]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/btrfs/inode.c
Btrfs: unreserve space if our ordered extent fails to work
[mirror_ubuntu-bionic-kernel.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM
23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM
29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
4b4e25f2 43#include "compat.h"
39279cc3
CM
44#include "ctree.h"
45#include "disk-io.h"
46#include "transaction.h"
47#include "btrfs_inode.h"
39279cc3 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;
8ccf6f19 74static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM
75struct kmem_cache *btrfs_trans_handle_cachep;
76struct kmem_cache *btrfs_transaction_cachep;
39279cc3 77struct kmem_cache *btrfs_path_cachep;
dc89e982 78struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM
79
80#define S_SHIFT 12
81static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89};
90
3972f260 91static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 92static int btrfs_truncate(struct inode *inode);
5fd02043 93static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM
94static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
70c8a91c
JB
98static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
99 u64 len, u64 orig_start,
100 u64 block_start, u64 block_len,
101 u64 orig_block_len, int type);
7b128766 102
f34f57a3 103static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP
104 struct inode *inode, struct inode *dir,
105 const struct qstr *qstr)
0279b4cd
JO
106{
107 int err;
108
f34f57a3 109 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 110 if (!err)
2a7dba39 111 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO
112 return err;
113}
114
c8b97818
CM
115/*
116 * this does all the hard work for inserting an inline extent into
117 * the btree. The caller should have done a btrfs_drop_extents so that
118 * no overlapping inline items exist in the btree
119 */
d397712b 120static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM
121 struct btrfs_root *root, struct inode *inode,
122 u64 start, size_t size, size_t compressed_size,
fe3f566c 123 int compress_type,
c8b97818
CM
124 struct page **compressed_pages)
125{
126 struct btrfs_key key;
127 struct btrfs_path *path;
128 struct extent_buffer *leaf;
129 struct page *page = NULL;
130 char *kaddr;
131 unsigned long ptr;
132 struct btrfs_file_extent_item *ei;
133 int err = 0;
134 int ret;
135 size_t cur_size = size;
136 size_t datasize;
137 unsigned long offset;
c8b97818 138
fe3f566c 139 if (compressed_size && compressed_pages)
c8b97818 140 cur_size = compressed_size;
c8b97818 141
d397712b
CM
142 path = btrfs_alloc_path();
143 if (!path)
c8b97818
CM
144 return -ENOMEM;
145
b9473439 146 path->leave_spinning = 1;
c8b97818 147
33345d01 148 key.objectid = btrfs_ino(inode);
c8b97818
CM
149 key.offset = start;
150 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM
151 datasize = btrfs_file_extent_calc_inline_size(cur_size);
152
153 inode_add_bytes(inode, size);
154 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 datasize);
c8b97818
CM
156 if (ret) {
157 err = ret;
c8b97818
CM
158 goto fail;
159 }
160 leaf = path->nodes[0];
161 ei = btrfs_item_ptr(leaf, path->slots[0],
162 struct btrfs_file_extent_item);
163 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
164 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
165 btrfs_set_file_extent_encryption(leaf, ei, 0);
166 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
167 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
168 ptr = btrfs_file_extent_inline_start(ei);
169
261507a0 170 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM
171 struct page *cpage;
172 int i = 0;
d397712b 173 while (compressed_size > 0) {
c8b97818 174 cpage = compressed_pages[i];
5b050f04 175 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM
176 PAGE_CACHE_SIZE);
177
7ac687d9 178 kaddr = kmap_atomic(cpage);
c8b97818 179 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 180 kunmap_atomic(kaddr);
c8b97818
CM
181
182 i++;
183 ptr += cur_size;
184 compressed_size -= cur_size;
185 }
186 btrfs_set_file_extent_compression(leaf, ei,
261507a0 187 compress_type);
c8b97818
CM
188 } else {
189 page = find_get_page(inode->i_mapping,
190 start >> PAGE_CACHE_SHIFT);
191 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 192 kaddr = kmap_atomic(page);
c8b97818
CM
193 offset = start & (PAGE_CACHE_SIZE - 1);
194 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 195 kunmap_atomic(kaddr);
c8b97818
CM
196 page_cache_release(page);
197 }
198 btrfs_mark_buffer_dirty(leaf);
199 btrfs_free_path(path);
200
c2167754
YZ
201 /*
202 * we're an inline extent, so nobody can
203 * extend the file past i_size without locking
204 * a page we already have locked.
205 *
206 * We must do any isize and inode updates
207 * before we unlock the pages. Otherwise we
208 * could end up racing with unlink.
209 */
c8b97818 210 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 211 ret = btrfs_update_inode(trans, root, inode);
c2167754 212
79787eaa 213 return ret;
c8b97818
CM
214fail:
215 btrfs_free_path(path);
216 return err;
217}
218
219
220/*
221 * conditionally insert an inline extent into the file. This
222 * does the checks required to make sure the data is small enough
223 * to fit as an inline extent.
224 */
7f366cfe 225static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM
226 struct btrfs_root *root,
227 struct inode *inode, u64 start, u64 end,
fe3f566c 228 size_t compressed_size, int compress_type,
c8b97818
CM
229 struct page **compressed_pages)
230{
231 u64 isize = i_size_read(inode);
232 u64 actual_end = min(end + 1, isize);
233 u64 inline_len = actual_end - start;
234 u64 aligned_end = (end + root->sectorsize - 1) &
235 ~((u64)root->sectorsize - 1);
c8b97818
CM
236 u64 data_len = inline_len;
237 int ret;
238
239 if (compressed_size)
240 data_len = compressed_size;
241
242 if (start > 0 ||
70b99e69 243 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM
244 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
245 (!compressed_size &&
246 (actual_end & (root->sectorsize - 1)) == 0) ||
247 end + 1 < isize ||
248 data_len > root->fs_info->max_inline) {
249 return 1;
250 }
251
2671485d 252 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM
253 if (ret)
254 return ret;
c8b97818
CM
255
256 if (isize > actual_end)
257 inline_len = min_t(u64, isize, actual_end);
258 ret = insert_inline_extent(trans, root, inode, start,
259 inline_len, compressed_size,
fe3f566c 260 compress_type, compressed_pages);
2adcac1a 261 if (ret && ret != -ENOSPC) {
79787eaa
JM
262 btrfs_abort_transaction(trans, root, ret);
263 return ret;
2adcac1a
JB
264 } else if (ret == -ENOSPC) {
265 return 1;
79787eaa 266 }
2adcac1a 267
0ca1f7ce 268 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 269 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM
270 return 0;
271}
272
771ed689
CM
273struct async_extent {
274 u64 start;
275 u64 ram_size;
276 u64 compressed_size;
277 struct page **pages;
278 unsigned long nr_pages;
261507a0 279 int compress_type;
771ed689
CM
280 struct list_head list;
281};
282
283struct async_cow {
284 struct inode *inode;
285 struct btrfs_root *root;
286 struct page *locked_page;
287 u64 start;
288 u64 end;
289 struct list_head extents;
290 struct btrfs_work work;
291};
292
293static noinline int add_async_extent(struct async_cow *cow,
294 u64 start, u64 ram_size,
295 u64 compressed_size,
296 struct page **pages,
261507a0
LZ
297 unsigned long nr_pages,
298 int compress_type)
771ed689
CM
299{
300 struct async_extent *async_extent;
301
302 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 303 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM
304 async_extent->start = start;
305 async_extent->ram_size = ram_size;
306 async_extent->compressed_size = compressed_size;
307 async_extent->pages = pages;
308 async_extent->nr_pages = nr_pages;
261507a0 309 async_extent->compress_type = compress_type;
771ed689
CM
310 list_add_tail(&async_extent->list, &cow->extents);
311 return 0;
312}
313
d352ac68 314/*
771ed689
CM
315 * we create compressed extents in two phases. The first
316 * phase compresses a range of pages that have already been
317 * locked (both pages and state bits are locked).
c8b97818 318 *
771ed689
CM
319 * This is done inside an ordered work queue, and the compression
320 * is spread across many cpus. The actual IO submission is step
321 * two, and the ordered work queue takes care of making sure that
322 * happens in the same order things were put onto the queue by
323 * writepages and friends.
c8b97818 324 *
771ed689
CM
325 * If this code finds it can't get good compression, it puts an
326 * entry onto the work queue to write the uncompressed bytes. This
327 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB
328 * are written in the same order that the flusher thread sent them
329 * down.
d352ac68 330 */
771ed689
CM
331static noinline int compress_file_range(struct inode *inode,
332 struct page *locked_page,
333 u64 start, u64 end,
334 struct async_cow *async_cow,
335 int *num_added)
b888db2b
CM
336{
337 struct btrfs_root *root = BTRFS_I(inode)->root;
338 struct btrfs_trans_handle *trans;
db94535d 339 u64 num_bytes;
db94535d 340 u64 blocksize = root->sectorsize;
c8b97818 341 u64 actual_end;
42dc7bab 342 u64 isize = i_size_read(inode);
e6dcd2dc 343 int ret = 0;
c8b97818
CM
344 struct page **pages = NULL;
345 unsigned long nr_pages;
346 unsigned long nr_pages_ret = 0;
347 unsigned long total_compressed = 0;
348 unsigned long total_in = 0;
349 unsigned long max_compressed = 128 * 1024;
771ed689 350 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM
351 int i;
352 int will_compress;
261507a0 353 int compress_type = root->fs_info->compress_type;
b888db2b 354
4cb13e5d
LB
355 /* if this is a small write inside eof, kick off a defrag */
356 if ((end - start + 1) < 16 * 1024 &&
357 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM
358 btrfs_add_inode_defrag(NULL, inode);
359
42dc7bab 360 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM
361again:
362 will_compress = 0;
363 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
364 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 365
f03d9301
CM
366 /*
367 * we don't want to send crud past the end of i_size through
368 * compression, that's just a waste of CPU time. So, if the
369 * end of the file is before the start of our current
370 * requested range of bytes, we bail out to the uncompressed
371 * cleanup code that can deal with all of this.
372 *
373 * It isn't really the fastest way to fix things, but this is a
374 * very uncommon corner.
375 */
376 if (actual_end <= start)
377 goto cleanup_and_bail_uncompressed;
378
c8b97818
CM
379 total_compressed = actual_end - start;
380
381 /* we want to make sure that amount of ram required to uncompress
382 * an extent is reasonable, so we limit the total size in ram
771ed689
CM
383 * of a compressed extent to 128k. This is a crucial number
384 * because it also controls how easily we can spread reads across
385 * cpus for decompression.
386 *
387 * We also want to make sure the amount of IO required to do
388 * a random read is reasonably small, so we limit the size of
389 * a compressed extent to 128k.
c8b97818
CM
390 */
391 total_compressed = min(total_compressed, max_uncompressed);
db94535d 392 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 393 num_bytes = max(blocksize, num_bytes);
c8b97818
CM
394 total_in = 0;
395 ret = 0;
db94535d 396
771ed689
CM
397 /*
398 * we do compression for mount -o compress and when the
399 * inode has not been flagged as nocompress. This flag can
400 * change at any time if we discover bad compression ratios.
c8b97818 401 */
6cbff00f 402 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 403 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB
404 (BTRFS_I(inode)->force_compress) ||
405 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 406 WARN_ON(pages);
cfbc246e 407 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ
408 if (!pages) {
409 /* just bail out to the uncompressed code */
410 goto cont;
411 }
c8b97818 412
261507a0
LZ
413 if (BTRFS_I(inode)->force_compress)
414 compress_type = BTRFS_I(inode)->force_compress;
415
416 ret = btrfs_compress_pages(compress_type,
417 inode->i_mapping, start,
418 total_compressed, pages,
419 nr_pages, &nr_pages_ret,
420 &total_in,
421 &total_compressed,
422 max_compressed);
c8b97818
CM
423
424 if (!ret) {
425 unsigned long offset = total_compressed &
426 (PAGE_CACHE_SIZE - 1);
427 struct page *page = pages[nr_pages_ret - 1];
428 char *kaddr;
429
430 /* zero the tail end of the last page, we might be
431 * sending it down to disk
432 */
433 if (offset) {
7ac687d9 434 kaddr = kmap_atomic(page);
c8b97818
CM
435 memset(kaddr + offset, 0,
436 PAGE_CACHE_SIZE - offset);
7ac687d9 437 kunmap_atomic(kaddr);
c8b97818
CM
438 }
439 will_compress = 1;
440 }
441 }
560f7d75 442cont:
c8b97818 443 if (start == 0) {
7a7eaa40 444 trans = btrfs_join_transaction(root);
79787eaa
JM
445 if (IS_ERR(trans)) {
446 ret = PTR_ERR(trans);
447 trans = NULL;
448 goto cleanup_and_out;
449 }
0ca1f7ce 450 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 451
c8b97818 452 /* lets try to make an inline extent */
771ed689 453 if (ret || total_in < (actual_end - start)) {
c8b97818 454 /* we didn't compress the entire range, try
771ed689 455 * to make an uncompressed inline extent.
c8b97818
CM
456 */
457 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 458 start, end, 0, 0, NULL);
c8b97818 459 } else {
771ed689 460 /* try making a compressed inline extent */
c8b97818
CM
461 ret = cow_file_range_inline(trans, root, inode,
462 start, end,
fe3f566c
LZ
463 total_compressed,
464 compress_type, pages);
c8b97818 465 }
79787eaa 466 if (ret <= 0) {
771ed689 467 /*
79787eaa
JM
468 * inline extent creation worked or returned error,
469 * we don't need to create any more async work items.
470 * Unlock and free up our temp pages.
771ed689 471 */
c8b97818 472 extent_clear_unlock_delalloc(inode,
a791e35e
CM
473 &BTRFS_I(inode)->io_tree,
474 start, end, NULL,
475 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 476 EXTENT_CLEAR_DELALLOC |
a791e35e 477 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ
478
479 btrfs_end_transaction(trans, root);
c8b97818
CM
480 goto free_pages_out;
481 }
c2167754 482 btrfs_end_transaction(trans, root);
c8b97818
CM
483 }
484
485 if (will_compress) {
486 /*
487 * we aren't doing an inline extent round the compressed size
488 * up to a block size boundary so the allocator does sane
489 * things
490 */
491 total_compressed = (total_compressed + blocksize - 1) &
492 ~(blocksize - 1);
493
494 /*
495 * one last check to make sure the compression is really a
496 * win, compare the page count read with the blocks on disk
497 */
498 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
499 ~(PAGE_CACHE_SIZE - 1);
500 if (total_compressed >= total_in) {
501 will_compress = 0;
502 } else {
c8b97818
CM
503 num_bytes = total_in;
504 }
505 }
506 if (!will_compress && pages) {
507 /*
508 * the compression code ran but failed to make things smaller,
509 * free any pages it allocated and our page pointer array
510 */
511 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 512 WARN_ON(pages[i]->mapping);
c8b97818
CM
513 page_cache_release(pages[i]);
514 }
515 kfree(pages);
516 pages = NULL;
517 total_compressed = 0;
518 nr_pages_ret = 0;
519
520 /* flag the file so we don't compress in the future */
1e701a32
CM
521 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
522 !(BTRFS_I(inode)->force_compress)) {
a555f810 523 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 524 }
c8b97818 525 }
771ed689
CM
526 if (will_compress) {
527 *num_added += 1;
c8b97818 528
771ed689
CM
529 /* the async work queues will take care of doing actual
530 * allocation on disk for these compressed pages,
531 * and will submit them to the elevator.
532 */
533 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ
534 total_compressed, pages, nr_pages_ret,
535 compress_type);
179e29e4 536
24ae6365 537 if (start + num_bytes < end) {
771ed689
CM
538 start += num_bytes;
539 pages = NULL;
540 cond_resched();
541 goto again;
542 }
543 } else {
f03d9301 544cleanup_and_bail_uncompressed:
771ed689
CM
545 /*
546 * No compression, but we still need to write the pages in
547 * the file we've been given so far. redirty the locked
548 * page if it corresponds to our extent and set things up
549 * for the async work queue to run cow_file_range to do
550 * the normal delalloc dance
551 */
552 if (page_offset(locked_page) >= start &&
553 page_offset(locked_page) <= end) {
554 __set_page_dirty_nobuffers(locked_page);
555 /* unlocked later on in the async handlers */
556 }
261507a0
LZ
557 add_async_extent(async_cow, start, end - start + 1,
558 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM
559 *num_added += 1;
560 }
3b951516 561
771ed689 562out:
79787eaa 563 return ret;
771ed689
CM
564
565free_pages_out:
566 for (i = 0; i < nr_pages_ret; i++) {
567 WARN_ON(pages[i]->mapping);
568 page_cache_release(pages[i]);
569 }
d397712b 570 kfree(pages);
771ed689
CM
571
572 goto out;
79787eaa
JM
573
574cleanup_and_out:
575 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
576 start, end, NULL,
577 EXTENT_CLEAR_UNLOCK_PAGE |
578 EXTENT_CLEAR_DIRTY |
579 EXTENT_CLEAR_DELALLOC |
580 EXTENT_SET_WRITEBACK |
581 EXTENT_END_WRITEBACK);
582 if (!trans || IS_ERR(trans))
583 btrfs_error(root->fs_info, ret, "Failed to join transaction");
584 else
585 btrfs_abort_transaction(trans, root, ret);
586 goto free_pages_out;
771ed689
CM
587}
588
589/*
590 * phase two of compressed writeback. This is the ordered portion
591 * of the code, which only gets called in the order the work was
592 * queued. We walk all the async extents created by compress_file_range
593 * and send them down to the disk.
594 */
595static noinline int submit_compressed_extents(struct inode *inode,
596 struct async_cow *async_cow)
597{
598 struct async_extent *async_extent;
599 u64 alloc_hint = 0;
600 struct btrfs_trans_handle *trans;
601 struct btrfs_key ins;
602 struct extent_map *em;
603 struct btrfs_root *root = BTRFS_I(inode)->root;
604 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
605 struct extent_io_tree *io_tree;
f5a84ee3 606 int ret = 0;
771ed689
CM
607
608 if (list_empty(&async_cow->extents))
609 return 0;
610
771ed689 611
d397712b 612 while (!list_empty(&async_cow->extents)) {
771ed689
CM
613 async_extent = list_entry(async_cow->extents.next,
614 struct async_extent, list);
615 list_del(&async_extent->list);
c8b97818 616
771ed689
CM
617 io_tree = &BTRFS_I(inode)->io_tree;
618
f5a84ee3 619retry:
771ed689
CM
620 /* did the compression code fall back to uncompressed IO? */
621 if (!async_extent->pages) {
622 int page_started = 0;
623 unsigned long nr_written = 0;
624
625 lock_extent(io_tree, async_extent->start,
2ac55d41 626 async_extent->start +
d0082371 627 async_extent->ram_size - 1);
771ed689
CM
628
629 /* allocate blocks */
f5a84ee3
JB
630 ret = cow_file_range(inode, async_cow->locked_page,
631 async_extent->start,
632 async_extent->start +
633 async_extent->ram_size - 1,
634 &page_started, &nr_written, 0);
771ed689 635
79787eaa
JM
636 /* JDM XXX */
637
771ed689
CM
638 /*
639 * if page_started, cow_file_range inserted an
640 * inline extent and took care of all the unlocking
641 * and IO for us. Otherwise, we need to submit
642 * all those pages down to the drive.
643 */
f5a84ee3 644 if (!page_started && !ret)
771ed689
CM
645 extent_write_locked_range(io_tree,
646 inode, async_extent->start,
d397712b 647 async_extent->start +
771ed689
CM
648 async_extent->ram_size - 1,
649 btrfs_get_extent,
650 WB_SYNC_ALL);
651 kfree(async_extent);
652 cond_resched();
653 continue;
654 }
655
656 lock_extent(io_tree, async_extent->start,
d0082371 657 async_extent->start + async_extent->ram_size - 1);
771ed689 658
7a7eaa40 659 trans = btrfs_join_transaction(root);
79787eaa
JM
660 if (IS_ERR(trans)) {
661 ret = PTR_ERR(trans);
662 } else {
663 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
664 ret = btrfs_reserve_extent(trans, root,
771ed689
CM
665 async_extent->compressed_size,
666 async_extent->compressed_size,
81c9ad23 667 0, alloc_hint, &ins, 1);
962197ba 668 if (ret && ret != -ENOSPC)
79787eaa
JM
669 btrfs_abort_transaction(trans, root, ret);
670 btrfs_end_transaction(trans, root);
671 }
c2167754 672
f5a84ee3
JB
673 if (ret) {
674 int i;
675 for (i = 0; i < async_extent->nr_pages; i++) {
676 WARN_ON(async_extent->pages[i]->mapping);
677 page_cache_release(async_extent->pages[i]);
678 }
679 kfree(async_extent->pages);
680 async_extent->nr_pages = 0;
681 async_extent->pages = NULL;
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
d0082371 684 async_extent->ram_size - 1);
79787eaa
JM
685 if (ret == -ENOSPC)
686 goto retry;
687 goto out_free; /* JDM: Requeue? */
f5a84ee3
JB
688 }
689
c2167754
YZ
690 /*
691 * here we're doing allocation and writeback of the
692 * compressed pages
693 */
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
697
172ddd60 698 em = alloc_extent_map();
79787eaa 699 BUG_ON(!em); /* -ENOMEM */
771ed689
CM
700 em->start = async_extent->start;
701 em->len = async_extent->ram_size;
445a6944 702 em->orig_start = em->start;
2ab28f32
JB
703 em->mod_start = em->start;
704 em->mod_len = em->len;
c8b97818 705
771ed689
CM
706 em->block_start = ins.objectid;
707 em->block_len = ins.offset;
b4939680 708 em->orig_block_len = ins.offset;
771ed689 709 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 710 em->compress_type = async_extent->compress_type;
771ed689
CM
711 set_bit(EXTENT_FLAG_PINNED, &em->flags);
712 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 713 em->generation = -1;
771ed689 714
d397712b 715 while (1) {
890871be 716 write_lock(&em_tree->lock);
771ed689 717 ret = add_extent_mapping(em_tree, em);
70c8a91c
JB
718 if (!ret)
719 list_move(&em->list,
720 &em_tree->modified_extents);
890871be 721 write_unlock(&em_tree->lock);
771ed689
CM
722 if (ret != -EEXIST) {
723 free_extent_map(em);
724 break;
725 }
726 btrfs_drop_extent_cache(inode, async_extent->start,
727 async_extent->start +
728 async_extent->ram_size - 1, 0);
729 }
730
261507a0
LZ
731 ret = btrfs_add_ordered_extent_compress(inode,
732 async_extent->start,
733 ins.objectid,
734 async_extent->ram_size,
735 ins.offset,
736 BTRFS_ORDERED_COMPRESSED,
737 async_extent->compress_type);
79787eaa 738 BUG_ON(ret); /* -ENOMEM */
771ed689 739
771ed689
CM
740 /*
741 * clear dirty, set writeback and unlock the pages.
742 */
743 extent_clear_unlock_delalloc(inode,
a791e35e
CM
744 &BTRFS_I(inode)->io_tree,
745 async_extent->start,
746 async_extent->start +
747 async_extent->ram_size - 1,
748 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
749 EXTENT_CLEAR_UNLOCK |
a3429ab7 750 EXTENT_CLEAR_DELALLOC |
a791e35e 751 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM
752
753 ret = btrfs_submit_compressed_write(inode,
d397712b
CM
754 async_extent->start,
755 async_extent->ram_size,
756 ins.objectid,
757 ins.offset, async_extent->pages,
758 async_extent->nr_pages);
771ed689 759
79787eaa 760 BUG_ON(ret); /* -ENOMEM */
771ed689
CM
761 alloc_hint = ins.objectid + ins.offset;
762 kfree(async_extent);
763 cond_resched();
764 }
79787eaa
JM
765 ret = 0;
766out:
767 return ret;
768out_free:
769 kfree(async_extent);
770 goto out;
771ed689
CM
771}
772
4b46fce2
JB
773static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
774 u64 num_bytes)
775{
776 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
777 struct extent_map *em;
778 u64 alloc_hint = 0;
779
780 read_lock(&em_tree->lock);
781 em = search_extent_mapping(em_tree, start, num_bytes);
782 if (em) {
783 /*
784 * if block start isn't an actual block number then find the
785 * first block in this inode and use that as a hint. If that
786 * block is also bogus then just don't worry about it.
787 */
788 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
789 free_extent_map(em);
790 em = search_extent_mapping(em_tree, 0, 0);
791 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
792 alloc_hint = em->block_start;
793 if (em)
794 free_extent_map(em);
795 } else {
796 alloc_hint = em->block_start;
797 free_extent_map(em);
798 }
799 }
800 read_unlock(&em_tree->lock);
801
802 return alloc_hint;
803}
804
771ed689
CM
805/*
806 * when extent_io.c finds a delayed allocation range in the file,
807 * the call backs end up in this code. The basic idea is to
808 * allocate extents on disk for the range, and create ordered data structs
809 * in ram to track those extents.
810 *
811 * locked_page is the page that writepage had locked already. We use
812 * it to make sure we don't do extra locks or unlocks.
813 *
814 * *page_started is set to one if we unlock locked_page and do everything
815 * required to start IO on it. It may be clean and already done with
816 * IO when we return.
817 */
b7d5b0a8
MX
818static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
819 struct inode *inode,
820 struct btrfs_root *root,
821 struct page *locked_page,
822 u64 start, u64 end, int *page_started,
823 unsigned long *nr_written,
824 int unlock)
771ed689 825{
771ed689
CM
826 u64 alloc_hint = 0;
827 u64 num_bytes;
828 unsigned long ram_size;
829 u64 disk_num_bytes;
830 u64 cur_alloc_size;
831 u64 blocksize = root->sectorsize;
771ed689
CM
832 struct btrfs_key ins;
833 struct extent_map *em;
834 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
835 int ret = 0;
836
83eea1f1 837 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 838
771ed689
CM
839 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
840 num_bytes = max(blocksize, num_bytes);
841 disk_num_bytes = num_bytes;
771ed689 842
4cb5300b 843 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB
844 if (num_bytes < 64 * 1024 &&
845 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM
846 btrfs_add_inode_defrag(trans, inode);
847
771ed689
CM
848 if (start == 0) {
849 /* lets try to make an inline extent */
850 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 851 start, end, 0, 0, NULL);
771ed689
CM
852 if (ret == 0) {
853 extent_clear_unlock_delalloc(inode,
a791e35e
CM
854 &BTRFS_I(inode)->io_tree,
855 start, end, NULL,
856 EXTENT_CLEAR_UNLOCK_PAGE |
857 EXTENT_CLEAR_UNLOCK |
858 EXTENT_CLEAR_DELALLOC |
859 EXTENT_CLEAR_DIRTY |
860 EXTENT_SET_WRITEBACK |
861 EXTENT_END_WRITEBACK);
c2167754 862
771ed689
CM
863 *nr_written = *nr_written +
864 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
865 *page_started = 1;
771ed689 866 goto out;
79787eaa
JM
867 } else if (ret < 0) {
868 btrfs_abort_transaction(trans, root, ret);
869 goto out_unlock;
771ed689
CM
870 }
871 }
872
873 BUG_ON(disk_num_bytes >
6c41761f 874 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 875
4b46fce2 876 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM
877 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
878
d397712b 879 while (disk_num_bytes > 0) {
a791e35e
CM
880 unsigned long op;
881
287a0ab9 882 cur_alloc_size = disk_num_bytes;
e6dcd2dc 883 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 884 root->sectorsize, 0, alloc_hint,
81c9ad23 885 &ins, 1);
79787eaa
JM
886 if (ret < 0) {
887 btrfs_abort_transaction(trans, root, ret);
888 goto out_unlock;
889 }
d397712b 890
172ddd60 891 em = alloc_extent_map();
79787eaa 892 BUG_ON(!em); /* -ENOMEM */
e6dcd2dc 893 em->start = start;
445a6944 894 em->orig_start = em->start;
771ed689
CM
895 ram_size = ins.offset;
896 em->len = ins.offset;
2ab28f32
JB
897 em->mod_start = em->start;
898 em->mod_len = em->len;
c8b97818 899
e6dcd2dc 900 em->block_start = ins.objectid;
c8b97818 901 em->block_len = ins.offset;
b4939680 902 em->orig_block_len = ins.offset;
e6dcd2dc 903 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 904 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 905 em->generation = -1;
c8b97818 906
d397712b 907 while (1) {
890871be 908 write_lock(&em_tree->lock);
e6dcd2dc 909 ret = add_extent_mapping(em_tree, em);
70c8a91c
JB
910 if (!ret)
911 list_move(&em->list,
912 &em_tree->modified_extents);
890871be 913 write_unlock(&em_tree->lock);
e6dcd2dc
CM
914 if (ret != -EEXIST) {
915 free_extent_map(em);
916 break;
917 }
918 btrfs_drop_extent_cache(inode, start,
c8b97818 919 start + ram_size - 1, 0);
e6dcd2dc
CM
920 }
921
98d20f67 922 cur_alloc_size = ins.offset;
e6dcd2dc 923 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 924 ram_size, cur_alloc_size, 0);
79787eaa 925 BUG_ON(ret); /* -ENOMEM */
c8b97818 926
17d217fe
YZ
927 if (root->root_key.objectid ==
928 BTRFS_DATA_RELOC_TREE_OBJECTID) {
929 ret = btrfs_reloc_clone_csums(inode, start,
930 cur_alloc_size);
79787eaa
JM
931 if (ret) {
932 btrfs_abort_transaction(trans, root, ret);
933 goto out_unlock;
934 }
17d217fe
YZ
935 }
936
d397712b 937 if (disk_num_bytes < cur_alloc_size)
3b951516 938 break;
d397712b 939
c8b97818
CM
940 /* we're not doing compressed IO, don't unlock the first
941 * page (which the caller expects to stay locked), don't
942 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM
943 *
944 * Do set the Private2 bit so we know this page was properly
945 * setup for writepage
c8b97818 946 */
a791e35e
CM
947 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
948 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
949 EXTENT_SET_PRIVATE2;
950
c8b97818
CM
951 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
952 start, start + ram_size - 1,
a791e35e 953 locked_page, op);
c8b97818 954 disk_num_bytes -= cur_alloc_size;
c59f8951
CM
955 num_bytes -= cur_alloc_size;
956 alloc_hint = ins.objectid + ins.offset;
957 start += cur_alloc_size;
b888db2b 958 }
79787eaa 959out:
be20aa9d 960 return ret;
b7d5b0a8 961
79787eaa
JM
962out_unlock:
963 extent_clear_unlock_delalloc(inode,
964 &BTRFS_I(inode)->io_tree,
beb42dd7 965 start, end, locked_page,
79787eaa
JM
966 EXTENT_CLEAR_UNLOCK_PAGE |
967 EXTENT_CLEAR_UNLOCK |
968 EXTENT_CLEAR_DELALLOC |
969 EXTENT_CLEAR_DIRTY |
970 EXTENT_SET_WRITEBACK |
971 EXTENT_END_WRITEBACK);
972
973 goto out;
771ed689 974}
c8b97818 975
b7d5b0a8
MX
976static noinline int cow_file_range(struct inode *inode,
977 struct page *locked_page,
978 u64 start, u64 end, int *page_started,
979 unsigned long *nr_written,
980 int unlock)
981{
982 struct btrfs_trans_handle *trans;
983 struct btrfs_root *root = BTRFS_I(inode)->root;
984 int ret;
985
986 trans = btrfs_join_transaction(root);
987 if (IS_ERR(trans)) {
988 extent_clear_unlock_delalloc(inode,
989 &BTRFS_I(inode)->io_tree,
990 start, end, locked_page,
991 EXTENT_CLEAR_UNLOCK_PAGE |
992 EXTENT_CLEAR_UNLOCK |
993 EXTENT_CLEAR_DELALLOC |
994 EXTENT_CLEAR_DIRTY |
995 EXTENT_SET_WRITEBACK |
996 EXTENT_END_WRITEBACK);
997 return PTR_ERR(trans);
998 }
999 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1000
1001 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1002 page_started, nr_written, unlock);
1003
1004 btrfs_end_transaction(trans, root);
1005
1006 return ret;
1007}
1008
771ed689
CM
1009/*
1010 * work queue call back to started compression on a file and pages
1011 */
1012static noinline void async_cow_start(struct btrfs_work *work)
1013{
1014 struct async_cow *async_cow;
1015 int num_added = 0;
1016 async_cow = container_of(work, struct async_cow, work);
1017
1018 compress_file_range(async_cow->inode, async_cow->locked_page,
1019 async_cow->start, async_cow->end, async_cow,
1020 &num_added);
8180ef88 1021 if (num_added == 0) {
cb77fcd8 1022 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1023 async_cow->inode = NULL;
8180ef88 1024 }
771ed689
CM
1025}
1026
1027/*
1028 * work queue call back to submit previously compressed pages
1029 */
1030static noinline void async_cow_submit(struct btrfs_work *work)
1031{
1032 struct async_cow *async_cow;
1033 struct btrfs_root *root;
1034 unsigned long nr_pages;
1035
1036 async_cow = container_of(work, struct async_cow, work);
1037
1038 root = async_cow->root;
1039 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1040 PAGE_CACHE_SHIFT;
1041
66657b31 1042 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1043 5 * 1024 * 1024 &&
771ed689
CM
1044 waitqueue_active(&root->fs_info->async_submit_wait))
1045 wake_up(&root->fs_info->async_submit_wait);
1046
d397712b 1047 if (async_cow->inode)
771ed689 1048 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1049}
c8b97818 1050
771ed689
CM
1051static noinline void async_cow_free(struct btrfs_work *work)
1052{
1053 struct async_cow *async_cow;
1054 async_cow = container_of(work, struct async_cow, work);
8180ef88 1055 if (async_cow->inode)
cb77fcd8 1056 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM
1057 kfree(async_cow);
1058}
1059
1060static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1061 u64 start, u64 end, int *page_started,
1062 unsigned long *nr_written)
1063{
1064 struct async_cow *async_cow;
1065 struct btrfs_root *root = BTRFS_I(inode)->root;
1066 unsigned long nr_pages;
1067 u64 cur_end;
287082b0 1068 int limit = 10 * 1024 * 1024;
771ed689 1069
a3429ab7
CM
1070 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1071 1, 0, NULL, GFP_NOFS);
d397712b 1072 while (start < end) {
771ed689 1073 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1074 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1075 async_cow->inode = igrab(inode);
771ed689
CM
1076 async_cow->root = root;
1077 async_cow->locked_page = locked_page;
1078 async_cow->start = start;
1079
6cbff00f 1080 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM
1081 cur_end = end;
1082 else
1083 cur_end = min(end, start + 512 * 1024 - 1);
1084
1085 async_cow->end = cur_end;
1086 INIT_LIST_HEAD(&async_cow->extents);
1087
1088 async_cow->work.func = async_cow_start;
1089 async_cow->work.ordered_func = async_cow_submit;
1090 async_cow->work.ordered_free = async_cow_free;
1091 async_cow->work.flags = 0;
1092
771ed689
CM
1093 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1094 PAGE_CACHE_SHIFT;
1095 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1096
1097 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1098 &async_cow->work);
1099
1100 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1101 wait_event(root->fs_info->async_submit_wait,
1102 (atomic_read(&root->fs_info->async_delalloc_pages) <
1103 limit));
1104 }
1105
d397712b 1106 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM
1107 atomic_read(&root->fs_info->async_delalloc_pages)) {
1108 wait_event(root->fs_info->async_submit_wait,
1109 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1110 0));
1111 }
1112
1113 *nr_written += nr_pages;
1114 start = cur_end + 1;
1115 }
1116 *page_started = 1;
1117 return 0;
be20aa9d
CM
1118}
1119
d397712b 1120static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ
1121 u64 bytenr, u64 num_bytes)
1122{
1123 int ret;
1124 struct btrfs_ordered_sum *sums;
1125 LIST_HEAD(list);
1126
07d400a6 1127 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1128 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ
1129 if (ret == 0 && list_empty(&list))
1130 return 0;
1131
1132 while (!list_empty(&list)) {
1133 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1134 list_del(&sums->list);
1135 kfree(sums);
1136 }
1137 return 1;
1138}
1139
d352ac68
CM
1140/*
1141 * when nowcow writeback call back. This checks for snapshots or COW copies
1142 * of the extents that exist in the file, and COWs the file as required.
1143 *
1144 * If no cow copies or snapshots exist, we write directly to the existing
1145 * blocks on disk
1146 */
7f366cfe
CM
1147static noinline int run_delalloc_nocow(struct inode *inode,
1148 struct page *locked_page,
771ed689
CM
1149 u64 start, u64 end, int *page_started, int force,
1150 unsigned long *nr_written)
be20aa9d 1151{
be20aa9d 1152 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1153 struct btrfs_trans_handle *trans;
be20aa9d 1154 struct extent_buffer *leaf;
be20aa9d 1155 struct btrfs_path *path;
80ff3856 1156 struct btrfs_file_extent_item *fi;
be20aa9d 1157 struct btrfs_key found_key;
80ff3856
YZ
1158 u64 cow_start;
1159 u64 cur_offset;
1160 u64 extent_end;
5d4f98a2 1161 u64 extent_offset;
80ff3856
YZ
1162 u64 disk_bytenr;
1163 u64 num_bytes;
b4939680 1164 u64 disk_num_bytes;
80ff3856 1165 int extent_type;
79787eaa 1166 int ret, err;
d899e052 1167 int type;
80ff3856
YZ
1168 int nocow;
1169 int check_prev = 1;
82d5902d 1170 bool nolock;
33345d01 1171 u64 ino = btrfs_ino(inode);
be20aa9d
CM
1172
1173 path = btrfs_alloc_path();
17ca04af
JB
1174 if (!path) {
1175 extent_clear_unlock_delalloc(inode,
1176 &BTRFS_I(inode)->io_tree,
1177 start, end, locked_page,
1178 EXTENT_CLEAR_UNLOCK_PAGE |
1179 EXTENT_CLEAR_UNLOCK |
1180 EXTENT_CLEAR_DELALLOC |
1181 EXTENT_CLEAR_DIRTY |
1182 EXTENT_SET_WRITEBACK |
1183 EXTENT_END_WRITEBACK);
d8926bb3 1184 return -ENOMEM;
17ca04af 1185 }
82d5902d 1186
83eea1f1 1187 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ
1188
1189 if (nolock)
7a7eaa40 1190 trans = btrfs_join_transaction_nolock(root);
82d5902d 1191 else
7a7eaa40 1192 trans = btrfs_join_transaction(root);
ff5714cc 1193
79787eaa 1194 if (IS_ERR(trans)) {
17ca04af
JB
1195 extent_clear_unlock_delalloc(inode,
1196 &BTRFS_I(inode)->io_tree,
1197 start, end, locked_page,
1198 EXTENT_CLEAR_UNLOCK_PAGE |
1199 EXTENT_CLEAR_UNLOCK |
1200 EXTENT_CLEAR_DELALLOC |
1201 EXTENT_CLEAR_DIRTY |
1202 EXTENT_SET_WRITEBACK |
1203 EXTENT_END_WRITEBACK);
79787eaa
JM
1204 btrfs_free_path(path);
1205 return PTR_ERR(trans);
1206 }
1207
74b21075 1208 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1209
80ff3856
YZ
1210 cow_start = (u64)-1;
1211 cur_offset = start;
1212 while (1) {
33345d01 1213 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1214 cur_offset, 0);
79787eaa
JM
1215 if (ret < 0) {
1216 btrfs_abort_transaction(trans, root, ret);
1217 goto error;
1218 }
80ff3856
YZ
1219 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1220 leaf = path->nodes[0];
1221 btrfs_item_key_to_cpu(leaf, &found_key,
1222 path->slots[0] - 1);
33345d01 1223 if (found_key.objectid == ino &&
80ff3856
YZ
1224 found_key.type == BTRFS_EXTENT_DATA_KEY)
1225 path->slots[0]--;
1226 }
1227 check_prev = 0;
1228next_slot:
1229 leaf = path->nodes[0];
1230 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1231 ret = btrfs_next_leaf(root, path);
79787eaa
JM
1232 if (ret < 0) {
1233 btrfs_abort_transaction(trans, root, ret);
1234 goto error;
1235 }
80ff3856
YZ
1236 if (ret > 0)
1237 break;
1238 leaf = path->nodes[0];
1239 }
be20aa9d 1240
80ff3856
YZ
1241 nocow = 0;
1242 disk_bytenr = 0;
17d217fe 1243 num_bytes = 0;
80ff3856
YZ
1244 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1245
33345d01 1246 if (found_key.objectid > ino ||
80ff3856
YZ
1247 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1248 found_key.offset > end)
1249 break;
1250
1251 if (found_key.offset > cur_offset) {
1252 extent_end = found_key.offset;
e9061e21 1253 extent_type = 0;
80ff3856
YZ
1254 goto out_check;
1255 }
1256
1257 fi = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_file_extent_item);
1259 extent_type = btrfs_file_extent_type(leaf, fi);
1260
d899e052
YZ
1261 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1262 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1263 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1264 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ
1265 extent_end = found_key.offset +
1266 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB
1267 disk_num_bytes =
1268 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ
1269 if (extent_end <= start) {
1270 path->slots[0]++;
1271 goto next_slot;
1272 }
17d217fe
YZ
1273 if (disk_bytenr == 0)
1274 goto out_check;
80ff3856
YZ
1275 if (btrfs_file_extent_compression(leaf, fi) ||
1276 btrfs_file_extent_encryption(leaf, fi) ||
1277 btrfs_file_extent_other_encoding(leaf, fi))
1278 goto out_check;
d899e052
YZ
1279 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1280 goto out_check;
d2fb3437 1281 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1282 goto out_check;
33345d01 1283 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ
1284 found_key.offset -
1285 extent_offset, disk_bytenr))
17d217fe 1286 goto out_check;
5d4f98a2 1287 disk_bytenr += extent_offset;
17d217fe
YZ
1288 disk_bytenr += cur_offset - found_key.offset;
1289 num_bytes = min(end + 1, extent_end) - cur_offset;
1290 /*
1291 * force cow if csum exists in the range.
1292 * this ensure that csum for a given extent are
1293 * either valid or do not exist.
1294 */
1295 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296 goto out_check;
80ff3856
YZ
1297 nocow = 1;
1298 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299 extent_end = found_key.offset +
1300 btrfs_file_extent_inline_len(leaf, fi);
1301 extent_end = ALIGN(extent_end, root->sectorsize);
1302 } else {
1303 BUG_ON(1);
1304 }
1305out_check:
1306 if (extent_end <= start) {
1307 path->slots[0]++;
1308 goto next_slot;
1309 }
1310 if (!nocow) {
1311 if (cow_start == (u64)-1)
1312 cow_start = cur_offset;
1313 cur_offset = extent_end;
1314 if (cur_offset > end)
1315 break;
1316 path->slots[0]++;
1317 goto next_slot;
7ea394f1
YZ
1318 }
1319
b3b4aa74 1320 btrfs_release_path(path);
80ff3856 1321 if (cow_start != (u64)-1) {
b7d5b0a8
MX
1322 ret = __cow_file_range(trans, inode, root, locked_page,
1323 cow_start, found_key.offset - 1,
1324 page_started, nr_written, 1);
79787eaa
JM
1325 if (ret) {
1326 btrfs_abort_transaction(trans, root, ret);
1327 goto error;
1328 }
80ff3856 1329 cow_start = (u64)-1;
7ea394f1 1330 }
80ff3856 1331
d899e052
YZ
1332 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1333 struct extent_map *em;
1334 struct extent_map_tree *em_tree;
1335 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1336 em = alloc_extent_map();
79787eaa 1337 BUG_ON(!em); /* -ENOMEM */
d899e052 1338 em->start = cur_offset;
70c8a91c 1339 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ
1340 em->len = num_bytes;
1341 em->block_len = num_bytes;
1342 em->block_start = disk_bytenr;
b4939680 1343 em->orig_block_len = disk_num_bytes;
d899e052 1344 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB
1345 em->mod_start = em->start;
1346 em->mod_len = em->len;
d899e052 1347 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1348 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1349 em->generation = -1;
d899e052 1350 while (1) {
890871be 1351 write_lock(&em_tree->lock);
d899e052 1352 ret = add_extent_mapping(em_tree, em);
70c8a91c
JB
1353 if (!ret)
1354 list_move(&em->list,
1355 &em_tree->modified_extents);
890871be 1356 write_unlock(&em_tree->lock);
d899e052
YZ
1357 if (ret != -EEXIST) {
1358 free_extent_map(em);
1359 break;
1360 }
1361 btrfs_drop_extent_cache(inode, em->start,
1362 em->start + em->len - 1, 0);
1363 }
1364 type = BTRFS_ORDERED_PREALLOC;
1365 } else {
1366 type = BTRFS_ORDERED_NOCOW;
1367 }
80ff3856
YZ
1368
1369 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1370 num_bytes, num_bytes, type);
79787eaa 1371 BUG_ON(ret); /* -ENOMEM */
771ed689 1372
efa56464
YZ
1373 if (root->root_key.objectid ==
1374 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1375 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1376 num_bytes);
79787eaa
JM
1377 if (ret) {
1378 btrfs_abort_transaction(trans, root, ret);
1379 goto error;
1380 }
efa56464
YZ
1381 }
1382
d899e052 1383 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM
1384 cur_offset, cur_offset + num_bytes - 1,
1385 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1386 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1387 EXTENT_SET_PRIVATE2);
80ff3856
YZ
1388 cur_offset = extent_end;
1389 if (cur_offset > end)
1390 break;
be20aa9d 1391 }
b3b4aa74 1392 btrfs_release_path(path);
80ff3856 1393
17ca04af 1394 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1395 cow_start = cur_offset;
17ca04af
JB
1396 cur_offset = end;
1397 }
1398
80ff3856 1399 if (cow_start != (u64)-1) {
b7d5b0a8
MX
1400 ret = __cow_file_range(trans, inode, root, locked_page,
1401 cow_start, end,
1402 page_started, nr_written, 1);
79787eaa
JM
1403 if (ret) {
1404 btrfs_abort_transaction(trans, root, ret);
1405 goto error;
1406 }
80ff3856
YZ
1407 }
1408
79787eaa 1409error:
a698d075 1410 err = btrfs_end_transaction(trans, root);
79787eaa
JM
1411 if (!ret)
1412 ret = err;
1413
17ca04af
JB
1414 if (ret && cur_offset < end)
1415 extent_clear_unlock_delalloc(inode,
1416 &BTRFS_I(inode)->io_tree,
1417 cur_offset, end, locked_page,
1418 EXTENT_CLEAR_UNLOCK_PAGE |
1419 EXTENT_CLEAR_UNLOCK |
1420 EXTENT_CLEAR_DELALLOC |
1421 EXTENT_CLEAR_DIRTY |
1422 EXTENT_SET_WRITEBACK |
1423 EXTENT_END_WRITEBACK);
1424
7ea394f1 1425 btrfs_free_path(path);
79787eaa 1426 return ret;
be20aa9d
CM
1427}
1428
d352ac68
CM
1429/*
1430 * extent_io.c call back to do delayed allocation processing
1431 */
c8b97818 1432static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM
1433 u64 start, u64 end, int *page_started,
1434 unsigned long *nr_written)
be20aa9d 1435{
be20aa9d 1436 int ret;
7f366cfe 1437 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1438
7ddf5a42 1439 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1440 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1441 page_started, 1, nr_written);
7ddf5a42 1442 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1443 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1444 page_started, 0, nr_written);
7ddf5a42
JB
1445 } else if (!btrfs_test_opt(root, COMPRESS) &&
1446 !(BTRFS_I(inode)->force_compress) &&
1447 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM
1448 ret = cow_file_range(inode, locked_page, start, end,
1449 page_started, nr_written, 1);
7ddf5a42
JB
1450 } else {
1451 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1452 &BTRFS_I(inode)->runtime_flags);
771ed689 1453 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1454 page_started, nr_written);
7ddf5a42 1455 }
b888db2b
CM
1456 return ret;
1457}
1458
1bf85046
JM
1459static void btrfs_split_extent_hook(struct inode *inode,
1460 struct extent_state *orig, u64 split)
9ed74f2d 1461{
0ca1f7ce 1462 /* not delalloc, ignore it */
9ed74f2d 1463 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1464 return;
9ed74f2d 1465
9e0baf60
JB
1466 spin_lock(&BTRFS_I(inode)->lock);
1467 BTRFS_I(inode)->outstanding_extents++;
1468 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB
1469}
1470
1471/*
1472 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1473 * extents so we can keep track of new extents that are just merged onto old
1474 * extents, such as when we are doing sequential writes, so we can properly
1475 * account for the metadata space we'll need.
1476 */
1bf85046
JM
1477static void btrfs_merge_extent_hook(struct inode *inode,
1478 struct extent_state *new,
1479 struct extent_state *other)
9ed74f2d 1480{
9ed74f2d
JB
1481 /* not delalloc, ignore it */
1482 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1483 return;
9ed74f2d 1484
9e0baf60
JB
1485 spin_lock(&BTRFS_I(inode)->lock);
1486 BTRFS_I(inode)->outstanding_extents--;
1487 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB
1488}
1489
d352ac68
CM
1490/*
1491 * extent_io.c set_bit_hook, used to track delayed allocation
1492 * bytes in this file, and to maintain the list of inodes that
1493 * have pending delalloc work to be done.
1494 */
1bf85046
JM
1495static void btrfs_set_bit_hook(struct inode *inode,
1496 struct extent_state *state, int *bits)
291d673e 1497{
9ed74f2d 1498
75eff68e
CM
1499 /*
1500 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1501 * but in this case, we are only testing for the DELALLOC
75eff68e
CM
1502 * bit, which is only set or cleared with irqs on
1503 */
0ca1f7ce 1504 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1505 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1506 u64 len = state->end + 1 - state->start;
83eea1f1 1507 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1508
9e0baf60 1509 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1510 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB
1511 } else {
1512 spin_lock(&BTRFS_I(inode)->lock);
1513 BTRFS_I(inode)->outstanding_extents++;
1514 spin_unlock(&BTRFS_I(inode)->lock);
1515 }
287a0ab9 1516
963d678b
MX
1517 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1518 root->fs_info->delalloc_batch);
df0af1a5
MX
1519 spin_lock(&BTRFS_I(inode)->lock);
1520 BTRFS_I(inode)->delalloc_bytes += len;
1521 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1522 &BTRFS_I(inode)->runtime_flags)) {
1523 spin_lock(&root->fs_info->delalloc_lock);
1524 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1525 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1526 &root->fs_info->delalloc_inodes);
1527 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1528 &BTRFS_I(inode)->runtime_flags);
1529 }
1530 spin_unlock(&root->fs_info->delalloc_lock);
ea8c2819 1531 }
df0af1a5 1532 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1533 }
291d673e
CM
1534}
1535
d352ac68
CM
1536/*
1537 * extent_io.c clear_bit_hook, see set_bit_hook for why
1538 */
1bf85046
JM
1539static void btrfs_clear_bit_hook(struct inode *inode,
1540 struct extent_state *state, int *bits)
291d673e 1541{
75eff68e
CM
1542 /*
1543 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1544 * but in this case, we are only testing for the DELALLOC
75eff68e
CM
1545 * bit, which is only set or cleared with irqs on
1546 */
0ca1f7ce 1547 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1548 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1549 u64 len = state->end + 1 - state->start;
83eea1f1 1550 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1551
9e0baf60 1552 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1553 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB
1554 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1555 spin_lock(&BTRFS_I(inode)->lock);
1556 BTRFS_I(inode)->outstanding_extents--;
1557 spin_unlock(&BTRFS_I(inode)->lock);
1558 }
0ca1f7ce
YZ
1559
1560 if (*bits & EXTENT_DO_ACCOUNTING)
1561 btrfs_delalloc_release_metadata(inode, len);
1562
0cb59c99
JB
1563 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1564 && do_list)
0ca1f7ce 1565 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1566
963d678b
MX
1567 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1568 root->fs_info->delalloc_batch);
df0af1a5 1569 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1570 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1571 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5
MX
1572 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1573 &BTRFS_I(inode)->runtime_flags)) {
1574 spin_lock(&root->fs_info->delalloc_lock);
1575 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1576 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1577 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1578 &BTRFS_I(inode)->runtime_flags);
1579 }
1580 spin_unlock(&root->fs_info->delalloc_lock);
ea8c2819 1581 }
df0af1a5 1582 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1583 }
291d673e
CM
1584}
1585
d352ac68
CM
1586/*
1587 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1588 * we don't create bios that span stripes or chunks
1589 */
239b14b3 1590int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM
1591 size_t size, struct bio *bio,
1592 unsigned long bio_flags)
239b14b3
CM
1593{
1594 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1595 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM
1596 u64 length = 0;
1597 u64 map_length;
239b14b3
CM
1598 int ret;
1599
771ed689
CM
1600 if (bio_flags & EXTENT_BIO_COMPRESSED)
1601 return 0;
1602
f2d8d74d 1603 length = bio->bi_size;
239b14b3 1604 map_length = length;
3ec706c8 1605 ret = btrfs_map_block(root->fs_info, READ, logical,
f188591e 1606 &map_length, NULL, 0);
3ec706c8 1607 /* Will always return 0 with map_multi == NULL */
3444a972 1608 BUG_ON(ret < 0);
d397712b 1609 if (map_length < length + size)
239b14b3 1610 return 1;
3444a972 1611 return 0;
239b14b3
CM
1612}
1613
d352ac68
CM
1614/*
1615 * in order to insert checksums into the metadata in large chunks,
1616 * we wait until bio submission time. All the pages in the bio are
1617 * checksummed and sums are attached onto the ordered extent record.
1618 *
1619 * At IO completion time the cums attached on the ordered extent record
1620 * are inserted into the btree
1621 */
d397712b
CM
1622static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1623 struct bio *bio, int mirror_num,
eaf25d93
CM
1624 unsigned long bio_flags,
1625 u64 bio_offset)
065631f6 1626{
065631f6 1627 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1628 int ret = 0;
e015640f 1629
d20f7043 1630 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1631 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM
1632 return 0;
1633}
e015640f 1634
4a69a410
CM
1635/*
1636 * in order to insert checksums into the metadata in large chunks,
1637 * we wait until bio submission time. All the pages in the bio are
1638 * checksummed and sums are attached onto the ordered extent record.
1639 *
1640 * At IO completion time the cums attached on the ordered extent record
1641 * are inserted into the btree
1642 */
b2950863 1643static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM
1644 int mirror_num, unsigned long bio_flags,
1645 u64 bio_offset)
4a69a410
CM
1646{
1647 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB
1648 int ret;
1649
1650 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1651 if (ret)
1652 bio_endio(bio, ret);
1653 return ret;
44b8bd7e
CM
1654}
1655
d352ac68 1656/*
cad321ad
CM
1657 * extent_io.c submission hook. This does the right thing for csum calculation
1658 * on write, or reading the csums from the tree before a read
d352ac68 1659 */
b2950863 1660static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM
1661 int mirror_num, unsigned long bio_flags,
1662 u64 bio_offset)
44b8bd7e
CM
1663{
1664 struct btrfs_root *root = BTRFS_I(inode)->root;
1665 int ret = 0;
19b9bdb0 1666 int skip_sum;
0417341e 1667 int metadata = 0;
b812ce28 1668 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1669
6cbff00f 1670 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1671
83eea1f1 1672 if (btrfs_is_free_space_inode(inode))
0417341e
JM
1673 metadata = 2;
1674
7b6d91da 1675 if (!(rw & REQ_WRITE)) {
5fd02043
JB
1676 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1677 if (ret)
61891923 1678 goto out;
5fd02043 1679
d20f7043 1680 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB
1681 ret = btrfs_submit_compressed_read(inode, bio,
1682 mirror_num,
1683 bio_flags);
1684 goto out;
c2db1073
TI
1685 } else if (!skip_sum) {
1686 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1687 if (ret)
61891923 1688 goto out;
c2db1073 1689 }
4d1b5fb4 1690 goto mapit;
b812ce28 1691 } else if (async && !skip_sum) {
17d217fe
YZ
1692 /* csum items have already been cloned */
1693 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1694 goto mapit;
19b9bdb0 1695 /* we're doing a write, do the async checksumming */
61891923 1696 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1697 inode, rw, bio, mirror_num,
eaf25d93
CM
1698 bio_flags, bio_offset,
1699 __btrfs_submit_bio_start,
4a69a410 1700 __btrfs_submit_bio_done);
61891923 1701 goto out;
b812ce28
JB
1702 } else if (!skip_sum) {
1703 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1704 if (ret)
1705 goto out;
19b9bdb0
CM
1706 }
1707
0b86a832 1708mapit:
61891923
SB
1709 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1710
1711out:
1712 if (ret < 0)
1713 bio_endio(bio, ret);
1714 return ret;
065631f6 1715}
6885f308 1716
d352ac68
CM
1717/*
1718 * given a list of ordered sums record them in the inode. This happens
1719 * at IO completion time based on sums calculated at bio submission time.
1720 */
ba1da2f4 1721static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM
1722 struct inode *inode, u64 file_offset,
1723 struct list_head *list)
1724{
e6dcd2dc
CM
1725 struct btrfs_ordered_sum *sum;
1726
c6e30871 1727 list_for_each_entry(sum, list, list) {
d20f7043
CM
1728 btrfs_csum_file_blocks(trans,
1729 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM
1730 }
1731 return 0;
1732}
1733
2ac55d41
JB
1734int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1735 struct extent_state **cached_state)
ea8c2819 1736{
6c1500f2 1737 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1738 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1739 cached_state, GFP_NOFS);
ea8c2819
CM
1740}
1741
d352ac68 1742/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM
1743struct btrfs_writepage_fixup {
1744 struct page *page;
1745 struct btrfs_work work;
1746};
1747
b2950863 1748static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM
1749{
1750 struct btrfs_writepage_fixup *fixup;
1751 struct btrfs_ordered_extent *ordered;
2ac55d41 1752 struct extent_state *cached_state = NULL;
247e743c
CM
1753 struct page *page;
1754 struct inode *inode;
1755 u64 page_start;
1756 u64 page_end;
87826df0 1757 int ret;
247e743c
CM
1758
1759 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1760 page = fixup->page;
4a096752 1761again:
247e743c
CM
1762 lock_page(page);
1763 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1764 ClearPageChecked(page);
1765 goto out_page;
1766 }
1767
1768 inode = page->mapping->host;
1769 page_start = page_offset(page);
1770 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1771
2ac55d41 1772 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1773 &cached_state);
4a096752
CM
1774
1775 /* already ordered? We're done */
8b62b72b 1776 if (PagePrivate2(page))
247e743c 1777 goto out;
4a096752
CM
1778
1779 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1780 if (ordered) {
2ac55d41
JB
1781 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1782 page_end, &cached_state, GFP_NOFS);
4a096752
CM
1783 unlock_page(page);
1784 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1785 btrfs_put_ordered_extent(ordered);
4a096752
CM
1786 goto again;
1787 }
247e743c 1788
87826df0
JM
1789 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1790 if (ret) {
1791 mapping_set_error(page->mapping, ret);
1792 end_extent_writepage(page, ret, page_start, page_end);
1793 ClearPageChecked(page);
1794 goto out;
1795 }
1796
2ac55d41 1797 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1798 ClearPageChecked(page);
87826df0 1799 set_page_dirty(page);
247e743c 1800out:
2ac55d41
JB
1801 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1802 &cached_state, GFP_NOFS);
247e743c
CM
1803out_page:
1804 unlock_page(page);
1805 page_cache_release(page);
b897abec 1806 kfree(fixup);
247e743c
CM
1807}
1808
1809/*
1810 * There are a few paths in the higher layers of the kernel that directly
1811 * set the page dirty bit without asking the filesystem if it is a
1812 * good idea. This causes problems because we want to make sure COW
1813 * properly happens and the data=ordered rules are followed.
1814 *
c8b97818 1815 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM
1816 * hasn't been properly setup for IO. We kick off an async process
1817 * to fix it up. The async helper will wait for ordered extents, set
1818 * the delalloc bit and make it safe to write the page.
1819 */
b2950863 1820static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM
1821{
1822 struct inode *inode = page->mapping->host;
1823 struct btrfs_writepage_fixup *fixup;
1824 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1825
8b62b72b
CM
1826 /* this page is properly in the ordered list */
1827 if (TestClearPagePrivate2(page))
247e743c
CM
1828 return 0;
1829
1830 if (PageChecked(page))
1831 return -EAGAIN;
1832
1833 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1834 if (!fixup)
1835 return -EAGAIN;
f421950f 1836
247e743c
CM
1837 SetPageChecked(page);
1838 page_cache_get(page);
1839 fixup->work.func = btrfs_writepage_fixup_worker;
1840 fixup->page = page;
1841 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1842 return -EBUSY;
247e743c
CM
1843}
1844
d899e052
YZ
1845static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1846 struct inode *inode, u64 file_pos,
1847 u64 disk_bytenr, u64 disk_num_bytes,
1848 u64 num_bytes, u64 ram_bytes,
1849 u8 compression, u8 encryption,
1850 u16 other_encoding, int extent_type)
1851{
1852 struct btrfs_root *root = BTRFS_I(inode)->root;
1853 struct btrfs_file_extent_item *fi;
1854 struct btrfs_path *path;
1855 struct extent_buffer *leaf;
1856 struct btrfs_key ins;
d899e052
YZ
1857 int ret;
1858
1859 path = btrfs_alloc_path();
d8926bb3
MF
1860 if (!path)
1861 return -ENOMEM;
d899e052 1862
b9473439 1863 path->leave_spinning = 1;
a1ed835e
CM
1864
1865 /*
1866 * we may be replacing one extent in the tree with another.
1867 * The new extent is pinned in the extent map, and we don't want
1868 * to drop it from the cache until it is completely in the btree.
1869 *
1870 * So, tell btrfs_drop_extents to leave this extent in the cache.
1871 * the caller is expected to unpin it and allow it to be merged
1872 * with the others.
1873 */
5dc562c5 1874 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1875 file_pos + num_bytes, 0);
79787eaa
JM
1876 if (ret)
1877 goto out;
d899e052 1878
33345d01 1879 ins.objectid = btrfs_ino(inode);
d899e052
YZ
1880 ins.offset = file_pos;
1881 ins.type = BTRFS_EXTENT_DATA_KEY;
1882 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM
1883 if (ret)
1884 goto out;
d899e052
YZ
1885 leaf = path->nodes[0];
1886 fi = btrfs_item_ptr(leaf, path->slots[0],
1887 struct btrfs_file_extent_item);
1888 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1889 btrfs_set_file_extent_type(leaf, fi, extent_type);
1890 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1891 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1892 btrfs_set_file_extent_offset(leaf, fi, 0);
1893 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1894 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1895 btrfs_set_file_extent_compression(leaf, fi, compression);
1896 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1897 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1898
d899e052 1899 btrfs_mark_buffer_dirty(leaf);
ce195332 1900 btrfs_release_path(path);
d899e052
YZ
1901
1902 inode_add_bytes(inode, num_bytes);
d899e052
YZ
1903
1904 ins.objectid = disk_bytenr;
1905 ins.offset = disk_num_bytes;
1906 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ
1907 ret = btrfs_alloc_reserved_file_extent(trans, root,
1908 root->root_key.objectid,
33345d01 1909 btrfs_ino(inode), file_pos, &ins);
79787eaa 1910out:
d899e052 1911 btrfs_free_path(path);
b9473439 1912
79787eaa 1913 return ret;
d899e052
YZ
1914}
1915
5d13a98f
CM
1916/*
1917 * helper function for btrfs_finish_ordered_io, this
1918 * just reads in some of the csum leaves to prime them into ram
1919 * before we start the transaction. It limits the amount of btree
1920 * reads required while inside the transaction.
1921 */
d352ac68
CM
1922/* as ordered data IO finishes, this gets called so we can finish
1923 * an ordered extent if the range of bytes in the file it covers are
1924 * fully written.
1925 */
5fd02043 1926static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 1927{
5fd02043 1928 struct inode *inode = ordered_extent->inode;
e6dcd2dc 1929 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1930 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 1931 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1932 struct extent_state *cached_state = NULL;
261507a0 1933 int compress_type = 0;
e6dcd2dc 1934 int ret;
82d5902d 1935 bool nolock;
e6dcd2dc 1936
83eea1f1 1937 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 1938
5fd02043
JB
1939 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1940 ret = -EIO;
1941 goto out;
1942 }
1943
c2167754 1944 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 1945 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB
1946 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1947 if (nolock)
1948 trans = btrfs_join_transaction_nolock(root);
1949 else
1950 trans = btrfs_join_transaction(root);
1951 if (IS_ERR(trans)) {
1952 ret = PTR_ERR(trans);
1953 trans = NULL;
1954 goto out;
c2167754 1955 }
6c760c07
JB
1956 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1957 ret = btrfs_update_inode_fallback(trans, root, inode);
1958 if (ret) /* -ENOMEM or corruption */
1959 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ
1960 goto out;
1961 }
e6dcd2dc 1962
2ac55d41
JB
1963 lock_extent_bits(io_tree, ordered_extent->file_offset,
1964 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 1965 0, &cached_state);
e6dcd2dc 1966
0cb59c99 1967 if (nolock)
7a7eaa40 1968 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1969 else
7a7eaa40 1970 trans = btrfs_join_transaction(root);
79787eaa
JM
1971 if (IS_ERR(trans)) {
1972 ret = PTR_ERR(trans);
1973 trans = NULL;
1974 goto out_unlock;
1975 }
0ca1f7ce 1976 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1977
c8b97818 1978 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1979 compress_type = ordered_extent->compress_type;
d899e052 1980 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1981 BUG_ON(compress_type);
920bbbfb 1982 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ
1983 ordered_extent->file_offset,
1984 ordered_extent->file_offset +
1985 ordered_extent->len);
d899e052 1986 } else {
0af3d00b 1987 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ
1988 ret = insert_reserved_file_extent(trans, inode,
1989 ordered_extent->file_offset,
1990 ordered_extent->start,
1991 ordered_extent->disk_len,
1992 ordered_extent->len,
1993 ordered_extent->len,
261507a0 1994 compress_type, 0, 0,
d899e052 1995 BTRFS_FILE_EXTENT_REG);
d899e052 1996 }
5dc562c5
JB
1997 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1998 ordered_extent->file_offset, ordered_extent->len,
1999 trans->transid);
79787eaa
JM
2000 if (ret < 0) {
2001 btrfs_abort_transaction(trans, root, ret);
5fd02043 2002 goto out_unlock;
79787eaa 2003 }
2ac55d41 2004
e6dcd2dc
CM
2005 add_pending_csums(trans, inode, ordered_extent->file_offset,
2006 &ordered_extent->list);
2007
6c760c07
JB
2008 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2009 ret = btrfs_update_inode_fallback(trans, root, inode);
2010 if (ret) { /* -ENOMEM or corruption */
2011 btrfs_abort_transaction(trans, root, ret);
2012 goto out_unlock;
1ef30be1
JB
2013 }
2014 ret = 0;
5fd02043
JB
2015out_unlock:
2016 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2017 ordered_extent->file_offset +
2018 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2019out:
5b0e95bf 2020 if (root != root->fs_info->tree_root)
0cb59c99 2021 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX
2022 if (trans)
2023 btrfs_end_transaction(trans, root);
0cb59c99 2024
0bec9ef5 2025 if (ret) {
5fd02043
JB
2026 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2027 ordered_extent->file_offset +
2028 ordered_extent->len - 1, NULL, GFP_NOFS);
2029
0bec9ef5
JB
2030 /*
2031 * If the ordered extent had an IOERR or something else went
2032 * wrong we need to return the space for this ordered extent
2033 * back to the allocator.
2034 */
2035 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2036 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2037 btrfs_free_reserved_extent(root, ordered_extent->start,
2038 ordered_extent->disk_len);
2039 }
2040
2041
5fd02043 2042 /*
8bad3c02
LB
2043 * This needs to be done to make sure anybody waiting knows we are done
2044 * updating everything for this ordered extent.
5fd02043
JB
2045 */
2046 btrfs_remove_ordered_extent(inode, ordered_extent);
2047
e6dcd2dc
CM
2048 /* once for us */
2049 btrfs_put_ordered_extent(ordered_extent);
2050 /* once for the tree */
2051 btrfs_put_ordered_extent(ordered_extent);
2052
5fd02043
JB
2053 return ret;
2054}
2055
2056static void finish_ordered_fn(struct btrfs_work *work)
2057{
2058 struct btrfs_ordered_extent *ordered_extent;
2059 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2060 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM
2061}
2062
b2950863 2063static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM
2064 struct extent_state *state, int uptodate)
2065{
5fd02043
JB
2066 struct inode *inode = page->mapping->host;
2067 struct btrfs_root *root = BTRFS_I(inode)->root;
2068 struct btrfs_ordered_extent *ordered_extent = NULL;
2069 struct btrfs_workers *workers;
2070
1abe9b8a 2071 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2072
8b62b72b 2073 ClearPagePrivate2(page);
5fd02043
JB
2074 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2075 end - start + 1, uptodate))
2076 return 0;
2077
2078 ordered_extent->work.func = finish_ordered_fn;
2079 ordered_extent->work.flags = 0;
2080
83eea1f1 2081 if (btrfs_is_free_space_inode(inode))
5fd02043
JB
2082 workers = &root->fs_info->endio_freespace_worker;
2083 else
2084 workers = &root->fs_info->endio_write_workers;
2085 btrfs_queue_worker(workers, &ordered_extent->work);
2086
2087 return 0;
211f90e6
CM
2088}
2089
d352ac68
CM
2090/*
2091 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS
2092 * if there's a match, we allow the bio to finish. If not, the code in
2093 * extent_io.c will try to find good copies for us.
d352ac68 2094 */
b2950863 2095static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2096 struct extent_state *state, int mirror)
07157aac 2097{
4eee4fa4 2098 size_t offset = start - page_offset(page);
07157aac 2099 struct inode *inode = page->mapping->host;
d1310b2e 2100 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2101 char *kaddr;
aadfeb6e 2102 u64 private = ~(u32)0;
07157aac 2103 int ret;
ff79f819
CM
2104 struct btrfs_root *root = BTRFS_I(inode)->root;
2105 u32 csum = ~(u32)0;
d1310b2e 2106
d20f7043
CM
2107 if (PageChecked(page)) {
2108 ClearPageChecked(page);
2109 goto good;
2110 }
6cbff00f
CH
2111
2112 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2113 goto good;
17d217fe
YZ
2114
2115 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2116 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ
2117 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2118 GFP_NOFS);
b6cda9bc 2119 return 0;
17d217fe 2120 }
d20f7043 2121
c2e639f0 2122 if (state && state->start == start) {
70dec807
CM
2123 private = state->private;
2124 ret = 0;
2125 } else {
2126 ret = get_state_private(io_tree, start, &private);
2127 }
7ac687d9 2128 kaddr = kmap_atomic(page);
d397712b 2129 if (ret)
07157aac 2130 goto zeroit;
d397712b 2131
ff79f819
CM
2132 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2133 btrfs_csum_final(csum, (char *)&csum);
d397712b 2134 if (csum != private)
07157aac 2135 goto zeroit;
d397712b 2136
7ac687d9 2137 kunmap_atomic(kaddr);
d20f7043 2138good:
07157aac
CM
2139 return 0;
2140
2141zeroit:
945d8962 2142 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ
2143 "private %llu\n",
2144 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM
2145 (unsigned long long)start, csum,
2146 (unsigned long long)private);
db94535d
CM
2147 memset(kaddr + offset, 1, end - start + 1);
2148 flush_dcache_page(page);
7ac687d9 2149 kunmap_atomic(kaddr);
3b951516
CM
2150 if (private == 0)
2151 return 0;
7e38326f 2152 return -EIO;
07157aac 2153}
b888db2b 2154
24bbcf04
YZ
2155struct delayed_iput {
2156 struct list_head list;
2157 struct inode *inode;
2158};
2159
79787eaa
JM
2160/* JDM: If this is fs-wide, why can't we add a pointer to
2161 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ
2162void btrfs_add_delayed_iput(struct inode *inode)
2163{
2164 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2165 struct delayed_iput *delayed;
2166
2167 if (atomic_add_unless(&inode->i_count, -1, 1))
2168 return;
2169
2170 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2171 delayed->inode = inode;
2172
2173 spin_lock(&fs_info->delayed_iput_lock);
2174 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2175 spin_unlock(&fs_info->delayed_iput_lock);
2176}
2177
2178void btrfs_run_delayed_iputs(struct btrfs_root *root)
2179{
2180 LIST_HEAD(list);
2181 struct btrfs_fs_info *fs_info = root->fs_info;
2182 struct delayed_iput *delayed;
2183 int empty;
2184
2185 spin_lock(&fs_info->delayed_iput_lock);
2186 empty = list_empty(&fs_info->delayed_iputs);
2187 spin_unlock(&fs_info->delayed_iput_lock);
2188 if (empty)
2189 return;
2190
24bbcf04
YZ
2191 spin_lock(&fs_info->delayed_iput_lock);
2192 list_splice_init(&fs_info->delayed_iputs, &list);
2193 spin_unlock(&fs_info->delayed_iput_lock);
2194
2195 while (!list_empty(&list)) {
2196 delayed = list_entry(list.next, struct delayed_iput, list);
2197 list_del(&delayed->list);
2198 iput(delayed->inode);
2199 kfree(delayed);
2200 }
24bbcf04
YZ
2201}
2202
d68fc57b
YZ
2203enum btrfs_orphan_cleanup_state {
2204 ORPHAN_CLEANUP_STARTED = 1,
2205 ORPHAN_CLEANUP_DONE = 2,
2206};
2207
2208/*
42b2aa86 2209 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ
2210 * files in the subvolume, it removes orphan item and frees block_rsv
2211 * structure.
2212 */
2213void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2214 struct btrfs_root *root)
2215{
90290e19 2216 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ
2217 int ret;
2218
8a35d95f 2219 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ
2220 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2221 return;
2222
90290e19 2223 spin_lock(&root->orphan_lock);
8a35d95f 2224 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB
2225 spin_unlock(&root->orphan_lock);
2226 return;
2227 }
2228
2229 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2230 spin_unlock(&root->orphan_lock);
2231 return;
2232 }
2233
2234 block_rsv = root->orphan_block_rsv;
2235 root->orphan_block_rsv = NULL;
2236 spin_unlock(&root->orphan_lock);
2237
d68fc57b
YZ
2238 if (root->orphan_item_inserted &&
2239 btrfs_root_refs(&root->root_item) > 0) {
2240 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2241 root->root_key.objectid);
2242 BUG_ON(ret);
2243 root->orphan_item_inserted = 0;
2244 }
2245
90290e19
JB
2246 if (block_rsv) {
2247 WARN_ON(block_rsv->size > 0);
2248 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ
2249 }
2250}
2251
7b128766
JB
2252/*
2253 * This creates an orphan entry for the given inode in case something goes
2254 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ
2255 *
2256 * NOTE: caller of this function should reserve 5 units of metadata for
2257 * this function.
7b128766
JB
2258 */
2259int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2260{
2261 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ
2262 struct btrfs_block_rsv *block_rsv = NULL;
2263 int reserve = 0;
2264 int insert = 0;
2265 int ret;
7b128766 2266
d68fc57b 2267 if (!root->orphan_block_rsv) {
66d8f3dd 2268 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI
2269 if (!block_rsv)
2270 return -ENOMEM;
d68fc57b 2271 }
7b128766 2272
d68fc57b
YZ
2273 spin_lock(&root->orphan_lock);
2274 if (!root->orphan_block_rsv) {
2275 root->orphan_block_rsv = block_rsv;
2276 } else if (block_rsv) {
2277 btrfs_free_block_rsv(root, block_rsv);
2278 block_rsv = NULL;
7b128766 2279 }
7b128766 2280
8a35d95f
JB
2281 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2282 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ
2283#if 0
2284 /*
2285 * For proper ENOSPC handling, we should do orphan
2286 * cleanup when mounting. But this introduces backward
2287 * compatibility issue.
2288 */
2289 if (!xchg(&root->orphan_item_inserted, 1))
2290 insert = 2;
2291 else
2292 insert = 1;
2293#endif
2294 insert = 1;
321f0e70 2295 atomic_inc(&root->orphan_inodes);
7b128766
JB
2296 }
2297
72ac3c0d
JB
2298 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2299 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2300 reserve = 1;
d68fc57b 2301 spin_unlock(&root->orphan_lock);
7b128766 2302
d68fc57b
YZ
2303 /* grab metadata reservation from transaction handle */
2304 if (reserve) {
2305 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 2306 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 2307 }
7b128766 2308
d68fc57b
YZ
2309 /* insert an orphan item to track this unlinked/truncated file */
2310 if (insert >= 1) {
33345d01 2311 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2312 if (ret && ret != -EEXIST) {
8a35d95f
JB
2313 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2314 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM
2315 btrfs_abort_transaction(trans, root, ret);
2316 return ret;
2317 }
2318 ret = 0;
d68fc57b
YZ
2319 }
2320
2321 /* insert an orphan item to track subvolume contains orphan files */
2322 if (insert >= 2) {
2323 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2324 root->root_key.objectid);
79787eaa
JM
2325 if (ret && ret != -EEXIST) {
2326 btrfs_abort_transaction(trans, root, ret);
2327 return ret;
2328 }
d68fc57b
YZ
2329 }
2330 return 0;
7b128766
JB
2331}
2332
2333/*
2334 * We have done the truncate/delete so we can go ahead and remove the orphan
2335 * item for this particular inode.
2336 */
2337int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2338{
2339 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ
2340 int delete_item = 0;
2341 int release_rsv = 0;
7b128766
JB
2342 int ret = 0;
2343
d68fc57b 2344 spin_lock(&root->orphan_lock);
8a35d95f
JB
2345 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2346 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2347 delete_item = 1;
7b128766 2348
72ac3c0d
JB
2349 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2350 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2351 release_rsv = 1;
d68fc57b 2352 spin_unlock(&root->orphan_lock);
7b128766 2353
d68fc57b 2354 if (trans && delete_item) {
33345d01 2355 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2356 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 2357 }
7b128766 2358
8a35d95f 2359 if (release_rsv) {
d68fc57b 2360 btrfs_orphan_release_metadata(inode);
8a35d95f
JB
2361 atomic_dec(&root->orphan_inodes);
2362 }
7b128766 2363
d68fc57b 2364 return 0;
7b128766
JB
2365}
2366
2367/*
2368 * this cleans up any orphans that may be left on the list from the last use
2369 * of this root.
2370 */
66b4ffd1 2371int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB
2372{
2373 struct btrfs_path *path;
2374 struct extent_buffer *leaf;
7b128766
JB
2375 struct btrfs_key key, found_key;
2376 struct btrfs_trans_handle *trans;
2377 struct inode *inode;
8f6d7f4f 2378 u64 last_objectid = 0;
7b128766
JB
2379 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2380
d68fc57b 2381 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2382 return 0;
c71bf099
YZ
2383
2384 path = btrfs_alloc_path();
66b4ffd1
JB
2385 if (!path) {
2386 ret = -ENOMEM;
2387 goto out;
2388 }
7b128766
JB
2389 path->reada = -1;
2390
2391 key.objectid = BTRFS_ORPHAN_OBJECTID;
2392 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2393 key.offset = (u64)-1;
2394
7b128766
JB
2395 while (1) {
2396 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB
2397 if (ret < 0)
2398 goto out;
7b128766
JB
2399
2400 /*
2401 * if ret == 0 means we found what we were searching for, which
25985edc 2402 * is weird, but possible, so only screw with path if we didn't
7b128766
JB
2403 * find the key and see if we have stuff that matches
2404 */
2405 if (ret > 0) {
66b4ffd1 2406 ret = 0;
7b128766
JB
2407 if (path->slots[0] == 0)
2408 break;
2409 path->slots[0]--;
2410 }
2411
2412 /* pull out the item */
2413 leaf = path->nodes[0];
7b128766
JB
2414 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2415
2416 /* make sure the item matches what we want */
2417 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2418 break;
2419 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2420 break;
2421
2422 /* release the path since we're done with it */
b3b4aa74 2423 btrfs_release_path(path);
7b128766
JB
2424
2425 /*
2426 * this is where we are basically btrfs_lookup, without the
2427 * crossing root thing. we store the inode number in the
2428 * offset of the orphan item.
2429 */
8f6d7f4f
JB
2430
2431 if (found_key.offset == last_objectid) {
2432 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2433 "stopping orphan cleanup\n");
2434 ret = -EINVAL;
2435 goto out;
2436 }
2437
2438 last_objectid = found_key.offset;
2439
5d4f98a2
YZ
2440 found_key.objectid = found_key.offset;
2441 found_key.type = BTRFS_INODE_ITEM_KEY;
2442 found_key.offset = 0;
73f73415 2443 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB
2444 ret = PTR_RET(inode);
2445 if (ret && ret != -ESTALE)
66b4ffd1 2446 goto out;
7b128766 2447
f8e9e0b0
AJ
2448 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2449 struct btrfs_root *dead_root;
2450 struct btrfs_fs_info *fs_info = root->fs_info;
2451 int is_dead_root = 0;
2452
2453 /*
2454 * this is an orphan in the tree root. Currently these
2455 * could come from 2 sources:
2456 * a) a snapshot deletion in progress
2457 * b) a free space cache inode
2458 * We need to distinguish those two, as the snapshot
2459 * orphan must not get deleted.
2460 * find_dead_roots already ran before us, so if this
2461 * is a snapshot deletion, we should find the root
2462 * in the dead_roots list
2463 */
2464 spin_lock(&fs_info->trans_lock);
2465 list_for_each_entry(dead_root, &fs_info->dead_roots,
2466 root_list) {
2467 if (dead_root->root_key.objectid ==
2468 found_key.objectid) {
2469 is_dead_root = 1;
2470 break;
2471 }
2472 }
2473 spin_unlock(&fs_info->trans_lock);
2474 if (is_dead_root) {
2475 /* prevent this orphan from being found again */
2476 key.offset = found_key.objectid - 1;
2477 continue;
2478 }
2479 }
7b128766 2480 /*
a8c9e576
JB
2481 * Inode is already gone but the orphan item is still there,
2482 * kill the orphan item.
7b128766 2483 */
a8c9e576
JB
2484 if (ret == -ESTALE) {
2485 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB
2486 if (IS_ERR(trans)) {
2487 ret = PTR_ERR(trans);
2488 goto out;
2489 }
8a35d95f
JB
2490 printk(KERN_ERR "auto deleting %Lu\n",
2491 found_key.objectid);
a8c9e576
JB
2492 ret = btrfs_del_orphan_item(trans, root,
2493 found_key.objectid);
79787eaa 2494 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 2495 btrfs_end_transaction(trans, root);
7b128766
JB
2496 continue;
2497 }
2498
a8c9e576
JB
2499 /*
2500 * add this inode to the orphan list so btrfs_orphan_del does
2501 * the proper thing when we hit it
2502 */
8a35d95f
JB
2503 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2504 &BTRFS_I(inode)->runtime_flags);
a8c9e576 2505
7b128766
JB
2506 /* if we have links, this was a truncate, lets do that */
2507 if (inode->i_nlink) {
a41ad394
JB
2508 if (!S_ISREG(inode->i_mode)) {
2509 WARN_ON(1);
2510 iput(inode);
2511 continue;
2512 }
7b128766 2513 nr_truncate++;
f3fe820c
JB
2514
2515 /* 1 for the orphan item deletion. */
2516 trans = btrfs_start_transaction(root, 1);
2517 if (IS_ERR(trans)) {
2518 ret = PTR_ERR(trans);
2519 goto out;
2520 }
2521 ret = btrfs_orphan_add(trans, inode);
2522 btrfs_end_transaction(trans, root);
2523 if (ret)
2524 goto out;
2525
66b4ffd1 2526 ret = btrfs_truncate(inode);
7b128766
JB
2527 } else {
2528 nr_unlink++;
2529 }
2530
2531 /* this will do delete_inode and everything for us */
2532 iput(inode);
66b4ffd1
JB
2533 if (ret)
2534 goto out;
7b128766 2535 }
3254c876
MX
2536 /* release the path since we're done with it */
2537 btrfs_release_path(path);
2538
d68fc57b
YZ
2539 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2540
2541 if (root->orphan_block_rsv)
2542 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2543 (u64)-1);
2544
2545 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2546 trans = btrfs_join_transaction(root);
66b4ffd1
JB
2547 if (!IS_ERR(trans))
2548 btrfs_end_transaction(trans, root);
d68fc57b 2549 }
7b128766
JB
2550
2551 if (nr_unlink)
2552 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2553 if (nr_truncate)
2554 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB
2555
2556out:
2557 if (ret)
2558 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2559 btrfs_free_path(path);
2560 return ret;
7b128766
JB
2561}
2562
46a53cca
CM
2563/*
2564 * very simple check to peek ahead in the leaf looking for xattrs. If we
2565 * don't find any xattrs, we know there can't be any acls.
2566 *
2567 * slot is the slot the inode is in, objectid is the objectid of the inode
2568 */
2569static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2570 int slot, u64 objectid)
2571{
2572 u32 nritems = btrfs_header_nritems(leaf);
2573 struct btrfs_key found_key;
2574 int scanned = 0;
2575
2576 slot++;
2577 while (slot < nritems) {
2578 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2579
2580 /* we found a different objectid, there must not be acls */
2581 if (found_key.objectid != objectid)
2582 return 0;
2583
2584 /* we found an xattr, assume we've got an acl */
2585 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2586 return 1;
2587
2588 /*
2589 * we found a key greater than an xattr key, there can't
2590 * be any acls later on
2591 */
2592 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2593 return 0;
2594
2595 slot++;
2596 scanned++;
2597
2598 /*
2599 * it goes inode, inode backrefs, xattrs, extents,
2600 * so if there are a ton of hard links to an inode there can
2601 * be a lot of backrefs. Don't waste time searching too hard,
2602 * this is just an optimization
2603 */
2604 if (scanned >= 8)
2605 break;
2606 }
2607 /* we hit the end of the leaf before we found an xattr or
2608 * something larger than an xattr. We have to assume the inode
2609 * has acls
2610 */
2611 return 1;
2612}
2613
d352ac68
CM
2614/*
2615 * read an inode from the btree into the in-memory inode
2616 */
5d4f98a2 2617static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM
2618{
2619 struct btrfs_path *path;
5f39d397 2620 struct extent_buffer *leaf;
39279cc3 2621 struct btrfs_inode_item *inode_item;
0b86a832 2622 struct btrfs_timespec *tspec;
39279cc3
CM
2623 struct btrfs_root *root = BTRFS_I(inode)->root;
2624 struct btrfs_key location;
46a53cca 2625 int maybe_acls;
618e21d5 2626 u32 rdev;
39279cc3 2627 int ret;
2f7e33d4
MX
2628 bool filled = false;
2629
2630 ret = btrfs_fill_inode(inode, &rdev);
2631 if (!ret)
2632 filled = true;
39279cc3
CM
2633
2634 path = btrfs_alloc_path();
1748f843
MF
2635 if (!path)
2636 goto make_bad;
2637
d90c7321 2638 path->leave_spinning = 1;
39279cc3 2639 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2640
39279cc3 2641 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2642 if (ret)
39279cc3 2643 goto make_bad;
39279cc3 2644
5f39d397 2645 leaf = path->nodes[0];
2f7e33d4
MX
2646
2647 if (filled)
2648 goto cache_acl;
2649
5f39d397
CM
2650 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2651 struct btrfs_inode_item);
5f39d397 2652 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 2653 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB
2654 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2655 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 2656 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM
2657
2658 tspec = btrfs_inode_atime(inode_item);
2659 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2660 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2661
2662 tspec = btrfs_inode_mtime(inode_item);
2663 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2664 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2665
2666 tspec = btrfs_inode_ctime(inode_item);
2667 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2668 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2669
a76a3cd4 2670 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2671 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB
2672 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2673
2674 /*
2675 * If we were modified in the current generation and evicted from memory
2676 * and then re-read we need to do a full sync since we don't have any
2677 * idea about which extents were modified before we were evicted from
2678 * cache.
2679 */
2680 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2681 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2682 &BTRFS_I(inode)->runtime_flags);
2683
0c4d2d95 2684 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2685 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2686 inode->i_rdev = 0;
5f39d397
CM
2687 rdev = btrfs_inode_rdev(leaf, inode_item);
2688
aec7477b 2689 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2690 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2691cache_acl:
46a53cca
CM
2692 /*
2693 * try to precache a NULL acl entry for files that don't have
2694 * any xattrs or acls
2695 */
33345d01
LZ
2696 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2697 btrfs_ino(inode));
72c04902
AV
2698 if (!maybe_acls)
2699 cache_no_acl(inode);
46a53cca 2700
39279cc3 2701 btrfs_free_path(path);
39279cc3 2702
39279cc3 2703 switch (inode->i_mode & S_IFMT) {
39279cc3
CM
2704 case S_IFREG:
2705 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2706 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2707 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM
2708 inode->i_fop = &btrfs_file_operations;
2709 inode->i_op = &btrfs_file_inode_operations;
2710 break;
2711 case S_IFDIR:
2712 inode->i_fop = &btrfs_dir_file_operations;
2713 if (root == root->fs_info->tree_root)
2714 inode->i_op = &btrfs_dir_ro_inode_operations;
2715 else
2716 inode->i_op = &btrfs_dir_inode_operations;
2717 break;
2718 case S_IFLNK:
2719 inode->i_op = &btrfs_symlink_inode_operations;
2720 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2721 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2722 break;
618e21d5 2723 default:
0279b4cd 2724 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB
2725 init_special_inode(inode, inode->i_mode, rdev);
2726 break;
39279cc3 2727 }
6cbff00f
CH
2728
2729 btrfs_update_iflags(inode);
39279cc3
CM
2730 return;
2731
2732make_bad:
39279cc3 2733 btrfs_free_path(path);
39279cc3
CM
2734 make_bad_inode(inode);
2735}
2736
d352ac68
CM
2737/*
2738 * given a leaf and an inode, copy the inode fields into the leaf
2739 */
e02119d5
CM
2740static void fill_inode_item(struct btrfs_trans_handle *trans,
2741 struct extent_buffer *leaf,
5f39d397 2742 struct btrfs_inode_item *item,
39279cc3
CM
2743 struct inode *inode)
2744{
51fab693
LB
2745 struct btrfs_map_token token;
2746
2747 btrfs_init_map_token(&token);
2748
2749 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2750 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2751 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
2752 &token);
2753 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2754 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2755
2756 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2757 inode->i_atime.tv_sec, &token);
2758 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2759 inode->i_atime.tv_nsec, &token);
2760
2761 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2762 inode->i_mtime.tv_sec, &token);
2763 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2764 inode->i_mtime.tv_nsec, &token);
2765
2766 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2767 inode->i_ctime.tv_sec, &token);
2768 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2769 inode->i_ctime.tv_nsec, &token);
2770
2771 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2772 &token);
2773 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
2774 &token);
2775 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2776 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2777 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
2778 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
2779 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM
2780}
2781
d352ac68
CM
2782/*
2783 * copy everything in the in-memory inode into the btree.
2784 */
2115133f 2785static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2786 struct btrfs_root *root, struct inode *inode)
39279cc3
CM
2787{
2788 struct btrfs_inode_item *inode_item;
2789 struct btrfs_path *path;
5f39d397 2790 struct extent_buffer *leaf;
39279cc3
CM
2791 int ret;
2792
2793 path = btrfs_alloc_path();
16cdcec7
MX
2794 if (!path)
2795 return -ENOMEM;
2796
b9473439 2797 path->leave_spinning = 1;
16cdcec7
MX
2798 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2799 1);
39279cc3
CM
2800 if (ret) {
2801 if (ret > 0)
2802 ret = -ENOENT;
2803 goto failed;
2804 }
2805
b4ce94de 2806 btrfs_unlock_up_safe(path, 1);
5f39d397
CM
2807 leaf = path->nodes[0];
2808 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2809 struct btrfs_inode_item);
39279cc3 2810
e02119d5 2811 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2812 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2813 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM
2814 ret = 0;
2815failed:
39279cc3
CM
2816 btrfs_free_path(path);
2817 return ret;
2818}
2819
2115133f
CM
2820/*
2821 * copy everything in the in-memory inode into the btree.
2822 */
2823noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2824 struct btrfs_root *root, struct inode *inode)
2825{
2826 int ret;
2827
2828 /*
2829 * If the inode is a free space inode, we can deadlock during commit
2830 * if we put it into the delayed code.
2831 *
2832 * The data relocation inode should also be directly updated
2833 * without delay
2834 */
83eea1f1 2835 if (!btrfs_is_free_space_inode(inode)
2115133f 2836 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB
2837 btrfs_update_root_times(trans, root);
2838
2115133f
CM
2839 ret = btrfs_delayed_update_inode(trans, root, inode);
2840 if (!ret)
2841 btrfs_set_inode_last_trans(trans, inode);
2842 return ret;
2843 }
2844
2845 return btrfs_update_inode_item(trans, root, inode);
2846}
2847
be6aef60
JB
2848noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2849 struct btrfs_root *root,
2850 struct inode *inode)
2115133f
CM
2851{
2852 int ret;
2853
2854 ret = btrfs_update_inode(trans, root, inode);
2855 if (ret == -ENOSPC)
2856 return btrfs_update_inode_item(trans, root, inode);
2857 return ret;
2858}
2859
d352ac68
CM
2860/*
2861 * unlink helper that gets used here in inode.c and in the tree logging
2862 * recovery code. It remove a link in a directory with a given name, and
2863 * also drops the back refs in the inode to the directory
2864 */
92986796
AV
2865static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2866 struct btrfs_root *root,
2867 struct inode *dir, struct inode *inode,
2868 const char *name, int name_len)
39279cc3
CM
2869{
2870 struct btrfs_path *path;
39279cc3 2871 int ret = 0;
5f39d397 2872 struct extent_buffer *leaf;
39279cc3 2873 struct btrfs_dir_item *di;
5f39d397 2874 struct btrfs_key key;
aec7477b 2875 u64 index;
33345d01
LZ
2876 u64 ino = btrfs_ino(inode);
2877 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM
2878
2879 path = btrfs_alloc_path();
54aa1f4d
CM
2880 if (!path) {
2881 ret = -ENOMEM;
554233a6 2882 goto out;
54aa1f4d
CM
2883 }
2884
b9473439 2885 path->leave_spinning = 1;
33345d01 2886 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM
2887 name, name_len, -1);
2888 if (IS_ERR(di)) {
2889 ret = PTR_ERR(di);
2890 goto err;
2891 }
2892 if (!di) {
2893 ret = -ENOENT;
2894 goto err;
2895 }
5f39d397
CM
2896 leaf = path->nodes[0];
2897 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2898 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM
2899 if (ret)
2900 goto err;
b3b4aa74 2901 btrfs_release_path(path);
39279cc3 2902
33345d01
LZ
2903 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2904 dir_ino, &index);
aec7477b 2905 if (ret) {
d397712b 2906 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ
2907 "inode %llu parent %llu\n", name_len, name,
2908 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 2909 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB
2910 goto err;
2911 }
2912
16cdcec7 2913 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM
2914 if (ret) {
2915 btrfs_abort_transaction(trans, root, ret);
39279cc3 2916 goto err;
79787eaa 2917 }
39279cc3 2918
e02119d5 2919 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2920 inode, dir_ino);
79787eaa
JM
2921 if (ret != 0 && ret != -ENOENT) {
2922 btrfs_abort_transaction(trans, root, ret);
2923 goto err;
2924 }
e02119d5
CM
2925
2926 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2927 dir, index);
6418c961
CM
2928 if (ret == -ENOENT)
2929 ret = 0;
39279cc3
CM
2930err:
2931 btrfs_free_path(path);
e02119d5
CM
2932 if (ret)
2933 goto out;
2934
2935 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB
2936 inode_inc_iversion(inode);
2937 inode_inc_iversion(dir);
e02119d5 2938 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 2939 ret = btrfs_update_inode(trans, root, dir);
e02119d5 2940out:
39279cc3
CM
2941 return ret;
2942}
2943
92986796
AV
2944int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2945 struct btrfs_root *root,
2946 struct inode *dir, struct inode *inode,
2947 const char *name, int name_len)
2948{
2949 int ret;
2950 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2951 if (!ret) {
2952 btrfs_drop_nlink(inode);
2953 ret = btrfs_update_inode(trans, root, inode);
2954 }
2955 return ret;
2956}
2957
2958
a22285a6
YZ
2959/* helper to check if there is any shared block in the path */
2960static int check_path_shared(struct btrfs_root *root,
2961 struct btrfs_path *path)
39279cc3 2962{
a22285a6
YZ
2963 struct extent_buffer *eb;
2964 int level;
0e4dcbef 2965 u64 refs = 1;
5df6a9f6 2966
a22285a6 2967 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB
2968 int ret;
2969
a22285a6
YZ
2970 if (!path->nodes[level])
2971 break;
2972 eb = path->nodes[level];
2973 if (!btrfs_block_can_be_shared(root, eb))
2974 continue;
2975 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2976 &refs, NULL);
2977 if (refs > 1)
2978 return 1;
5df6a9f6 2979 }
dedefd72 2980 return 0;
39279cc3
CM
2981}
2982
a22285a6
YZ
2983/*
2984 * helper to start transaction for unlink and rmdir.
2985 *
2986 * unlink and rmdir are special in btrfs, they do not always free space.
2987 * so in enospc case, we should make sure they will free space before
2988 * allowing them to use the global metadata reservation.
2989 */
2990static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2991 struct dentry *dentry)
4df27c4d 2992{
39279cc3 2993 struct btrfs_trans_handle *trans;
a22285a6 2994 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2995 struct btrfs_path *path;
4df27c4d 2996 struct btrfs_dir_item *di;
7b128766 2997 struct inode *inode = dentry->d_inode;
4df27c4d 2998 u64 index;
a22285a6
YZ
2999 int check_link = 1;
3000 int err = -ENOSPC;
4df27c4d 3001 int ret;
33345d01
LZ
3002 u64 ino = btrfs_ino(inode);
3003 u64 dir_ino = btrfs_ino(dir);
4df27c4d 3004
e70bea5f
JB
3005 /*
3006 * 1 for the possible orphan item
3007 * 1 for the dir item
3008 * 1 for the dir index
3009 * 1 for the inode ref
3010 * 1 for the inode ref in the tree log
3011 * 2 for the dir entries in the log
3012 * 1 for the inode
3013 */
3014 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ
3015 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3016 return trans;
4df27c4d 3017
33345d01 3018 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 3019 return ERR_PTR(-ENOSPC);
4df27c4d 3020
a22285a6
YZ
3021 /* check if there is someone else holds reference */
3022 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3023 return ERR_PTR(-ENOSPC);
4df27c4d 3024
a22285a6
YZ
3025 if (atomic_read(&inode->i_count) > 2)
3026 return ERR_PTR(-ENOSPC);
4df27c4d 3027
a22285a6
YZ
3028 if (xchg(&root->fs_info->enospc_unlink, 1))
3029 return ERR_PTR(-ENOSPC);
3030
3031 path = btrfs_alloc_path();
3032 if (!path) {
3033 root->fs_info->enospc_unlink = 0;
3034 return ERR_PTR(-ENOMEM);
4df27c4d
YZ
3035 }
3036
3880a1b4
JB
3037 /* 1 for the orphan item */
3038 trans = btrfs_start_transaction(root, 1);
5df6a9f6 3039 if (IS_ERR(trans)) {
a22285a6
YZ
3040 btrfs_free_path(path);
3041 root->fs_info->enospc_unlink = 0;
3042 return trans;
3043 }
4df27c4d 3044
a22285a6
YZ
3045 path->skip_locking = 1;
3046 path->search_commit_root = 1;
4df27c4d 3047
a22285a6
YZ
3048 ret = btrfs_lookup_inode(trans, root, path,
3049 &BTRFS_I(dir)->location, 0);
3050 if (ret < 0) {
3051 err = ret;
3052 goto out;
3053 }
3054 if (ret == 0) {
3055 if (check_path_shared(root, path))
3056 goto out;
3057 } else {
3058 check_link = 0;
5df6a9f6 3059 }
b3b4aa74 3060 btrfs_release_path(path);
a22285a6
YZ
3061
3062 ret = btrfs_lookup_inode(trans, root, path,
3063 &BTRFS_I(inode)->location, 0);
3064 if (ret < 0) {
3065 err = ret;
3066 goto out;
3067 }
3068 if (ret == 0) {
3069 if (check_path_shared(root, path))
3070 goto out;
3071 } else {
3072 check_link = 0;
3073 }
b3b4aa74 3074 btrfs_release_path(path);
a22285a6
YZ
3075
3076 if (ret == 0 && S_ISREG(inode->i_mode)) {
3077 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 3078 ino, (u64)-1, 0);
a22285a6
YZ
3079 if (ret < 0) {
3080 err = ret;
3081 goto out;
3082 }
79787eaa 3083 BUG_ON(ret == 0); /* Corruption */
a22285a6
YZ
3084 if (check_path_shared(root, path))
3085 goto out;
b3b4aa74 3086 btrfs_release_path(path);
a22285a6
YZ
3087 }
3088
3089 if (!check_link) {
3090 err = 0;
3091 goto out;
3092 }
3093
33345d01 3094 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ
3095 dentry->d_name.name, dentry->d_name.len, 0);
3096 if (IS_ERR(di)) {
3097 err = PTR_ERR(di);
3098 goto out;
3099 }
3100 if (di) {
3101 if (check_path_shared(root, path))
3102 goto out;
3103 } else {
3104 err = 0;
3105 goto out;
3106 }
b3b4aa74 3107 btrfs_release_path(path);
a22285a6 3108
f186373f
MF
3109 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3110 dentry->d_name.len, ino, dir_ino, 0,
3111 &index);
3112 if (ret) {
3113 err = ret;
a22285a6
YZ
3114 goto out;
3115 }
f186373f 3116
a22285a6
YZ
3117 if (check_path_shared(root, path))
3118 goto out;
f186373f 3119
b3b4aa74 3120 btrfs_release_path(path);
a22285a6 3121
16cdcec7
MX
3122 /*
3123 * This is a commit root search, if we can lookup inode item and other
3124 * relative items in the commit root, it means the transaction of
3125 * dir/file creation has been committed, and the dir index item that we
3126 * delay to insert has also been inserted into the commit root. So
3127 * we needn't worry about the delayed insertion of the dir index item
3128 * here.
3129 */
33345d01 3130 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ
3131 dentry->d_name.name, dentry->d_name.len, 0);
3132 if (IS_ERR(di)) {
3133 err = PTR_ERR(di);
3134 goto out;
3135 }
3136 BUG_ON(ret == -ENOENT);
3137 if (check_path_shared(root, path))
3138 goto out;
3139
3140 err = 0;
3141out:
3142 btrfs_free_path(path);
3880a1b4
JB
3143 /* Migrate the orphan reservation over */
3144 if (!err)
3145 err = btrfs_block_rsv_migrate(trans->block_rsv,
3146 &root->fs_info->global_block_rsv,
5a77d76c 3147 trans->bytes_reserved);
3880a1b4 3148
a22285a6
YZ
3149 if (err) {
3150 btrfs_end_transaction(trans, root);
3151 root->fs_info->enospc_unlink = 0;
3152 return ERR_PTR(err);
3153 }
3154
3155 trans->block_rsv = &root->fs_info->global_block_rsv;
3156 return trans;
3157}
3158
3159static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3160 struct btrfs_root *root)
3161{
66d8f3dd 3162 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
5a77d76c
JB
3163 btrfs_block_rsv_release(root, trans->block_rsv,
3164 trans->bytes_reserved);
3165 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ
3166 BUG_ON(!root->fs_info->enospc_unlink);
3167 root->fs_info->enospc_unlink = 0;
3168 }
7ad85bb7 3169 btrfs_end_transaction(trans, root);
a22285a6
YZ
3170}
3171
3172static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3173{
3174 struct btrfs_root *root = BTRFS_I(dir)->root;
3175 struct btrfs_trans_handle *trans;
3176 struct inode *inode = dentry->d_inode;
3177 int ret;
a22285a6
YZ
3178
3179 trans = __unlink_start_trans(dir, dentry);
3180 if (IS_ERR(trans))
3181 return PTR_ERR(trans);
5f39d397 3182
12fcfd22
CM
3183 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3184
e02119d5
CM
3185 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3186 dentry->d_name.name, dentry->d_name.len);
b532402e
TI
3187 if (ret)
3188 goto out;
7b128766 3189
a22285a6 3190 if (inode->i_nlink == 0) {
7b128766 3191 ret = btrfs_orphan_add(trans, inode);
b532402e
TI
3192 if (ret)
3193 goto out;
a22285a6 3194 }
7b128766 3195
b532402e 3196out:
a22285a6 3197 __unlink_end_trans(trans, root);
b53d3f5d 3198 btrfs_btree_balance_dirty(root);
39279cc3
CM
3199 return ret;
3200}
3201
4df27c4d
YZ
3202int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3203 struct btrfs_root *root,
3204 struct inode *dir, u64 objectid,
3205 const char *name, int name_len)
3206{
3207 struct btrfs_path *path;
3208 struct extent_buffer *leaf;
3209 struct btrfs_dir_item *di;
3210 struct btrfs_key key;
3211 u64 index;
3212 int ret;
33345d01 3213 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ
3214
3215 path = btrfs_alloc_path();
3216 if (!path)
3217 return -ENOMEM;
3218
33345d01 3219 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3220 name, name_len, -1);
79787eaa
JM
3221 if (IS_ERR_OR_NULL(di)) {
3222 if (!di)
3223 ret = -ENOENT;
3224 else
3225 ret = PTR_ERR(di);
3226 goto out;
3227 }
4df27c4d
YZ
3228
3229 leaf = path->nodes[0];
3230 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3231 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3232 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM
3233 if (ret) {
3234 btrfs_abort_transaction(trans, root, ret);
3235 goto out;
3236 }
b3b4aa74 3237 btrfs_release_path(path);
4df27c4d
YZ
3238
3239 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3240 objectid, root->root_key.objectid,
33345d01 3241 dir_ino, &index, name, name_len);
4df27c4d 3242 if (ret < 0) {
79787eaa
JM
3243 if (ret != -ENOENT) {
3244 btrfs_abort_transaction(trans, root, ret);
3245 goto out;
3246 }
33345d01 3247 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3248 name, name_len);
79787eaa
JM
3249 if (IS_ERR_OR_NULL(di)) {
3250 if (!di)
3251 ret = -ENOENT;
3252 else
3253 ret = PTR_ERR(di);
3254 btrfs_abort_transaction(trans, root, ret);
3255 goto out;
3256 }
4df27c4d
YZ
3257
3258 leaf = path->nodes[0];
3259 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3260 btrfs_release_path(path);
4df27c4d
YZ
3261 index = key.offset;
3262 }
945d8962 3263 btrfs_release_path(path);
4df27c4d 3264
16cdcec7 3265 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM
3266 if (ret) {
3267 btrfs_abort_transaction(trans, root, ret);
3268 goto out;
3269 }
4df27c4d
YZ
3270
3271 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3272 inode_inc_iversion(dir);
4df27c4d 3273 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3274 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM
3275 if (ret)
3276 btrfs_abort_transaction(trans, root, ret);
3277out:
71d7aed0 3278 btrfs_free_path(path);
79787eaa 3279 return ret;
4df27c4d
YZ
3280}
3281
39279cc3
CM
3282static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3283{
3284 struct inode *inode = dentry->d_inode;
1832a6d5 3285 int err = 0;
39279cc3 3286 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3287 struct btrfs_trans_handle *trans;
39279cc3 3288
b3ae244e 3289 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3290 return -ENOTEMPTY;
b3ae244e
DS
3291 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3292 return -EPERM;
134d4512 3293
a22285a6
YZ
3294 trans = __unlink_start_trans(dir, dentry);
3295 if (IS_ERR(trans))
5df6a9f6 3296 return PTR_ERR(trans);
5df6a9f6 3297
33345d01 3298 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ
3299 err = btrfs_unlink_subvol(trans, root, dir,
3300 BTRFS_I(inode)->location.objectid,
3301 dentry->d_name.name,
3302 dentry->d_name.len);
3303 goto out;
3304 }
3305
7b128766
JB
3306 err = btrfs_orphan_add(trans, inode);
3307 if (err)
4df27c4d 3308 goto out;
7b128766 3309
39279cc3 3310 /* now the directory is empty */
e02119d5
CM
3311 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3312 dentry->d_name.name, dentry->d_name.len);
d397712b 3313 if (!err)
dbe674a9 3314 btrfs_i_size_write(inode, 0);
4df27c4d 3315out:
a22285a6 3316 __unlink_end_trans(trans, root);
b53d3f5d 3317 btrfs_btree_balance_dirty(root);
3954401f 3318
39279cc3
CM
3319 return err;
3320}
3321
39279cc3
CM
3322/*
3323 * this can truncate away extent items, csum items and directory items.
3324 * It starts at a high offset and removes keys until it can't find
d352ac68 3325 * any higher than new_size
39279cc3
CM
3326 *
3327 * csum items that cross the new i_size are truncated to the new size
3328 * as well.
7b128766
JB
3329 *
3330 * min_type is the minimum key type to truncate down to. If set to 0, this
3331 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3332 */
8082510e
YZ
3333int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3334 struct btrfs_root *root,
3335 struct inode *inode,
3336 u64 new_size, u32 min_type)
39279cc3 3337{
39279cc3 3338 struct btrfs_path *path;
5f39d397 3339 struct extent_buffer *leaf;
39279cc3 3340 struct btrfs_file_extent_item *fi;
8082510e
YZ
3341 struct btrfs_key key;
3342 struct btrfs_key found_key;
39279cc3 3343 u64 extent_start = 0;
db94535d 3344 u64 extent_num_bytes = 0;
5d4f98a2 3345 u64 extent_offset = 0;
39279cc3 3346 u64 item_end = 0;
8082510e
YZ
3347 u64 mask = root->sectorsize - 1;
3348 u32 found_type = (u8)-1;
39279cc3
CM
3349 int found_extent;
3350 int del_item;
85e21bac
CM
3351 int pending_del_nr = 0;
3352 int pending_del_slot = 0;
179e29e4 3353 int extent_type = -1;
8082510e
YZ
3354 int ret;
3355 int err = 0;
33345d01 3356 u64 ino = btrfs_ino(inode);
8082510e
YZ
3357
3358 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3359
0eb0e19c
MF
3360 path = btrfs_alloc_path();
3361 if (!path)
3362 return -ENOMEM;
3363 path->reada = -1;
3364
5dc562c5
JB
3365 /*
3366 * We want to drop from the next block forward in case this new size is
3367 * not block aligned since we will be keeping the last block of the
3368 * extent just the way it is.
3369 */
0af3d00b 3370 if (root->ref_cows || root == root->fs_info->tree_root)
5dc562c5 3371 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
8082510e 3372
16cdcec7
MX
3373 /*
3374 * This function is also used to drop the items in the log tree before
3375 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3376 * it is used to drop the loged items. So we shouldn't kill the delayed
3377 * items.
3378 */
3379 if (min_type == 0 && root == BTRFS_I(inode)->root)
3380 btrfs_kill_delayed_inode_items(inode);
3381
33345d01 3382 key.objectid = ino;
39279cc3 3383 key.offset = (u64)-1;
5f39d397
CM
3384 key.type = (u8)-1;
3385
85e21bac 3386search_again:
b9473439 3387 path->leave_spinning = 1;
85e21bac 3388 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ
3389 if (ret < 0) {
3390 err = ret;
3391 goto out;
3392 }
d397712b 3393
85e21bac 3394 if (ret > 0) {
e02119d5
CM
3395 /* there are no items in the tree for us to truncate, we're
3396 * done
3397 */
8082510e
YZ
3398 if (path->slots[0] == 0)
3399 goto out;
85e21bac
CM
3400 path->slots[0]--;
3401 }
3402
d397712b 3403 while (1) {
39279cc3 3404 fi = NULL;
5f39d397
CM
3405 leaf = path->nodes[0];
3406 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3407 found_type = btrfs_key_type(&found_key);
39279cc3 3408
33345d01 3409 if (found_key.objectid != ino)
39279cc3 3410 break;
5f39d397 3411
85e21bac 3412 if (found_type < min_type)
39279cc3
CM
3413 break;
3414
5f39d397 3415 item_end = found_key.offset;
39279cc3 3416 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3417 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3418 struct btrfs_file_extent_item);
179e29e4
CM
3419 extent_type = btrfs_file_extent_type(leaf, fi);
3420 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3421 item_end +=
db94535d 3422 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3423 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3424 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3425 fi);
39279cc3 3426 }
008630c1 3427 item_end--;
39279cc3 3428 }
8082510e
YZ
3429 if (found_type > min_type) {
3430 del_item = 1;
3431 } else {
3432 if (item_end < new_size)
b888db2b 3433 break;
8082510e
YZ
3434 if (found_key.offset >= new_size)
3435 del_item = 1;
3436 else
3437 del_item = 0;
39279cc3 3438 }
39279cc3 3439 found_extent = 0;
39279cc3 3440 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM
3441 if (found_type != BTRFS_EXTENT_DATA_KEY)
3442 goto delete;
3443
3444 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3445 u64 num_dec;
db94535d 3446 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3447 if (!del_item) {
db94535d
CM
3448 u64 orig_num_bytes =
3449 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3450 extent_num_bytes = new_size -
5f39d397 3451 found_key.offset + root->sectorsize - 1;
b1632b10
Y
3452 extent_num_bytes = extent_num_bytes &
3453 ~((u64)root->sectorsize - 1);
db94535d
CM
3454 btrfs_set_file_extent_num_bytes(leaf, fi,
3455 extent_num_bytes);
3456 num_dec = (orig_num_bytes -
9069218d 3457 extent_num_bytes);
e02119d5 3458 if (root->ref_cows && extent_start != 0)
a76a3cd4 3459 inode_sub_bytes(inode, num_dec);
5f39d397 3460 btrfs_mark_buffer_dirty(leaf);
39279cc3 3461 } else {
db94535d
CM
3462 extent_num_bytes =
3463 btrfs_file_extent_disk_num_bytes(leaf,
3464 fi);
5d4f98a2
YZ
3465 extent_offset = found_key.offset -
3466 btrfs_file_extent_offset(leaf, fi);
3467
39279cc3 3468 /* FIXME blocksize != 4096 */
9069218d 3469 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM
3470 if (extent_start != 0) {
3471 found_extent = 1;
e02119d5 3472 if (root->ref_cows)
a76a3cd4 3473 inode_sub_bytes(inode, num_dec);
e02119d5 3474 }
39279cc3 3475 }
9069218d 3476 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM
3477 /*
3478 * we can't truncate inline items that have had
3479 * special encodings
3480 */
3481 if (!del_item &&
3482 btrfs_file_extent_compression(leaf, fi) == 0 &&
3483 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3484 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM
3485 u32 size = new_size - found_key.offset;
3486
3487 if (root->ref_cows) {
a76a3cd4
YZ
3488 inode_sub_bytes(inode, item_end + 1 -
3489 new_size);
e02119d5
CM
3490 }
3491 size =
3492 btrfs_file_extent_calc_inline_size(size);
143bede5
JM
3493 btrfs_truncate_item(trans, root, path,
3494 size, 1);
e02119d5 3495 } else if (root->ref_cows) {
a76a3cd4
YZ
3496 inode_sub_bytes(inode, item_end + 1 -
3497 found_key.offset);
9069218d 3498 }
39279cc3 3499 }
179e29e4 3500delete:
39279cc3 3501 if (del_item) {
85e21bac
CM
3502 if (!pending_del_nr) {
3503 /* no pending yet, add ourselves */
3504 pending_del_slot = path->slots[0];
3505 pending_del_nr = 1;
3506 } else if (pending_del_nr &&
3507 path->slots[0] + 1 == pending_del_slot) {
3508 /* hop on the pending chunk */
3509 pending_del_nr++;
3510 pending_del_slot = path->slots[0];
3511 } else {
d397712b 3512 BUG();
85e21bac 3513 }
39279cc3
CM
3514 } else {
3515 break;
3516 }
0af3d00b
JB
3517 if (found_extent && (root->ref_cows ||
3518 root == root->fs_info->tree_root)) {
b9473439 3519 btrfs_set_path_blocking(path);
39279cc3 3520 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ
3521 extent_num_bytes, 0,
3522 btrfs_header_owner(leaf),
66d7e7f0 3523 ino, extent_offset, 0);
39279cc3
CM
3524 BUG_ON(ret);
3525 }
85e21bac 3526
8082510e
YZ
3527 if (found_type == BTRFS_INODE_ITEM_KEY)
3528 break;
3529
3530 if (path->slots[0] == 0 ||
3531 path->slots[0] != pending_del_slot) {
8082510e
YZ
3532 if (pending_del_nr) {
3533 ret = btrfs_del_items(trans, root, path,
3534 pending_del_slot,
3535 pending_del_nr);
79787eaa
JM
3536 if (ret) {
3537 btrfs_abort_transaction(trans,
3538 root, ret);
3539 goto error;
3540 }
8082510e
YZ
3541 pending_del_nr = 0;
3542 }
b3b4aa74 3543 btrfs_release_path(path);
85e21bac 3544 goto search_again;
8082510e
YZ
3545 } else {
3546 path->slots[0]--;
85e21bac 3547 }
39279cc3 3548 }
8082510e 3549out:
85e21bac
CM
3550 if (pending_del_nr) {
3551 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3552 pending_del_nr);
79787eaa
JM
3553 if (ret)
3554 btrfs_abort_transaction(trans, root, ret);
85e21bac 3555 }
79787eaa 3556error:
39279cc3 3557 btrfs_free_path(path);
8082510e 3558 return err;
39279cc3
CM
3559}
3560
3561/*
2aaa6655
JB
3562 * btrfs_truncate_page - read, zero a chunk and write a page
3563 * @inode - inode that we're zeroing
3564 * @from - the offset to start zeroing
3565 * @len - the length to zero, 0 to zero the entire range respective to the
3566 * offset
3567 * @front - zero up to the offset instead of from the offset on
3568 *
3569 * This will find the page for the "from" offset and cow the page and zero the
3570 * part we want to zero. This is used with truncate and hole punching.
39279cc3 3571 */
2aaa6655
JB
3572int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3573 int front)
39279cc3 3574{
2aaa6655 3575 struct address_space *mapping = inode->i_mapping;
db94535d 3576 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM
3577 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3578 struct btrfs_ordered_extent *ordered;
2ac55d41 3579 struct extent_state *cached_state = NULL;
e6dcd2dc 3580 char *kaddr;
db94535d 3581 u32 blocksize = root->sectorsize;
39279cc3
CM
3582 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3583 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3584 struct page *page;
3b16a4e3 3585 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3586 int ret = 0;
a52d9a80 3587 u64 page_start;
e6dcd2dc 3588 u64 page_end;
39279cc3 3589
2aaa6655
JB
3590 if ((offset & (blocksize - 1)) == 0 &&
3591 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 3592 goto out;
0ca1f7ce 3593 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB
3594 if (ret)
3595 goto out;
39279cc3 3596
211c17f5 3597again:
3b16a4e3 3598 page = find_or_create_page(mapping, index, mask);
5d5e103a 3599 if (!page) {
0ca1f7ce 3600 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 3601 ret = -ENOMEM;
39279cc3 3602 goto out;
5d5e103a 3603 }
e6dcd2dc
CM
3604
3605 page_start = page_offset(page);
3606 page_end = page_start + PAGE_CACHE_SIZE - 1;
3607
39279cc3 3608 if (!PageUptodate(page)) {
9ebefb18 3609 ret = btrfs_readpage(NULL, page);
39279cc3 3610 lock_page(page);
211c17f5
CM
3611 if (page->mapping != mapping) {
3612 unlock_page(page);
3613 page_cache_release(page);
3614 goto again;
3615 }
39279cc3
CM
3616 if (!PageUptodate(page)) {
3617 ret = -EIO;
89642229 3618 goto out_unlock;
39279cc3
CM
3619 }
3620 }
211c17f5 3621 wait_on_page_writeback(page);
e6dcd2dc 3622
d0082371 3623 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM
3624 set_page_extent_mapped(page);
3625
3626 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3627 if (ordered) {
2ac55d41
JB
3628 unlock_extent_cached(io_tree, page_start, page_end,
3629 &cached_state, GFP_NOFS);
e6dcd2dc
CM
3630 unlock_page(page);
3631 page_cache_release(page);
eb84ae03 3632 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM
3633 btrfs_put_ordered_extent(ordered);
3634 goto again;
3635 }
3636
2ac55d41 3637 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB
3638 EXTENT_DIRTY | EXTENT_DELALLOC |
3639 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 3640 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3641
2ac55d41
JB
3642 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3643 &cached_state);
9ed74f2d 3644 if (ret) {
2ac55d41
JB
3645 unlock_extent_cached(io_tree, page_start, page_end,
3646 &cached_state, GFP_NOFS);
9ed74f2d
JB
3647 goto out_unlock;
3648 }
3649
e6dcd2dc 3650 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB
3651 if (!len)
3652 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 3653 kaddr = kmap(page);
2aaa6655
JB
3654 if (front)
3655 memset(kaddr, 0, offset);
3656 else
3657 memset(kaddr + offset, 0, len);
e6dcd2dc
CM
3658 flush_dcache_page(page);
3659 kunmap(page);
3660 }
247e743c 3661 ClearPageChecked(page);
e6dcd2dc 3662 set_page_dirty(page);
2ac55d41
JB
3663 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3664 GFP_NOFS);
39279cc3 3665
89642229 3666out_unlock:
5d5e103a 3667 if (ret)
0ca1f7ce 3668 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM
3669 unlock_page(page);
3670 page_cache_release(page);
3671out:
3672 return ret;
3673}
3674
695a0d0d
JB
3675/*
3676 * This function puts in dummy file extents for the area we're creating a hole
3677 * for. So if we are truncating this file to a larger size we need to insert
3678 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3679 * the range between oldsize and size
3680 */
a41ad394 3681int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3682{
9036c102
YZ
3683 struct btrfs_trans_handle *trans;
3684 struct btrfs_root *root = BTRFS_I(inode)->root;
3685 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3686 struct extent_map *em = NULL;
2ac55d41 3687 struct extent_state *cached_state = NULL;
5dc562c5 3688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9036c102 3689 u64 mask = root->sectorsize - 1;
a41ad394 3690 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ
3691 u64 block_end = (size + mask) & ~mask;
3692 u64 last_byte;
3693 u64 cur_offset;
3694 u64 hole_size;
9ed74f2d 3695 int err = 0;
39279cc3 3696
9036c102
YZ
3697 if (size <= hole_start)
3698 return 0;
3699
9036c102
YZ
3700 while (1) {
3701 struct btrfs_ordered_extent *ordered;
3702 btrfs_wait_ordered_range(inode, hole_start,
3703 block_end - hole_start);
2ac55d41 3704 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 3705 &cached_state);
9036c102
YZ
3706 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3707 if (!ordered)
3708 break;
2ac55d41
JB
3709 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3710 &cached_state, GFP_NOFS);
9036c102
YZ
3711 btrfs_put_ordered_extent(ordered);
3712 }
39279cc3 3713
9036c102
YZ
3714 cur_offset = hole_start;
3715 while (1) {
3716 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3717 block_end - cur_offset, 0);
79787eaa
JM
3718 if (IS_ERR(em)) {
3719 err = PTR_ERR(em);
f2767956 3720 em = NULL;
79787eaa
JM
3721 break;
3722 }
9036c102
YZ
3723 last_byte = min(extent_map_end(em), block_end);
3724 last_byte = (last_byte + mask) & ~mask;
8082510e 3725 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 3726 struct extent_map *hole_em;
9036c102 3727 hole_size = last_byte - cur_offset;
9ed74f2d 3728
3642320e 3729 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ
3730 if (IS_ERR(trans)) {
3731 err = PTR_ERR(trans);
9ed74f2d 3732 break;
a22285a6 3733 }
8082510e 3734
5dc562c5
JB
3735 err = btrfs_drop_extents(trans, root, inode,
3736 cur_offset,
2671485d 3737 cur_offset + hole_size, 1);
5b397377 3738 if (err) {
79787eaa 3739 btrfs_abort_transaction(trans, root, err);
5b397377 3740 btrfs_end_transaction(trans, root);
3893e33b 3741 break;
5b397377 3742 }
8082510e 3743
9036c102 3744 err = btrfs_insert_file_extent(trans, root,
33345d01 3745 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ
3746 0, hole_size, 0, hole_size,
3747 0, 0, 0);
5b397377 3748 if (err) {
79787eaa 3749 btrfs_abort_transaction(trans, root, err);
5b397377 3750 btrfs_end_transaction(trans, root);
3893e33b 3751 break;
5b397377 3752 }
8082510e 3753
5dc562c5
JB
3754 btrfs_drop_extent_cache(inode, cur_offset,
3755 cur_offset + hole_size - 1, 0);
3756 hole_em = alloc_extent_map();
3757 if (!hole_em) {
3758 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3759 &BTRFS_I(inode)->runtime_flags);
3760 goto next;
3761 }
3762 hole_em->start = cur_offset;
3763 hole_em->len = hole_size;
3764 hole_em->orig_start = cur_offset;
8082510e 3765
5dc562c5
JB
3766 hole_em->block_start = EXTENT_MAP_HOLE;
3767 hole_em->block_len = 0;
b4939680 3768 hole_em->orig_block_len = 0;
5dc562c5
JB
3769 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3770 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3771 hole_em->generation = trans->transid;
8082510e 3772
5dc562c5
JB
3773 while (1) {
3774 write_lock(&em_tree->lock);
3775 err = add_extent_mapping(em_tree, hole_em);
3776 if (!err)
3777 list_move(&hole_em->list,
3778 &em_tree->modified_extents);
3779 write_unlock(&em_tree->lock);
3780 if (err != -EEXIST)
3781 break;
3782 btrfs_drop_extent_cache(inode, cur_offset,
3783 cur_offset +
3784 hole_size - 1, 0);
3785 }
3786 free_extent_map(hole_em);
3787next:
3642320e 3788 btrfs_update_inode(trans, root, inode);
8082510e 3789 btrfs_end_transaction(trans, root);
9036c102
YZ
3790 }
3791 free_extent_map(em);
a22285a6 3792 em = NULL;
9036c102 3793 cur_offset = last_byte;
8082510e 3794 if (cur_offset >= block_end)
9036c102
YZ
3795 break;
3796 }
1832a6d5 3797
a22285a6 3798 free_extent_map(em);
2ac55d41
JB
3799 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3800 GFP_NOFS);
9036c102
YZ
3801 return err;
3802}
39279cc3 3803
3972f260 3804static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 3805{
f4a2f4c5
MX
3806 struct btrfs_root *root = BTRFS_I(inode)->root;
3807 struct btrfs_trans_handle *trans;
a41ad394 3808 loff_t oldsize = i_size_read(inode);
3972f260
ES
3809 loff_t newsize = attr->ia_size;
3810 int mask = attr->ia_valid;
8082510e
YZ
3811 int ret;
3812
a41ad394 3813 if (newsize == oldsize)
8082510e
YZ
3814 return 0;
3815
3972f260
ES
3816 /*
3817 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3818 * special case where we need to update the times despite not having
3819 * these flags set. For all other operations the VFS set these flags
3820 * explicitly if it wants a timestamp update.
3821 */
3822 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3823 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3824
a41ad394 3825 if (newsize > oldsize) {
a41ad394
JB
3826 truncate_pagecache(inode, oldsize, newsize);
3827 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 3828 if (ret)
8082510e 3829 return ret;
8082510e 3830
f4a2f4c5
MX
3831 trans = btrfs_start_transaction(root, 1);
3832 if (IS_ERR(trans))
3833 return PTR_ERR(trans);
3834
3835 i_size_write(inode, newsize);
3836 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3837 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 3838 btrfs_end_transaction(trans, root);
a41ad394 3839 } else {
8082510e 3840
a41ad394
JB
3841 /*
3842 * We're truncating a file that used to have good data down to
3843 * zero. Make sure it gets into the ordered flush list so that
3844 * any new writes get down to disk quickly.
3845 */
3846 if (newsize == 0)
72ac3c0d
JB
3847 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3848 &BTRFS_I(inode)->runtime_flags);
8082510e 3849
f3fe820c
JB
3850 /*
3851 * 1 for the orphan item we're going to add
3852 * 1 for the orphan item deletion.
3853 */
3854 trans = btrfs_start_transaction(root, 2);
3855 if (IS_ERR(trans))
3856 return PTR_ERR(trans);
3857
3858 /*
3859 * We need to do this in case we fail at _any_ point during the
3860 * actual truncate. Once we do the truncate_setsize we could
3861 * invalidate pages which forces any outstanding ordered io to
3862 * be instantly completed which will give us extents that need
3863 * to be truncated. If we fail to get an orphan inode down we
3864 * could have left over extents that were never meant to live,
3865 * so we need to garuntee from this point on that everything
3866 * will be consistent.
3867 */
3868 ret = btrfs_orphan_add(trans, inode);
3869 btrfs_end_transaction(trans, root);
3870 if (ret)
3871 return ret;
3872
a41ad394
JB
3873 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3874 truncate_setsize(inode, newsize);
3875 ret = btrfs_truncate(inode);
f3fe820c
JB
3876 if (ret && inode->i_nlink)
3877 btrfs_orphan_del(NULL, inode);
8082510e
YZ
3878 }
3879
a41ad394 3880 return ret;
8082510e
YZ
3881}
3882
9036c102
YZ
3883static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3884{
3885 struct inode *inode = dentry->d_inode;
b83cc969 3886 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3887 int err;
39279cc3 3888
b83cc969
LZ
3889 if (btrfs_root_readonly(root))
3890 return -EROFS;
3891
9036c102
YZ
3892 err = inode_change_ok(inode, attr);
3893 if (err)
3894 return err;
2bf5a725 3895
5a3f23d5 3896 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 3897 err = btrfs_setsize(inode, attr);
8082510e
YZ
3898 if (err)
3899 return err;
39279cc3 3900 }
9036c102 3901
1025774c
CH
3902 if (attr->ia_valid) {
3903 setattr_copy(inode, attr);
0c4d2d95 3904 inode_inc_iversion(inode);
22c44fe6 3905 err = btrfs_dirty_inode(inode);
1025774c 3906
22c44fe6 3907 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH
3908 err = btrfs_acl_chmod(inode);
3909 }
33268eaf 3910
39279cc3
CM
3911 return err;
3912}
61295eb8 3913
bd555975 3914void btrfs_evict_inode(struct inode *inode)
39279cc3
CM
3915{
3916 struct btrfs_trans_handle *trans;
3917 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3918 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3919 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM
3920 int ret;
3921
1abe9b8a 3922 trace_btrfs_inode_evict(inode);
3923
39279cc3 3924 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3925 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 3926 btrfs_is_free_space_inode(inode)))
bd555975
AV
3927 goto no_delete;
3928
39279cc3 3929 if (is_bad_inode(inode)) {
7b128766 3930 btrfs_orphan_del(NULL, inode);
39279cc3
CM
3931 goto no_delete;
3932 }
bd555975 3933 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3934 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3935
c71bf099 3936 if (root->fs_info->log_root_recovering) {
6bf02314 3937 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 3938 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ
3939 goto no_delete;
3940 }
3941
76dda93c
YZ
3942 if (inode->i_nlink > 0) {
3943 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3944 goto no_delete;
3945 }
3946
0e8c36a9
MX
3947 ret = btrfs_commit_inode_delayed_inode(inode);
3948 if (ret) {
3949 btrfs_orphan_del(NULL, inode);
3950 goto no_delete;
3951 }
3952
66d8f3dd 3953 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB
3954 if (!rsv) {
3955 btrfs_orphan_del(NULL, inode);
3956 goto no_delete;
3957 }
4a338542 3958 rsv->size = min_size;
ca7e70f5 3959 rsv->failfast = 1;
726c35fa 3960 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3961
dbe674a9 3962 btrfs_i_size_write(inode, 0);
5f39d397 3963
4289a667 3964 /*
8407aa46
MX
3965 * This is a bit simpler than btrfs_truncate since we've already
3966 * reserved our space for our orphan item in the unlink, so we just
3967 * need to reserve some slack space in case we add bytes and update
3968 * inode item when doing the truncate.
4289a667 3969 */
8082510e 3970 while (1) {
08e007d2
MX
3971 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3972 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB
3973
3974 /*
3975 * Try and steal from the global reserve since we will
3976 * likely not use this space anyway, we want to try as
3977 * hard as possible to get this to work.
3978 */
3979 if (ret)
3980 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 3981
d68fc57b 3982 if (ret) {
4289a667 3983 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3984 "delete, will truncate on mount %d\n", ret);
4289a667
JB
3985 btrfs_orphan_del(NULL, inode);
3986 btrfs_free_block_rsv(root, rsv);
3987 goto no_delete;
d68fc57b 3988 }
7b128766 3989
0e8c36a9 3990 trans = btrfs_join_transaction(root);
4289a667
JB
3991 if (IS_ERR(trans)) {
3992 btrfs_orphan_del(NULL, inode);
3993 btrfs_free_block_rsv(root, rsv);
3994 goto no_delete;
d68fc57b 3995 }
7b128766 3996
4289a667
JB
3997 trans->block_rsv = rsv;
3998
d68fc57b 3999 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4000 if (ret != -ENOSPC)
8082510e 4001 break;
85e21bac 4002
8407aa46 4003 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ
4004 btrfs_end_transaction(trans, root);
4005 trans = NULL;
b53d3f5d 4006 btrfs_btree_balance_dirty(root);
8082510e 4007 }
5f39d397 4008
4289a667
JB
4009 btrfs_free_block_rsv(root, rsv);
4010
8082510e 4011 if (ret == 0) {
4289a667 4012 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ
4013 ret = btrfs_orphan_del(trans, inode);
4014 BUG_ON(ret);
4015 }
54aa1f4d 4016
4289a667 4017 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ
4018 if (!(root == root->fs_info->tree_root ||
4019 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4020 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4021
54aa1f4d 4022 btrfs_end_transaction(trans, root);
b53d3f5d 4023 btrfs_btree_balance_dirty(root);
39279cc3 4024no_delete:
dbd5768f 4025 clear_inode(inode);
8082510e 4026 return;
39279cc3
CM
4027}
4028
4029/*
4030 * this returns the key found in the dir entry in the location pointer.
4031 * If no dir entries were found, location->objectid is 0.
4032 */
4033static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4034 struct btrfs_key *location)
4035{
4036 const char *name = dentry->d_name.name;
4037 int namelen = dentry->d_name.len;
4038 struct btrfs_dir_item *di;
4039 struct btrfs_path *path;
4040 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4041 int ret = 0;
39279cc3
CM
4042
4043 path = btrfs_alloc_path();
d8926bb3
MF
4044 if (!path)
4045 return -ENOMEM;
3954401f 4046
33345d01 4047 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4048 namelen, 0);
0d9f7f3e
Y
4049 if (IS_ERR(di))
4050 ret = PTR_ERR(di);
d397712b 4051
c704005d 4052 if (IS_ERR_OR_NULL(di))
3954401f 4053 goto out_err;
d397712b 4054
5f39d397 4055 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4056out:
39279cc3
CM
4057 btrfs_free_path(path);
4058 return ret;
3954401f
CM
4059out_err:
4060 location->objectid = 0;
4061 goto out;
39279cc3
CM
4062}
4063
4064/*
4065 * when we hit a tree root in a directory, the btrfs part of the inode
4066 * needs to be changed to reflect the root directory of the tree root. This
4067 * is kind of like crossing a mount point.
4068 */
4069static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ
4070 struct inode *dir,
4071 struct dentry *dentry,
4072 struct btrfs_key *location,
4073 struct btrfs_root **sub_root)
39279cc3 4074{
4df27c4d
YZ
4075 struct btrfs_path *path;
4076 struct btrfs_root *new_root;
4077 struct btrfs_root_ref *ref;
4078 struct extent_buffer *leaf;
4079 int ret;
4080 int err = 0;
39279cc3 4081
4df27c4d
YZ
4082 path = btrfs_alloc_path();
4083 if (!path) {
4084 err = -ENOMEM;
4085 goto out;
4086 }
39279cc3 4087
4df27c4d
YZ
4088 err = -ENOENT;
4089 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4090 BTRFS_I(dir)->root->root_key.objectid,
4091 location->objectid);
4092 if (ret) {
4093 if (ret < 0)
4094 err = ret;
4095 goto out;
4096 }
39279cc3 4097
4df27c4d
YZ
4098 leaf = path->nodes[0];
4099 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4100 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ
4101 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4102 goto out;
39279cc3 4103
4df27c4d
YZ
4104 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4105 (unsigned long)(ref + 1),
4106 dentry->d_name.len);
4107 if (ret)
4108 goto out;
4109
b3b4aa74 4110 btrfs_release_path(path);
4df27c4d
YZ
4111
4112 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4113 if (IS_ERR(new_root)) {
4114 err = PTR_ERR(new_root);
4115 goto out;
4116 }
4117
4118 if (btrfs_root_refs(&new_root->root_item) == 0) {
4119 err = -ENOENT;
4120 goto out;
4121 }
4122
4123 *sub_root = new_root;
4124 location->objectid = btrfs_root_dirid(&new_root->root_item);
4125 location->type = BTRFS_INODE_ITEM_KEY;
4126 location->offset = 0;
4127 err = 0;
4128out:
4129 btrfs_free_path(path);
4130 return err;
39279cc3
CM
4131}
4132
5d4f98a2
YZ
4133static void inode_tree_add(struct inode *inode)
4134{
4135 struct btrfs_root *root = BTRFS_I(inode)->root;
4136 struct btrfs_inode *entry;
03e860bd
FNP
4137 struct rb_node **p;
4138 struct rb_node *parent;
33345d01 4139 u64 ino = btrfs_ino(inode);
03e860bd
FNP
4140again:
4141 p = &root->inode_tree.rb_node;
4142 parent = NULL;
5d4f98a2 4143
1d3382cb 4144 if (inode_unhashed(inode))
76dda93c
YZ
4145 return;
4146
5d4f98a2
YZ
4147 spin_lock(&root->inode_lock);
4148 while (*p) {
4149 parent = *p;
4150 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4151
33345d01 4152 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4153 p = &parent->rb_left;
33345d01 4154 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4155 p = &parent->rb_right;
5d4f98a2
YZ
4156 else {
4157 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4158 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP
4159 rb_erase(parent, &root->inode_tree);
4160 RB_CLEAR_NODE(parent);
4161 spin_unlock(&root->inode_lock);
4162 goto again;
5d4f98a2
YZ
4163 }
4164 }
4165 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4166 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4167 spin_unlock(&root->inode_lock);
4168}
4169
4170static void inode_tree_del(struct inode *inode)
4171{
4172 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4173 int empty = 0;
5d4f98a2 4174
03e860bd 4175 spin_lock(&root->inode_lock);
5d4f98a2 4176 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4177 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4178 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4179 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4180 }
03e860bd 4181 spin_unlock(&root->inode_lock);
76dda93c 4182
0af3d00b
JB
4183 /*
4184 * Free space cache has inodes in the tree root, but the tree root has a
4185 * root_refs of 0, so this could end up dropping the tree root as a
4186 * snapshot, so we need the extra !root->fs_info->tree_root check to
4187 * make sure we don't drop it.
4188 */
4189 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4190 root != root->fs_info->tree_root) {
76dda93c
YZ
4191 synchronize_srcu(&root->fs_info->subvol_srcu);
4192 spin_lock(&root->inode_lock);
4193 empty = RB_EMPTY_ROOT(&root->inode_tree);
4194 spin_unlock(&root->inode_lock);
4195 if (empty)
4196 btrfs_add_dead_root(root);
4197 }
4198}
4199
143bede5 4200void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ
4201{
4202 struct rb_node *node;
4203 struct rb_node *prev;
4204 struct btrfs_inode *entry;
4205 struct inode *inode;
4206 u64 objectid = 0;
4207
4208 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4209
4210 spin_lock(&root->inode_lock);
4211again:
4212 node = root->inode_tree.rb_node;
4213 prev = NULL;
4214 while (node) {
4215 prev = node;
4216 entry = rb_entry(node, struct btrfs_inode, rb_node);
4217
33345d01 4218 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4219 node = node->rb_left;
33345d01 4220 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ
4221 node = node->rb_right;
4222 else
4223 break;
4224 }
4225 if (!node) {
4226 while (prev) {
4227 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4228 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ
4229 node = prev;
4230 break;
4231 }
4232 prev = rb_next(prev);
4233 }
4234 }
4235 while (node) {
4236 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4237 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ
4238 inode = igrab(&entry->vfs_inode);
4239 if (inode) {
4240 spin_unlock(&root->inode_lock);
4241 if (atomic_read(&inode->i_count) > 1)
4242 d_prune_aliases(inode);
4243 /*
45321ac5 4244 * btrfs_drop_inode will have it removed from
76dda93c
YZ
4245 * the inode cache when its usage count
4246 * hits zero.
4247 */
4248 iput(inode);
4249 cond_resched();
4250 spin_lock(&root->inode_lock);
4251 goto again;
4252 }
4253
4254 if (cond_resched_lock(&root->inode_lock))
4255 goto again;
4256
4257 node = rb_next(node);
4258 }
4259 spin_unlock(&root->inode_lock);
5d4f98a2
YZ
4260}
4261
e02119d5
CM
4262static int btrfs_init_locked_inode(struct inode *inode, void *p)
4263{
4264 struct btrfs_iget_args *args = p;
4265 inode->i_ino = args->ino;
e02119d5 4266 BTRFS_I(inode)->root = args->root;
39279cc3
CM
4267 return 0;
4268}
4269
4270static int btrfs_find_actor(struct inode *inode, void *opaque)
4271{
4272 struct btrfs_iget_args *args = opaque;
33345d01 4273 return args->ino == btrfs_ino(inode) &&
d397712b 4274 args->root == BTRFS_I(inode)->root;
39279cc3
CM
4275}
4276
5d4f98a2
YZ
4277static struct inode *btrfs_iget_locked(struct super_block *s,
4278 u64 objectid,
4279 struct btrfs_root *root)
39279cc3
CM
4280{
4281 struct inode *inode;
4282 struct btrfs_iget_args args;
4283 args.ino = objectid;
4284 args.root = root;
4285
4286 inode = iget5_locked(s, objectid, btrfs_find_actor,
4287 btrfs_init_locked_inode,
4288 (void *)&args);
4289 return inode;
4290}
4291
1a54ef8c
BR
4292/* Get an inode object given its location and corresponding root.
4293 * Returns in *is_new if the inode was read from disk
4294 */
4295struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4296 struct btrfs_root *root, int *new)
1a54ef8c
BR
4297{
4298 struct inode *inode;
4299
4300 inode = btrfs_iget_locked(s, location->objectid, root);
4301 if (!inode)
5d4f98a2 4302 return ERR_PTR(-ENOMEM);
1a54ef8c
BR
4303
4304 if (inode->i_state & I_NEW) {
4305 BTRFS_I(inode)->root = root;
4306 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4307 btrfs_read_locked_inode(inode);
1748f843
MF
4308 if (!is_bad_inode(inode)) {
4309 inode_tree_add(inode);
4310 unlock_new_inode(inode);
4311 if (new)
4312 *new = 1;
4313 } else {
e0b6d65b
ST
4314 unlock_new_inode(inode);
4315 iput(inode);
4316 inode = ERR_PTR(-ESTALE);
1748f843
MF
4317 }
4318 }
4319
1a54ef8c
BR
4320 return inode;
4321}
4322
4df27c4d
YZ
4323static struct inode *new_simple_dir(struct super_block *s,
4324 struct btrfs_key *key,
4325 struct btrfs_root *root)
4326{
4327 struct inode *inode = new_inode(s);
4328
4329 if (!inode)
4330 return ERR_PTR(-ENOMEM);
4331
4df27c4d
YZ
4332 BTRFS_I(inode)->root = root;
4333 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4334 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ
4335
4336 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4337 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ
4338 inode->i_fop = &simple_dir_operations;
4339 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4340 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4341
4342 return inode;
4343}
4344
3de4586c 4345struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4346{
d397712b 4347 struct inode *inode;
4df27c4d 4348 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM
4349 struct btrfs_root *sub_root = root;
4350 struct btrfs_key location;
76dda93c 4351 int index;
b4aff1f8 4352 int ret = 0;
39279cc3
CM
4353
4354 if (dentry->d_name.len > BTRFS_NAME_LEN)
4355 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4356
b4aff1f8
JB
4357 if (unlikely(d_need_lookup(dentry))) {
4358 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4359 kfree(dentry->d_fsdata);
4360 dentry->d_fsdata = NULL;
a66e7cc6
JB
4361 /* This thing is hashed, drop it for now */
4362 d_drop(dentry);
b4aff1f8
JB
4363 } else {
4364 ret = btrfs_inode_by_name(dir, dentry, &location);
4365 }
5f39d397 4366
39279cc3
CM
4367 if (ret < 0)
4368 return ERR_PTR(ret);
5f39d397 4369
4df27c4d
YZ
4370 if (location.objectid == 0)
4371 return NULL;
4372
4373 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4374 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ
4375 return inode;
4376 }
4377
4378 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4379
76dda93c 4380 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ
4381 ret = fixup_tree_root_location(root, dir, dentry,
4382 &location, &sub_root);
4383 if (ret < 0) {
4384 if (ret != -ENOENT)
4385 inode = ERR_PTR(ret);
4386 else
4387 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4388 } else {
73f73415 4389 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4390 }
76dda93c
YZ
4391 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4392
34d19bad 4393 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ
4394 down_read(&root->fs_info->cleanup_work_sem);
4395 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4396 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4397 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB
4398 if (ret)
4399 inode = ERR_PTR(ret);
c71bf099
YZ
4400 }
4401
3de4586c
CM
4402 return inode;
4403}
4404
fe15ce44 4405static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ
4406{
4407 struct btrfs_root *root;
848cce0d 4408 struct inode *inode = dentry->d_inode;
76dda93c 4409
848cce0d
LZ
4410 if (!inode && !IS_ROOT(dentry))
4411 inode = dentry->d_parent->d_inode;
76dda93c 4412
848cce0d
LZ
4413 if (inode) {
4414 root = BTRFS_I(inode)->root;
efefb143
YZ
4415 if (btrfs_root_refs(&root->root_item) == 0)
4416 return 1;
848cce0d
LZ
4417
4418 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4419 return 1;
efefb143 4420 }
76dda93c
YZ
4421 return 0;
4422}
4423
b4aff1f8
JB
4424static void btrfs_dentry_release(struct dentry *dentry)
4425{
4426 if (dentry->d_fsdata)
4427 kfree(dentry->d_fsdata);
4428}
4429
3de4586c 4430static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 4431 unsigned int flags)
3de4586c 4432{
a66e7cc6
JB
4433 struct dentry *ret;
4434
4435 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4436 if (unlikely(d_need_lookup(dentry))) {
4437 spin_lock(&dentry->d_lock);
4438 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4439 spin_unlock(&dentry->d_lock);
4440 }
4441 return ret;
39279cc3
CM
4442}
4443
16cdcec7 4444unsigned char btrfs_filetype_table[] = {
39279cc3
CM
4445 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4446};
4447
cbdf5a24
DW
4448static int btrfs_real_readdir(struct file *filp, void *dirent,
4449 filldir_t filldir)
39279cc3 4450{
6da6abae 4451 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM
4452 struct btrfs_root *root = BTRFS_I(inode)->root;
4453 struct btrfs_item *item;
4454 struct btrfs_dir_item *di;
4455 struct btrfs_key key;
5f39d397 4456 struct btrfs_key found_key;
39279cc3 4457 struct btrfs_path *path;
16cdcec7
MX
4458 struct list_head ins_list;
4459 struct list_head del_list;
39279cc3 4460 int ret;
5f39d397 4461 struct extent_buffer *leaf;
39279cc3 4462 int slot;
39279cc3
CM
4463 unsigned char d_type;
4464 int over = 0;
4465 u32 di_cur;
4466 u32 di_total;
4467 u32 di_len;
4468 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM
4469 char tmp_name[32];
4470 char *name_ptr;
4471 int name_len;
16cdcec7 4472 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM
4473
4474 /* FIXME, use a real flag for deciding about the key type */
4475 if (root->fs_info->tree_root == root)
4476 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4477
3954401f
CM
4478 /* special case for "." */
4479 if (filp->f_pos == 0) {
3765fefa
HS
4480 over = filldir(dirent, ".", 1,
4481 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM
4482 if (over)
4483 return 0;
4484 filp->f_pos = 1;
4485 }
3954401f
CM
4486 /* special case for .., just use the back ref */
4487 if (filp->f_pos == 1) {
5ecc7e5d 4488 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4489 over = filldir(dirent, "..", 2,
3765fefa 4490 filp->f_pos, pino, DT_DIR);
3954401f 4491 if (over)
49593bfa 4492 return 0;
3954401f
CM
4493 filp->f_pos = 2;
4494 }
49593bfa 4495 path = btrfs_alloc_path();
16cdcec7
MX
4496 if (!path)
4497 return -ENOMEM;
ff5714cc 4498
026fd317 4499 path->reada = 1;
49593bfa 4500
16cdcec7
MX
4501 if (key_type == BTRFS_DIR_INDEX_KEY) {
4502 INIT_LIST_HEAD(&ins_list);
4503 INIT_LIST_HEAD(&del_list);
4504 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4505 }
4506
39279cc3
CM
4507 btrfs_set_key_type(&key, key_type);
4508 key.offset = filp->f_pos;
33345d01 4509 key.objectid = btrfs_ino(inode);
5f39d397 4510
39279cc3
CM
4511 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4512 if (ret < 0)
4513 goto err;
49593bfa
DW
4514
4515 while (1) {
5f39d397 4516 leaf = path->nodes[0];
39279cc3 4517 slot = path->slots[0];
b9e03af0
LZ
4518 if (slot >= btrfs_header_nritems(leaf)) {
4519 ret = btrfs_next_leaf(root, path);
4520 if (ret < 0)
4521 goto err;
4522 else if (ret > 0)
4523 break;
4524 continue;
39279cc3 4525 }
3de4586c 4526
5f39d397
CM
4527 item = btrfs_item_nr(leaf, slot);
4528 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4529
4530 if (found_key.objectid != key.objectid)
39279cc3 4531 break;
5f39d397 4532 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4533 break;
5f39d397 4534 if (found_key.offset < filp->f_pos)
b9e03af0 4535 goto next;
16cdcec7
MX
4536 if (key_type == BTRFS_DIR_INDEX_KEY &&
4537 btrfs_should_delete_dir_index(&del_list,
4538 found_key.offset))
4539 goto next;
5f39d397
CM
4540
4541 filp->f_pos = found_key.offset;
16cdcec7 4542 is_curr = 1;
49593bfa 4543
39279cc3
CM
4544 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4545 di_cur = 0;
5f39d397 4546 di_total = btrfs_item_size(leaf, item);
49593bfa
DW
4547
4548 while (di_cur < di_total) {
5f39d397
CM
4549 struct btrfs_key location;
4550
22a94d44
JB
4551 if (verify_dir_item(root, leaf, di))
4552 break;
4553
5f39d397 4554 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4555 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM
4556 name_ptr = tmp_name;
4557 } else {
4558 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW
4559 if (!name_ptr) {
4560 ret = -ENOMEM;
4561 goto err;
4562 }
5f39d397
CM
4563 }
4564 read_extent_buffer(leaf, name_ptr,
4565 (unsigned long)(di + 1), name_len);
4566
4567 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4568 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4569
fede766f 4570
3de4586c 4571 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ
4572 * skip it.
4573 *
4574 * In contrast to old kernels, we insert the snapshot's
4575 * dir item and dir index after it has been created, so
4576 * we won't find a reference to our own snapshot. We
4577 * still keep the following code for backward
4578 * compatibility.
3de4586c
CM
4579 */
4580 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4581 location.objectid == root->root_key.objectid) {
4582 over = 0;
4583 goto skip;
4584 }
5f39d397 4585 over = filldir(dirent, name_ptr, name_len,
49593bfa 4586 found_key.offset, location.objectid,
39279cc3 4587 d_type);
5f39d397 4588
3de4586c 4589skip:
5f39d397
CM
4590 if (name_ptr != tmp_name)
4591 kfree(name_ptr);
4592
39279cc3
CM
4593 if (over)
4594 goto nopos;
5103e947 4595 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4596 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM
4597 di_cur += di_len;
4598 di = (struct btrfs_dir_item *)((char *)di + di_len);
4599 }
b9e03af0
LZ
4600next:
4601 path->slots[0]++;
39279cc3 4602 }
49593bfa 4603
16cdcec7
MX
4604 if (key_type == BTRFS_DIR_INDEX_KEY) {
4605 if (is_curr)
4606 filp->f_pos++;
4607 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4608 &ins_list);
4609 if (ret)
4610 goto nopos;
4611 }
4612
49593bfa 4613 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4614 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE
4615 /*
4616 * 32-bit glibc will use getdents64, but then strtol -
4617 * so the last number we can serve is this.
4618 */
4619 filp->f_pos = 0x7fffffff;
5e591a07
YZ
4620 else
4621 filp->f_pos++;
39279cc3
CM
4622nopos:
4623 ret = 0;
4624err:
16cdcec7
MX
4625 if (key_type == BTRFS_DIR_INDEX_KEY)
4626 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4627 btrfs_free_path(path);
39279cc3
CM
4628 return ret;
4629}
4630
a9185b41 4631int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM
4632{
4633 struct btrfs_root *root = BTRFS_I(inode)->root;
4634 struct btrfs_trans_handle *trans;
4635 int ret = 0;
0af3d00b 4636 bool nolock = false;
39279cc3 4637
72ac3c0d 4638 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM
4639 return 0;
4640
83eea1f1 4641 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 4642 nolock = true;
0af3d00b 4643
a9185b41 4644 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4645 if (nolock)
7a7eaa40 4646 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4647 else
7a7eaa40 4648 trans = btrfs_join_transaction(root);
3612b495
TI
4649 if (IS_ERR(trans))
4650 return PTR_ERR(trans);
a698d075 4651 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM
4652 }
4653 return ret;
4654}
4655
4656/*
54aa1f4d 4657 * This is somewhat expensive, updating the tree every time the
39279cc3
CM
4658 * inode changes. But, it is most likely to find the inode in cache.
4659 * FIXME, needs more benchmarking...there are no reasons other than performance
4660 * to keep or drop this code.
4661 */
22c44fe6 4662int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM
4663{
4664 struct btrfs_root *root = BTRFS_I(inode)->root;
4665 struct btrfs_trans_handle *trans;
8929ecfa
YZ
4666 int ret;
4667
72ac3c0d 4668 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 4669 return 0;
39279cc3 4670
7a7eaa40 4671 trans = btrfs_join_transaction(root);
22c44fe6
JB
4672 if (IS_ERR(trans))
4673 return PTR_ERR(trans);
8929ecfa
YZ
4674
4675 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM
4676 if (ret && ret == -ENOSPC) {
4677 /* whoops, lets try again with the full transaction */
4678 btrfs_end_transaction(trans, root);
4679 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB
4680 if (IS_ERR(trans))
4681 return PTR_ERR(trans);
8929ecfa 4682
94b60442 4683 ret = btrfs_update_inode(trans, root, inode);
94b60442 4684 }
39279cc3 4685 btrfs_end_transaction(trans, root);
16cdcec7
MX
4686 if (BTRFS_I(inode)->delayed_node)
4687 btrfs_balance_delayed_items(root);
22c44fe6
JB
4688
4689 return ret;
4690}
4691
4692/*
4693 * This is a copy of file_update_time. We need this so we can return error on
4694 * ENOSPC for updating the inode in the case of file write and mmap writes.
4695 */
e41f941a
JB
4696static int btrfs_update_time(struct inode *inode, struct timespec *now,
4697 int flags)
22c44fe6 4698{
2bc55652
AB
4699 struct btrfs_root *root = BTRFS_I(inode)->root;
4700
4701 if (btrfs_root_readonly(root))
4702 return -EROFS;
4703
e41f941a 4704 if (flags & S_VERSION)
22c44fe6 4705 inode_inc_iversion(inode);
e41f941a
JB
4706 if (flags & S_CTIME)
4707 inode->i_ctime = *now;
4708 if (flags & S_MTIME)
4709 inode->i_mtime = *now;
4710 if (flags & S_ATIME)
4711 inode->i_atime = *now;
4712 return btrfs_dirty_inode(inode);
39279cc3
CM
4713}
4714
d352ac68
CM
4715/*
4716 * find the highest existing sequence number in a directory
4717 * and then set the in-memory index_cnt variable to reflect
4718 * free sequence numbers
4719 */
aec7477b
JB
4720static int btrfs_set_inode_index_count(struct inode *inode)
4721{
4722 struct btrfs_root *root = BTRFS_I(inode)->root;
4723 struct btrfs_key key, found_key;
4724 struct btrfs_path *path;
4725 struct extent_buffer *leaf;
4726 int ret;
4727
33345d01 4728 key.objectid = btrfs_ino(inode);
aec7477b
JB
4729 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4730 key.offset = (u64)-1;
4731
4732 path = btrfs_alloc_path();
4733 if (!path)
4734 return -ENOMEM;
4735
4736 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4737 if (ret < 0)
4738 goto out;
4739 /* FIXME: we should be able to handle this */
4740 if (ret == 0)
4741 goto out;
4742 ret = 0;
4743
4744 /*
4745 * MAGIC NUMBER EXPLANATION:
4746 * since we search a directory based on f_pos we have to start at 2
4747 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4748 * else has to start at 2
4749 */
4750 if (path->slots[0] == 0) {
4751 BTRFS_I(inode)->index_cnt = 2;
4752 goto out;
4753 }
4754
4755 path->slots[0]--;
4756
4757 leaf = path->nodes[0];
4758 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4759
33345d01 4760 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB
4761 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4762 BTRFS_I(inode)->index_cnt = 2;
4763 goto out;
4764 }
4765
4766 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4767out:
4768 btrfs_free_path(path);
4769 return ret;
4770}
4771
d352ac68
CM
4772/*
4773 * helper to find a free sequence number in a given directory. This current
4774 * code is very simple, later versions will do smarter things in the btree
4775 */
3de4586c 4776int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB
4777{
4778 int ret = 0;
4779
4780 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX
4781 ret = btrfs_inode_delayed_dir_index_count(dir);
4782 if (ret) {
4783 ret = btrfs_set_inode_index_count(dir);
4784 if (ret)
4785 return ret;
4786 }
aec7477b
JB
4787 }
4788
00e4e6b3 4789 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB
4790 BTRFS_I(dir)->index_cnt++;
4791
4792 return ret;
4793}
4794
39279cc3
CM
4795static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4796 struct btrfs_root *root,
aec7477b 4797 struct inode *dir,
9c58309d 4798 const char *name, int name_len,
175a4eb7
AV
4799 u64 ref_objectid, u64 objectid,
4800 umode_t mode, u64 *index)
39279cc3
CM
4801{
4802 struct inode *inode;
5f39d397 4803 struct btrfs_inode_item *inode_item;
39279cc3 4804 struct btrfs_key *location;
5f39d397 4805 struct btrfs_path *path;
9c58309d
CM
4806 struct btrfs_inode_ref *ref;
4807 struct btrfs_key key[2];
4808 u32 sizes[2];
4809 unsigned long ptr;
39279cc3
CM
4810 int ret;
4811 int owner;
4812
5f39d397 4813 path = btrfs_alloc_path();
d8926bb3
MF
4814 if (!path)
4815 return ERR_PTR(-ENOMEM);
5f39d397 4816
39279cc3 4817 inode = new_inode(root->fs_info->sb);
8fb27640
YS
4818 if (!inode) {
4819 btrfs_free_path(path);
39279cc3 4820 return ERR_PTR(-ENOMEM);
8fb27640 4821 }
39279cc3 4822
581bb050
LZ
4823 /*
4824 * we have to initialize this early, so we can reclaim the inode
4825 * number if we fail afterwards in this function.
4826 */
4827 inode->i_ino = objectid;
4828
aec7477b 4829 if (dir) {
1abe9b8a 4830 trace_btrfs_inode_request(dir);
4831
3de4586c 4832 ret = btrfs_set_inode_index(dir, index);
09771430 4833 if (ret) {
8fb27640 4834 btrfs_free_path(path);
09771430 4835 iput(inode);
aec7477b 4836 return ERR_PTR(ret);
09771430 4837 }
aec7477b
JB
4838 }
4839 /*
4840 * index_cnt is ignored for everything but a dir,
4841 * btrfs_get_inode_index_count has an explanation for the magic
4842 * number
4843 */
4844 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4845 BTRFS_I(inode)->root = root;
e02119d5 4846 BTRFS_I(inode)->generation = trans->transid;
76195853 4847 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 4848
5dc562c5
JB
4849 /*
4850 * We could have gotten an inode number from somebody who was fsynced
4851 * and then removed in this same transaction, so let's just set full
4852 * sync since it will be a full sync anyway and this will blow away the
4853 * old info in the log.
4854 */
4855 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4856
569254b0 4857 if (S_ISDIR(mode))
39279cc3
CM
4858 owner = 0;
4859 else
4860 owner = 1;
9c58309d
CM
4861
4862 key[0].objectid = objectid;
4863 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4864 key[0].offset = 0;
4865
f186373f
MF
4866 /*
4867 * Start new inodes with an inode_ref. This is slightly more
4868 * efficient for small numbers of hard links since they will
4869 * be packed into one item. Extended refs will kick in if we
4870 * add more hard links than can fit in the ref item.
4871 */
9c58309d
CM
4872 key[1].objectid = objectid;
4873 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4874 key[1].offset = ref_objectid;
4875
4876 sizes[0] = sizeof(struct btrfs_inode_item);
4877 sizes[1] = name_len + sizeof(*ref);
4878
b9473439 4879 path->leave_spinning = 1;
9c58309d
CM
4880 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4881 if (ret != 0)
5f39d397
CM
4882 goto fail;
4883
ecc11fab 4884 inode_init_owner(inode, dir, mode);
a76a3cd4 4885 inode_set_bytes(inode, 0);
39279cc3 4886 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM
4887 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4888 struct btrfs_inode_item);
293f7e07
LZ
4889 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4890 sizeof(*inode_item));
e02119d5 4891 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM
4892
4893 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4894 struct btrfs_inode_ref);
4895 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4896 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM
4897 ptr = (unsigned long)(ref + 1);
4898 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4899
5f39d397
CM
4900 btrfs_mark_buffer_dirty(path->nodes[0]);
4901 btrfs_free_path(path);
4902
39279cc3
CM
4903 location = &BTRFS_I(inode)->location;
4904 location->objectid = objectid;
39279cc3
CM
4905 location->offset = 0;
4906 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4907
6cbff00f
CH
4908 btrfs_inherit_iflags(inode, dir);
4909
569254b0 4910 if (S_ISREG(mode)) {
94272164
CM
4911 if (btrfs_test_opt(root, NODATASUM))
4912 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 4913 if (btrfs_test_opt(root, NODATACOW))
94272164
CM
4914 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4915 }
4916
39279cc3 4917 insert_inode_hash(inode);
5d4f98a2 4918 inode_tree_add(inode);
1abe9b8a 4919
4920 trace_btrfs_inode_new(inode);
1973f0fa 4921 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4922
8ea05e3a
AB
4923 btrfs_update_root_times(trans, root);
4924
39279cc3 4925 return inode;
5f39d397 4926fail:
aec7477b
JB
4927 if (dir)
4928 BTRFS_I(dir)->index_cnt--;
5f39d397 4929 btrfs_free_path(path);
09771430 4930 iput(inode);
5f39d397 4931 return ERR_PTR(ret);
39279cc3
CM
4932}
4933
4934static inline u8 btrfs_inode_type(struct inode *inode)
4935{
4936 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4937}
4938
d352ac68
CM
4939/*
4940 * utility function to add 'inode' into 'parent_inode' with
4941 * a give name and a given sequence number.
4942 * if 'add_backref' is true, also insert a backref from the
4943 * inode to the parent directory.
4944 */
e02119d5
CM
4945int btrfs_add_link(struct btrfs_trans_handle *trans,
4946 struct inode *parent_inode, struct inode *inode,
4947 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4948{
4df27c4d 4949 int ret = 0;
39279cc3 4950 struct btrfs_key key;
e02119d5 4951 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ
4952 u64 ino = btrfs_ino(inode);
4953 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4954
33345d01 4955 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ
4956 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4957 } else {
33345d01 4958 key.objectid = ino;
4df27c4d
YZ
4959 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4960 key.offset = 0;
4961 }
4962
33345d01 4963 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ
4964 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4965 key.objectid, root->root_key.objectid,
33345d01 4966 parent_ino, index, name, name_len);
4df27c4d 4967 } else if (add_backref) {
33345d01
LZ
4968 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4969 parent_ino, index);
4df27c4d 4970 }
39279cc3 4971
79787eaa
JM
4972 /* Nothing to clean up yet */
4973 if (ret)
4974 return ret;
4df27c4d 4975
79787eaa
JM
4976 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4977 parent_inode, &key,
4978 btrfs_inode_type(inode), index);
9c52057c 4979 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM
4980 goto fail_dir_item;
4981 else if (ret) {
4982 btrfs_abort_transaction(trans, root, ret);
4983 return ret;
39279cc3 4984 }
79787eaa
JM
4985
4986 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4987 name_len * 2);
0c4d2d95 4988 inode_inc_iversion(parent_inode);
79787eaa
JM
4989 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4990 ret = btrfs_update_inode(trans, root, parent_inode);
4991 if (ret)
4992 btrfs_abort_transaction(trans, root, ret);
39279cc3 4993 return ret;
fe66a05a
CM
4994
4995fail_dir_item:
4996 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4997 u64 local_index;
4998 int err;
4999 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5000 key.objectid, root->root_key.objectid,
5001 parent_ino, &local_index, name, name_len);
5002
5003 } else if (add_backref) {
5004 u64 local_index;
5005 int err;
5006
5007 err = btrfs_del_inode_ref(trans, root, name, name_len,
5008 ino, parent_ino, &local_index);
5009 }
5010 return ret;
39279cc3
CM
5011}
5012
5013static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB
5014 struct inode *dir, struct dentry *dentry,
5015 struct inode *inode, int backref, u64 index)
39279cc3 5016{
a1b075d2
JB
5017 int err = btrfs_add_link(trans, dir, inode,
5018 dentry->d_name.name, dentry->d_name.len,
5019 backref, index);
39279cc3
CM
5020 if (err > 0)
5021 err = -EEXIST;
5022 return err;
5023}
5024
618e21d5 5025static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5026 umode_t mode, dev_t rdev)
618e21d5
JB
5027{
5028 struct btrfs_trans_handle *trans;
5029 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5030 struct inode *inode = NULL;
618e21d5
JB
5031 int err;
5032 int drop_inode = 0;
5033 u64 objectid;
00e4e6b3 5034 u64 index = 0;
618e21d5
JB
5035
5036 if (!new_valid_dev(rdev))
5037 return -EINVAL;
5038
9ed74f2d
JB
5039 /*
5040 * 2 for inode item and ref
5041 * 2 for dir items
5042 * 1 for xattr if selinux is on
5043 */
a22285a6
YZ
5044 trans = btrfs_start_transaction(root, 5);
5045 if (IS_ERR(trans))
5046 return PTR_ERR(trans);
1832a6d5 5047
581bb050
LZ
5048 err = btrfs_find_free_ino(root, &objectid);
5049 if (err)
5050 goto out_unlock;
5051
aec7477b 5052 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5053 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5054 mode, &index);
7cf96da3
TI
5055 if (IS_ERR(inode)) {
5056 err = PTR_ERR(inode);
618e21d5 5057 goto out_unlock;
7cf96da3 5058 }
618e21d5 5059
2a7dba39 5060 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB
5061 if (err) {
5062 drop_inode = 1;
5063 goto out_unlock;
5064 }
5065
ad19db71
CS
5066 /*
5067 * If the active LSM wants to access the inode during
5068 * d_instantiate it needs these. Smack checks to see
5069 * if the filesystem supports xattrs by looking at the
5070 * ops vector.
5071 */
5072
5073 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5074 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB
5075 if (err)
5076 drop_inode = 1;
5077 else {
618e21d5 5078 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5079 btrfs_update_inode(trans, root, inode);
08c422c2 5080 d_instantiate(dentry, inode);
618e21d5 5081 }
618e21d5 5082out_unlock:
7ad85bb7 5083 btrfs_end_transaction(trans, root);
b53d3f5d 5084 btrfs_btree_balance_dirty(root);
618e21d5
JB
5085 if (drop_inode) {
5086 inode_dec_link_count(inode);
5087 iput(inode);
5088 }
618e21d5
JB
5089 return err;
5090}
5091
39279cc3 5092static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5093 umode_t mode, bool excl)
39279cc3
CM
5094{
5095 struct btrfs_trans_handle *trans;
5096 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5097 struct inode *inode = NULL;
43baa579 5098 int drop_inode_on_err = 0;
a22285a6 5099 int err;
39279cc3 5100 u64 objectid;
00e4e6b3 5101 u64 index = 0;
39279cc3 5102
9ed74f2d
JB
5103 /*
5104 * 2 for inode item and ref
5105 * 2 for dir items
5106 * 1 for xattr if selinux is on
5107 */
a22285a6
YZ
5108 trans = btrfs_start_transaction(root, 5);
5109 if (IS_ERR(trans))
5110 return PTR_ERR(trans);
9ed74f2d 5111
581bb050
LZ
5112 err = btrfs_find_free_ino(root, &objectid);
5113 if (err)
5114 goto out_unlock;
5115
aec7477b 5116 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5117 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5118 mode, &index);
7cf96da3
TI
5119 if (IS_ERR(inode)) {
5120 err = PTR_ERR(inode);
39279cc3 5121 goto out_unlock;
7cf96da3 5122 }
43baa579 5123 drop_inode_on_err = 1;
39279cc3 5124
2a7dba39 5125 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5126 if (err)
33268eaf 5127 goto out_unlock;
33268eaf 5128
9185aa58
FB
5129 err = btrfs_update_inode(trans, root, inode);
5130 if (err)
5131 goto out_unlock;
5132
ad19db71
CS
5133 /*
5134 * If the active LSM wants to access the inode during
5135 * d_instantiate it needs these. Smack checks to see
5136 * if the filesystem supports xattrs by looking at the
5137 * ops vector.
5138 */
5139 inode->i_fop = &btrfs_file_operations;
5140 inode->i_op = &btrfs_file_inode_operations;
5141
a1b075d2 5142 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5143 if (err)
43baa579
FB
5144 goto out_unlock;
5145
5146 inode->i_mapping->a_ops = &btrfs_aops;
5147 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5148 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5149 d_instantiate(dentry, inode);
5150
39279cc3 5151out_unlock:
7ad85bb7 5152 btrfs_end_transaction(trans, root);
43baa579 5153 if (err && drop_inode_on_err) {
39279cc3
CM
5154 inode_dec_link_count(inode);
5155 iput(inode);
5156 }
b53d3f5d 5157 btrfs_btree_balance_dirty(root);
39279cc3
CM
5158 return err;
5159}
5160
5161static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5162 struct dentry *dentry)
5163{
5164 struct btrfs_trans_handle *trans;
5165 struct btrfs_root *root = BTRFS_I(dir)->root;
5166 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5167 u64 index;
39279cc3
CM
5168 int err;
5169 int drop_inode = 0;
5170
4a8be425
TH
5171 /* do not allow sys_link's with other subvols of the same device */
5172 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5173 return -EXDEV;
4a8be425 5174
f186373f 5175 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5176 return -EMLINK;
4a8be425 5177
3de4586c 5178 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB
5179 if (err)
5180 goto fail;
5181
a22285a6 5182 /*
7e6b6465 5183 * 2 items for inode and inode ref
a22285a6 5184 * 2 items for dir items
7e6b6465 5185 * 1 item for parent inode
a22285a6 5186 */
7e6b6465 5187 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ
5188 if (IS_ERR(trans)) {
5189 err = PTR_ERR(trans);
5190 goto fail;
5191 }
5f39d397 5192
3153495d 5193 btrfs_inc_nlink(inode);
0c4d2d95 5194 inode_inc_iversion(inode);
3153495d 5195 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5196 ihold(inode);
e9976151 5197 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5198
a1b075d2 5199 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5200
a5719521 5201 if (err) {
54aa1f4d 5202 drop_inode = 1;
a5719521 5203 } else {
10d9f309 5204 struct dentry *parent = dentry->d_parent;
a5719521 5205 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM
5206 if (err)
5207 goto fail;
08c422c2 5208 d_instantiate(dentry, inode);
6a912213 5209 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5210 }
39279cc3 5211
7ad85bb7 5212 btrfs_end_transaction(trans, root);
1832a6d5 5213fail:
39279cc3
CM
5214 if (drop_inode) {
5215 inode_dec_link_count(inode);
5216 iput(inode);
5217 }
b53d3f5d 5218 btrfs_btree_balance_dirty(root);
39279cc3
CM
5219 return err;
5220}
5221
18bb1db3 5222static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5223{
b9d86667 5224 struct inode *inode = NULL;
39279cc3
CM
5225 struct btrfs_trans_handle *trans;
5226 struct btrfs_root *root = BTRFS_I(dir)->root;
5227 int err = 0;
5228 int drop_on_err = 0;
b9d86667 5229 u64 objectid = 0;
00e4e6b3 5230 u64 index = 0;
39279cc3 5231
9ed74f2d
JB
5232 /*
5233 * 2 items for inode and ref
5234 * 2 items for dir items
5235 * 1 for xattr if selinux is on
5236 */
a22285a6
YZ
5237 trans = btrfs_start_transaction(root, 5);
5238 if (IS_ERR(trans))
5239 return PTR_ERR(trans);
39279cc3 5240
581bb050
LZ
5241 err = btrfs_find_free_ino(root, &objectid);
5242 if (err)
5243 goto out_fail;
5244
aec7477b 5245 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5246 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5247 S_IFDIR | mode, &index);
39279cc3
CM
5248 if (IS_ERR(inode)) {
5249 err = PTR_ERR(inode);
5250 goto out_fail;
5251 }
5f39d397 5252
39279cc3 5253 drop_on_err = 1;
33268eaf 5254
2a7dba39 5255 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB
5256 if (err)
5257 goto out_fail;
5258
39279cc3
CM
5259 inode->i_op = &btrfs_dir_inode_operations;
5260 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5261
dbe674a9 5262 btrfs_i_size_write(inode, 0);
39279cc3
CM
5263 err = btrfs_update_inode(trans, root, inode);
5264 if (err)
5265 goto out_fail;
5f39d397 5266
a1b075d2
JB
5267 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5268 dentry->d_name.len, 0, index);
39279cc3
CM
5269 if (err)
5270 goto out_fail;
5f39d397 5271
39279cc3
CM
5272 d_instantiate(dentry, inode);
5273 drop_on_err = 0;
39279cc3
CM
5274
5275out_fail:
7ad85bb7 5276 btrfs_end_transaction(trans, root);
39279cc3
CM
5277 if (drop_on_err)
5278 iput(inode);
b53d3f5d 5279 btrfs_btree_balance_dirty(root);
39279cc3
CM
5280 return err;
5281}
5282
d352ac68
CM
5283/* helper for btfs_get_extent. Given an existing extent in the tree,
5284 * and an extent that you want to insert, deal with overlap and insert
5285 * the new extent into the tree.
5286 */
3b951516
CM
5287static int merge_extent_mapping(struct extent_map_tree *em_tree,
5288 struct extent_map *existing,
e6dcd2dc
CM
5289 struct extent_map *em,
5290 u64 map_start, u64 map_len)
3b951516
CM
5291{
5292 u64 start_diff;
3b951516 5293
e6dcd2dc
CM
5294 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5295 start_diff = map_start - em->start;
5296 em->start = map_start;
5297 em->len = map_len;
c8b97818
CM
5298 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5299 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5300 em->block_start += start_diff;
c8b97818
CM
5301 em->block_len -= start_diff;
5302 }
e6dcd2dc 5303 return add_extent_mapping(em_tree, em);
3b951516
CM
5304}
5305
c8b97818
CM
5306static noinline int uncompress_inline(struct btrfs_path *path,
5307 struct inode *inode, struct page *page,
5308 size_t pg_offset, u64 extent_offset,
5309 struct btrfs_file_extent_item *item)
5310{
5311 int ret;
5312 struct extent_buffer *leaf = path->nodes[0];
5313 char *tmp;
5314 size_t max_size;
5315 unsigned long inline_size;
5316 unsigned long ptr;
261507a0 5317 int compress_type;
c8b97818
CM
5318
5319 WARN_ON(pg_offset != 0);
261507a0 5320 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM
5321 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5322 inline_size = btrfs_file_extent_inline_item_len(leaf,
5323 btrfs_item_nr(leaf, path->slots[0]));
5324 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI
5325 if (!tmp)
5326 return -ENOMEM;
c8b97818
CM
5327 ptr = btrfs_file_extent_inline_start(item);
5328
5329 read_extent_buffer(leaf, tmp, ptr, inline_size);
5330
5b050f04 5331 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ
5332 ret = btrfs_decompress(compress_type, tmp, page,
5333 extent_offset, inline_size, max_size);
c8b97818 5334 if (ret) {
7ac687d9 5335 char *kaddr = kmap_atomic(page);
c8b97818
CM
5336 unsigned long copy_size = min_t(u64,
5337 PAGE_CACHE_SIZE - pg_offset,
5338 max_size - extent_offset);
5339 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5340 kunmap_atomic(kaddr);
c8b97818
CM
5341 }
5342 kfree(tmp);
5343 return 0;
5344}
5345
d352ac68
CM
5346/*
5347 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM
5348 * the ugly parts come from merging extents from the disk with the in-ram
5349 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM
5350 * where the in-ram extents might be locked pending data=ordered completion.
5351 *
5352 * This also copies inline extents directly into the page.
5353 */
d397712b 5354
a52d9a80 5355struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5356 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM
5357 int create)
5358{
5359 int ret;
5360 int err = 0;
db94535d 5361 u64 bytenr;
a52d9a80
CM
5362 u64 extent_start = 0;
5363 u64 extent_end = 0;
33345d01 5364 u64 objectid = btrfs_ino(inode);
a52d9a80 5365 u32 found_type;
f421950f 5366 struct btrfs_path *path = NULL;
a52d9a80
CM
5367 struct btrfs_root *root = BTRFS_I(inode)->root;
5368 struct btrfs_file_extent_item *item;
5f39d397
CM
5369 struct extent_buffer *leaf;
5370 struct btrfs_key found_key;
a52d9a80
CM
5371 struct extent_map *em = NULL;
5372 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5373 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5374 struct btrfs_trans_handle *trans = NULL;
261507a0 5375 int compress_type;
a52d9a80 5376
a52d9a80 5377again:
890871be 5378 read_lock(&em_tree->lock);
d1310b2e 5379 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM
5380 if (em)
5381 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5382 read_unlock(&em_tree->lock);
d1310b2e 5383
a52d9a80 5384 if (em) {
e1c4b745
CM
5385 if (em->start > start || em->start + em->len <= start)
5386 free_extent_map(em);
5387 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM
5388 free_extent_map(em);
5389 else
5390 goto out;
a52d9a80 5391 }
172ddd60 5392 em = alloc_extent_map();
a52d9a80 5393 if (!em) {
d1310b2e
CM
5394 err = -ENOMEM;
5395 goto out;
a52d9a80 5396 }
e6dcd2dc 5397 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5398 em->start = EXTENT_MAP_HOLE;
445a6944 5399 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5400 em->len = (u64)-1;
c8b97818 5401 em->block_len = (u64)-1;
f421950f
CM
5402
5403 if (!path) {
5404 path = btrfs_alloc_path();
026fd317
JB
5405 if (!path) {
5406 err = -ENOMEM;
5407 goto out;
5408 }
5409 /*
5410 * Chances are we'll be called again, so go ahead and do
5411 * readahead
5412 */
5413 path->reada = 1;
f421950f
CM
5414 }
5415
179e29e4
CM
5416 ret = btrfs_lookup_file_extent(trans, root, path,
5417 objectid, start, trans != NULL);
a52d9a80
CM
5418 if (ret < 0) {
5419 err = ret;
5420 goto out;
5421 }
5422
5423 if (ret != 0) {
5424 if (path->slots[0] == 0)
5425 goto not_found;
5426 path->slots[0]--;
5427 }
5428
5f39d397
CM
5429 leaf = path->nodes[0];
5430 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5431 struct btrfs_file_extent_item);
a52d9a80 5432 /* are we inside the extent that was found? */
5f39d397
CM
5433 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5434 found_type = btrfs_key_type(&found_key);
5435 if (found_key.objectid != objectid ||
a52d9a80
CM
5436 found_type != BTRFS_EXTENT_DATA_KEY) {
5437 goto not_found;
5438 }
5439
5f39d397
CM
5440 found_type = btrfs_file_extent_type(leaf, item);
5441 extent_start = found_key.offset;
261507a0 5442 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ
5443 if (found_type == BTRFS_FILE_EXTENT_REG ||
5444 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5445 extent_end = extent_start +
db94535d 5446 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ
5447 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5448 size_t size;
5449 size = btrfs_file_extent_inline_len(leaf, item);
5450 extent_end = (extent_start + size + root->sectorsize - 1) &
5451 ~((u64)root->sectorsize - 1);
5452 }
5453
5454 if (start >= extent_end) {
5455 path->slots[0]++;
5456 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5457 ret = btrfs_next_leaf(root, path);
5458 if (ret < 0) {
5459 err = ret;
5460 goto out;
a52d9a80 5461 }
9036c102
YZ
5462 if (ret > 0)
5463 goto not_found;
5464 leaf = path->nodes[0];
a52d9a80 5465 }
9036c102
YZ
5466 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5467 if (found_key.objectid != objectid ||
5468 found_key.type != BTRFS_EXTENT_DATA_KEY)
5469 goto not_found;
5470 if (start + len <= found_key.offset)
5471 goto not_found;
5472 em->start = start;
70c8a91c 5473 em->orig_start = start;
9036c102
YZ
5474 em->len = found_key.offset - start;
5475 goto not_found_em;
5476 }
5477
d899e052
YZ
5478 if (found_type == BTRFS_FILE_EXTENT_REG ||
5479 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ
5480 em->start = extent_start;
5481 em->len = extent_end - extent_start;
ff5b7ee3
YZ
5482 em->orig_start = extent_start -
5483 btrfs_file_extent_offset(leaf, item);
b4939680
JB
5484 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5485 item);
db94535d
CM
5486 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5487 if (bytenr == 0) {
5f39d397 5488 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM
5489 goto insert;
5490 }
261507a0 5491 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5492 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5493 em->compress_type = compress_type;
c8b97818 5494 em->block_start = bytenr;
b4939680 5495 em->block_len = em->orig_block_len;
c8b97818
CM
5496 } else {
5497 bytenr += btrfs_file_extent_offset(leaf, item);
5498 em->block_start = bytenr;
5499 em->block_len = em->len;
d899e052
YZ
5500 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5501 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5502 }
a52d9a80
CM
5503 goto insert;
5504 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5505 unsigned long ptr;
a52d9a80 5506 char *map;
3326d1b0
CM
5507 size_t size;
5508 size_t extent_offset;
5509 size_t copy_size;
a52d9a80 5510
689f9346 5511 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5512 if (!page || create) {
689f9346 5513 em->start = extent_start;
9036c102 5514 em->len = extent_end - extent_start;
689f9346
Y
5515 goto out;
5516 }
5f39d397 5517
9036c102
YZ
5518 size = btrfs_file_extent_inline_len(leaf, item);
5519 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5520 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5521 size - extent_offset);
3326d1b0 5522 em->start = extent_start + extent_offset;
70dec807
CM
5523 em->len = (copy_size + root->sectorsize - 1) &
5524 ~((u64)root->sectorsize - 1);
b4939680 5525 em->orig_block_len = em->len;
70c8a91c 5526 em->orig_start = em->start;
261507a0 5527 if (compress_type) {
c8b97818 5528 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ
5529 em->compress_type = compress_type;
5530 }
689f9346 5531 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5532 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ
5533 if (btrfs_file_extent_compression(leaf, item) !=
5534 BTRFS_COMPRESS_NONE) {
c8b97818
CM
5535 ret = uncompress_inline(path, inode, page,
5536 pg_offset,
5537 extent_offset, item);
79787eaa 5538 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM
5539 } else {
5540 map = kmap(page);
5541 read_extent_buffer(leaf, map + pg_offset, ptr,
5542 copy_size);
93c82d57
CM
5543 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5544 memset(map + pg_offset + copy_size, 0,
5545 PAGE_CACHE_SIZE - pg_offset -
5546 copy_size);
5547 }
c8b97818
CM
5548 kunmap(page);
5549 }
179e29e4
CM
5550 flush_dcache_page(page);
5551 } else if (create && PageUptodate(page)) {
6bf7e080 5552 BUG();
179e29e4
CM
5553 if (!trans) {
5554 kunmap(page);
5555 free_extent_map(em);
5556 em = NULL;
ff5714cc 5557
b3b4aa74 5558 btrfs_release_path(path);
7a7eaa40 5559 trans = btrfs_join_transaction(root);
ff5714cc 5560
3612b495
TI
5561 if (IS_ERR(trans))
5562 return ERR_CAST(trans);
179e29e4
CM
5563 goto again;
5564 }
c8b97818 5565 map = kmap(page);
70dec807 5566 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5567 copy_size);
c8b97818 5568 kunmap(page);
179e29e4 5569 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5570 }
d1310b2e 5571 set_extent_uptodate(io_tree, em->start,
507903b8 5572 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM
5573 goto insert;
5574 } else {
31b1a2bd 5575 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM
5576 }
5577not_found:
5578 em->start = start;
70c8a91c 5579 em->orig_start = start;
d1310b2e 5580 em->len = len;
a52d9a80 5581not_found_em:
5f39d397 5582 em->block_start = EXTENT_MAP_HOLE;
9036c102 5583 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5584insert:
b3b4aa74 5585 btrfs_release_path(path);
d1310b2e 5586 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM
5587 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5588 "[%llu %llu]\n", (unsigned long long)em->start,
5589 (unsigned long long)em->len,
5590 (unsigned long long)start,
5591 (unsigned long long)len);
a52d9a80
CM
5592 err = -EIO;
5593 goto out;
5594 }
d1310b2e
CM
5595
5596 err = 0;
890871be 5597 write_lock(&em_tree->lock);
a52d9a80 5598 ret = add_extent_mapping(em_tree, em);
3b951516
CM
5599 /* it is possible that someone inserted the extent into the tree
5600 * while we had the lock dropped. It is also possible that
5601 * an overlapping map exists in the tree
5602 */
a52d9a80 5603 if (ret == -EEXIST) {
3b951516 5604 struct extent_map *existing;
e6dcd2dc
CM
5605
5606 ret = 0;
5607
3b951516 5608 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM
5609 if (existing && (existing->start > start ||
5610 existing->start + existing->len <= start)) {
5611 free_extent_map(existing);
5612 existing = NULL;
5613 }
3b951516
CM
5614 if (!existing) {
5615 existing = lookup_extent_mapping(em_tree, em->start,
5616 em->len);
5617 if (existing) {
5618 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM
5619 em, start,
5620 root->sectorsize);
3b951516
CM
5621 free_extent_map(existing);
5622 if (err) {
5623 free_extent_map(em);
5624 em = NULL;
5625 }
5626 } else {
5627 err = -EIO;
3b951516
CM
5628 free_extent_map(em);
5629 em = NULL;
5630 }
5631 } else {
5632 free_extent_map(em);
5633 em = existing;
e6dcd2dc 5634 err = 0;
a52d9a80 5635 }
a52d9a80 5636 }
890871be 5637 write_unlock(&em_tree->lock);
a52d9a80 5638out:
1abe9b8a 5639
f0bd95ea
TI
5640 if (em)
5641 trace_btrfs_get_extent(root, em);
1abe9b8a 5642
f421950f
CM
5643 if (path)
5644 btrfs_free_path(path);
a52d9a80
CM
5645 if (trans) {
5646 ret = btrfs_end_transaction(trans, root);
d397712b 5647 if (!err)
a52d9a80
CM
5648 err = ret;
5649 }
a52d9a80
CM
5650 if (err) {
5651 free_extent_map(em);
a52d9a80
CM
5652 return ERR_PTR(err);
5653 }
79787eaa 5654 BUG_ON(!em); /* Error is always set */
a52d9a80
CM
5655 return em;
5656}
5657
ec29ed5b
CM
5658struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5659 size_t pg_offset, u64 start, u64 len,
5660 int create)
5661{
5662 struct extent_map *em;
5663 struct extent_map *hole_em = NULL;
5664 u64 range_start = start;
5665 u64 end;
5666 u64 found;
5667 u64 found_end;
5668 int err = 0;
5669
5670 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5671 if (IS_ERR(em))
5672 return em;
5673 if (em) {
5674 /*
f9e4fb53
LB
5675 * if our em maps to
5676 * - a hole or
5677 * - a pre-alloc extent,
5678 * there might actually be delalloc bytes behind it.
ec29ed5b 5679 */
f9e4fb53
LB
5680 if (em->block_start != EXTENT_MAP_HOLE &&
5681 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM
5682 return em;
5683 else
5684 hole_em = em;
5685 }
5686
5687 /* check to see if we've wrapped (len == -1 or similar) */
5688 end = start + len;
5689 if (end < start)
5690 end = (u64)-1;
5691 else
5692 end -= 1;
5693
5694 em = NULL;
5695
5696 /* ok, we didn't find anything, lets look for delalloc */
5697 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5698 end, len, EXTENT_DELALLOC, 1);
5699 found_end = range_start + found;
5700 if (found_end < range_start)
5701 found_end = (u64)-1;
5702
5703 /*
5704 * we didn't find anything useful, return
5705 * the original results from get_extent()
5706 */
5707 if (range_start > end || found_end <= start) {
5708 em = hole_em;
5709 hole_em = NULL;
5710 goto out;
5711 }
5712
5713 /* adjust the range_start to make sure it doesn't
5714 * go backwards from the start they passed in
5715 */
5716 range_start = max(start,range_start);
5717 found = found_end - range_start;
5718
5719 if (found > 0) {
5720 u64 hole_start = start;
5721 u64 hole_len = len;
5722
172ddd60 5723 em = alloc_extent_map();
ec29ed5b
CM
5724 if (!em) {
5725 err = -ENOMEM;
5726 goto out;
5727 }
5728 /*
5729 * when btrfs_get_extent can't find anything it
5730 * returns one huge hole
5731 *
5732 * make sure what it found really fits our range, and
5733 * adjust to make sure it is based on the start from
5734 * the caller
5735 */
5736 if (hole_em) {
5737 u64 calc_end = extent_map_end(hole_em);
5738
5739 if (calc_end <= start || (hole_em->start > end)) {
5740 free_extent_map(hole_em);
5741 hole_em = NULL;
5742 } else {
5743 hole_start = max(hole_em->start, start);
5744 hole_len = calc_end - hole_start;
5745 }
5746 }
5747 em->bdev = NULL;
5748 if (hole_em && range_start > hole_start) {
5749 /* our hole starts before our delalloc, so we
5750 * have to return just the parts of the hole
5751 * that go until the delalloc starts
5752 */
5753 em->len = min(hole_len,
5754 range_start - hole_start);
5755 em->start = hole_start;
5756 em->orig_start = hole_start;
5757 /*
5758 * don't adjust block start at all,
5759 * it is fixed at EXTENT_MAP_HOLE
5760 */
5761 em->block_start = hole_em->block_start;
5762 em->block_len = hole_len;
f9e4fb53
LB
5763 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5764 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM
5765 } else {
5766 em->start = range_start;
5767 em->len = found;
5768 em->orig_start = range_start;
5769 em->block_start = EXTENT_MAP_DELALLOC;
5770 em->block_len = found;
5771 }
5772 } else if (hole_em) {
5773 return hole_em;
5774 }
5775out:
5776
5777 free_extent_map(hole_em);
5778 if (err) {
5779 free_extent_map(em);
5780 return ERR_PTR(err);
5781 }
5782 return em;
5783}
5784
4b46fce2
JB
5785static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5786 u64 start, u64 len)
5787{
5788 struct btrfs_root *root = BTRFS_I(inode)->root;
5789 struct btrfs_trans_handle *trans;
70c8a91c 5790 struct extent_map *em;
4b46fce2
JB
5791 struct btrfs_key ins;
5792 u64 alloc_hint;
5793 int ret;
4b46fce2 5794
7a7eaa40 5795 trans = btrfs_join_transaction(root);
3612b495
TI
5796 if (IS_ERR(trans))
5797 return ERR_CAST(trans);
4b46fce2
JB
5798
5799 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5800
5801 alloc_hint = get_extent_allocation_hint(inode, start, len);
5802 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 5803 alloc_hint, &ins, 1);
4b46fce2
JB
5804 if (ret) {
5805 em = ERR_PTR(ret);
5806 goto out;
5807 }
5808
70c8a91c
JB
5809 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5810 ins.offset, ins.offset, 0);
5811 if (IS_ERR(em))
5812 goto out;
4b46fce2
JB
5813
5814 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5815 ins.offset, ins.offset, 0);
5816 if (ret) {
5817 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5818 em = ERR_PTR(ret);
5819 }
5820out:
5821 btrfs_end_transaction(trans, root);
5822 return em;
5823}
5824
46bfbb5c
CM
5825/*
5826 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5827 * block must be cow'd
5828 */
5829static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5830 struct inode *inode, u64 offset, u64 len)
5831{
5832 struct btrfs_path *path;
5833 int ret;
5834 struct extent_buffer *leaf;
5835 struct btrfs_root *root = BTRFS_I(inode)->root;
5836 struct btrfs_file_extent_item *fi;
5837 struct btrfs_key key;
5838 u64 disk_bytenr;
5839 u64 backref_offset;
5840 u64 extent_end;
5841 u64 num_bytes;
5842 int slot;
5843 int found_type;
5844
5845 path = btrfs_alloc_path();
5846 if (!path)
5847 return -ENOMEM;
5848
33345d01 5849 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM
5850 offset, 0);
5851 if (ret < 0)
5852 goto out;
5853
5854 slot = path->slots[0];
5855 if (ret == 1) {
5856 if (slot == 0) {
5857 /* can't find the item, must cow */
5858 ret = 0;
5859 goto out;
5860 }
5861 slot--;
5862 }
5863 ret = 0;
5864 leaf = path->nodes[0];
5865 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5866 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM
5867 key.type != BTRFS_EXTENT_DATA_KEY) {
5868 /* not our file or wrong item type, must cow */
5869 goto out;
5870 }
5871
5872 if (key.offset > offset) {
5873 /* Wrong offset, must cow */
5874 goto out;
5875 }
5876
5877 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5878 found_type = btrfs_file_extent_type(leaf, fi);
5879 if (found_type != BTRFS_FILE_EXTENT_REG &&
5880 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5881 /* not a regular extent, must cow */
5882 goto out;
5883 }
5884 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5885 backref_offset = btrfs_file_extent_offset(leaf, fi);
5886
5887 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5888 if (extent_end < offset + len) {
5889 /* extent doesn't include our full range, must cow */
5890 goto out;
5891 }
5892
5893 if (btrfs_extent_readonly(root, disk_bytenr))
5894 goto out;
5895
5896 /*
5897 * look for other files referencing this extent, if we
5898 * find any we must cow
5899 */
33345d01 5900 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM
5901 key.offset - backref_offset, disk_bytenr))
5902 goto out;
5903
5904 /*
5905 * adjust disk_bytenr and num_bytes to cover just the bytes
5906 * in this extent we are about to write. If there
5907 * are any csums in that range we have to cow in order
5908 * to keep the csums correct
5909 */
5910 disk_bytenr += backref_offset;
5911 disk_bytenr += offset - key.offset;
5912 num_bytes = min(offset + len, extent_end) - offset;
5913 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5914 goto out;
5915 /*
5916 * all of the above have passed, it is safe to overwrite this extent
5917 * without cow
5918 */
5919 ret = 1;
5920out:
5921 btrfs_free_path(path);
5922 return ret;
5923}
5924
eb838e73
JB
5925static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5926 struct extent_state **cached_state, int writing)
5927{
5928 struct btrfs_ordered_extent *ordered;
5929 int ret = 0;
5930
5931 while (1) {
5932 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5933 0, cached_state);
5934 /*
5935 * We're concerned with the entire range that we're going to be
5936 * doing DIO to, so we need to make sure theres no ordered
5937 * extents in this range.
5938 */
5939 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5940 lockend - lockstart + 1);
5941
5942 /*
5943 * We need to make sure there are no buffered pages in this
5944 * range either, we could have raced between the invalidate in
5945 * generic_file_direct_write and locking the extent. The
5946 * invalidate needs to happen so that reads after a write do not
5947 * get stale data.
5948 */
5949 if (!ordered && (!writing ||
5950 !test_range_bit(&BTRFS_I(inode)->io_tree,
5951 lockstart, lockend, EXTENT_UPTODATE, 0,
5952 *cached_state)))
5953 break;
5954
5955 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5956 cached_state, GFP_NOFS);
5957
5958 if (ordered) {
5959 btrfs_start_ordered_extent(inode, ordered, 1);
5960 btrfs_put_ordered_extent(ordered);
5961 } else {
5962 /* Screw you mmap */
5963 ret = filemap_write_and_wait_range(inode->i_mapping,
5964 lockstart,
5965 lockend);
5966 if (ret)
5967 break;
5968
5969 /*
5970 * If we found a page that couldn't be invalidated just
5971 * fall back to buffered.
5972 */
5973 ret = invalidate_inode_pages2_range(inode->i_mapping,
5974 lockstart >> PAGE_CACHE_SHIFT,
5975 lockend >> PAGE_CACHE_SHIFT);
5976 if (ret)
5977 break;
5978 }
5979
5980 cond_resched();
5981 }
5982
5983 return ret;
5984}
5985
69ffb543
JB
5986static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5987 u64 len, u64 orig_start,
5988 u64 block_start, u64 block_len,
b4939680 5989 u64 orig_block_len, int type)
69ffb543
JB
5990{
5991 struct extent_map_tree *em_tree;
5992 struct extent_map *em;
5993 struct btrfs_root *root = BTRFS_I(inode)->root;
5994 int ret;
5995
5996 em_tree = &BTRFS_I(inode)->extent_tree;
5997 em = alloc_extent_map();
5998 if (!em)
5999 return ERR_PTR(-ENOMEM);
6000
6001 em->start = start;
6002 em->orig_start = orig_start;
2ab28f32
JB
6003 em->mod_start = start;
6004 em->mod_len = len;
69ffb543
JB
6005 em->len = len;
6006 em->block_len = block_len;
6007 em->block_start = block_start;
6008 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6009 em->orig_block_len = orig_block_len;
70c8a91c 6010 em->generation = -1;
69ffb543
JB
6011 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6012 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6013 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB
6014
6015 do {
6016 btrfs_drop_extent_cache(inode, em->start,
6017 em->start + em->len - 1, 0);
6018 write_lock(&em_tree->lock);
6019 ret = add_extent_mapping(em_tree, em);
70c8a91c
JB
6020 if (!ret)
6021 list_move(&em->list,
6022 &em_tree->modified_extents);
69ffb543
JB
6023 write_unlock(&em_tree->lock);
6024 } while (ret == -EEXIST);
6025
6026 if (ret) {
6027 free_extent_map(em);
6028 return ERR_PTR(ret);
6029 }
6030
6031 return em;
6032}
6033
6034
4b46fce2
JB
6035static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6036 struct buffer_head *bh_result, int create)
6037{
6038 struct extent_map *em;
6039 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6040 struct extent_state *cached_state = NULL;
4b46fce2 6041 u64 start = iblock << inode->i_blkbits;
eb838e73 6042 u64 lockstart, lockend;
4b46fce2 6043 u64 len = bh_result->b_size;
46bfbb5c 6044 struct btrfs_trans_handle *trans;
eb838e73
JB
6045 int unlock_bits = EXTENT_LOCKED;
6046 int ret;
6047
eb838e73
JB
6048 if (create) {
6049 ret = btrfs_delalloc_reserve_space(inode, len);
6050 if (ret)
6051 return ret;
6052 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
c329861d
JB
6053 } else {
6054 len = min_t(u64, len, root->sectorsize);
eb838e73
JB
6055 }
6056
c329861d
JB
6057 lockstart = start;
6058 lockend = start + len - 1;
6059
eb838e73
JB
6060 /*
6061 * If this errors out it's because we couldn't invalidate pagecache for
6062 * this range and we need to fallback to buffered.
6063 */
6064 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6065 return -ENOTBLK;
6066
6067 if (create) {
6068 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6069 lockend, EXTENT_DELALLOC, NULL,
6070 &cached_state, GFP_NOFS);
6071 if (ret)
6072 goto unlock_err;
6073 }
4b46fce2
JB
6074
6075 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB
6076 if (IS_ERR(em)) {
6077 ret = PTR_ERR(em);
6078 goto unlock_err;
6079 }
4b46fce2
JB
6080
6081 /*
6082 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6083 * io. INLINE is special, and we could probably kludge it in here, but
6084 * it's still buffered so for safety lets just fall back to the generic
6085 * buffered path.
6086 *
6087 * For COMPRESSED we _have_ to read the entire extent in so we can
6088 * decompress it, so there will be buffering required no matter what we
6089 * do, so go ahead and fallback to buffered.
6090 *
6091 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6092 * to buffered IO. Don't blame me, this is the price we pay for using
6093 * the generic code.
6094 */
6095 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6096 em->block_start == EXTENT_MAP_INLINE) {
6097 free_extent_map(em);
eb838e73
JB
6098 ret = -ENOTBLK;
6099 goto unlock_err;
4b46fce2
JB
6100 }
6101
6102 /* Just a good old fashioned hole, return */
6103 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6104 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6105 free_extent_map(em);
eb838e73
JB
6106 ret = 0;
6107 goto unlock_err;
4b46fce2
JB
6108 }
6109
6110 /*
6111 * We don't allocate a new extent in the following cases
6112 *
6113 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6114 * existing extent.
6115 * 2) The extent is marked as PREALLOC. We're good to go here and can
6116 * just use the extent.
6117 *
6118 */
46bfbb5c 6119 if (!create) {
eb838e73
JB
6120 len = min(len, em->len - (start - em->start));
6121 lockstart = start + len;
6122 goto unlock;
46bfbb5c 6123 }
4b46fce2
JB
6124
6125 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6126 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6127 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB
6128 int type;
6129 int ret;
46bfbb5c 6130 u64 block_start;
4b46fce2
JB
6131
6132 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6133 type = BTRFS_ORDERED_PREALLOC;
6134 else
6135 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6136 len = min(len, em->len - (start - em->start));
4b46fce2 6137 block_start = em->block_start + (start - em->start);
46bfbb5c
CM
6138
6139 /*
6140 * we're not going to log anything, but we do need
6141 * to make sure the current transaction stays open
6142 * while we look for nocow cross refs
6143 */
7a7eaa40 6144 trans = btrfs_join_transaction(root);
3612b495 6145 if (IS_ERR(trans))
46bfbb5c
CM
6146 goto must_cow;
6147
6148 if (can_nocow_odirect(trans, inode, start, len) == 1) {
70c8a91c 6149 u64 orig_start = em->orig_start;
b4939680 6150 u64 orig_block_len = em->orig_block_len;
69ffb543
JB
6151
6152 if (type == BTRFS_ORDERED_PREALLOC) {
6153 free_extent_map(em);
6154 em = create_pinned_em(inode, start, len,
6155 orig_start,
b4939680
JB
6156 block_start, len,
6157 orig_block_len, type);
69ffb543
JB
6158 if (IS_ERR(em)) {
6159 btrfs_end_transaction(trans, root);
6160 goto unlock_err;
6161 }
6162 }
6163
46bfbb5c
CM
6164 ret = btrfs_add_ordered_extent_dio(inode, start,
6165 block_start, len, len, type);
6166 btrfs_end_transaction(trans, root);
6167 if (ret) {
6168 free_extent_map(em);
eb838e73 6169 goto unlock_err;
46bfbb5c
CM
6170 }
6171 goto unlock;
4b46fce2 6172 }
46bfbb5c 6173 btrfs_end_transaction(trans, root);
4b46fce2 6174 }
46bfbb5c
CM
6175must_cow:
6176 /*
6177 * this will cow the extent, reset the len in case we changed
6178 * it above
6179 */
6180 len = bh_result->b_size;
70c8a91c
JB
6181 free_extent_map(em);
6182 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB
6183 if (IS_ERR(em)) {
6184 ret = PTR_ERR(em);
6185 goto unlock_err;
6186 }
46bfbb5c
CM
6187 len = min(len, em->len - (start - em->start));
6188unlock:
4b46fce2
JB
6189 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6190 inode->i_blkbits;
46bfbb5c 6191 bh_result->b_size = len;
4b46fce2
JB
6192 bh_result->b_bdev = em->bdev;
6193 set_buffer_mapped(bh_result);
c3473e83
JB
6194 if (create) {
6195 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6196 set_buffer_new(bh_result);
6197
6198 /*
6199 * Need to update the i_size under the extent lock so buffered
6200 * readers will get the updated i_size when we unlock.
6201 */
6202 if (start + len > i_size_read(inode))
6203 i_size_write(inode, start + len);
6204 }
4b46fce2 6205
eb838e73
JB
6206 /*
6207 * In the case of write we need to clear and unlock the entire range,
6208 * in the case of read we need to unlock only the end area that we
6209 * aren't using if there is any left over space.
6210 */
24c03fa5
LB
6211 if (lockstart < lockend) {
6212 if (create && len < lockend - lockstart) {
6213 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
9e8a4a8b
LB
6214 lockstart + len - 1,
6215 unlock_bits | EXTENT_DEFRAG, 1, 0,
24c03fa5
LB
6216 &cached_state, GFP_NOFS);
6217 /*
6218 * Beside unlock, we also need to cleanup reserved space
6219 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6220 */
6221 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6222 lockstart + len, lockend,
9e8a4a8b
LB
6223 unlock_bits | EXTENT_DO_ACCOUNTING |
6224 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
24c03fa5
LB
6225 } else {
6226 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6227 lockend, unlock_bits, 1, 0,
6228 &cached_state, GFP_NOFS);
6229 }
6230 } else {
eb838e73 6231 free_extent_state(cached_state);
24c03fa5 6232 }
eb838e73 6233
4b46fce2
JB
6234 free_extent_map(em);
6235
6236 return 0;
eb838e73
JB
6237
6238unlock_err:
6239 if (create)
6240 unlock_bits |= EXTENT_DO_ACCOUNTING;
6241
6242 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6243 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6244 return ret;
4b46fce2
JB
6245}
6246
6247struct btrfs_dio_private {
6248 struct inode *inode;
6249 u64 logical_offset;
6250 u64 disk_bytenr;
6251 u64 bytes;
4b46fce2 6252 void *private;
e65e1535
MX
6253
6254 /* number of bios pending for this dio */
6255 atomic_t pending_bios;
6256
6257 /* IO errors */
6258 int errors;
6259
6260 struct bio *orig_bio;
4b46fce2
JB
6261};
6262
6263static void btrfs_endio_direct_read(struct bio *bio, int err)
6264{
e65e1535 6265 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB
6266 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6267 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB
6268 struct inode *inode = dip->inode;
6269 struct btrfs_root *root = BTRFS_I(inode)->root;
6270 u64 start;
4b46fce2
JB
6271
6272 start = dip->logical_offset;
6273 do {
6274 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6275 struct page *page = bvec->bv_page;
6276 char *kaddr;
6277 u32 csum = ~(u32)0;
c329861d 6278 u64 private = ~(u32)0;
4b46fce2
JB
6279 unsigned long flags;
6280
c329861d
JB
6281 if (get_state_private(&BTRFS_I(inode)->io_tree,
6282 start, &private))
6283 goto failed;
4b46fce2 6284 local_irq_save(flags);
7ac687d9 6285 kaddr = kmap_atomic(page);
4b46fce2
JB
6286 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6287 csum, bvec->bv_len);
6288 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6289 kunmap_atomic(kaddr);
4b46fce2
JB
6290 local_irq_restore(flags);
6291
6292 flush_dcache_page(bvec->bv_page);
c329861d
JB
6293 if (csum != private) {
6294failed:
33345d01 6295 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 6296 " %llu csum %u private %u\n",
33345d01
LZ
6297 (unsigned long long)btrfs_ino(inode),
6298 (unsigned long long)start,
c329861d 6299 csum, (unsigned)private);
4b46fce2
JB
6300 err = -EIO;
6301 }
6302 }
6303
6304 start += bvec->bv_len;
4b46fce2
JB
6305 bvec++;
6306 } while (bvec <= bvec_end);
6307
6308 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6309 dip->logical_offset + dip->bytes - 1);
4b46fce2
JB
6310 bio->bi_private = dip->private;
6311
4b46fce2 6312 kfree(dip);
c0da7aa1
JB
6313
6314 /* If we had a csum failure make sure to clear the uptodate flag */
6315 if (err)
6316 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB
6317 dio_end_io(bio, err);
6318}
6319
6320static void btrfs_endio_direct_write(struct bio *bio, int err)
6321{
6322 struct btrfs_dio_private *dip = bio->bi_private;
6323 struct inode *inode = dip->inode;
6324 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6325 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM
6326 u64 ordered_offset = dip->logical_offset;
6327 u64 ordered_bytes = dip->bytes;
4b46fce2
JB
6328 int ret;
6329
6330 if (err)
6331 goto out_done;
163cf09c
CM
6332again:
6333 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6334 &ordered_offset,
5fd02043 6335 ordered_bytes, !err);
4b46fce2 6336 if (!ret)
163cf09c 6337 goto out_test;
4b46fce2 6338
5fd02043
JB
6339 ordered->work.func = finish_ordered_fn;
6340 ordered->work.flags = 0;
6341 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6342 &ordered->work);
163cf09c
CM
6343out_test:
6344 /*
6345 * our bio might span multiple ordered extents. If we haven't
6346 * completed the accounting for the whole dio, go back and try again
6347 */
6348 if (ordered_offset < dip->logical_offset + dip->bytes) {
6349 ordered_bytes = dip->logical_offset + dip->bytes -
6350 ordered_offset;
5fd02043 6351 ordered = NULL;
163cf09c
CM
6352 goto again;
6353 }
4b46fce2
JB
6354out_done:
6355 bio->bi_private = dip->private;
6356
4b46fce2 6357 kfree(dip);
c0da7aa1
JB
6358
6359 /* If we had an error make sure to clear the uptodate flag */
6360 if (err)
6361 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB
6362 dio_end_io(bio, err);
6363}
6364
eaf25d93
CM
6365static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6366 struct bio *bio, int mirror_num,
6367 unsigned long bio_flags, u64 offset)
6368{
6369 int ret;
6370 struct btrfs_root *root = BTRFS_I(inode)->root;
6371 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6372 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM
6373 return 0;
6374}
6375
e65e1535
MX
6376static void btrfs_end_dio_bio(struct bio *bio, int err)
6377{
6378 struct btrfs_dio_private *dip = bio->bi_private;
6379
6380 if (err) {
33345d01 6381 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6382 "sector %#Lx len %u err no %d\n",
33345d01 6383 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6384 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX
6385 dip->errors = 1;
6386
6387 /*
6388 * before atomic variable goto zero, we must make sure
6389 * dip->errors is perceived to be set.
6390 */
6391 smp_mb__before_atomic_dec();
6392 }
6393
6394 /* if there are more bios still pending for this dio, just exit */
6395 if (!atomic_dec_and_test(&dip->pending_bios))
6396 goto out;
6397
6398 if (dip->errors)
6399 bio_io_error(dip->orig_bio);
6400 else {
6401 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6402 bio_endio(dip->orig_bio, 0);
6403 }
6404out:
6405 bio_put(bio);
6406}
6407
6408static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6409 u64 first_sector, gfp_t gfp_flags)
6410{
6411 int nr_vecs = bio_get_nr_vecs(bdev);
6412 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6413}
6414
6415static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6416 int rw, u64 file_offset, int skip_sum,
c329861d 6417 int async_submit)
e65e1535
MX
6418{
6419 int write = rw & REQ_WRITE;
6420 struct btrfs_root *root = BTRFS_I(inode)->root;
6421 int ret;
6422
b812ce28
JB
6423 if (async_submit)
6424 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6425
e65e1535 6426 bio_get(bio);
5fd02043
JB
6427
6428 if (!write) {
6429 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6430 if (ret)
6431 goto err;
6432 }
e65e1535 6433
1ae39938
JB
6434 if (skip_sum)
6435 goto map;
6436
6437 if (write && async_submit) {
e65e1535
MX
6438 ret = btrfs_wq_submit_bio(root->fs_info,
6439 inode, rw, bio, 0, 0,
6440 file_offset,
6441 __btrfs_submit_bio_start_direct_io,
6442 __btrfs_submit_bio_done);
6443 goto err;
1ae39938
JB
6444 } else if (write) {
6445 /*
6446 * If we aren't doing async submit, calculate the csum of the
6447 * bio now.
6448 */
6449 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6450 if (ret)
6451 goto err;
c2db1073 6452 } else if (!skip_sum) {
c329861d 6453 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI
6454 if (ret)
6455 goto err;
6456 }
e65e1535 6457
1ae39938
JB
6458map:
6459 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX
6460err:
6461 bio_put(bio);
6462 return ret;
6463}
6464
6465static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6466 int skip_sum)
6467{
6468 struct inode *inode = dip->inode;
6469 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX
6470 struct bio *bio;
6471 struct bio *orig_bio = dip->orig_bio;
6472 struct bio_vec *bvec = orig_bio->bi_io_vec;
6473 u64 start_sector = orig_bio->bi_sector;
6474 u64 file_offset = dip->logical_offset;
6475 u64 submit_len = 0;
6476 u64 map_length;
6477 int nr_pages = 0;
e65e1535 6478 int ret = 0;
1ae39938 6479 int async_submit = 0;
e65e1535 6480
e65e1535 6481 map_length = orig_bio->bi_size;
3ec706c8 6482 ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
e65e1535
MX
6483 &map_length, NULL, 0);
6484 if (ret) {
64728bbb 6485 bio_put(orig_bio);
e65e1535
MX
6486 return -EIO;
6487 }
6488
02f57c7a
JB
6489 if (map_length >= orig_bio->bi_size) {
6490 bio = orig_bio;
6491 goto submit;
6492 }
6493
1ae39938 6494 async_submit = 1;
02f57c7a
JB
6495 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6496 if (!bio)
6497 return -ENOMEM;
6498 bio->bi_private = dip;
6499 bio->bi_end_io = btrfs_end_dio_bio;
6500 atomic_inc(&dip->pending_bios);
6501
e65e1535
MX
6502 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6503 if (unlikely(map_length < submit_len + bvec->bv_len ||
6504 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6505 bvec->bv_offset) < bvec->bv_len)) {
6506 /*
6507 * inc the count before we submit the bio so
6508 * we know the end IO handler won't happen before
6509 * we inc the count. Otherwise, the dip might get freed
6510 * before we're done setting it up
6511 */
6512 atomic_inc(&dip->pending_bios);
6513 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6514 file_offset, skip_sum,
c329861d 6515 async_submit);
e65e1535
MX
6516 if (ret) {
6517 bio_put(bio);
6518 atomic_dec(&dip->pending_bios);
6519 goto out_err;
6520 }
6521
e65e1535
MX
6522 start_sector += submit_len >> 9;
6523 file_offset += submit_len;
6524
6525 submit_len = 0;
6526 nr_pages = 0;
6527
6528 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6529 start_sector, GFP_NOFS);
6530 if (!bio)
6531 goto out_err;
6532 bio->bi_private = dip;
6533 bio->bi_end_io = btrfs_end_dio_bio;
6534
6535 map_length = orig_bio->bi_size;
3ec706c8
SB
6536 ret = btrfs_map_block(root->fs_info, READ,
6537 start_sector << 9,
e65e1535
MX
6538 &map_length, NULL, 0);
6539 if (ret) {
6540 bio_put(bio);
6541 goto out_err;
6542 }
6543 } else {
6544 submit_len += bvec->bv_len;
6545 nr_pages ++;
6546 bvec++;
6547 }
6548 }
6549
02f57c7a 6550submit:
e65e1535 6551 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 6552 async_submit);
e65e1535
MX
6553 if (!ret)
6554 return 0;
6555
6556 bio_put(bio);
6557out_err:
6558 dip->errors = 1;
6559 /*
6560 * before atomic variable goto zero, we must
6561 * make sure dip->errors is perceived to be set.
6562 */
6563 smp_mb__before_atomic_dec();
6564 if (atomic_dec_and_test(&dip->pending_bios))
6565 bio_io_error(dip->orig_bio);
6566
6567 /* bio_end_io() will handle error, so we needn't return it */
6568 return 0;
6569}
6570
4b46fce2
JB
6571static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6572 loff_t file_offset)
6573{
6574 struct btrfs_root *root = BTRFS_I(inode)->root;
6575 struct btrfs_dio_private *dip;
6576 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6577 int skip_sum;
7b6d91da 6578 int write = rw & REQ_WRITE;
4b46fce2
JB
6579 int ret = 0;
6580
6581 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6582
6583 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6584 if (!dip) {
6585 ret = -ENOMEM;
6586 goto free_ordered;
6587 }
4b46fce2
JB
6588
6589 dip->private = bio->bi_private;
6590 dip->inode = inode;
6591 dip->logical_offset = file_offset;
6592
4b46fce2
JB
6593 dip->bytes = 0;
6594 do {
6595 dip->bytes += bvec->bv_len;
6596 bvec++;
6597 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6598
46bfbb5c 6599 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6600 bio->bi_private = dip;
e65e1535
MX
6601 dip->errors = 0;
6602 dip->orig_bio = bio;
6603 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB
6604
6605 if (write)
6606 bio->bi_end_io = btrfs_endio_direct_write;
6607 else
6608 bio->bi_end_io = btrfs_endio_direct_read;
6609
e65e1535
MX
6610 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6611 if (!ret)
eaf25d93 6612 return;
4b46fce2
JB
6613free_ordered:
6614 /*
6615 * If this is a write, we need to clean up the reserved space and kill
6616 * the ordered extent.
6617 */
6618 if (write) {
6619 struct btrfs_ordered_extent *ordered;
955256f2 6620 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB
6621 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6622 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6623 btrfs_free_reserved_extent(root, ordered->start,
6624 ordered->disk_len);
6625 btrfs_put_ordered_extent(ordered);
6626 btrfs_put_ordered_extent(ordered);
6627 }
6628 bio_endio(bio, ret);
6629}
6630
5a5f79b5
CM
6631static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6632 const struct iovec *iov, loff_t offset,
6633 unsigned long nr_segs)
6634{
6635 int seg;
a1b75f7d 6636 int i;
5a5f79b5
CM
6637 size_t size;
6638 unsigned long addr;
6639 unsigned blocksize_mask = root->sectorsize - 1;
6640 ssize_t retval = -EINVAL;
6641 loff_t end = offset;
6642
6643 if (offset & blocksize_mask)
6644 goto out;
6645
6646 /* Check the memory alignment. Blocks cannot straddle pages */
6647 for (seg = 0; seg < nr_segs; seg++) {
6648 addr = (unsigned long)iov[seg].iov_base;
6649 size = iov[seg].iov_len;
6650 end += size;
a1b75f7d 6651 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6652 goto out;
a1b75f7d
JB
6653
6654 /* If this is a write we don't need to check anymore */
6655 if (rw & WRITE)
6656 continue;
6657
6658 /*
6659 * Check to make sure we don't have duplicate iov_base's in this
6660 * iovec, if so return EINVAL, otherwise we'll get csum errors
6661 * when reading back.
6662 */
6663 for (i = seg + 1; i < nr_segs; i++) {
6664 if (iov[seg].iov_base == iov[i].iov_base)
6665 goto out;
6666 }
5a5f79b5
CM
6667 }
6668 retval = 0;
6669out:
6670 return retval;
6671}
eb838e73 6672
16432985
CM
6673static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6674 const struct iovec *iov, loff_t offset,
6675 unsigned long nr_segs)
6676{
4b46fce2
JB
6677 struct file *file = iocb->ki_filp;
6678 struct inode *inode = file->f_mapping->host;
4b46fce2 6679
5a5f79b5 6680 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 6681 offset, nr_segs))
5a5f79b5 6682 return 0;
3f7c579c 6683
eb838e73 6684 return __blockdev_direct_IO(rw, iocb, inode,
5a5f79b5
CM
6685 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6686 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6687 btrfs_submit_direct, 0);
16432985
CM
6688}
6689
05dadc09
TI
6690#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
6691
1506fcc8
YS
6692static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6693 __u64 start, __u64 len)
6694{
05dadc09
TI
6695 int ret;
6696
6697 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6698 if (ret)
6699 return ret;
6700
ec29ed5b 6701 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS
6702}
6703
a52d9a80 6704int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6705{
d1310b2e
CM
6706 struct extent_io_tree *tree;
6707 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6708 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6709}
1832a6d5 6710
a52d9a80 6711static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6712{
d1310b2e 6713 struct extent_io_tree *tree;
b888db2b
CM
6714
6715
6716 if (current->flags & PF_MEMALLOC) {
6717 redirty_page_for_writepage(wbc, page);
6718 unlock_page(page);
6719 return 0;
6720 }
d1310b2e 6721 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6722 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM
6723}
6724
f421950f
CM
6725int btrfs_writepages(struct address_space *mapping,
6726 struct writeback_control *wbc)
b293f02e 6727{
d1310b2e 6728 struct extent_io_tree *tree;
771ed689 6729
d1310b2e 6730 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM
6731 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6732}
6733
3ab2fb5a
CM
6734static int
6735btrfs_readpages(struct file *file, struct address_space *mapping,
6736 struct list_head *pages, unsigned nr_pages)
6737{
d1310b2e
CM
6738 struct extent_io_tree *tree;
6739 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM
6740 return extent_readpages(tree, mapping, pages, nr_pages,
6741 btrfs_get_extent);
6742}
e6dcd2dc 6743static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6744{
d1310b2e
CM
6745 struct extent_io_tree *tree;
6746 struct extent_map_tree *map;
a52d9a80 6747 int ret;
8c2383c3 6748
d1310b2e
CM
6749 tree = &BTRFS_I(page->mapping->host)->io_tree;
6750 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6751 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM
6752 if (ret == 1) {
6753 ClearPagePrivate(page);
6754 set_page_private(page, 0);
6755 page_cache_release(page);
39279cc3 6756 }
a52d9a80 6757 return ret;
39279cc3
CM
6758}
6759
e6dcd2dc
CM
6760static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6761{
98509cfc
CM
6762 if (PageWriteback(page) || PageDirty(page))
6763 return 0;
b335b003 6764 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM
6765}
6766
a52d9a80 6767static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6768{
5fd02043 6769 struct inode *inode = page->mapping->host;
d1310b2e 6770 struct extent_io_tree *tree;
e6dcd2dc 6771 struct btrfs_ordered_extent *ordered;
2ac55d41 6772 struct extent_state *cached_state = NULL;
e6dcd2dc
CM
6773 u64 page_start = page_offset(page);
6774 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6775
8b62b72b
CM
6776 /*
6777 * we have the page locked, so new writeback can't start,
6778 * and the dirty bit won't be cleared while we are here.
6779 *
6780 * Wait for IO on this page so that we can safely clear
6781 * the PagePrivate2 bit and do ordered accounting
6782 */
e6dcd2dc 6783 wait_on_page_writeback(page);
8b62b72b 6784
5fd02043 6785 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM
6786 if (offset) {
6787 btrfs_releasepage(page, GFP_NOFS);
6788 return;
6789 }
d0082371 6790 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 6791 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 6792 if (ordered) {
eb84ae03
CM
6793 /*
6794 * IO on this page will never be started, so we need
6795 * to account for any ordered extents now
6796 */
e6dcd2dc
CM
6797 clear_extent_bit(tree, page_start, page_end,
6798 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB
6799 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6800 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM
6801 /*
6802 * whoever cleared the private bit is responsible
6803 * for the finish_ordered_io
6804 */
5fd02043
JB
6805 if (TestClearPagePrivate2(page) &&
6806 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6807 PAGE_CACHE_SIZE, 1)) {
6808 btrfs_finish_ordered_io(ordered);
8b62b72b 6809 }
e6dcd2dc 6810 btrfs_put_ordered_extent(ordered);
2ac55d41 6811 cached_state = NULL;
d0082371 6812 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM
6813 }
6814 clear_extent_bit(tree, page_start, page_end,
32c00aff 6815 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB
6816 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6817 &cached_state, GFP_NOFS);
e6dcd2dc
CM
6818 __btrfs_releasepage(page, GFP_NOFS);
6819
4a096752 6820 ClearPageChecked(page);
9ad6b7bc 6821 if (PagePrivate(page)) {
9ad6b7bc
CM
6822 ClearPagePrivate(page);
6823 set_page_private(page, 0);
6824 page_cache_release(page);
6825 }
39279cc3
CM
6826}
6827
9ebefb18
CM
6828/*
6829 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6830 * called from a page fault handler when a page is first dirtied. Hence we must
6831 * be careful to check for EOF conditions here. We set the page up correctly
6832 * for a written page which means we get ENOSPC checking when writing into
6833 * holes and correct delalloc and unwritten extent mapping on filesystems that
6834 * support these features.
6835 *
6836 * We are not allowed to take the i_mutex here so we have to play games to
6837 * protect against truncate races as the page could now be beyond EOF. Because
6838 * vmtruncate() writes the inode size before removing pages, once we have the
6839 * page lock we can determine safely if the page is beyond EOF. If it is not
6840 * beyond EOF, then the page is guaranteed safe against truncation until we
6841 * unlock the page.
6842 */
c2ec175c 6843int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6844{
c2ec175c 6845 struct page *page = vmf->page;
6da6abae 6846 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6847 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM
6848 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6849 struct btrfs_ordered_extent *ordered;
2ac55d41 6850 struct extent_state *cached_state = NULL;
e6dcd2dc
CM
6851 char *kaddr;
6852 unsigned long zero_start;
9ebefb18 6853 loff_t size;
1832a6d5 6854 int ret;
9998eb70 6855 int reserved = 0;
a52d9a80 6856 u64 page_start;
e6dcd2dc 6857 u64 page_end;
9ebefb18 6858
b2b5ef5c 6859 sb_start_pagefault(inode->i_sb);
0ca1f7ce 6860 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 6861 if (!ret) {
e41f941a 6862 ret = file_update_time(vma->vm_file);
9998eb70
CM
6863 reserved = 1;
6864 }
56a76f82
NP
6865 if (ret) {
6866 if (ret == -ENOMEM)
6867 ret = VM_FAULT_OOM;
6868 else /* -ENOSPC, -EIO, etc */
6869 ret = VM_FAULT_SIGBUS;
9998eb70
CM
6870 if (reserved)
6871 goto out;
6872 goto out_noreserve;
56a76f82 6873 }
1832a6d5 6874
56a76f82 6875 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6876again:
9ebefb18 6877 lock_page(page);
9ebefb18 6878 size = i_size_read(inode);
e6dcd2dc
CM
6879 page_start = page_offset(page);
6880 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6881
9ebefb18 6882 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6883 (page_start >= size)) {
9ebefb18
CM
6884 /* page got truncated out from underneath us */
6885 goto out_unlock;
6886 }
e6dcd2dc
CM
6887 wait_on_page_writeback(page);
6888
d0082371 6889 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM
6890 set_page_extent_mapped(page);
6891
eb84ae03
CM
6892 /*
6893 * we can't set the delalloc bits if there are pending ordered
6894 * extents. Drop our locks and wait for them to finish
6895 */
e6dcd2dc
CM
6896 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6897 if (ordered) {
2ac55d41
JB
6898 unlock_extent_cached(io_tree, page_start, page_end,
6899 &cached_state, GFP_NOFS);
e6dcd2dc 6900 unlock_page(page);
eb84ae03 6901 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM
6902 btrfs_put_ordered_extent(ordered);
6903 goto again;
6904 }
6905
fbf19087
JB
6906 /*
6907 * XXX - page_mkwrite gets called every time the page is dirtied, even
6908 * if it was already dirty, so for space accounting reasons we need to
6909 * clear any delalloc bits for the range we are fixing to save. There
6910 * is probably a better way to do this, but for now keep consistent with
6911 * prepare_pages in the normal write path.
6912 */
2ac55d41 6913 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB
6914 EXTENT_DIRTY | EXTENT_DELALLOC |
6915 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 6916 0, 0, &cached_state, GFP_NOFS);
fbf19087 6917
2ac55d41
JB
6918 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6919 &cached_state);
9ed74f2d 6920 if (ret) {
2ac55d41
JB
6921 unlock_extent_cached(io_tree, page_start, page_end,
6922 &cached_state, GFP_NOFS);
9ed74f2d
JB
6923 ret = VM_FAULT_SIGBUS;
6924 goto out_unlock;
6925 }
e6dcd2dc 6926 ret = 0;
9ebefb18
CM
6927
6928 /* page is wholly or partially inside EOF */
a52d9a80 6929 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6930 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6931 else
e6dcd2dc 6932 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6933
e6dcd2dc
CM
6934 if (zero_start != PAGE_CACHE_SIZE) {
6935 kaddr = kmap(page);
6936 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6937 flush_dcache_page(page);
6938 kunmap(page);
6939 }
247e743c 6940 ClearPageChecked(page);
e6dcd2dc 6941 set_page_dirty(page);
50a9b214 6942 SetPageUptodate(page);
5a3f23d5 6943
257c62e1
CM
6944 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6945 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 6946 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 6947
2ac55d41 6948 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM
6949
6950out_unlock:
b2b5ef5c
JK
6951 if (!ret) {
6952 sb_end_pagefault(inode->i_sb);
50a9b214 6953 return VM_FAULT_LOCKED;
b2b5ef5c 6954 }
9ebefb18 6955 unlock_page(page);
1832a6d5 6956out:
ec39e180 6957 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 6958out_noreserve:
b2b5ef5c 6959 sb_end_pagefault(inode->i_sb);
9ebefb18
CM
6960 return ret;
6961}
6962
a41ad394 6963static int btrfs_truncate(struct inode *inode)
39279cc3
CM
6964{
6965 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6966 struct btrfs_block_rsv *rsv;
39279cc3 6967 int ret;
3893e33b 6968 int err = 0;
39279cc3 6969 struct btrfs_trans_handle *trans;
dbe674a9 6970 u64 mask = root->sectorsize - 1;
07127184 6971 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6972
2aaa6655 6973 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
5d5e103a 6974 if (ret)
a41ad394 6975 return ret;
8082510e 6976
4a096752 6977 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6978 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6979
fcb80c2a
JB
6980 /*
6981 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6982 * 3 things going on here
6983 *
6984 * 1) We need to reserve space for our orphan item and the space to
6985 * delete our orphan item. Lord knows we don't want to have a dangling
6986 * orphan item because we didn't reserve space to remove it.
6987 *
6988 * 2) We need to reserve space to update our inode.
6989 *
6990 * 3) We need to have something to cache all the space that is going to
6991 * be free'd up by the truncate operation, but also have some slack
6992 * space reserved in case it uses space during the truncate (thank you
6993 * very much snapshotting).
6994 *
6995 * And we need these to all be seperate. The fact is we can use alot of
6996 * space doing the truncate, and we have no earthly idea how much space
6997 * we will use, so we need the truncate reservation to be seperate so it
6998 * doesn't end up using space reserved for updating the inode or
6999 * removing the orphan item. We also need to be able to stop the
7000 * transaction and start a new one, which means we need to be able to
7001 * update the inode several times, and we have no idea of knowing how
7002 * many times that will be, so we can't just reserve 1 item for the
7003 * entirety of the opration, so that has to be done seperately as well.
7004 * Then there is the orphan item, which does indeed need to be held on
7005 * to for the whole operation, and we need nobody to touch this reserved
7006 * space except the orphan code.
7007 *
7008 * So that leaves us with
7009 *
7010 * 1) root->orphan_block_rsv - for the orphan deletion.
7011 * 2) rsv - for the truncate reservation, which we will steal from the
7012 * transaction reservation.
7013 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7014 * updating the inode.
7015 */
66d8f3dd 7016 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB
7017 if (!rsv)
7018 return -ENOMEM;
4a338542 7019 rsv->size = min_size;
ca7e70f5 7020 rsv->failfast = 1;
f0cd846e 7021
907cbceb 7022 /*
07127184 7023 * 1 for the truncate slack space
907cbceb
JB
7024 * 1 for updating the inode.
7025 */
f3fe820c 7026 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB
7027 if (IS_ERR(trans)) {
7028 err = PTR_ERR(trans);
7029 goto out;
7030 }
f0cd846e 7031
907cbceb
JB
7032 /* Migrate the slack space for the truncate to our reserve */
7033 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7034 min_size);
fcb80c2a 7035 BUG_ON(ret);
f0cd846e 7036
5a3f23d5
CM
7037 /*
7038 * setattr is responsible for setting the ordered_data_close flag,
7039 * but that is only tested during the last file release. That
7040 * could happen well after the next commit, leaving a great big
7041 * window where new writes may get lost if someone chooses to write
7042 * to this file after truncating to zero
7043 *
7044 * The inode doesn't have any dirty data here, and so if we commit
7045 * this is a noop. If someone immediately starts writing to the inode
7046 * it is very likely we'll catch some of their writes in this
7047 * transaction, and the commit will find this file on the ordered
7048 * data list with good things to send down.
7049 *
7050 * This is a best effort solution, there is still a window where
7051 * using truncate to replace the contents of the file will
7052 * end up with a zero length file after a crash.
7053 */
72ac3c0d
JB
7054 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7055 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM
7056 btrfs_add_ordered_operation(trans, root, inode);
7057
5dc562c5
JB
7058 /*
7059 * So if we truncate and then write and fsync we normally would just
7060 * write the extents that changed, which is a problem if we need to
7061 * first truncate that entire inode. So set this flag so we write out
7062 * all of the extents in the inode to the sync log so we're completely
7063 * safe.
7064 */
7065 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7066 trans->block_rsv = rsv;
907cbceb 7067
8082510e
YZ
7068 while (1) {
7069 ret = btrfs_truncate_inode_items(trans, root, inode,
7070 inode->i_size,
7071 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7072 if (ret != -ENOSPC) {
3893e33b 7073 err = ret;
8082510e 7074 break;
3893e33b 7075 }
39279cc3 7076
fcb80c2a 7077 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7078 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB
7079 if (ret) {
7080 err = ret;
7081 break;
7082 }
ca7e70f5 7083
8082510e 7084 btrfs_end_transaction(trans, root);
b53d3f5d 7085 btrfs_btree_balance_dirty(root);
ca7e70f5
JB
7086
7087 trans = btrfs_start_transaction(root, 2);
7088 if (IS_ERR(trans)) {
7089 ret = err = PTR_ERR(trans);
7090 trans = NULL;
7091 break;
7092 }
7093
7094 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7095 rsv, min_size);
7096 BUG_ON(ret); /* shouldn't happen */
7097 trans->block_rsv = rsv;
8082510e
YZ
7098 }
7099
7100 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7101 trans->block_rsv = root->orphan_block_rsv;
8082510e 7102 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB
7103 if (ret)
7104 err = ret;
8082510e
YZ
7105 }
7106
917c16b2
CM
7107 if (trans) {
7108 trans->block_rsv = &root->fs_info->trans_block_rsv;
7109 ret = btrfs_update_inode(trans, root, inode);
7110 if (ret && !err)
7111 err = ret;
7b128766 7112
7ad85bb7 7113 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7114 btrfs_btree_balance_dirty(root);
917c16b2 7115 }
fcb80c2a
JB
7116
7117out:
7118 btrfs_free_block_rsv(root, rsv);
7119
3893e33b
JB
7120 if (ret && !err)
7121 err = ret;
a41ad394 7122
3893e33b 7123 return err;
39279cc3
CM
7124}
7125
d352ac68
CM
7126/*
7127 * create a new subvolume directory/inode (helper for the ioctl).
7128 */
d2fb3437 7129int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7130 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7131{
39279cc3 7132 struct inode *inode;
76dda93c 7133 int err;
00e4e6b3 7134 u64 index = 0;
39279cc3 7135
12fc9d09
FA
7136 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7137 new_dirid, new_dirid,
7138 S_IFDIR | (~current_umask() & S_IRWXUGO),
7139 &index);
54aa1f4d 7140 if (IS_ERR(inode))
f46b5a66 7141 return PTR_ERR(inode);
39279cc3
CM
7142 inode->i_op = &btrfs_dir_inode_operations;
7143 inode->i_fop = &btrfs_dir_file_operations;
7144
bfe86848 7145 set_nlink(inode, 1);
dbe674a9 7146 btrfs_i_size_write(inode, 0);
3b96362c 7147
76dda93c 7148 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7149
76dda93c 7150 iput(inode);
ce598979 7151 return err;
39279cc3
CM
7152}
7153
39279cc3
CM
7154struct inode *btrfs_alloc_inode(struct super_block *sb)
7155{
7156 struct btrfs_inode *ei;
2ead6ae7 7157 struct inode *inode;
39279cc3
CM
7158
7159 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7160 if (!ei)
7161 return NULL;
2ead6ae7
YZ
7162
7163 ei->root = NULL;
2ead6ae7 7164 ei->generation = 0;
15ee9bc7 7165 ei->last_trans = 0;
257c62e1 7166 ei->last_sub_trans = 0;
e02119d5 7167 ei->logged_trans = 0;
2ead6ae7 7168 ei->delalloc_bytes = 0;
2ead6ae7
YZ
7169 ei->disk_i_size = 0;
7170 ei->flags = 0;
7709cde3 7171 ei->csum_bytes = 0;
2ead6ae7
YZ
7172 ei->index_cnt = (u64)-1;
7173 ei->last_unlink_trans = 0;
46d8bc34 7174 ei->last_log_commit = 0;
2ead6ae7 7175
9e0baf60
JB
7176 spin_lock_init(&ei->lock);
7177 ei->outstanding_extents = 0;
7178 ei->reserved_extents = 0;
2ead6ae7 7179
72ac3c0d 7180 ei->runtime_flags = 0;
261507a0 7181 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7182
16cdcec7
MX
7183 ei->delayed_node = NULL;
7184
2ead6ae7 7185 inode = &ei->vfs_inode;
a8067e02 7186 extent_map_tree_init(&ei->extent_tree);
f993c883
DS
7187 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7188 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB
7189 ei->io_tree.track_uptodate = 1;
7190 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7191 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7192 mutex_init(&ei->log_mutex);
f248679e 7193 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7194 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7195 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7196 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ
7197 RB_CLEAR_NODE(&ei->rb_node);
7198
7199 return inode;
39279cc3
CM
7200}
7201
fa0d7e3d
NP
7202static void btrfs_i_callback(struct rcu_head *head)
7203{
7204 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP
7205 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7206}
7207
39279cc3
CM
7208void btrfs_destroy_inode(struct inode *inode)
7209{
e6dcd2dc 7210 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM
7211 struct btrfs_root *root = BTRFS_I(inode)->root;
7212
b3d9b7a3 7213 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7214 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB
7215 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7216 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB
7217 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7218 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7219
a6dbd429
JB
7220 /*
7221 * This can happen where we create an inode, but somebody else also
7222 * created the same inode and we need to destroy the one we already
7223 * created.
7224 */
7225 if (!root)
7226 goto free;
7227
5a3f23d5
CM
7228 /*
7229 * Make sure we're properly removed from the ordered operation
7230 * lists.
7231 */
7232 smp_mb();
7233 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7234 spin_lock(&root->fs_info->ordered_extent_lock);
7235 list_del_init(&BTRFS_I(inode)->ordered_operations);
7236 spin_unlock(&root->fs_info->ordered_extent_lock);
7237 }
7238
8a35d95f
JB
7239 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7240 &BTRFS_I(inode)->runtime_flags)) {
33345d01
LZ
7241 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7242 (unsigned long long)btrfs_ino(inode));
8a35d95f 7243 atomic_dec(&root->orphan_inodes);
7b128766 7244 }
7b128766 7245
d397712b 7246 while (1) {
e6dcd2dc
CM
7247 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7248 if (!ordered)
7249 break;
7250 else {
d397712b
CM
7251 printk(KERN_ERR "btrfs found ordered "
7252 "extent %llu %llu on inode cleanup\n",
7253 (unsigned long long)ordered->file_offset,
7254 (unsigned long long)ordered->len);
e6dcd2dc
CM
7255 btrfs_remove_ordered_extent(inode, ordered);
7256 btrfs_put_ordered_extent(ordered);
7257 btrfs_put_ordered_extent(ordered);
7258 }
7259 }
5d4f98a2 7260 inode_tree_del(inode);
5b21f2ed 7261 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7262free:
16cdcec7 7263 btrfs_remove_delayed_node(inode);
fa0d7e3d 7264 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM
7265}
7266
45321ac5 7267int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ
7268{
7269 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7270
0af3d00b 7271 if (btrfs_root_refs(&root->root_item) == 0 &&
83eea1f1 7272 !btrfs_is_free_space_inode(inode))
45321ac5 7273 return 1;
76dda93c 7274 else
45321ac5 7275 return generic_drop_inode(inode);
76dda93c
YZ
7276}
7277
0ee0fda0 7278static void init_once(void *foo)
39279cc3
CM
7279{
7280 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7281
7282 inode_init_once(&ei->vfs_inode);
7283}
7284
7285void btrfs_destroy_cachep(void)
7286{
8c0a8537
KS
7287 /*
7288 * Make sure all delayed rcu free inodes are flushed before we
7289 * destroy cache.
7290 */
7291 rcu_barrier();
39279cc3
CM
7292 if (btrfs_inode_cachep)
7293 kmem_cache_destroy(btrfs_inode_cachep);
7294 if (btrfs_trans_handle_cachep)
7295 kmem_cache_destroy(btrfs_trans_handle_cachep);
7296 if (btrfs_transaction_cachep)
7297 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM
7298 if (btrfs_path_cachep)
7299 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB
7300 if (btrfs_free_space_cachep)
7301 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX
7302 if (btrfs_delalloc_work_cachep)
7303 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM
7304}
7305
7306int btrfs_init_cachep(void)
7307{
837e1972 7308 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH
7309 sizeof(struct btrfs_inode), 0,
7310 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM
7311 if (!btrfs_inode_cachep)
7312 goto fail;
9601e3f6 7313
837e1972 7314 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH
7315 sizeof(struct btrfs_trans_handle), 0,
7316 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM
7317 if (!btrfs_trans_handle_cachep)
7318 goto fail;
9601e3f6 7319
837e1972 7320 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH
7321 sizeof(struct btrfs_transaction), 0,
7322 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM
7323 if (!btrfs_transaction_cachep)
7324 goto fail;
9601e3f6 7325
837e1972 7326 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH
7327 sizeof(struct btrfs_path), 0,
7328 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM
7329 if (!btrfs_path_cachep)
7330 goto fail;
9601e3f6 7331
837e1972 7332 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB
7333 sizeof(struct btrfs_free_space), 0,
7334 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7335 if (!btrfs_free_space_cachep)
7336 goto fail;
7337
8ccf6f19
MX
7338 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7339 sizeof(struct btrfs_delalloc_work), 0,
7340 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7341 NULL);
7342 if (!btrfs_delalloc_work_cachep)
7343 goto fail;
7344
39279cc3
CM
7345 return 0;
7346fail:
7347 btrfs_destroy_cachep();
7348 return -ENOMEM;
7349}
7350
7351static int btrfs_getattr(struct vfsmount *mnt,
7352 struct dentry *dentry, struct kstat *stat)
7353{
df0af1a5 7354 u64 delalloc_bytes;
39279cc3 7355 struct inode *inode = dentry->d_inode;
fadc0d8b
DS
7356 u32 blocksize = inode->i_sb->s_blocksize;
7357
39279cc3 7358 generic_fillattr(inode, stat);
0ee5dc67 7359 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7360 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX
7361
7362 spin_lock(&BTRFS_I(inode)->lock);
7363 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7364 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 7365 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 7366 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM
7367 return 0;
7368}
7369
75e7cb7f
LB
7370/*
7371 * If a file is moved, it will inherit the cow and compression flags of the new
7372 * directory.
7373 */
7374static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7375{
7376 struct btrfs_inode *b_dir = BTRFS_I(dir);
7377 struct btrfs_inode *b_inode = BTRFS_I(inode);
7378
7379 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7380 b_inode->flags |= BTRFS_INODE_NODATACOW;
7381 else
7382 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7383
bc178237 7384 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
75e7cb7f 7385 b_inode->flags |= BTRFS_INODE_COMPRESS;
bc178237
LB
7386 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7387 } else {
7388 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7389 BTRFS_INODE_NOCOMPRESS);
7390 }
75e7cb7f
LB
7391}
7392
d397712b
CM
7393static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7394 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM
7395{
7396 struct btrfs_trans_handle *trans;
7397 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7398 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM
7399 struct inode *new_inode = new_dentry->d_inode;
7400 struct inode *old_inode = old_dentry->d_inode;
7401 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7402 u64 index = 0;
4df27c4d 7403 u64 root_objectid;
39279cc3 7404 int ret;
33345d01 7405 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7406
33345d01 7407 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ
7408 return -EPERM;
7409
4df27c4d 7410 /* we only allow rename subvolume link between subvolumes */
33345d01 7411 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM
7412 return -EXDEV;
7413
33345d01
LZ
7414 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7415 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 7416 return -ENOTEMPTY;
5f39d397 7417
4df27c4d
YZ
7418 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7419 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7420 return -ENOTEMPTY;
9c52057c
CM
7421
7422
7423 /* check for collisions, even if the name isn't there */
7424 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7425 new_dentry->d_name.name,
7426 new_dentry->d_name.len);
7427
7428 if (ret) {
7429 if (ret == -EEXIST) {
7430 /* we shouldn't get
7431 * eexist without a new_inode */
7432 if (!new_inode) {
7433 WARN_ON(1);
7434 return ret;
7435 }
7436 } else {
7437 /* maybe -EOVERFLOW */
7438 return ret;
7439 }
7440 }
7441 ret = 0;
7442
5a3f23d5
CM
7443 /*
7444 * we're using rename to replace one file with another.
7445 * and the replacement file is large. Start IO on it now so
7446 * we don't add too much work to the end of the transaction
7447 */
4baf8c92 7448 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM
7449 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7450 filemap_flush(old_inode->i_mapping);
7451
76dda93c 7452 /* close the racy window with snapshot create/destroy ioctl */
33345d01 7453 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7454 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ
7455 /*
7456 * We want to reserve the absolute worst case amount of items. So if
7457 * both inodes are subvols and we need to unlink them then that would
7458 * require 4 item modifications, but if they are both normal inodes it
7459 * would require 5 item modifications, so we'll assume their normal
7460 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7461 * should cover the worst case number of items we'll modify.
7462 */
7463 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL
7464 if (IS_ERR(trans)) {
7465 ret = PTR_ERR(trans);
7466 goto out_notrans;
7467 }
76dda93c 7468
4df27c4d
YZ
7469 if (dest != root)
7470 btrfs_record_root_in_trans(trans, dest);
5f39d397 7471
a5719521
YZ
7472 ret = btrfs_set_inode_index(new_dir, &index);
7473 if (ret)
7474 goto out_fail;
5a3f23d5 7475
33345d01 7476 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ
7477 /* force full log commit if subvolume involved. */
7478 root->fs_info->last_trans_log_full_commit = trans->transid;
7479 } else {
a5719521
YZ
7480 ret = btrfs_insert_inode_ref(trans, dest,
7481 new_dentry->d_name.name,
7482 new_dentry->d_name.len,
33345d01
LZ
7483 old_ino,
7484 btrfs_ino(new_dir), index);
a5719521
YZ
7485 if (ret)
7486 goto out_fail;
4df27c4d
YZ
7487 /*
7488 * this is an ugly little race, but the rename is required
7489 * to make sure that if we crash, the inode is either at the
7490 * old name or the new one. pinning the log transaction lets
7491 * us make sure we don't allow a log commit to come in after
7492 * we unlink the name but before we add the new name back in.
7493 */
7494 btrfs_pin_log_trans(root);
7495 }
5a3f23d5
CM
7496 /*
7497 * make sure the inode gets flushed if it is replacing
7498 * something.
7499 */
33345d01 7500 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7501 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7502
0c4d2d95
JB
7503 inode_inc_iversion(old_dir);
7504 inode_inc_iversion(new_dir);
7505 inode_inc_iversion(old_inode);
39279cc3
CM
7506 old_dir->i_ctime = old_dir->i_mtime = ctime;
7507 new_dir->i_ctime = new_dir->i_mtime = ctime;
7508 old_inode->i_ctime = ctime;
5f39d397 7509
12fcfd22
CM
7510 if (old_dentry->d_parent != new_dentry->d_parent)
7511 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7512
33345d01 7513 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ
7514 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7515 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7516 old_dentry->d_name.name,
7517 old_dentry->d_name.len);
7518 } else {
92986796
AV
7519 ret = __btrfs_unlink_inode(trans, root, old_dir,
7520 old_dentry->d_inode,
7521 old_dentry->d_name.name,
7522 old_dentry->d_name.len);
7523 if (!ret)
7524 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 7525 }
79787eaa
JM
7526 if (ret) {
7527 btrfs_abort_transaction(trans, root, ret);
7528 goto out_fail;
7529 }
39279cc3
CM
7530
7531 if (new_inode) {
0c4d2d95 7532 inode_inc_iversion(new_inode);
39279cc3 7533 new_inode->i_ctime = CURRENT_TIME;
33345d01 7534 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ
7535 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7536 root_objectid = BTRFS_I(new_inode)->location.objectid;
7537 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7538 root_objectid,
7539 new_dentry->d_name.name,
7540 new_dentry->d_name.len);
7541 BUG_ON(new_inode->i_nlink == 0);
7542 } else {
7543 ret = btrfs_unlink_inode(trans, dest, new_dir,
7544 new_dentry->d_inode,
7545 new_dentry->d_name.name,
7546 new_dentry->d_name.len);
7547 }
79787eaa 7548 if (!ret && new_inode->i_nlink == 0) {
e02119d5 7549 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7550 BUG_ON(ret);
7b128766 7551 }
79787eaa
JM
7552 if (ret) {
7553 btrfs_abort_transaction(trans, root, ret);
7554 goto out_fail;
7555 }
39279cc3 7556 }
aec7477b 7557
75e7cb7f
LB
7558 fixup_inode_flags(new_dir, old_inode);
7559
4df27c4d
YZ
7560 ret = btrfs_add_link(trans, new_dir, old_inode,
7561 new_dentry->d_name.name,
a5719521 7562 new_dentry->d_name.len, 0, index);
79787eaa
JM
7563 if (ret) {
7564 btrfs_abort_transaction(trans, root, ret);
7565 goto out_fail;
7566 }
39279cc3 7567
33345d01 7568 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7569 struct dentry *parent = new_dentry->d_parent;
6a912213 7570 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ
7571 btrfs_end_log_trans(root);
7572 }
39279cc3 7573out_fail:
7ad85bb7 7574 btrfs_end_transaction(trans, root);
b44c59a8 7575out_notrans:
33345d01 7576 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7577 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7578
39279cc3
CM
7579 return ret;
7580}
7581
8ccf6f19
MX
7582static void btrfs_run_delalloc_work(struct btrfs_work *work)
7583{
7584 struct btrfs_delalloc_work *delalloc_work;
7585
7586 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7587 work);
7588 if (delalloc_work->wait)
7589 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7590 else
7591 filemap_flush(delalloc_work->inode->i_mapping);
7592
7593 if (delalloc_work->delay_iput)
7594 btrfs_add_delayed_iput(delalloc_work->inode);
7595 else
7596 iput(delalloc_work->inode);
7597 complete(&delalloc_work->completion);
7598}
7599
7600struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7601 int wait, int delay_iput)
7602{
7603 struct btrfs_delalloc_work *work;
7604
7605 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7606 if (!work)
7607 return NULL;
7608
7609 init_completion(&work->completion);
7610 INIT_LIST_HEAD(&work->list);
7611 work->inode = inode;
7612 work->wait = wait;
7613 work->delay_iput = delay_iput;
7614 work->work.func = btrfs_run_delalloc_work;
7615
7616 return work;
7617}
7618
7619void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7620{
7621 wait_for_completion(&work->completion);
7622 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7623}
7624
d352ac68
CM
7625/*
7626 * some fairly slow code that needs optimization. This walks the list
7627 * of all the inodes with pending delalloc and forces them to disk.
7628 */
24bbcf04 7629int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 7630{
ea8c2819 7631 struct btrfs_inode *binode;
5b21f2ed 7632 struct inode *inode;
8ccf6f19
MX
7633 struct btrfs_delalloc_work *work, *next;
7634 struct list_head works;
1eafa6c7 7635 struct list_head splice;
8ccf6f19 7636 int ret = 0;
ea8c2819 7637
c146afad
YZ
7638 if (root->fs_info->sb->s_flags & MS_RDONLY)
7639 return -EROFS;
7640
8ccf6f19 7641 INIT_LIST_HEAD(&works);
1eafa6c7 7642 INIT_LIST_HEAD(&splice);
63607cc8 7643
75eff68e 7644 spin_lock(&root->fs_info->delalloc_lock);
1eafa6c7
MX
7645 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7646 while (!list_empty(&splice)) {
7647 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 7648 delalloc_inodes);
1eafa6c7
MX
7649
7650 list_del_init(&binode->delalloc_inodes);
7651
5b21f2ed 7652 inode = igrab(&binode->vfs_inode);
df0af1a5
MX
7653 if (!inode) {
7654 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
7655 &binode->runtime_flags);
1eafa6c7 7656 continue;
df0af1a5 7657 }
1eafa6c7
MX
7658
7659 list_add_tail(&binode->delalloc_inodes,
7660 &root->fs_info->delalloc_inodes);
75eff68e 7661 spin_unlock(&root->fs_info->delalloc_lock);
1eafa6c7
MX
7662
7663 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7664 if (unlikely(!work)) {
7665 ret = -ENOMEM;
7666 goto out;
5b21f2ed 7667 }
1eafa6c7
MX
7668 list_add_tail(&work->list, &works);
7669 btrfs_queue_worker(&root->fs_info->flush_workers,
7670 &work->work);
7671
5b21f2ed 7672 cond_resched();
75eff68e 7673 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7674 }
75eff68e 7675 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d 7676
1eafa6c7
MX
7677 list_for_each_entry_safe(work, next, &works, list) {
7678 list_del_init(&work->list);
7679 btrfs_wait_and_free_delalloc_work(work);
7680 }
7681
8c8bee1d
CM
7682 /* the filemap_flush will queue IO into the worker threads, but
7683 * we have to make sure the IO is actually started and that
7684 * ordered extents get created before we return
7685 */
7686 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7687 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7688 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7689 wait_event(root->fs_info->async_submit_wait,
771ed689
CM
7690 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7691 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM
7692 }
7693 atomic_dec(&root->fs_info->async_submit_draining);
1eafa6c7 7694 return 0;
8ccf6f19
MX
7695out:
7696 list_for_each_entry_safe(work, next, &works, list) {
7697 list_del_init(&work->list);
7698 btrfs_wait_and_free_delalloc_work(work);
7699 }
1eafa6c7
MX
7700
7701 if (!list_empty_careful(&splice)) {
7702 spin_lock(&root->fs_info->delalloc_lock);
7703 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7704 spin_unlock(&root->fs_info->delalloc_lock);
7705 }
8ccf6f19 7706 return ret;
ea8c2819
CM
7707}
7708
39279cc3
CM
7709static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7710 const char *symname)
7711{
7712 struct btrfs_trans_handle *trans;
7713 struct btrfs_root *root = BTRFS_I(dir)->root;
7714 struct btrfs_path *path;
7715 struct btrfs_key key;
1832a6d5 7716 struct inode *inode = NULL;
39279cc3
CM
7717 int err;
7718 int drop_inode = 0;
7719 u64 objectid;
00e4e6b3 7720 u64 index = 0 ;
39279cc3
CM
7721 int name_len;
7722 int datasize;
5f39d397 7723 unsigned long ptr;
39279cc3 7724 struct btrfs_file_extent_item *ei;
5f39d397 7725 struct extent_buffer *leaf;
39279cc3
CM
7726
7727 name_len = strlen(symname) + 1;
7728 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7729 return -ENAMETOOLONG;
1832a6d5 7730
9ed74f2d
JB
7731 /*
7732 * 2 items for inode item and ref
7733 * 2 items for dir items
7734 * 1 item for xattr if selinux is on
7735 */
a22285a6
YZ
7736 trans = btrfs_start_transaction(root, 5);
7737 if (IS_ERR(trans))
7738 return PTR_ERR(trans);
1832a6d5 7739
581bb050
LZ
7740 err = btrfs_find_free_ino(root, &objectid);
7741 if (err)
7742 goto out_unlock;
7743
aec7477b 7744 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7745 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7746 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI
7747 if (IS_ERR(inode)) {
7748 err = PTR_ERR(inode);
39279cc3 7749 goto out_unlock;
7cf96da3 7750 }
39279cc3 7751
2a7dba39 7752 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB
7753 if (err) {
7754 drop_inode = 1;
7755 goto out_unlock;
7756 }
7757
ad19db71
CS
7758 /*
7759 * If the active LSM wants to access the inode during
7760 * d_instantiate it needs these. Smack checks to see
7761 * if the filesystem supports xattrs by looking at the
7762 * ops vector.
7763 */
7764 inode->i_fop = &btrfs_file_operations;
7765 inode->i_op = &btrfs_file_inode_operations;
7766
a1b075d2 7767 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM
7768 if (err)
7769 drop_inode = 1;
7770 else {
7771 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7772 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 7773 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7774 }
39279cc3
CM
7775 if (drop_inode)
7776 goto out_unlock;
7777
7778 path = btrfs_alloc_path();
d8926bb3
MF
7779 if (!path) {
7780 err = -ENOMEM;
7781 drop_inode = 1;
7782 goto out_unlock;
7783 }
33345d01 7784 key.objectid = btrfs_ino(inode);
39279cc3 7785 key.offset = 0;
39279cc3
CM
7786 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7787 datasize = btrfs_file_extent_calc_inline_size(name_len);
7788 err = btrfs_insert_empty_item(trans, root, path, &key,
7789 datasize);
54aa1f4d
CM
7790 if (err) {
7791 drop_inode = 1;
b0839166 7792 btrfs_free_path(path);
54aa1f4d
CM
7793 goto out_unlock;
7794 }
5f39d397
CM
7795 leaf = path->nodes[0];
7796 ei = btrfs_item_ptr(leaf, path->slots[0],
7797 struct btrfs_file_extent_item);
7798 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7799 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7800 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM
7801 btrfs_set_file_extent_encryption(leaf, ei, 0);
7802 btrfs_set_file_extent_compression(leaf, ei, 0);
7803 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7804 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7805
39279cc3 7806 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM
7807 write_extent_buffer(leaf, symname, ptr, name_len);
7808 btrfs_mark_buffer_dirty(leaf);
39279cc3 7809 btrfs_free_path(path);
5f39d397 7810
39279cc3
CM
7811 inode->i_op = &btrfs_symlink_inode_operations;
7812 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7813 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7814 inode_set_bytes(inode, name_len);
dbe674a9 7815 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM
7816 err = btrfs_update_inode(trans, root, inode);
7817 if (err)
7818 drop_inode = 1;
39279cc3
CM
7819
7820out_unlock:
08c422c2
AV
7821 if (!err)
7822 d_instantiate(dentry, inode);
7ad85bb7 7823 btrfs_end_transaction(trans, root);
39279cc3
CM
7824 if (drop_inode) {
7825 inode_dec_link_count(inode);
7826 iput(inode);
7827 }
b53d3f5d 7828 btrfs_btree_balance_dirty(root);
39279cc3
CM
7829 return err;
7830}
16432985 7831
0af3d00b
JB
7832static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7833 u64 start, u64 num_bytes, u64 min_size,
7834 loff_t actual_len, u64 *alloc_hint,
7835 struct btrfs_trans_handle *trans)
d899e052 7836{
5dc562c5
JB
7837 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7838 struct extent_map *em;
d899e052
YZ
7839 struct btrfs_root *root = BTRFS_I(inode)->root;
7840 struct btrfs_key ins;
d899e052 7841 u64 cur_offset = start;
55a61d1d 7842 u64 i_size;
d899e052 7843 int ret = 0;
0af3d00b 7844 bool own_trans = true;
d899e052 7845
0af3d00b
JB
7846 if (trans)
7847 own_trans = false;
d899e052 7848 while (num_bytes > 0) {
0af3d00b
JB
7849 if (own_trans) {
7850 trans = btrfs_start_transaction(root, 3);
7851 if (IS_ERR(trans)) {
7852 ret = PTR_ERR(trans);
7853 break;
7854 }
5a303d5d
YZ
7855 }
7856
efa56464 7857 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
81c9ad23 7858 0, *alloc_hint, &ins, 1);
5a303d5d 7859 if (ret) {
0af3d00b
JB
7860 if (own_trans)
7861 btrfs_end_transaction(trans, root);
a22285a6 7862 break;
d899e052 7863 }
5a303d5d 7864
d899e052
YZ
7865 ret = insert_reserved_file_extent(trans, inode,
7866 cur_offset, ins.objectid,
7867 ins.offset, ins.offset,
920bbbfb 7868 ins.offset, 0, 0, 0,
d899e052 7869 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM
7870 if (ret) {
7871 btrfs_abort_transaction(trans, root, ret);
7872 if (own_trans)
7873 btrfs_end_transaction(trans, root);
7874 break;
7875 }
a1ed835e
CM
7876 btrfs_drop_extent_cache(inode, cur_offset,
7877 cur_offset + ins.offset -1, 0);
5a303d5d 7878
5dc562c5
JB
7879 em = alloc_extent_map();
7880 if (!em) {
7881 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7882 &BTRFS_I(inode)->runtime_flags);
7883 goto next;
7884 }
7885
7886 em->start = cur_offset;
7887 em->orig_start = cur_offset;
7888 em->len = ins.offset;
7889 em->block_start = ins.objectid;
7890 em->block_len = ins.offset;
b4939680 7891 em->orig_block_len = ins.offset;
5dc562c5
JB
7892 em->bdev = root->fs_info->fs_devices->latest_bdev;
7893 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7894 em->generation = trans->transid;
7895
7896 while (1) {
7897 write_lock(&em_tree->lock);
7898 ret = add_extent_mapping(em_tree, em);
7899 if (!ret)
7900 list_move(&em->list,
7901 &em_tree->modified_extents);
7902 write_unlock(&em_tree->lock);
7903 if (ret != -EEXIST)
7904 break;
7905 btrfs_drop_extent_cache(inode, cur_offset,
7906 cur_offset + ins.offset - 1,
7907 0);
7908 }
7909 free_extent_map(em);
7910next:
d899e052
YZ
7911 num_bytes -= ins.offset;
7912 cur_offset += ins.offset;
efa56464 7913 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7914
0c4d2d95 7915 inode_inc_iversion(inode);
d899e052 7916 inode->i_ctime = CURRENT_TIME;
6cbff00f 7917 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7918 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ
7919 (actual_len > inode->i_size) &&
7920 (cur_offset > inode->i_size)) {
d1ea6a61 7921 if (cur_offset > actual_len)
55a61d1d 7922 i_size = actual_len;
d1ea6a61 7923 else
55a61d1d
JB
7924 i_size = cur_offset;
7925 i_size_write(inode, i_size);
7926 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ
7927 }
7928
d899e052 7929 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM
7930
7931 if (ret) {
7932 btrfs_abort_transaction(trans, root, ret);
7933 if (own_trans)
7934 btrfs_end_transaction(trans, root);
7935 break;
7936 }
d899e052 7937
0af3d00b
JB
7938 if (own_trans)
7939 btrfs_end_transaction(trans, root);
5a303d5d 7940 }
d899e052
YZ
7941 return ret;
7942}
7943
0af3d00b
JB
7944int btrfs_prealloc_file_range(struct inode *inode, int mode,
7945 u64 start, u64 num_bytes, u64 min_size,
7946 loff_t actual_len, u64 *alloc_hint)
7947{
7948 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7949 min_size, actual_len, alloc_hint,
7950 NULL);
7951}
7952
7953int btrfs_prealloc_file_range_trans(struct inode *inode,
7954 struct btrfs_trans_handle *trans, int mode,
7955 u64 start, u64 num_bytes, u64 min_size,
7956 loff_t actual_len, u64 *alloc_hint)
7957{
7958 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7959 min_size, actual_len, alloc_hint, trans);
7960}
7961
e6dcd2dc
CM
7962static int btrfs_set_page_dirty(struct page *page)
7963{
e6dcd2dc
CM
7964 return __set_page_dirty_nobuffers(page);
7965}
7966
10556cb2 7967static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7968{
b83cc969 7969 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7970 umode_t mode = inode->i_mode;
b83cc969 7971
cb6db4e5
JM
7972 if (mask & MAY_WRITE &&
7973 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7974 if (btrfs_root_readonly(root))
7975 return -EROFS;
7976 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7977 return -EACCES;
7978 }
2830ba7f 7979 return generic_permission(inode, mask);
fdebe2bd 7980}
39279cc3 7981
6e1d5dcc 7982static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7983 .getattr = btrfs_getattr,
39279cc3
CM
7984 .lookup = btrfs_lookup,
7985 .create = btrfs_create,
7986 .unlink = btrfs_unlink,
7987 .link = btrfs_link,
7988 .mkdir = btrfs_mkdir,
7989 .rmdir = btrfs_rmdir,
7990 .rename = btrfs_rename,
7991 .symlink = btrfs_symlink,
7992 .setattr = btrfs_setattr,
618e21d5 7993 .mknod = btrfs_mknod,
95819c05
CH
7994 .setxattr = btrfs_setxattr,
7995 .getxattr = btrfs_getxattr,
5103e947 7996 .listxattr = btrfs_listxattr,
95819c05 7997 .removexattr = btrfs_removexattr,
fdebe2bd 7998 .permission = btrfs_permission,
4e34e719 7999 .get_acl = btrfs_get_acl,
39279cc3 8000};
6e1d5dcc 8001static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8002 .lookup = btrfs_lookup,
fdebe2bd 8003 .permission = btrfs_permission,
4e34e719 8004 .get_acl = btrfs_get_acl,
39279cc3 8005};
76dda93c 8006
828c0950 8007static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM
8008 .llseek = generic_file_llseek,
8009 .read = generic_read_dir,
cbdf5a24 8010 .readdir = btrfs_real_readdir,
34287aa3 8011 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8012#ifdef CONFIG_COMPAT
34287aa3 8013 .compat_ioctl = btrfs_ioctl,
39279cc3 8014#endif
6bf13c0c 8015 .release = btrfs_release_file,
e02119d5 8016 .fsync = btrfs_sync_file,
39279cc3
CM
8017};
8018
d1310b2e 8019static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8020 .fill_delalloc = run_delalloc_range,
065631f6 8021 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8022 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8023 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8024 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8025 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM
8026 .set_bit_hook = btrfs_set_bit_hook,
8027 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB
8028 .merge_extent_hook = btrfs_merge_extent_hook,
8029 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM
8030};
8031
35054394
CM
8032/*
8033 * btrfs doesn't support the bmap operation because swapfiles
8034 * use bmap to make a mapping of extents in the file. They assume
8035 * these extents won't change over the life of the file and they
8036 * use the bmap result to do IO directly to the drive.
8037 *
8038 * the btrfs bmap call would return logical addresses that aren't
8039 * suitable for IO and they also will change frequently as COW
8040 * operations happen. So, swapfile + btrfs == corruption.
8041 *
8042 * For now we're avoiding this by dropping bmap.
8043 */
7f09410b 8044static const struct address_space_operations btrfs_aops = {
39279cc3
CM
8045 .readpage = btrfs_readpage,
8046 .writepage = btrfs_writepage,
b293f02e 8047 .writepages = btrfs_writepages,
3ab2fb5a 8048 .readpages = btrfs_readpages,
16432985 8049 .direct_IO = btrfs_direct_IO,
a52d9a80
CM
8050 .invalidatepage = btrfs_invalidatepage,
8051 .releasepage = btrfs_releasepage,
e6dcd2dc 8052 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8053 .error_remove_page = generic_error_remove_page,
39279cc3
CM
8054};
8055
7f09410b 8056static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM
8057 .readpage = btrfs_readpage,
8058 .writepage = btrfs_writepage,
2bf5a725
CM
8059 .invalidatepage = btrfs_invalidatepage,
8060 .releasepage = btrfs_releasepage,
39279cc3
CM
8061};
8062
6e1d5dcc 8063static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM
8064 .getattr = btrfs_getattr,
8065 .setattr = btrfs_setattr,
95819c05
CH
8066 .setxattr = btrfs_setxattr,
8067 .getxattr = btrfs_getxattr,
5103e947 8068 .listxattr = btrfs_listxattr,
95819c05 8069 .removexattr = btrfs_removexattr,
fdebe2bd 8070 .permission = btrfs_permission,
1506fcc8 8071 .fiemap = btrfs_fiemap,
4e34e719 8072 .get_acl = btrfs_get_acl,
e41f941a 8073 .update_time = btrfs_update_time,
39279cc3 8074};
6e1d5dcc 8075static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB
8076 .getattr = btrfs_getattr,
8077 .setattr = btrfs_setattr,
fdebe2bd 8078 .permission = btrfs_permission,
95819c05
CH
8079 .setxattr = btrfs_setxattr,
8080 .getxattr = btrfs_getxattr,
33268eaf 8081 .listxattr = btrfs_listxattr,
95819c05 8082 .removexattr = btrfs_removexattr,
4e34e719 8083 .get_acl = btrfs_get_acl,
e41f941a 8084 .update_time = btrfs_update_time,
618e21d5 8085};
6e1d5dcc 8086static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM
8087 .readlink = generic_readlink,
8088 .follow_link = page_follow_link_light,
8089 .put_link = page_put_link,
f209561a 8090 .getattr = btrfs_getattr,
22c44fe6 8091 .setattr = btrfs_setattr,
fdebe2bd 8092 .permission = btrfs_permission,
0279b4cd
JO
8093 .setxattr = btrfs_setxattr,
8094 .getxattr = btrfs_getxattr,
8095 .listxattr = btrfs_listxattr,
8096 .removexattr = btrfs_removexattr,
4e34e719 8097 .get_acl = btrfs_get_acl,
e41f941a 8098 .update_time = btrfs_update_time,
39279cc3 8099};
76dda93c 8100
82d339d9 8101const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8102 .d_delete = btrfs_dentry_delete,
b4aff1f8 8103 .d_release = btrfs_dentry_release,
76dda93c 8104};