root, rsv);
6711
3893e33b
JB		 6712 if (ret && !err)
6713 err = ret;
a41ad394 6714
3893e33b 6715 return err;
39279cc3
CM		 6716}
6717
d352ac68
CM		 6718/*
6719 * create a new subvolume directory/inode (helper for the ioctl).
6720 */
d2fb3437 6721int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 6722 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 6723{
39279cc3 6724 struct inode *inode;
76dda93c 6725 int err;
00e4e6b3 6726 u64 index = 0;
39279cc3 6727
12fc9d09
FA		 6728 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6729 new_dirid, new_dirid,
6730 S_IFDIR | (~current_umask() & S_IRWXUGO),
6731 &index);
54aa1f4d 6732 if (IS_ERR(inode))
f46b5a66 6733 return PTR_ERR(inode);
39279cc3
CM		 6734 inode->i_op = &btrfs_dir_inode_operations;
6735 inode->i_fop = &btrfs_dir_file_operations;
6736
bfe86848 6737 set_nlink(inode, 1);
dbe674a9 6738 btrfs_i_size_write(inode, 0);
3b96362c 6739
76dda93c
YZ		 6740 err = btrfs_update_inode(trans, new_root, inode);
6741 BUG_ON(err);
cb8e7090 6742
76dda93c 6743 iput(inode);
cb8e7090 6744 return 0;
39279cc3
CM		 6745}
6746
39279cc3
CM		 6747struct inode *btrfs_alloc_inode(struct super_block *sb)
6748{
6749 struct btrfs_inode *ei;
2ead6ae7 6750 struct inode *inode;
39279cc3
CM		 6751
6752 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6753 if (!ei)
6754 return NULL;
2ead6ae7
YZ		 6755
6756 ei->root = NULL;
6757 ei->space_info = NULL;
6758 ei->generation = 0;
6759 ei->sequence = 0;
15ee9bc7 6760 ei->last_trans = 0;
257c62e1 6761 ei->last_sub_trans = 0;
e02119d5 6762 ei->logged_trans = 0;
2ead6ae7 6763 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 6764 ei->disk_i_size = 0;
6765 ei->flags = 0;
7709cde3 6766 ei->csum_bytes = 0;
2ead6ae7
YZ		 6767 ei->index_cnt = (u64)-1;
6768 ei->last_unlink_trans = 0;
6769
9e0baf60
JB		 6770 spin_lock_init(&ei->lock);
6771 ei->outstanding_extents = 0;
6772 ei->reserved_extents = 0;
2ead6ae7
YZ		 6773
6774 ei->ordered_data_close = 0;
d68fc57b 6775 ei->orphan_meta_reserved = 0;
2ead6ae7 6776 ei->dummy_inode = 0;
4cb5300b 6777 ei->in_defrag = 0;
7fd2ae21 6778 ei->delalloc_meta_reserved = 0;
261507a0 6779 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 6780
16cdcec7
MX		 6781 ei->delayed_node = NULL;
6782
2ead6ae7 6783 inode = &ei->vfs_inode;
a8067e02 6784 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 6785 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6786 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
2ead6ae7 6787 mutex_init(&ei->log_mutex);
f248679e 6788 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 6789 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7b128766 6790 INIT_LIST_HEAD(&ei->i_orphan);
2ead6ae7 6791 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 6792 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 6793 RB_CLEAR_NODE(&ei->rb_node);
6794
6795 return inode;
39279cc3
CM		 6796}
6797
fa0d7e3d
NP		 6798static void btrfs_i_callback(struct rcu_head *head)
6799{
6800 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 6801 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6802}
6803
39279cc3
CM		 6804void btrfs_destroy_inode(struct inode *inode)
6805{
e6dcd2dc 6806 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 6807 struct btrfs_root *root = BTRFS_I(inode)->root;
6808
39279cc3
CM		 6809 WARN_ON(!list_empty(&inode->i_dentry));
6810 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 6811 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6812 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 6813 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6814 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 6815
a6dbd429
JB		 6816 /*
6817 * This can happen where we create an inode, but somebody else also
6818 * created the same inode and we need to destroy the one we already
6819 * created.
6820 */
6821 if (!root)
6822 goto free;
6823
5a3f23d5
CM		 6824 /*
6825 * Make sure we're properly removed from the ordered operation
6826 * lists.
6827 */
6828 smp_mb();
6829 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6830 spin_lock(&root->fs_info->ordered_extent_lock);
6831 list_del_init(&BTRFS_I(inode)->ordered_operations);
6832 spin_unlock(&root->fs_info->ordered_extent_lock);
6833 }
6834
d68fc57b 6835 spin_lock(&root->orphan_lock);
7b128766 6836 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
33345d01
LZ		 6837 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6838 (unsigned long long)btrfs_ino(inode));
8082510e 6839 list_del_init(&BTRFS_I(inode)->i_orphan);
7b128766 6840 }
d68fc57b 6841 spin_unlock(&root->orphan_lock);
7b128766 6842
d397712b 6843 while (1) {
e6dcd2dc
CM		 6844 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6845 if (!ordered)
6846 break;
6847 else {
d397712b
CM		 6848 printk(KERN_ERR "btrfs found ordered "
6849 "extent %llu %llu on inode cleanup\n",
6850 (unsigned long long)ordered->file_offset,
6851 (unsigned long long)ordered->len);
e6dcd2dc
CM		 6852 btrfs_remove_ordered_extent(inode, ordered);
6853 btrfs_put_ordered_extent(ordered);
6854 btrfs_put_ordered_extent(ordered);
6855 }
6856 }
5d4f98a2 6857 inode_tree_del(inode);
5b21f2ed 6858 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 6859free:
16cdcec7 6860 btrfs_remove_delayed_node(inode);
fa0d7e3d 6861 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 6862}
6863
45321ac5 6864int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 6865{
6866 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 6867
0af3d00b 6868 if (btrfs_root_refs(&root->root_item) == 0 &&
2cf8572d 6869 !btrfs_is_free_space_inode(root, inode))
45321ac5 6870 return 1;
76dda93c 6871 else
45321ac5 6872 return generic_drop_inode(inode);
76dda93c
YZ		 6873}
6874
0ee0fda0 6875static void init_once(void *foo)
39279cc3
CM		 6876{
6877 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6878
6879 inode_init_once(&ei->vfs_inode);
6880}
6881
6882void btrfs_destroy_cachep(void)
6883{
6884 if (btrfs_inode_cachep)
6885 kmem_cache_destroy(btrfs_inode_cachep);
6886 if (btrfs_trans_handle_cachep)
6887 kmem_cache_destroy(btrfs_trans_handle_cachep);
6888 if (btrfs_transaction_cachep)
6889 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 6890 if (btrfs_path_cachep)
6891 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 6892 if (btrfs_free_space_cachep)
6893 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 6894}
6895
6896int btrfs_init_cachep(void)
6897{
9601e3f6
CH		 6898 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6899 sizeof(struct btrfs_inode), 0,
6900 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 6901 if (!btrfs_inode_cachep)
6902 goto fail;
9601e3f6
CH		 6903
6904 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6905 sizeof(struct btrfs_trans_handle), 0,
6906 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6907 if (!btrfs_trans_handle_cachep)
6908 goto fail;
9601e3f6
CH		 6909
6910 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6911 sizeof(struct btrfs_transaction), 0,
6912 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6913 if (!btrfs_transaction_cachep)
6914 goto fail;
9601e3f6
CH		 6915
6916 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6917 sizeof(struct btrfs_path), 0,
6918 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 6919 if (!btrfs_path_cachep)
6920 goto fail;
9601e3f6 6921
dc89e982
JB		 6922 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6923 sizeof(struct btrfs_free_space), 0,
6924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6925 if (!btrfs_free_space_cachep)
6926 goto fail;
6927
39279cc3
CM		 6928 return 0;
6929fail:
6930 btrfs_destroy_cachep();
6931 return -ENOMEM;
6932}
6933
6934static int btrfs_getattr(struct vfsmount *mnt,
6935 struct dentry *dentry, struct kstat *stat)
6936{
6937 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 6938 u32 blocksize = inode->i_sb->s_blocksize;
6939
39279cc3 6940 generic_fillattr(inode, stat);
0ee5dc67 6941 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 6942 stat->blksize = PAGE_CACHE_SIZE;
fadc0d8b
DS		 6943 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6944 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 6945 return 0;
6946}
6947
75e7cb7f
LB		 6948/*
6949 * If a file is moved, it will inherit the cow and compression flags of the new
6950 * directory.
6951 */
6952static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6953{
6954 struct btrfs_inode *b_dir = BTRFS_I(dir);
6955 struct btrfs_inode *b_inode = BTRFS_I(inode);
6956
6957 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6958 b_inode->flags |= BTRFS_INODE_NODATACOW;
6959 else
6960 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6961
6962 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6963 b_inode->flags |= BTRFS_INODE_COMPRESS;
6964 else
6965 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6966}
6967
d397712b
CM		 6968static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6969 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 6970{
6971 struct btrfs_trans_handle *trans;
6972 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 6973 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 6974 struct inode *new_inode = new_dentry->d_inode;
6975 struct inode *old_inode = old_dentry->d_inode;
6976 struct timespec ctime = CURRENT_TIME;
00e4e6b3 6977 u64 index = 0;
4df27c4d 6978 u64 root_objectid;
39279cc3 6979 int ret;
33345d01 6980 u64 old_ino = btrfs_ino(old_inode);
39279cc3 6981
33345d01 6982 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 6983 return -EPERM;
6984
4df27c4d 6985 /* we only allow rename subvolume link between subvolumes */
33345d01 6986 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 6987 return -EXDEV;
6988
33345d01
LZ		 6989 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6990 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 6991 return -ENOTEMPTY;
5f39d397 6992
4df27c4d
YZ		 6993 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6994 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6995 return -ENOTEMPTY;
5a3f23d5
CM		 6996 /*
6997 * we're using rename to replace one file with another.
6998 * and the replacement file is large. Start IO on it now so
6999 * we don't add too much work to the end of the transaction
7000 */
4baf8c92 7001 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 7002 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7003 filemap_flush(old_inode->i_mapping);
7004
76dda93c 7005 /* close the racy window with snapshot create/destroy ioctl */
33345d01 7006 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7007 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 7008 /*
7009 * We want to reserve the absolute worst case amount of items. So if
7010 * both inodes are subvols and we need to unlink them then that would
7011 * require 4 item modifications, but if they are both normal inodes it
7012 * would require 5 item modifications, so we'll assume their normal
7013 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7014 * should cover the worst case number of items we'll modify.
7015 */
7016 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 7017 if (IS_ERR(trans)) {
7018 ret = PTR_ERR(trans);
7019 goto out_notrans;
7020 }
76dda93c 7021
4df27c4d
YZ		 7022 if (dest != root)
7023 btrfs_record_root_in_trans(trans, dest);
5f39d397 7024
a5719521
YZ		 7025 ret = btrfs_set_inode_index(new_dir, &index);
7026 if (ret)
7027 goto out_fail;
5a3f23d5 7028
33345d01 7029 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7030 /* force full log commit if subvolume involved. */
7031 root->fs_info->last_trans_log_full_commit = trans->transid;
7032 } else {
a5719521
YZ		 7033 ret = btrfs_insert_inode_ref(trans, dest,
7034 new_dentry->d_name.name,
7035 new_dentry->d_name.len,
33345d01
LZ		 7036 old_ino,
7037 btrfs_ino(new_dir), index);
a5719521
YZ		 7038 if (ret)
7039 goto out_fail;
4df27c4d
YZ		 7040 /*
7041 * this is an ugly little race, but the rename is required
7042 * to make sure that if we crash, the inode is either at the
7043 * old name or the new one. pinning the log transaction lets
7044 * us make sure we don't allow a log commit to come in after
7045 * we unlink the name but before we add the new name back in.
7046 */
7047 btrfs_pin_log_trans(root);
7048 }
5a3f23d5
CM		 7049 /*
7050 * make sure the inode gets flushed if it is replacing
7051 * something.
7052 */
33345d01 7053 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7054 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7055
39279cc3
CM		 7056 old_dir->i_ctime = old_dir->i_mtime = ctime;
7057 new_dir->i_ctime = new_dir->i_mtime = ctime;
7058 old_inode->i_ctime = ctime;
5f39d397 7059
12fcfd22
CM		 7060 if (old_dentry->d_parent != new_dentry->d_parent)
7061 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7062
33345d01 7063 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7064 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7065 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7066 old_dentry->d_name.name,
7067 old_dentry->d_name.len);
7068 } else {
92986796
AV		 7069 ret = __btrfs_unlink_inode(trans, root, old_dir,
7070 old_dentry->d_inode,
7071 old_dentry->d_name.name,
7072 old_dentry->d_name.len);
7073 if (!ret)
7074 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d
YZ		 7075 }
7076 BUG_ON(ret);
39279cc3
CM		 7077
7078 if (new_inode) {
7079 new_inode->i_ctime = CURRENT_TIME;
33345d01 7080 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 7081 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7082 root_objectid = BTRFS_I(new_inode)->location.objectid;
7083 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7084 root_objectid,
7085 new_dentry->d_name.name,
7086 new_dentry->d_name.len);
7087 BUG_ON(new_inode->i_nlink == 0);
7088 } else {
7089 ret = btrfs_unlink_inode(trans, dest, new_dir,
7090 new_dentry->d_inode,
7091 new_dentry->d_name.name,
7092 new_dentry->d_name.len);
7093 }
7094 BUG_ON(ret);
7b128766 7095 if (new_inode->i_nlink == 0) {
e02119d5 7096 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7097 BUG_ON(ret);
7b128766 7098 }
39279cc3 7099 }
aec7477b 7100
75e7cb7f
LB		 7101 fixup_inode_flags(new_dir, old_inode);
7102
4df27c4d
YZ		 7103 ret = btrfs_add_link(trans, new_dir, old_inode,
7104 new_dentry->d_name.name,
a5719521 7105 new_dentry->d_name.len, 0, index);
4df27c4d 7106 BUG_ON(ret);
39279cc3 7107
33345d01 7108 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7109 struct dentry *parent = new_dentry->d_parent;
6a912213 7110 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7111 btrfs_end_log_trans(root);
7112 }
39279cc3 7113out_fail:
7ad85bb7 7114 btrfs_end_transaction(trans, root);
b44c59a8 7115out_notrans:
33345d01 7116 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7117 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7118
39279cc3
CM		 7119 return ret;
7120}
7121
d352ac68
CM		 7122/*
7123 * some fairly slow code that needs optimization. This walks the list
7124 * of all the inodes with pending delalloc and forces them to disk.
7125 */
24bbcf04 7126int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7127{
7128 struct list_head *head = &root->fs_info->delalloc_inodes;
7129 struct btrfs_inode *binode;
5b21f2ed 7130 struct inode *inode;
ea8c2819 7131
c146afad
YZ		 7132 if (root->fs_info->sb->s_flags & MS_RDONLY)
7133 return -EROFS;
7134
75eff68e 7135 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7136 while (!list_empty(head)) {
ea8c2819
CM		 7137 binode = list_entry(head->next, struct btrfs_inode,
7138 delalloc_inodes);
5b21f2ed
ZY		 7139 inode = igrab(&binode->vfs_inode);
7140 if (!inode)
7141 list_del_init(&binode->delalloc_inodes);
75eff68e 7142 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7143 if (inode) {
8c8bee1d 7144 filemap_flush(inode->i_mapping);
24bbcf04
YZ		 7145 if (delay_iput)
7146 btrfs_add_delayed_iput(inode);
7147 else
7148 iput(inode);
5b21f2ed
ZY		 7149 }
7150 cond_resched();
75eff68e 7151 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7152 }
75eff68e 7153 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7154
7155 /* the filemap_flush will queue IO into the worker threads, but
7156 * we have to make sure the IO is actually started and that
7157 * ordered extents get created before we return
7158 */
7159 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7160 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7161 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7162 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7163 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7164 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7165 }
7166 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 7167 return 0;
7168}
7169
39279cc3
CM		 7170static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7171 const char *symname)
7172{
7173 struct btrfs_trans_handle *trans;
7174 struct btrfs_root *root = BTRFS_I(dir)->root;
7175 struct btrfs_path *path;
7176 struct btrfs_key key;
1832a6d5 7177 struct inode *inode = NULL;
39279cc3
CM		 7178 int err;
7179 int drop_inode = 0;
7180 u64 objectid;
00e4e6b3 7181 u64 index = 0 ;
39279cc3
CM		 7182 int name_len;
7183 int datasize;
5f39d397 7184 unsigned long ptr;
39279cc3 7185 struct btrfs_file_extent_item *ei;
5f39d397 7186 struct extent_buffer *leaf;
1832a6d5 7187 unsigned long nr = 0;
39279cc3
CM		 7188
7189 name_len = strlen(symname) + 1;
7190 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7191 return -ENAMETOOLONG;
1832a6d5 7192
9ed74f2d
JB		 7193 /*
7194 * 2 items for inode item and ref
7195 * 2 items for dir items
7196 * 1 item for xattr if selinux is on
7197 */
a22285a6
YZ		 7198 trans = btrfs_start_transaction(root, 5);
7199 if (IS_ERR(trans))
7200 return PTR_ERR(trans);
1832a6d5 7201
581bb050
LZ		 7202 err = btrfs_find_free_ino(root, &objectid);
7203 if (err)
7204 goto out_unlock;
7205
aec7477b 7206 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7207 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7208 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7209 if (IS_ERR(inode)) {
7210 err = PTR_ERR(inode);
39279cc3 7211 goto out_unlock;
7cf96da3 7212 }
39279cc3 7213
2a7dba39 7214 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7215 if (err) {
7216 drop_inode = 1;
7217 goto out_unlock;
7218 }
7219
ad19db71
CS		 7220 /*
7221 * If the active LSM wants to access the inode during
7222 * d_instantiate it needs these. Smack checks to see
7223 * if the filesystem supports xattrs by looking at the
7224 * ops vector.
7225 */
7226 inode->i_fop = &btrfs_file_operations;
7227 inode->i_op = &btrfs_file_inode_operations;
7228
a1b075d2 7229 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7230 if (err)
7231 drop_inode = 1;
7232 else {
7233 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7234 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 7235 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7236 }
39279cc3
CM		 7237 if (drop_inode)
7238 goto out_unlock;
7239
7240 path = btrfs_alloc_path();
d8926bb3
MF		 7241 if (!path) {
7242 err = -ENOMEM;
7243 drop_inode = 1;
7244 goto out_unlock;
7245 }
33345d01 7246 key.objectid = btrfs_ino(inode);
39279cc3 7247 key.offset = 0;
39279cc3
CM		 7248 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7249 datasize = btrfs_file_extent_calc_inline_size(name_len);
7250 err = btrfs_insert_empty_item(trans, root, path, &key,
7251 datasize);
54aa1f4d
CM		 7252 if (err) {
7253 drop_inode = 1;
b0839166 7254 btrfs_free_path(path);
54aa1f4d
CM		 7255 goto out_unlock;
7256 }
5f39d397
CM		 7257 leaf = path->nodes[0];
7258 ei = btrfs_item_ptr(leaf, path->slots[0],
7259 struct btrfs_file_extent_item);
7260 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7261 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7262 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7263 btrfs_set_file_extent_encryption(leaf, ei, 0);
7264 btrfs_set_file_extent_compression(leaf, ei, 0);
7265 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7266 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7267
39279cc3 7268 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7269 write_extent_buffer(leaf, symname, ptr, name_len);
7270 btrfs_mark_buffer_dirty(leaf);
39279cc3 7271 btrfs_free_path(path);
5f39d397 7272
39279cc3
CM		 7273 inode->i_op = &btrfs_symlink_inode_operations;
7274 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7275 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7276 inode_set_bytes(inode, name_len);
dbe674a9 7277 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7278 err = btrfs_update_inode(trans, root, inode);
7279 if (err)
7280 drop_inode = 1;
39279cc3
CM		 7281
7282out_unlock:
08c422c2
AV		 7283 if (!err)
7284 d_instantiate(dentry, inode);
d3c2fdcf 7285 nr = trans->blocks_used;
7ad85bb7 7286 btrfs_end_transaction(trans, root);
39279cc3
CM		 7287 if (drop_inode) {
7288 inode_dec_link_count(inode);
7289 iput(inode);
7290 }
d3c2fdcf 7291 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7292 return err;
7293}
16432985 7294
0af3d00b
JB		 7295static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7296 u64 start, u64 num_bytes, u64 min_size,
7297 loff_t actual_len, u64 *alloc_hint,
7298 struct btrfs_trans_handle *trans)
d899e052 7299{
d899e052
YZ		 7300 struct btrfs_root *root = BTRFS_I(inode)->root;
7301 struct btrfs_key ins;
d899e052 7302 u64 cur_offset = start;
55a61d1d 7303 u64 i_size;
d899e052 7304 int ret = 0;
0af3d00b 7305 bool own_trans = true;
d899e052 7306
0af3d00b
JB		 7307 if (trans)
7308 own_trans = false;
d899e052 7309 while (num_bytes > 0) {
0af3d00b
JB		 7310 if (own_trans) {
7311 trans = btrfs_start_transaction(root, 3);
7312 if (IS_ERR(trans)) {
7313 ret = PTR_ERR(trans);
7314 break;
7315 }
5a303d5d
YZ		 7316 }
7317
efa56464
YZ		 7318 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7319 0, *alloc_hint, (u64)-1, &ins, 1);
5a303d5d 7320 if (ret) {
0af3d00b
JB		 7321 if (own_trans)
7322 btrfs_end_transaction(trans, root);
a22285a6 7323 break;
d899e052 7324 }
5a303d5d 7325
d899e052
YZ		 7326 ret = insert_reserved_file_extent(trans, inode,
7327 cur_offset, ins.objectid,
7328 ins.offset, ins.offset,
920bbbfb 7329 ins.offset, 0, 0, 0,
d899e052
YZ		 7330 BTRFS_FILE_EXTENT_PREALLOC);
7331 BUG_ON(ret);
a1ed835e
CM		 7332 btrfs_drop_extent_cache(inode, cur_offset,
7333 cur_offset + ins.offset -1, 0);
5a303d5d 7334
d899e052
YZ		 7335 num_bytes -= ins.offset;
7336 cur_offset += ins.offset;
efa56464 7337 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7338
d899e052 7339 inode->i_ctime = CURRENT_TIME;
6cbff00f 7340 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7341 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7342 (actual_len > inode->i_size) &&
7343 (cur_offset > inode->i_size)) {
d1ea6a61 7344 if (cur_offset > actual_len)
55a61d1d 7345 i_size = actual_len;
d1ea6a61 7346 else
55a61d1d
JB		 7347 i_size = cur_offset;
7348 i_size_write(inode, i_size);
7349 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7350 }
7351
d899e052
YZ		 7352 ret = btrfs_update_inode(trans, root, inode);
7353 BUG_ON(ret);
d899e052 7354
0af3d00b
JB		 7355 if (own_trans)
7356 btrfs_end_transaction(trans, root);
5a303d5d 7357 }
d899e052
YZ		 7358 return ret;
7359}
7360
0af3d00b
JB		 7361int btrfs_prealloc_file_range(struct inode *inode, int mode,
7362 u64 start, u64 num_bytes, u64 min_size,
7363 loff_t actual_len, u64 *alloc_hint)
7364{
7365 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7366 min_size, actual_len, alloc_hint,
7367 NULL);
7368}
7369
7370int btrfs_prealloc_file_range_trans(struct inode *inode,
7371 struct btrfs_trans_handle *trans, int mode,
7372 u64 start, u64 num_bytes, u64 min_size,
7373 loff_t actual_len, u64 *alloc_hint)
7374{
7375 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7376 min_size, actual_len, alloc_hint, trans);
7377}
7378
e6dcd2dc
CM		 7379static int btrfs_set_page_dirty(struct page *page)
7380{
e6dcd2dc
CM		 7381 return __set_page_dirty_nobuffers(page);
7382}
7383
10556cb2 7384static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7385{
b83cc969 7386 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7387 umode_t mode = inode->i_mode;
b83cc969 7388
cb6db4e5
JM		 7389 if (mask & MAY_WRITE &&
7390 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7391 if (btrfs_root_readonly(root))
7392 return -EROFS;
7393 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7394 return -EACCES;
7395 }
2830ba7f 7396 return generic_permission(inode, mask);
fdebe2bd 7397}
39279cc3 7398
6e1d5dcc 7399static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7400 .getattr = btrfs_getattr,
39279cc3
CM		 7401 .lookup = btrfs_lookup,
7402 .create = btrfs_create,
7403 .unlink = btrfs_unlink,
7404 .link = btrfs_link,
7405 .mkdir = btrfs_mkdir,
7406 .rmdir = btrfs_rmdir,
7407 .rename = btrfs_rename,
7408 .symlink = btrfs_symlink,
7409 .setattr = btrfs_setattr,
618e21d5 7410 .mknod = btrfs_mknod,
95819c05
CH		 7411 .setxattr = btrfs_setxattr,
7412 .getxattr = btrfs_getxattr,
5103e947 7413 .listxattr = btrfs_listxattr,
95819c05 7414 .removexattr = btrfs_removexattr,
fdebe2bd 7415 .permission = btrfs_permission,
4e34e719 7416 .get_acl = btrfs_get_acl,
39279cc3 7417};
6e1d5dcc 7418static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7419 .lookup = btrfs_lookup,
fdebe2bd 7420 .permission = btrfs_permission,
4e34e719 7421 .get_acl = btrfs_get_acl,
39279cc3 7422};
76dda93c 7423
828c0950 7424static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7425 .llseek = generic_file_llseek,
7426 .read = generic_read_dir,
cbdf5a24 7427 .readdir = btrfs_real_readdir,
34287aa3 7428 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7429#ifdef CONFIG_COMPAT
34287aa3 7430 .compat_ioctl = btrfs_ioctl,
39279cc3 7431#endif
6bf13c0c 7432 .release = btrfs_release_file,
e02119d5 7433 .fsync = btrfs_sync_file,
39279cc3
CM		 7434};
7435
d1310b2e 7436static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7437 .fill_delalloc = run_delalloc_range,
065631f6 7438 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7439 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7440 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7441 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7442 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 7443 .set_bit_hook = btrfs_set_bit_hook,
7444 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7445 .merge_extent_hook = btrfs_merge_extent_hook,
7446 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7447};
7448
35054394
CM		 7449/*
7450 * btrfs doesn't support the bmap operation because swapfiles
7451 * use bmap to make a mapping of extents in the file. They assume
7452 * these extents won't change over the life of the file and they
7453 * use the bmap result to do IO directly to the drive.
7454 *
7455 * the btrfs bmap call would return logical addresses that aren't
7456 * suitable for IO and they also will change frequently as COW
7457 * operations happen. So, swapfile + btrfs == corruption.
7458 *
7459 * For now we're avoiding this by dropping bmap.
7460 */
7f09410b 7461static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7462 .readpage = btrfs_readpage,
7463 .writepage = btrfs_writepage,
b293f02e 7464 .writepages = btrfs_writepages,
3ab2fb5a 7465 .readpages = btrfs_readpages,
16432985 7466 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7467 .invalidatepage = btrfs_invalidatepage,
7468 .releasepage = btrfs_releasepage,
e6dcd2dc 7469 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7470 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7471};
7472
7f09410b 7473static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7474 .readpage = btrfs_readpage,
7475 .writepage = btrfs_writepage,
2bf5a725
CM		 7476 .invalidatepage = btrfs_invalidatepage,
7477 .releasepage = btrfs_releasepage,
39279cc3
CM		 7478};
7479
6e1d5dcc 7480static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7481 .getattr = btrfs_getattr,
7482 .setattr = btrfs_setattr,
95819c05
CH		 7483 .setxattr = btrfs_setxattr,
7484 .getxattr = btrfs_getxattr,
5103e947 7485 .listxattr = btrfs_listxattr,
95819c05 7486 .removexattr = btrfs_removexattr,
fdebe2bd 7487 .permission = btrfs_permission,
1506fcc8 7488 .fiemap = btrfs_fiemap,
4e34e719 7489 .get_acl = btrfs_get_acl,
39279cc3 7490};
6e1d5dcc 7491static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7492 .getattr = btrfs_getattr,
7493 .setattr = btrfs_setattr,
fdebe2bd 7494 .permission = btrfs_permission,
95819c05
CH		 7495 .setxattr = btrfs_setxattr,
7496 .getxattr = btrfs_getxattr,
33268eaf 7497 .listxattr = btrfs_listxattr,
95819c05 7498 .removexattr = btrfs_removexattr,
4e34e719 7499 .get_acl = btrfs_get_acl,
618e21d5 7500};
6e1d5dcc 7501static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7502 .readlink = generic_readlink,
7503 .follow_link = page_follow_link_light,
7504 .put_link = page_put_link,
f209561a 7505 .getattr = btrfs_getattr,
22c44fe6 7506 .setattr = btrfs_setattr,
fdebe2bd 7507 .permission = btrfs_permission,
0279b4cd
JO		 7508 .setxattr = btrfs_setxattr,
7509 .getxattr = btrfs_getxattr,
7510 .listxattr = btrfs_listxattr,
7511 .removexattr = btrfs_removexattr,
4e34e719 7512 .get_acl = btrfs_get_acl,
39279cc3 7513};
76dda93c 7514
82d339d9 7515const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7516 .d_delete = btrfs_dentry_delete,
b4aff1f8 7517 .d_release = btrfs_dentry_release,
76dda93c 7518};
Linux 6.x block layer and io_uring tree(s)