projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix tree logs parallel sync
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
29#include <linux/smp_lock.h>
30#include <linux/backing-dev.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
9ebefb18 36#include <linux/bit_spinlock.h>
5103e947 37#include <linux/xattr.h>
33268eaf 38#include <linux/posix_acl.h>
d899e052 39#include <linux/falloc.h>
4b4e25f2 40#include "compat.h"
39279cc3
CM		 41#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
0b86a832 47#include "volumes.h"
e6dcd2dc 48#include "ordered-data.h"
95819c05 49#include "xattr.h"
e02119d5 50#include "tree-log.h"
5b84e8d6 51#include "ref-cache.h"
c8b97818 52#include "compression.h"
39279cc3
CM		 53
54struct btrfs_iget_args {
55 u64 ino;
56 struct btrfs_root *root;
57};
58
59static struct inode_operations btrfs_dir_inode_operations;
60static struct inode_operations btrfs_symlink_inode_operations;
61static struct inode_operations btrfs_dir_ro_inode_operations;
618e21d5 62static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM		 63static struct inode_operations btrfs_file_inode_operations;
64static struct address_space_operations btrfs_aops;
65static struct address_space_operations btrfs_symlink_aops;
66static struct file_operations btrfs_dir_file_operations;
d1310b2e 67static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 68
69static struct kmem_cache *btrfs_inode_cachep;
70struct kmem_cache *btrfs_trans_handle_cachep;
71struct kmem_cache *btrfs_transaction_cachep;
72struct kmem_cache *btrfs_bit_radix_cachep;
73struct kmem_cache *btrfs_path_cachep;
74
75#define S_SHIFT 12
76static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
78 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
79 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
80 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
81 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
82 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
83 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84};
85
7b128766 86static void btrfs_truncate(struct inode *inode);
c8b97818 87static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 88static noinline int cow_file_range(struct inode *inode,
89 struct page *locked_page,
90 u64 start, u64 end, int *page_started,
91 unsigned long *nr_written, int unlock);
7b128766 92
d352ac68
CM		 93/*
94 * a very lame attempt at stopping writes when the FS is 85% full. There
95 * are countless ways this is incorrect, but it is better than nothing.
96 */
1832a6d5
CM		 97int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
98 int for_del)
99{
a2135011
CM		 100 u64 total;
101 u64 used;
1832a6d5
CM		 102 u64 thresh;
103 int ret = 0;
104
75eff68e 105 spin_lock(&root->fs_info->delalloc_lock);
a2135011
CM		 106 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
107 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
1832a6d5 108 if (for_del)
f9ef6604 109 thresh = total * 90;
1832a6d5 110 else
f9ef6604
CM		 111 thresh = total * 85;
112
113 do_div(thresh, 100);
1832a6d5 114
1832a6d5
CM		 115 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
116 ret = -ENOSPC;
75eff68e 117 spin_unlock(&root->fs_info->delalloc_lock);
1832a6d5
CM		 118 return ret;
119}
120
c8b97818
CM		 121/*
122 * this does all the hard work for inserting an inline extent into
123 * the btree. The caller should have done a btrfs_drop_extents so that
124 * no overlapping inline items exist in the btree
125 */
d397712b 126static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 127 struct btrfs_root *root, struct inode *inode,
128 u64 start, size_t size, size_t compressed_size,
129 struct page **compressed_pages)
130{
131 struct btrfs_key key;
132 struct btrfs_path *path;
133 struct extent_buffer *leaf;
134 struct page *page = NULL;
135 char *kaddr;
136 unsigned long ptr;
137 struct btrfs_file_extent_item *ei;
138 int err = 0;
139 int ret;
140 size_t cur_size = size;
141 size_t datasize;
142 unsigned long offset;
143 int use_compress = 0;
144
145 if (compressed_size && compressed_pages) {
146 use_compress = 1;
147 cur_size = compressed_size;
148 }
149
d397712b
CM		 150 path = btrfs_alloc_path();
151 if (!path)
c8b97818
CM		 152 return -ENOMEM;
153
154 btrfs_set_trans_block_group(trans, inode);
155
156 key.objectid = inode->i_ino;
157 key.offset = start;
158 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 159 datasize = btrfs_file_extent_calc_inline_size(cur_size);
160
161 inode_add_bytes(inode, size);
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
163 datasize);
164 BUG_ON(ret);
165 if (ret) {
166 err = ret;
c8b97818
CM		 167 goto fail;
168 }
169 leaf = path->nodes[0];
170 ei = btrfs_item_ptr(leaf, path->slots[0],
171 struct btrfs_file_extent_item);
172 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
173 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
174 btrfs_set_file_extent_encryption(leaf, ei, 0);
175 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
176 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
177 ptr = btrfs_file_extent_inline_start(ei);
178
179 if (use_compress) {
180 struct page *cpage;
181 int i = 0;
d397712b 182 while (compressed_size > 0) {
c8b97818 183 cpage = compressed_pages[i];
5b050f04 184 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 185 PAGE_CACHE_SIZE);
186
187 kaddr = kmap(cpage);
188 write_extent_buffer(leaf, kaddr, ptr, cur_size);
189 kunmap(cpage);
190
191 i++;
192 ptr += cur_size;
193 compressed_size -= cur_size;
194 }
195 btrfs_set_file_extent_compression(leaf, ei,
196 BTRFS_COMPRESS_ZLIB);
197 } else {
198 page = find_get_page(inode->i_mapping,
199 start >> PAGE_CACHE_SHIFT);
200 btrfs_set_file_extent_compression(leaf, ei, 0);
201 kaddr = kmap_atomic(page, KM_USER0);
202 offset = start & (PAGE_CACHE_SIZE - 1);
203 write_extent_buffer(leaf, kaddr + offset, ptr, size);
204 kunmap_atomic(kaddr, KM_USER0);
205 page_cache_release(page);
206 }
207 btrfs_mark_buffer_dirty(leaf);
208 btrfs_free_path(path);
209
210 BTRFS_I(inode)->disk_i_size = inode->i_size;
211 btrfs_update_inode(trans, root, inode);
212 return 0;
213fail:
214 btrfs_free_path(path);
215 return err;
216}
217
218
219/*
220 * conditionally insert an inline extent into the file. This
221 * does the checks required to make sure the data is small enough
222 * to fit as an inline extent.
223 */
224static int cow_file_range_inline(struct btrfs_trans_handle *trans,
225 struct btrfs_root *root,
226 struct inode *inode, u64 start, u64 end,
227 size_t compressed_size,
228 struct page **compressed_pages)
229{
230 u64 isize = i_size_read(inode);
231 u64 actual_end = min(end + 1, isize);
232 u64 inline_len = actual_end - start;
233 u64 aligned_end = (end + root->sectorsize - 1) &
234 ~((u64)root->sectorsize - 1);
235 u64 hint_byte;
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
c8b97818 252 ret = btrfs_drop_extents(trans, root, inode, start,
70b99e69 253 aligned_end, start, &hint_byte);
c8b97818
CM		 254 BUG_ON(ret);
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,
260 compressed_pages);
261 BUG_ON(ret);
262 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 263 return 0;
264}
265
771ed689
CM		 266struct async_extent {
267 u64 start;
268 u64 ram_size;
269 u64 compressed_size;
270 struct page **pages;
271 unsigned long nr_pages;
272 struct list_head list;
273};
274
275struct async_cow {
276 struct inode *inode;
277 struct btrfs_root *root;
278 struct page *locked_page;
279 u64 start;
280 u64 end;
281 struct list_head extents;
282 struct btrfs_work work;
283};
284
285static noinline int add_async_extent(struct async_cow *cow,
286 u64 start, u64 ram_size,
287 u64 compressed_size,
288 struct page **pages,
289 unsigned long nr_pages)
290{
291 struct async_extent *async_extent;
292
293 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
299 list_add_tail(&async_extent->list, &cow->extents);
300 return 0;
301}
302
d352ac68 303/*
771ed689
CM		 304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
c8b97818 307 *
771ed689
CM		 308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
c8b97818 313 *
771ed689
CM		 314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
d352ac68 318 */
771ed689
CM		 319static noinline int compress_file_range(struct inode *inode,
320 struct page *locked_page,
321 u64 start, u64 end,
322 struct async_cow *async_cow,
323 int *num_added)
b888db2b
CM		 324{
325 struct btrfs_root *root = BTRFS_I(inode)->root;
326 struct btrfs_trans_handle *trans;
db94535d 327 u64 num_bytes;
c8b97818
CM		 328 u64 orig_start;
329 u64 disk_num_bytes;
db94535d 330 u64 blocksize = root->sectorsize;
c8b97818 331 u64 actual_end;
42dc7bab 332 u64 isize = i_size_read(inode);
e6dcd2dc 333 int ret = 0;
c8b97818
CM		 334 struct page **pages = NULL;
335 unsigned long nr_pages;
336 unsigned long nr_pages_ret = 0;
337 unsigned long total_compressed = 0;
338 unsigned long total_in = 0;
339 unsigned long max_compressed = 128 * 1024;
771ed689 340 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 341 int i;
342 int will_compress;
b888db2b 343
c8b97818
CM		 344 orig_start = start;
345
42dc7bab 346 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 347again:
348 will_compress = 0;
349 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
350 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 351
c8b97818
CM		 352 total_compressed = actual_end - start;
353
354 /* we want to make sure that amount of ram required to uncompress
355 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 356 * of a compressed extent to 128k. This is a crucial number
357 * because it also controls how easily we can spread reads across
358 * cpus for decompression.
359 *
360 * We also want to make sure the amount of IO required to do
361 * a random read is reasonably small, so we limit the size of
362 * a compressed extent to 128k.
c8b97818
CM		 363 */
364 total_compressed = min(total_compressed, max_uncompressed);
db94535d 365 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 366 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 367 disk_num_bytes = num_bytes;
368 total_in = 0;
369 ret = 0;
db94535d 370
771ed689
CM		 371 /*
372 * we do compression for mount -o compress and when the
373 * inode has not been flagged as nocompress. This flag can
374 * change at any time if we discover bad compression ratios.
c8b97818
CM		 375 */
376 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
377 btrfs_test_opt(root, COMPRESS)) {
378 WARN_ON(pages);
cfbc246e 379 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 380
c8b97818
CM		 381 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
382 total_compressed, pages,
383 nr_pages, &nr_pages_ret,
384 &total_in,
385 &total_compressed,
386 max_compressed);
387
388 if (!ret) {
389 unsigned long offset = total_compressed &
390 (PAGE_CACHE_SIZE - 1);
391 struct page *page = pages[nr_pages_ret - 1];
392 char *kaddr;
393
394 /* zero the tail end of the last page, we might be
395 * sending it down to disk
396 */
397 if (offset) {
398 kaddr = kmap_atomic(page, KM_USER0);
399 memset(kaddr + offset, 0,
400 PAGE_CACHE_SIZE - offset);
401 kunmap_atomic(kaddr, KM_USER0);
402 }
403 will_compress = 1;
404 }
405 }
406 if (start == 0) {
771ed689
CM		 407 trans = btrfs_join_transaction(root, 1);
408 BUG_ON(!trans);
409 btrfs_set_trans_block_group(trans, inode);
410
c8b97818 411 /* lets try to make an inline extent */
771ed689 412 if (ret || total_in < (actual_end - start)) {
c8b97818 413 /* we didn't compress the entire range, try
771ed689 414 * to make an uncompressed inline extent.
c8b97818
CM		 415 */
416 ret = cow_file_range_inline(trans, root, inode,
417 start, end, 0, NULL);
418 } else {
771ed689 419 /* try making a compressed inline extent */
c8b97818
CM		 420 ret = cow_file_range_inline(trans, root, inode,
421 start, end,
422 total_compressed, pages);
423 }
771ed689 424 btrfs_end_transaction(trans, root);
c8b97818 425 if (ret == 0) {
771ed689
CM		 426 /*
427 * inline extent creation worked, we don't need
428 * to create any more async work items. Unlock
429 * and free up our temp pages.
430 */
c8b97818
CM		 431 extent_clear_unlock_delalloc(inode,
432 &BTRFS_I(inode)->io_tree,
771ed689
CM		 433 start, end, NULL, 1, 0,
434 0, 1, 1, 1);
c8b97818
CM		 435 ret = 0;
436 goto free_pages_out;
437 }
438 }
439
440 if (will_compress) {
441 /*
442 * we aren't doing an inline extent round the compressed size
443 * up to a block size boundary so the allocator does sane
444 * things
445 */
446 total_compressed = (total_compressed + blocksize - 1) &
447 ~(blocksize - 1);
448
449 /*
450 * one last check to make sure the compression is really a
451 * win, compare the page count read with the blocks on disk
452 */
453 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
454 ~(PAGE_CACHE_SIZE - 1);
455 if (total_compressed >= total_in) {
456 will_compress = 0;
457 } else {
458 disk_num_bytes = total_compressed;
459 num_bytes = total_in;
460 }
461 }
462 if (!will_compress && pages) {
463 /*
464 * the compression code ran but failed to make things smaller,
465 * free any pages it allocated and our page pointer array
466 */
467 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 468 WARN_ON(pages[i]->mapping);
c8b97818
CM		 469 page_cache_release(pages[i]);
470 }
471 kfree(pages);
472 pages = NULL;
473 total_compressed = 0;
474 nr_pages_ret = 0;
475
476 /* flag the file so we don't compress in the future */
477 btrfs_set_flag(inode, NOCOMPRESS);
478 }
771ed689
CM		 479 if (will_compress) {
480 *num_added += 1;
c8b97818 481
771ed689
CM		 482 /* the async work queues will take care of doing actual
483 * allocation on disk for these compressed pages,
484 * and will submit them to the elevator.
485 */
486 add_async_extent(async_cow, start, num_bytes,
487 total_compressed, pages, nr_pages_ret);
179e29e4 488
42dc7bab 489 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM		 490 start += num_bytes;
491 pages = NULL;
492 cond_resched();
493 goto again;
494 }
495 } else {
496 /*
497 * No compression, but we still need to write the pages in
498 * the file we've been given so far. redirty the locked
499 * page if it corresponds to our extent and set things up
500 * for the async work queue to run cow_file_range to do
501 * the normal delalloc dance
502 */
503 if (page_offset(locked_page) >= start &&
504 page_offset(locked_page) <= end) {
505 __set_page_dirty_nobuffers(locked_page);
506 /* unlocked later on in the async handlers */
507 }
508 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
509 *num_added += 1;
510 }
3b951516 511
771ed689
CM		 512out:
513 return 0;
514
515free_pages_out:
516 for (i = 0; i < nr_pages_ret; i++) {
517 WARN_ON(pages[i]->mapping);
518 page_cache_release(pages[i]);
519 }
d397712b 520 kfree(pages);
771ed689
CM		 521
522 goto out;
523}
524
525/*
526 * phase two of compressed writeback. This is the ordered portion
527 * of the code, which only gets called in the order the work was
528 * queued. We walk all the async extents created by compress_file_range
529 * and send them down to the disk.
530 */
531static noinline int submit_compressed_extents(struct inode *inode,
532 struct async_cow *async_cow)
533{
534 struct async_extent *async_extent;
535 u64 alloc_hint = 0;
536 struct btrfs_trans_handle *trans;
537 struct btrfs_key ins;
538 struct extent_map *em;
539 struct btrfs_root *root = BTRFS_I(inode)->root;
540 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
541 struct extent_io_tree *io_tree;
542 int ret;
543
544 if (list_empty(&async_cow->extents))
545 return 0;
546
547 trans = btrfs_join_transaction(root, 1);
548
d397712b 549 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 550 async_extent = list_entry(async_cow->extents.next,
551 struct async_extent, list);
552 list_del(&async_extent->list);
c8b97818 553
771ed689
CM		 554 io_tree = &BTRFS_I(inode)->io_tree;
555
556 /* did the compression code fall back to uncompressed IO? */
557 if (!async_extent->pages) {
558 int page_started = 0;
559 unsigned long nr_written = 0;
560
561 lock_extent(io_tree, async_extent->start,
d397712b
CM		 562 async_extent->start +
563 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 564
565 /* allocate blocks */
566 cow_file_range(inode, async_cow->locked_page,
567 async_extent->start,
568 async_extent->start +
569 async_extent->ram_size - 1,
570 &page_started, &nr_written, 0);
571
572 /*
573 * if page_started, cow_file_range inserted an
574 * inline extent and took care of all the unlocking
575 * and IO for us. Otherwise, we need to submit
576 * all those pages down to the drive.
577 */
578 if (!page_started)
579 extent_write_locked_range(io_tree,
580 inode, async_extent->start,
d397712b 581 async_extent->start +
771ed689
CM		 582 async_extent->ram_size - 1,
583 btrfs_get_extent,
584 WB_SYNC_ALL);
585 kfree(async_extent);
586 cond_resched();
587 continue;
588 }
589
590 lock_extent(io_tree, async_extent->start,
591 async_extent->start + async_extent->ram_size - 1,
592 GFP_NOFS);
c8b97818 593 /*
771ed689
CM		 594 * here we're doing allocation and writeback of the
595 * compressed pages
c8b97818 596 */
771ed689
CM		 597 btrfs_drop_extent_cache(inode, async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1, 0);
600
601 ret = btrfs_reserve_extent(trans, root,
602 async_extent->compressed_size,
603 async_extent->compressed_size,
604 0, alloc_hint,
605 (u64)-1, &ins, 1);
606 BUG_ON(ret);
607 em = alloc_extent_map(GFP_NOFS);
608 em->start = async_extent->start;
609 em->len = async_extent->ram_size;
445a6944 610 em->orig_start = em->start;
c8b97818 611
771ed689
CM		 612 em->block_start = ins.objectid;
613 em->block_len = ins.offset;
614 em->bdev = root->fs_info->fs_devices->latest_bdev;
615 set_bit(EXTENT_FLAG_PINNED, &em->flags);
616 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
617
d397712b 618 while (1) {
771ed689
CM		 619 spin_lock(&em_tree->lock);
620 ret = add_extent_mapping(em_tree, em);
621 spin_unlock(&em_tree->lock);
622 if (ret != -EEXIST) {
623 free_extent_map(em);
624 break;
625 }
626 btrfs_drop_extent_cache(inode, async_extent->start,
627 async_extent->start +
628 async_extent->ram_size - 1, 0);
629 }
630
631 ret = btrfs_add_ordered_extent(inode, async_extent->start,
632 ins.objectid,
633 async_extent->ram_size,
634 ins.offset,
635 BTRFS_ORDERED_COMPRESSED);
636 BUG_ON(ret);
637
638 btrfs_end_transaction(trans, root);
639
640 /*
641 * clear dirty, set writeback and unlock the pages.
642 */
643 extent_clear_unlock_delalloc(inode,
644 &BTRFS_I(inode)->io_tree,
645 async_extent->start,
646 async_extent->start +
647 async_extent->ram_size - 1,
648 NULL, 1, 1, 0, 1, 1, 0);
649
650 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 651 async_extent->start,
652 async_extent->ram_size,
653 ins.objectid,
654 ins.offset, async_extent->pages,
655 async_extent->nr_pages);
771ed689
CM		 656
657 BUG_ON(ret);
658 trans = btrfs_join_transaction(root, 1);
659 alloc_hint = ins.objectid + ins.offset;
660 kfree(async_extent);
661 cond_resched();
662 }
663
664 btrfs_end_transaction(trans, root);
665 return 0;
666}
667
668/*
669 * when extent_io.c finds a delayed allocation range in the file,
670 * the call backs end up in this code. The basic idea is to
671 * allocate extents on disk for the range, and create ordered data structs
672 * in ram to track those extents.
673 *
674 * locked_page is the page that writepage had locked already. We use
675 * it to make sure we don't do extra locks or unlocks.
676 *
677 * *page_started is set to one if we unlock locked_page and do everything
678 * required to start IO on it. It may be clean and already done with
679 * IO when we return.
680 */
681static noinline int cow_file_range(struct inode *inode,
682 struct page *locked_page,
683 u64 start, u64 end, int *page_started,
684 unsigned long *nr_written,
685 int unlock)
686{
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct btrfs_trans_handle *trans;
689 u64 alloc_hint = 0;
690 u64 num_bytes;
691 unsigned long ram_size;
692 u64 disk_num_bytes;
693 u64 cur_alloc_size;
694 u64 blocksize = root->sectorsize;
695 u64 actual_end;
42dc7bab 696 u64 isize = i_size_read(inode);
771ed689
CM		 697 struct btrfs_key ins;
698 struct extent_map *em;
699 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
700 int ret = 0;
701
702 trans = btrfs_join_transaction(root, 1);
703 BUG_ON(!trans);
704 btrfs_set_trans_block_group(trans, inode);
705
42dc7bab 706 actual_end = min_t(u64, isize, end + 1);
771ed689
CM		 707
708 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
709 num_bytes = max(blocksize, num_bytes);
710 disk_num_bytes = num_bytes;
711 ret = 0;
712
713 if (start == 0) {
714 /* lets try to make an inline extent */
715 ret = cow_file_range_inline(trans, root, inode,
716 start, end, 0, NULL);
717 if (ret == 0) {
718 extent_clear_unlock_delalloc(inode,
719 &BTRFS_I(inode)->io_tree,
720 start, end, NULL, 1, 1,
721 1, 1, 1, 1);
722 *nr_written = *nr_written +
723 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
724 *page_started = 1;
725 ret = 0;
726 goto out;
727 }
728 }
729
730 BUG_ON(disk_num_bytes >
731 btrfs_super_total_bytes(&root->fs_info->super_copy));
732
733 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
734
d397712b 735 while (disk_num_bytes > 0) {
c8b97818 736 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 737 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 738 root->sectorsize, 0, alloc_hint,
e6dcd2dc 739 (u64)-1, &ins, 1);
d397712b
CM		 740 BUG_ON(ret);
741
e6dcd2dc
CM		 742 em = alloc_extent_map(GFP_NOFS);
743 em->start = start;
445a6944 744 em->orig_start = em->start;
c8b97818 745
771ed689
CM		 746 ram_size = ins.offset;
747 em->len = ins.offset;
c8b97818 748
e6dcd2dc 749 em->block_start = ins.objectid;
c8b97818 750 em->block_len = ins.offset;
e6dcd2dc 751 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 752 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 753
d397712b 754 while (1) {
e6dcd2dc
CM		 755 spin_lock(&em_tree->lock);
756 ret = add_extent_mapping(em_tree, em);
757 spin_unlock(&em_tree->lock);
758 if (ret != -EEXIST) {
759 free_extent_map(em);
760 break;
761 }
762 btrfs_drop_extent_cache(inode, start,
c8b97818 763 start + ram_size - 1, 0);
e6dcd2dc
CM		 764 }
765
98d20f67 766 cur_alloc_size = ins.offset;
e6dcd2dc 767 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 768 ram_size, cur_alloc_size, 0);
e6dcd2dc 769 BUG_ON(ret);
c8b97818 770
17d217fe
YZ		 771 if (root->root_key.objectid ==
772 BTRFS_DATA_RELOC_TREE_OBJECTID) {
773 ret = btrfs_reloc_clone_csums(inode, start,
774 cur_alloc_size);
775 BUG_ON(ret);
776 }
777
d397712b 778 if (disk_num_bytes < cur_alloc_size)
3b951516 779 break;
d397712b 780
c8b97818
CM		 781 /* we're not doing compressed IO, don't unlock the first
782 * page (which the caller expects to stay locked), don't
783 * clear any dirty bits and don't set any writeback bits
784 */
785 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
786 start, start + ram_size - 1,
771ed689
CM		 787 locked_page, unlock, 1,
788 1, 0, 0, 0);
c8b97818 789 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 790 num_bytes -= cur_alloc_size;
791 alloc_hint = ins.objectid + ins.offset;
792 start += cur_alloc_size;
b888db2b 793 }
b888db2b 794out:
771ed689 795 ret = 0;
b888db2b 796 btrfs_end_transaction(trans, root);
c8b97818 797
be20aa9d 798 return ret;
771ed689 799}
c8b97818 800
771ed689
CM		 801/*
802 * work queue call back to started compression on a file and pages
803 */
804static noinline void async_cow_start(struct btrfs_work *work)
805{
806 struct async_cow *async_cow;
807 int num_added = 0;
808 async_cow = container_of(work, struct async_cow, work);
809
810 compress_file_range(async_cow->inode, async_cow->locked_page,
811 async_cow->start, async_cow->end, async_cow,
812 &num_added);
813 if (num_added == 0)
814 async_cow->inode = NULL;
815}
816
817/*
818 * work queue call back to submit previously compressed pages
819 */
820static noinline void async_cow_submit(struct btrfs_work *work)
821{
822 struct async_cow *async_cow;
823 struct btrfs_root *root;
824 unsigned long nr_pages;
825
826 async_cow = container_of(work, struct async_cow, work);
827
828 root = async_cow->root;
829 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
830 PAGE_CACHE_SHIFT;
831
832 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
833
834 if (atomic_read(&root->fs_info->async_delalloc_pages) <
835 5 * 1042 * 1024 &&
836 waitqueue_active(&root->fs_info->async_submit_wait))
837 wake_up(&root->fs_info->async_submit_wait);
838
d397712b 839 if (async_cow->inode)
771ed689 840 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 841}
c8b97818 842
771ed689
CM		 843static noinline void async_cow_free(struct btrfs_work *work)
844{
845 struct async_cow *async_cow;
846 async_cow = container_of(work, struct async_cow, work);
847 kfree(async_cow);
848}
849
850static int cow_file_range_async(struct inode *inode, struct page *locked_page,
851 u64 start, u64 end, int *page_started,
852 unsigned long *nr_written)
853{
854 struct async_cow *async_cow;
855 struct btrfs_root *root = BTRFS_I(inode)->root;
856 unsigned long nr_pages;
857 u64 cur_end;
858 int limit = 10 * 1024 * 1042;
859
860 if (!btrfs_test_opt(root, COMPRESS)) {
861 return cow_file_range(inode, locked_page, start, end,
862 page_started, nr_written, 1);
863 }
864
865 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
866 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
d397712b 867 while (start < end) {
771ed689
CM		 868 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
869 async_cow->inode = inode;
870 async_cow->root = root;
871 async_cow->locked_page = locked_page;
872 async_cow->start = start;
873
874 if (btrfs_test_flag(inode, NOCOMPRESS))
875 cur_end = end;
876 else
877 cur_end = min(end, start + 512 * 1024 - 1);
878
879 async_cow->end = cur_end;
880 INIT_LIST_HEAD(&async_cow->extents);
881
882 async_cow->work.func = async_cow_start;
883 async_cow->work.ordered_func = async_cow_submit;
884 async_cow->work.ordered_free = async_cow_free;
885 async_cow->work.flags = 0;
886
771ed689
CM		 887 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
888 PAGE_CACHE_SHIFT;
889 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
890
891 btrfs_queue_worker(&root->fs_info->delalloc_workers,
892 &async_cow->work);
893
894 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
895 wait_event(root->fs_info->async_submit_wait,
896 (atomic_read(&root->fs_info->async_delalloc_pages) <
897 limit));
898 }
899
d397712b 900 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 901 atomic_read(&root->fs_info->async_delalloc_pages)) {
902 wait_event(root->fs_info->async_submit_wait,
903 (atomic_read(&root->fs_info->async_delalloc_pages) ==
904 0));
905 }
906
907 *nr_written += nr_pages;
908 start = cur_end + 1;
909 }
910 *page_started = 1;
911 return 0;
be20aa9d
CM		 912}
913
d397712b 914static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 915 u64 bytenr, u64 num_bytes)
916{
917 int ret;
918 struct btrfs_ordered_sum *sums;
919 LIST_HEAD(list);
920
07d400a6
YZ		 921 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
922 bytenr + num_bytes - 1, &list);
17d217fe
YZ		 923 if (ret == 0 && list_empty(&list))
924 return 0;
925
926 while (!list_empty(&list)) {
927 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
928 list_del(&sums->list);
929 kfree(sums);
930 }
931 return 1;
932}
933
d352ac68
CM		 934/*
935 * when nowcow writeback call back. This checks for snapshots or COW copies
936 * of the extents that exist in the file, and COWs the file as required.
937 *
938 * If no cow copies or snapshots exist, we write directly to the existing
939 * blocks on disk
940 */
c8b97818 941static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM		 942 u64 start, u64 end, int *page_started, int force,
943 unsigned long *nr_written)
be20aa9d 944{
be20aa9d 945 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 946 struct btrfs_trans_handle *trans;
be20aa9d 947 struct extent_buffer *leaf;
be20aa9d 948 struct btrfs_path *path;
80ff3856 949 struct btrfs_file_extent_item *fi;
be20aa9d 950 struct btrfs_key found_key;
80ff3856
YZ		 951 u64 cow_start;
952 u64 cur_offset;
953 u64 extent_end;
954 u64 disk_bytenr;
955 u64 num_bytes;
956 int extent_type;
957 int ret;
d899e052 958 int type;
80ff3856
YZ		 959 int nocow;
960 int check_prev = 1;
be20aa9d
CM		 961
962 path = btrfs_alloc_path();
963 BUG_ON(!path);
7ea394f1
YZ		 964 trans = btrfs_join_transaction(root, 1);
965 BUG_ON(!trans);
be20aa9d 966
80ff3856
YZ		 967 cow_start = (u64)-1;
968 cur_offset = start;
969 while (1) {
970 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
971 cur_offset, 0);
972 BUG_ON(ret < 0);
973 if (ret > 0 && path->slots[0] > 0 && check_prev) {
974 leaf = path->nodes[0];
975 btrfs_item_key_to_cpu(leaf, &found_key,
976 path->slots[0] - 1);
977 if (found_key.objectid == inode->i_ino &&
978 found_key.type == BTRFS_EXTENT_DATA_KEY)
979 path->slots[0]--;
980 }
981 check_prev = 0;
982next_slot:
983 leaf = path->nodes[0];
984 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
985 ret = btrfs_next_leaf(root, path);
986 if (ret < 0)
987 BUG_ON(1);
988 if (ret > 0)
989 break;
990 leaf = path->nodes[0];
991 }
be20aa9d 992
80ff3856
YZ		 993 nocow = 0;
994 disk_bytenr = 0;
17d217fe 995 num_bytes = 0;
80ff3856
YZ		 996 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
997
998 if (found_key.objectid > inode->i_ino ||
999 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1000 found_key.offset > end)
1001 break;
1002
1003 if (found_key.offset > cur_offset) {
1004 extent_end = found_key.offset;
1005 goto out_check;
1006 }
1007
1008 fi = btrfs_item_ptr(leaf, path->slots[0],
1009 struct btrfs_file_extent_item);
1010 extent_type = btrfs_file_extent_type(leaf, fi);
1011
d899e052
YZ		 1012 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1013 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 1014 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1015 extent_end = found_key.offset +
1016 btrfs_file_extent_num_bytes(leaf, fi);
1017 if (extent_end <= start) {
1018 path->slots[0]++;
1019 goto next_slot;
1020 }
17d217fe
YZ		 1021 if (disk_bytenr == 0)
1022 goto out_check;
80ff3856
YZ		 1023 if (btrfs_file_extent_compression(leaf, fi) ||
1024 btrfs_file_extent_encryption(leaf, fi) ||
1025 btrfs_file_extent_other_encoding(leaf, fi))
1026 goto out_check;
d899e052
YZ		 1027 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1028 goto out_check;
d2fb3437 1029 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1030 goto out_check;
17d217fe
YZ		 1031 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1032 disk_bytenr))
1033 goto out_check;
80ff3856 1034 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
17d217fe
YZ		 1035 disk_bytenr += cur_offset - found_key.offset;
1036 num_bytes = min(end + 1, extent_end) - cur_offset;
1037 /*
1038 * force cow if csum exists in the range.
1039 * this ensure that csum for a given extent are
1040 * either valid or do not exist.
1041 */
1042 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1043 goto out_check;
80ff3856
YZ		 1044 nocow = 1;
1045 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1046 extent_end = found_key.offset +
1047 btrfs_file_extent_inline_len(leaf, fi);
1048 extent_end = ALIGN(extent_end, root->sectorsize);
1049 } else {
1050 BUG_ON(1);
1051 }
1052out_check:
1053 if (extent_end <= start) {
1054 path->slots[0]++;
1055 goto next_slot;
1056 }
1057 if (!nocow) {
1058 if (cow_start == (u64)-1)
1059 cow_start = cur_offset;
1060 cur_offset = extent_end;
1061 if (cur_offset > end)
1062 break;
1063 path->slots[0]++;
1064 goto next_slot;
7ea394f1
YZ		 1065 }
1066
1067 btrfs_release_path(root, path);
80ff3856
YZ		 1068 if (cow_start != (u64)-1) {
1069 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1070 found_key.offset - 1, page_started,
1071 nr_written, 1);
80ff3856
YZ		 1072 BUG_ON(ret);
1073 cow_start = (u64)-1;
7ea394f1 1074 }
80ff3856 1075
d899e052
YZ		 1076 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1077 struct extent_map *em;
1078 struct extent_map_tree *em_tree;
1079 em_tree = &BTRFS_I(inode)->extent_tree;
1080 em = alloc_extent_map(GFP_NOFS);
1081 em->start = cur_offset;
445a6944 1082 em->orig_start = em->start;
d899e052
YZ		 1083 em->len = num_bytes;
1084 em->block_len = num_bytes;
1085 em->block_start = disk_bytenr;
1086 em->bdev = root->fs_info->fs_devices->latest_bdev;
1087 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1088 while (1) {
1089 spin_lock(&em_tree->lock);
1090 ret = add_extent_mapping(em_tree, em);
1091 spin_unlock(&em_tree->lock);
1092 if (ret != -EEXIST) {
1093 free_extent_map(em);
1094 break;
1095 }
1096 btrfs_drop_extent_cache(inode, em->start,
1097 em->start + em->len - 1, 0);
1098 }
1099 type = BTRFS_ORDERED_PREALLOC;
1100 } else {
1101 type = BTRFS_ORDERED_NOCOW;
1102 }
80ff3856
YZ		 1103
1104 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1105 num_bytes, num_bytes, type);
1106 BUG_ON(ret);
771ed689 1107
d899e052
YZ		 1108 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1109 cur_offset, cur_offset + num_bytes - 1,
771ed689 1110 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1111 cur_offset = extent_end;
1112 if (cur_offset > end)
1113 break;
be20aa9d 1114 }
80ff3856
YZ		 1115 btrfs_release_path(root, path);
1116
1117 if (cur_offset <= end && cow_start == (u64)-1)
1118 cow_start = cur_offset;
1119 if (cow_start != (u64)-1) {
1120 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1121 page_started, nr_written, 1);
80ff3856
YZ		 1122 BUG_ON(ret);
1123 }
1124
1125 ret = btrfs_end_transaction(trans, root);
1126 BUG_ON(ret);
7ea394f1 1127 btrfs_free_path(path);
80ff3856 1128 return 0;
be20aa9d
CM		 1129}
1130
d352ac68
CM		 1131/*
1132 * extent_io.c call back to do delayed allocation processing
1133 */
c8b97818 1134static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1135 u64 start, u64 end, int *page_started,
1136 unsigned long *nr_written)
be20aa9d 1137{
be20aa9d 1138 int ret;
a2135011 1139
17d217fe 1140 if (btrfs_test_flag(inode, NODATACOW))
c8b97818 1141 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1142 page_started, 1, nr_written);
d899e052
YZ		 1143 else if (btrfs_test_flag(inode, PREALLOC))
1144 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1145 page_started, 0, nr_written);
be20aa9d 1146 else
771ed689 1147 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1148 page_started, nr_written);
1832a6d5 1149
b888db2b
CM		 1150 return ret;
1151}
1152
d352ac68
CM		 1153/*
1154 * extent_io.c set_bit_hook, used to track delayed allocation
1155 * bytes in this file, and to maintain the list of inodes that
1156 * have pending delalloc work to be done.
1157 */
b2950863 1158static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1159 unsigned long old, unsigned long bits)
291d673e 1160{
75eff68e
CM		 1161 /*
1162 * set_bit and clear bit hooks normally require _irqsave/restore
1163 * but in this case, we are only testeing for the DELALLOC
1164 * bit, which is only set or cleared with irqs on
1165 */
b0c68f8b 1166 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1167 struct btrfs_root *root = BTRFS_I(inode)->root;
75eff68e 1168 spin_lock(&root->fs_info->delalloc_lock);
9069218d 1169 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1170 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1171 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1172 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1173 &root->fs_info->delalloc_inodes);
1174 }
75eff68e 1175 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1176 }
1177 return 0;
1178}
1179
d352ac68
CM		 1180/*
1181 * extent_io.c clear_bit_hook, see set_bit_hook for why
1182 */
b2950863 1183static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1184 unsigned long old, unsigned long bits)
291d673e 1185{
75eff68e
CM		 1186 /*
1187 * set_bit and clear bit hooks normally require _irqsave/restore
1188 * but in this case, we are only testeing for the DELALLOC
1189 * bit, which is only set or cleared with irqs on
1190 */
b0c68f8b 1191 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1192 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1193
75eff68e 1194 spin_lock(&root->fs_info->delalloc_lock);
b0c68f8b 1195 if (end - start + 1 > root->fs_info->delalloc_bytes) {
d397712b
CM		 1196 printk(KERN_INFO "btrfs warning: delalloc account "
1197 "%llu %llu\n",
1198 (unsigned long long)end - start + 1,
1199 (unsigned long long)
1200 root->fs_info->delalloc_bytes);
b0c68f8b 1201 root->fs_info->delalloc_bytes = 0;
9069218d 1202 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 1203 } else {
1204 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1205 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1206 }
ea8c2819
CM		 1207 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1208 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1209 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1210 }
75eff68e 1211 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1212 }
1213 return 0;
1214}
1215
d352ac68
CM		 1216/*
1217 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1218 * we don't create bios that span stripes or chunks
1219 */
239b14b3 1220int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1221 size_t size, struct bio *bio,
1222 unsigned long bio_flags)
239b14b3
CM		 1223{
1224 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1225 struct btrfs_mapping_tree *map_tree;
a62b9401 1226 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1227 u64 length = 0;
1228 u64 map_length;
239b14b3
CM		 1229 int ret;
1230
771ed689
CM		 1231 if (bio_flags & EXTENT_BIO_COMPRESSED)
1232 return 0;
1233
f2d8d74d 1234 length = bio->bi_size;
239b14b3
CM		 1235 map_tree = &root->fs_info->mapping_tree;
1236 map_length = length;
cea9e445 1237 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1238 &map_length, NULL, 0);
cea9e445 1239
d397712b 1240 if (map_length < length + size)
239b14b3 1241 return 1;
239b14b3
CM		 1242 return 0;
1243}
1244
d352ac68
CM		 1245/*
1246 * in order to insert checksums into the metadata in large chunks,
1247 * we wait until bio submission time. All the pages in the bio are
1248 * checksummed and sums are attached onto the ordered extent record.
1249 *
1250 * At IO completion time the cums attached on the ordered extent record
1251 * are inserted into the btree
1252 */
d397712b
CM		 1253static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1254 struct bio *bio, int mirror_num,
1255 unsigned long bio_flags)
065631f6 1256{
065631f6 1257 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1258 int ret = 0;
e015640f 1259
d20f7043 1260 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1261 BUG_ON(ret);
4a69a410
CM		 1262 return 0;
1263}
e015640f 1264
4a69a410
CM		 1265/*
1266 * in order to insert checksums into the metadata in large chunks,
1267 * we wait until bio submission time. All the pages in the bio are
1268 * checksummed and sums are attached onto the ordered extent record.
1269 *
1270 * At IO completion time the cums attached on the ordered extent record
1271 * are inserted into the btree
1272 */
b2950863 1273static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM		 1274 int mirror_num, unsigned long bio_flags)
1275{
1276 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1277 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1278}
1279
d352ac68 1280/*
cad321ad
CM		 1281 * extent_io.c submission hook. This does the right thing for csum calculation
1282 * on write, or reading the csums from the tree before a read
d352ac68 1283 */
b2950863 1284static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1285 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1286{
1287 struct btrfs_root *root = BTRFS_I(inode)->root;
1288 int ret = 0;
19b9bdb0 1289 int skip_sum;
44b8bd7e 1290
cad321ad
CM		 1291 skip_sum = btrfs_test_flag(inode, NODATASUM);
1292
e6dcd2dc
CM		 1293 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1294 BUG_ON(ret);
065631f6 1295
4d1b5fb4 1296 if (!(rw & (1 << BIO_RW))) {
d20f7043 1297 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1298 return btrfs_submit_compressed_read(inode, bio,
1299 mirror_num, bio_flags);
d20f7043
CM		 1300 } else if (!skip_sum)
1301 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1302 goto mapit;
19b9bdb0 1303 } else if (!skip_sum) {
17d217fe
YZ		 1304 /* csum items have already been cloned */
1305 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1306 goto mapit;
19b9bdb0
CM		 1307 /* we're doing a write, do the async checksumming */
1308 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1309 inode, rw, bio, mirror_num,
4a69a410
CM		 1310 bio_flags, __btrfs_submit_bio_start,
1311 __btrfs_submit_bio_done);
19b9bdb0
CM		 1312 }
1313
0b86a832 1314mapit:
8b712842 1315 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1316}
6885f308 1317
d352ac68
CM		 1318/*
1319 * given a list of ordered sums record them in the inode. This happens
1320 * at IO completion time based on sums calculated at bio submission time.
1321 */
ba1da2f4 1322static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1323 struct inode *inode, u64 file_offset,
1324 struct list_head *list)
1325{
e6dcd2dc
CM		 1326 struct btrfs_ordered_sum *sum;
1327
1328 btrfs_set_trans_block_group(trans, inode);
c6e30871
QF		 1329
1330 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1331 btrfs_csum_file_blocks(trans,
1332 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1333 }
1334 return 0;
1335}
1336
ea8c2819
CM		 1337int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1338{
d397712b 1339 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1340 WARN_ON(1);
ea8c2819
CM		 1341 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1342 GFP_NOFS);
1343}
1344
d352ac68 1345/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1346struct btrfs_writepage_fixup {
1347 struct page *page;
1348 struct btrfs_work work;
1349};
1350
b2950863 1351static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1352{
1353 struct btrfs_writepage_fixup *fixup;
1354 struct btrfs_ordered_extent *ordered;
1355 struct page *page;
1356 struct inode *inode;
1357 u64 page_start;
1358 u64 page_end;
1359
1360 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1361 page = fixup->page;
4a096752 1362again:
247e743c
CM		 1363 lock_page(page);
1364 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1365 ClearPageChecked(page);
1366 goto out_page;
1367 }
1368
1369 inode = page->mapping->host;
1370 page_start = page_offset(page);
1371 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1372
1373 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1374
1375 /* already ordered? We're done */
1376 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1377 EXTENT_ORDERED, 0)) {
247e743c 1378 goto out;
4a096752
CM		 1379 }
1380
1381 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1382 if (ordered) {
1383 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1384 page_end, GFP_NOFS);
1385 unlock_page(page);
1386 btrfs_start_ordered_extent(inode, ordered, 1);
1387 goto again;
1388 }
247e743c 1389
ea8c2819 1390 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1391 ClearPageChecked(page);
1392out:
1393 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1394out_page:
1395 unlock_page(page);
1396 page_cache_release(page);
1397}
1398
1399/*
1400 * There are a few paths in the higher layers of the kernel that directly
1401 * set the page dirty bit without asking the filesystem if it is a
1402 * good idea. This causes problems because we want to make sure COW
1403 * properly happens and the data=ordered rules are followed.
1404 *
c8b97818 1405 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1406 * hasn't been properly setup for IO. We kick off an async process
1407 * to fix it up. The async helper will wait for ordered extents, set
1408 * the delalloc bit and make it safe to write the page.
1409 */
b2950863 1410static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1411{
1412 struct inode *inode = page->mapping->host;
1413 struct btrfs_writepage_fixup *fixup;
1414 struct btrfs_root *root = BTRFS_I(inode)->root;
1415 int ret;
1416
1417 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1418 EXTENT_ORDERED, 0);
1419 if (ret)
1420 return 0;
1421
1422 if (PageChecked(page))
1423 return -EAGAIN;
1424
1425 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1426 if (!fixup)
1427 return -EAGAIN;
f421950f 1428
247e743c
CM		 1429 SetPageChecked(page);
1430 page_cache_get(page);
1431 fixup->work.func = btrfs_writepage_fixup_worker;
1432 fixup->page = page;
1433 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1434 return -EAGAIN;
1435}
1436
d899e052
YZ		 1437static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1438 struct inode *inode, u64 file_pos,
1439 u64 disk_bytenr, u64 disk_num_bytes,
1440 u64 num_bytes, u64 ram_bytes,
1441 u8 compression, u8 encryption,
1442 u16 other_encoding, int extent_type)
1443{
1444 struct btrfs_root *root = BTRFS_I(inode)->root;
1445 struct btrfs_file_extent_item *fi;
1446 struct btrfs_path *path;
1447 struct extent_buffer *leaf;
1448 struct btrfs_key ins;
1449 u64 hint;
1450 int ret;
1451
1452 path = btrfs_alloc_path();
1453 BUG_ON(!path);
1454
1455 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1456 file_pos + num_bytes, file_pos, &hint);
1457 BUG_ON(ret);
1458
1459 ins.objectid = inode->i_ino;
1460 ins.offset = file_pos;
1461 ins.type = BTRFS_EXTENT_DATA_KEY;
1462 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1463 BUG_ON(ret);
1464 leaf = path->nodes[0];
1465 fi = btrfs_item_ptr(leaf, path->slots[0],
1466 struct btrfs_file_extent_item);
1467 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1468 btrfs_set_file_extent_type(leaf, fi, extent_type);
1469 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1470 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1471 btrfs_set_file_extent_offset(leaf, fi, 0);
1472 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1473 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1474 btrfs_set_file_extent_compression(leaf, fi, compression);
1475 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1476 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1477 btrfs_mark_buffer_dirty(leaf);
1478
1479 inode_add_bytes(inode, num_bytes);
1480 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1481
1482 ins.objectid = disk_bytenr;
1483 ins.offset = disk_num_bytes;
1484 ins.type = BTRFS_EXTENT_ITEM_KEY;
1485 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1486 root->root_key.objectid,
1487 trans->transid, inode->i_ino, &ins);
1488 BUG_ON(ret);
1489
1490 btrfs_free_path(path);
1491 return 0;
1492}
1493
d352ac68
CM		 1494/* as ordered data IO finishes, this gets called so we can finish
1495 * an ordered extent if the range of bytes in the file it covers are
1496 * fully written.
1497 */
211f90e6 1498static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1499{
e6dcd2dc
CM		 1500 struct btrfs_root *root = BTRFS_I(inode)->root;
1501 struct btrfs_trans_handle *trans;
1502 struct btrfs_ordered_extent *ordered_extent;
1503 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1504 int compressed = 0;
e6dcd2dc
CM		 1505 int ret;
1506
1507 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1508 if (!ret)
e6dcd2dc 1509 return 0;
e6dcd2dc 1510
f9295749 1511 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1512
1513 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1514 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1515 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1516 goto nocow;
e6dcd2dc
CM		 1517
1518 lock_extent(io_tree, ordered_extent->file_offset,
1519 ordered_extent->file_offset + ordered_extent->len - 1,
1520 GFP_NOFS);
1521
c8b97818 1522 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1523 compressed = 1;
1524 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1525 BUG_ON(compressed);
1526 ret = btrfs_mark_extent_written(trans, root, inode,
1527 ordered_extent->file_offset,
1528 ordered_extent->file_offset +
1529 ordered_extent->len);
1530 BUG_ON(ret);
1531 } else {
1532 ret = insert_reserved_file_extent(trans, inode,
1533 ordered_extent->file_offset,
1534 ordered_extent->start,
1535 ordered_extent->disk_len,
1536 ordered_extent->len,
1537 ordered_extent->len,
1538 compressed, 0, 0,
1539 BTRFS_FILE_EXTENT_REG);
1540 BUG_ON(ret);
1541 }
e6dcd2dc
CM		 1542 unlock_extent(io_tree, ordered_extent->file_offset,
1543 ordered_extent->file_offset + ordered_extent->len - 1,
1544 GFP_NOFS);
7ea394f1 1545nocow:
e6dcd2dc
CM		 1546 add_pending_csums(trans, inode, ordered_extent->file_offset,
1547 &ordered_extent->list);
1548
34353029 1549 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1550 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1551 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1552 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1553 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1554
e6dcd2dc
CM		 1555 /* once for us */
1556 btrfs_put_ordered_extent(ordered_extent);
1557 /* once for the tree */
1558 btrfs_put_ordered_extent(ordered_extent);
1559
e6dcd2dc
CM		 1560 btrfs_end_transaction(trans, root);
1561 return 0;
1562}
1563
b2950863 1564static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1565 struct extent_state *state, int uptodate)
1566{
1567 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1568}
1569
d352ac68
CM		 1570/*
1571 * When IO fails, either with EIO or csum verification fails, we
1572 * try other mirrors that might have a good copy of the data. This
1573 * io_failure_record is used to record state as we go through all the
1574 * mirrors. If another mirror has good data, the page is set up to date
1575 * and things continue. If a good mirror can't be found, the original
1576 * bio end_io callback is called to indicate things have failed.
1577 */
7e38326f
CM		 1578struct io_failure_record {
1579 struct page *page;
1580 u64 start;
1581 u64 len;
1582 u64 logical;
d20f7043 1583 unsigned long bio_flags;
7e38326f
CM		 1584 int last_mirror;
1585};
1586
b2950863 1587static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1588 struct page *page, u64 start, u64 end,
1589 struct extent_state *state)
7e38326f
CM		 1590{
1591 struct io_failure_record *failrec = NULL;
1592 u64 private;
1593 struct extent_map *em;
1594 struct inode *inode = page->mapping->host;
1595 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1596 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1597 struct bio *bio;
1598 int num_copies;
1599 int ret;
1259ab75 1600 int rw;
7e38326f
CM		 1601 u64 logical;
1602
1603 ret = get_state_private(failure_tree, start, &private);
1604 if (ret) {
7e38326f
CM		 1605 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1606 if (!failrec)
1607 return -ENOMEM;
1608 failrec->start = start;
1609 failrec->len = end - start + 1;
1610 failrec->last_mirror = 0;
d20f7043 1611 failrec->bio_flags = 0;
7e38326f 1612
3b951516
CM		 1613 spin_lock(&em_tree->lock);
1614 em = lookup_extent_mapping(em_tree, start, failrec->len);
1615 if (em->start > start || em->start + em->len < start) {
1616 free_extent_map(em);
1617 em = NULL;
1618 }
1619 spin_unlock(&em_tree->lock);
7e38326f
CM		 1620
1621 if (!em || IS_ERR(em)) {
1622 kfree(failrec);
1623 return -EIO;
1624 }
1625 logical = start - em->start;
1626 logical = em->block_start + logical;
d20f7043
CM		 1627 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1628 logical = em->block_start;
1629 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1630 }
7e38326f
CM		 1631 failrec->logical = logical;
1632 free_extent_map(em);
1633 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1634 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1635 set_state_private(failure_tree, start,
1636 (u64)(unsigned long)failrec);
7e38326f 1637 } else {
587f7704 1638 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1639 }
1640 num_copies = btrfs_num_copies(
1641 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1642 failrec->logical, failrec->len);
1643 failrec->last_mirror++;
1644 if (!state) {
cad321ad 1645 spin_lock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1646 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1647 failrec->start,
1648 EXTENT_LOCKED);
1649 if (state && state->start != failrec->start)
1650 state = NULL;
cad321ad 1651 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1652 }
1653 if (!state || failrec->last_mirror > num_copies) {
1654 set_state_private(failure_tree, failrec->start, 0);
1655 clear_extent_bits(failure_tree, failrec->start,
1656 failrec->start + failrec->len - 1,
1657 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1658 kfree(failrec);
1659 return -EIO;
1660 }
1661 bio = bio_alloc(GFP_NOFS, 1);
1662 bio->bi_private = state;
1663 bio->bi_end_io = failed_bio->bi_end_io;
1664 bio->bi_sector = failrec->logical >> 9;
1665 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1666 bio->bi_size = 0;
d20f7043 1667
7e38326f 1668 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1669 if (failed_bio->bi_rw & (1 << BIO_RW))
1670 rw = WRITE;
1671 else
1672 rw = READ;
1673
1674 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1675 failrec->last_mirror,
d20f7043 1676 failrec->bio_flags);
1259ab75
CM		 1677 return 0;
1678}
1679
d352ac68
CM		 1680/*
1681 * each time an IO finishes, we do a fast check in the IO failure tree
1682 * to see if we need to process or clean up an io_failure_record
1683 */
b2950863 1684static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1685{
1686 u64 private;
1687 u64 private_failure;
1688 struct io_failure_record *failure;
1689 int ret;
1690
1691 private = 0;
1692 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1693 (u64)-1, 1, EXTENT_DIRTY)) {
1694 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1695 start, &private_failure);
1696 if (ret == 0) {
1697 failure = (struct io_failure_record *)(unsigned long)
1698 private_failure;
1699 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1700 failure->start, 0);
1701 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1702 failure->start,
1703 failure->start + failure->len - 1,
1704 EXTENT_DIRTY | EXTENT_LOCKED,
1705 GFP_NOFS);
1706 kfree(failure);
1707 }
1708 }
7e38326f
CM		 1709 return 0;
1710}
1711
d352ac68
CM		 1712/*
1713 * when reads are done, we need to check csums to verify the data is correct
1714 * if there's a match, we allow the bio to finish. If not, we go through
1715 * the io_failure_record routines to find good copies
1716 */
b2950863 1717static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1718 struct extent_state *state)
07157aac 1719{
35ebb934 1720 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1721 struct inode *inode = page->mapping->host;
d1310b2e 1722 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1723 char *kaddr;
aadfeb6e 1724 u64 private = ~(u32)0;
07157aac 1725 int ret;
ff79f819
CM		 1726 struct btrfs_root *root = BTRFS_I(inode)->root;
1727 u32 csum = ~(u32)0;
d1310b2e 1728
d20f7043
CM		 1729 if (PageChecked(page)) {
1730 ClearPageChecked(page);
1731 goto good;
1732 }
17d217fe
YZ		 1733 if (btrfs_test_flag(inode, NODATASUM))
1734 return 0;
1735
1736 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1737 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1738 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1739 GFP_NOFS);
b6cda9bc 1740 return 0;
17d217fe 1741 }
d20f7043 1742
c2e639f0 1743 if (state && state->start == start) {
70dec807
CM		 1744 private = state->private;
1745 ret = 0;
1746 } else {
1747 ret = get_state_private(io_tree, start, &private);
1748 }
9ab86c8e 1749 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1750 if (ret)
07157aac 1751 goto zeroit;
d397712b 1752
ff79f819
CM		 1753 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1754 btrfs_csum_final(csum, (char *)&csum);
d397712b 1755 if (csum != private)
07157aac 1756 goto zeroit;
d397712b 1757
9ab86c8e 1758 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1759good:
7e38326f
CM		 1760 /* if the io failure tree for this inode is non-empty,
1761 * check to see if we've recovered from a failed IO
1762 */
1259ab75 1763 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1764 return 0;
1765
1766zeroit:
d397712b
CM		 1767 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1768 "private %llu\n", page->mapping->host->i_ino,
1769 (unsigned long long)start, csum,
1770 (unsigned long long)private);
db94535d
CM		 1771 memset(kaddr + offset, 1, end - start + 1);
1772 flush_dcache_page(page);
9ab86c8e 1773 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1774 if (private == 0)
1775 return 0;
7e38326f 1776 return -EIO;
07157aac 1777}
b888db2b 1778
7b128766
JB		 1779/*
1780 * This creates an orphan entry for the given inode in case something goes
1781 * wrong in the middle of an unlink/truncate.
1782 */
1783int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1784{
1785 struct btrfs_root *root = BTRFS_I(inode)->root;
1786 int ret = 0;
1787
bcc63abb 1788 spin_lock(&root->list_lock);
7b128766
JB		 1789
1790 /* already on the orphan list, we're good */
1791 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1792 spin_unlock(&root->list_lock);
7b128766
JB		 1793 return 0;
1794 }
1795
1796 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1797
bcc63abb 1798 spin_unlock(&root->list_lock);
7b128766
JB		 1799
1800 /*
1801 * insert an orphan item to track this unlinked/truncated file
1802 */
1803 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1804
1805 return ret;
1806}
1807
1808/*
1809 * We have done the truncate/delete so we can go ahead and remove the orphan
1810 * item for this particular inode.
1811 */
1812int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1813{
1814 struct btrfs_root *root = BTRFS_I(inode)->root;
1815 int ret = 0;
1816
bcc63abb 1817 spin_lock(&root->list_lock);
7b128766
JB		 1818
1819 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1820 spin_unlock(&root->list_lock);
7b128766
JB		 1821 return 0;
1822 }
1823
1824 list_del_init(&BTRFS_I(inode)->i_orphan);
1825 if (!trans) {
bcc63abb 1826 spin_unlock(&root->list_lock);
7b128766
JB		 1827 return 0;
1828 }
1829
bcc63abb 1830 spin_unlock(&root->list_lock);
7b128766
JB		 1831
1832 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1833
1834 return ret;
1835}
1836
1837/*
1838 * this cleans up any orphans that may be left on the list from the last use
1839 * of this root.
1840 */
1841void btrfs_orphan_cleanup(struct btrfs_root *root)
1842{
1843 struct btrfs_path *path;
1844 struct extent_buffer *leaf;
1845 struct btrfs_item *item;
1846 struct btrfs_key key, found_key;
1847 struct btrfs_trans_handle *trans;
1848 struct inode *inode;
1849 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1850
7b128766
JB		 1851 path = btrfs_alloc_path();
1852 if (!path)
1853 return;
1854 path->reada = -1;
1855
1856 key.objectid = BTRFS_ORPHAN_OBJECTID;
1857 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1858 key.offset = (u64)-1;
1859
7b128766
JB		 1860
1861 while (1) {
1862 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1863 if (ret < 0) {
1864 printk(KERN_ERR "Error searching slot for orphan: %d"
1865 "\n", ret);
1866 break;
1867 }
1868
1869 /*
1870 * if ret == 0 means we found what we were searching for, which
1871 * is weird, but possible, so only screw with path if we didnt
1872 * find the key and see if we have stuff that matches
1873 */
1874 if (ret > 0) {
1875 if (path->slots[0] == 0)
1876 break;
1877 path->slots[0]--;
1878 }
1879
1880 /* pull out the item */
1881 leaf = path->nodes[0];
1882 item = btrfs_item_nr(leaf, path->slots[0]);
1883 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1884
1885 /* make sure the item matches what we want */
1886 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1887 break;
1888 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1889 break;
1890
1891 /* release the path since we're done with it */
1892 btrfs_release_path(root, path);
1893
1894 /*
1895 * this is where we are basically btrfs_lookup, without the
1896 * crossing root thing. we store the inode number in the
1897 * offset of the orphan item.
1898 */
5b21f2ed 1899 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1900 found_key.offset, root);
1901 if (!inode)
1902 break;
1903
1904 if (inode->i_state & I_NEW) {
1905 BTRFS_I(inode)->root = root;
1906
1907 /* have to set the location manually */
1908 BTRFS_I(inode)->location.objectid = inode->i_ino;
1909 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1910 BTRFS_I(inode)->location.offset = 0;
1911
1912 btrfs_read_locked_inode(inode);
1913 unlock_new_inode(inode);
1914 }
1915
1916 /*
1917 * add this inode to the orphan list so btrfs_orphan_del does
1918 * the proper thing when we hit it
1919 */
bcc63abb 1920 spin_lock(&root->list_lock);
7b128766 1921 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1922 spin_unlock(&root->list_lock);
7b128766
JB		 1923
1924 /*
1925 * if this is a bad inode, means we actually succeeded in
1926 * removing the inode, but not the orphan record, which means
1927 * we need to manually delete the orphan since iput will just
1928 * do a destroy_inode
1929 */
1930 if (is_bad_inode(inode)) {
5b21f2ed 1931 trans = btrfs_start_transaction(root, 1);
7b128766 1932 btrfs_orphan_del(trans, inode);
5b21f2ed 1933 btrfs_end_transaction(trans, root);
7b128766
JB		 1934 iput(inode);
1935 continue;
1936 }
1937
1938 /* if we have links, this was a truncate, lets do that */
1939 if (inode->i_nlink) {
1940 nr_truncate++;
1941 btrfs_truncate(inode);
1942 } else {
1943 nr_unlink++;
1944 }
1945
1946 /* this will do delete_inode and everything for us */
1947 iput(inode);
1948 }
1949
1950 if (nr_unlink)
1951 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1952 if (nr_truncate)
1953 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1954
1955 btrfs_free_path(path);
7b128766
JB		 1956}
1957
d352ac68
CM		 1958/*
1959 * read an inode from the btree into the in-memory inode
1960 */
39279cc3
CM		 1961void btrfs_read_locked_inode(struct inode *inode)
1962{
1963 struct btrfs_path *path;
5f39d397 1964 struct extent_buffer *leaf;
39279cc3 1965 struct btrfs_inode_item *inode_item;
0b86a832 1966 struct btrfs_timespec *tspec;
39279cc3
CM		 1967 struct btrfs_root *root = BTRFS_I(inode)->root;
1968 struct btrfs_key location;
1969 u64 alloc_group_block;
618e21d5 1970 u32 rdev;
39279cc3
CM		 1971 int ret;
1972
1973 path = btrfs_alloc_path();
1974 BUG_ON(!path);
39279cc3 1975 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1976
39279cc3 1977 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1978 if (ret)
39279cc3 1979 goto make_bad;
39279cc3 1980
5f39d397
CM		 1981 leaf = path->nodes[0];
1982 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1983 struct btrfs_inode_item);
1984
1985 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1986 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1987 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1988 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1989 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1990
1991 tspec = btrfs_inode_atime(inode_item);
1992 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1993 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1994
1995 tspec = btrfs_inode_mtime(inode_item);
1996 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1997 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1998
1999 tspec = btrfs_inode_ctime(inode_item);
2000 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2001 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2002
a76a3cd4 2003 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2004 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2005 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2006 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2007 inode->i_rdev = 0;
5f39d397
CM		 2008 rdev = btrfs_inode_rdev(leaf, inode_item);
2009
aec7477b 2010 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2011 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2012
5f39d397 2013 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
d2fb3437
YZ		 2014 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2015 alloc_group_block, 0);
39279cc3
CM		 2016 btrfs_free_path(path);
2017 inode_item = NULL;
2018
39279cc3 2019 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2020 case S_IFREG:
2021 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2022 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2023 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2024 inode->i_fop = &btrfs_file_operations;
2025 inode->i_op = &btrfs_file_inode_operations;
2026 break;
2027 case S_IFDIR:
2028 inode->i_fop = &btrfs_dir_file_operations;
2029 if (root == root->fs_info->tree_root)
2030 inode->i_op = &btrfs_dir_ro_inode_operations;
2031 else
2032 inode->i_op = &btrfs_dir_inode_operations;
2033 break;
2034 case S_IFLNK:
2035 inode->i_op = &btrfs_symlink_inode_operations;
2036 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2037 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2038 break;
618e21d5
JB		 2039 default:
2040 init_special_inode(inode, inode->i_mode, rdev);
2041 break;
39279cc3
CM		 2042 }
2043 return;
2044
2045make_bad:
39279cc3 2046 btrfs_free_path(path);
39279cc3
CM		 2047 make_bad_inode(inode);
2048}
2049
d352ac68
CM		 2050/*
2051 * given a leaf and an inode, copy the inode fields into the leaf
2052 */
e02119d5
CM		 2053static void fill_inode_item(struct btrfs_trans_handle *trans,
2054 struct extent_buffer *leaf,
5f39d397 2055 struct btrfs_inode_item *item,
39279cc3
CM		 2056 struct inode *inode)
2057{
5f39d397
CM		 2058 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2059 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2060 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2061 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2062 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2063
2064 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2065 inode->i_atime.tv_sec);
2066 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2067 inode->i_atime.tv_nsec);
2068
2069 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2070 inode->i_mtime.tv_sec);
2071 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2072 inode->i_mtime.tv_nsec);
2073
2074 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2075 inode->i_ctime.tv_sec);
2076 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2077 inode->i_ctime.tv_nsec);
2078
a76a3cd4 2079 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2080 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2081 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2082 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2083 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2084 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2085 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2086}
2087
d352ac68
CM		 2088/*
2089 * copy everything in the in-memory inode into the btree.
2090 */
d397712b
CM		 2091noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2092 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2093{
2094 struct btrfs_inode_item *inode_item;
2095 struct btrfs_path *path;
5f39d397 2096 struct extent_buffer *leaf;
39279cc3
CM		 2097 int ret;
2098
2099 path = btrfs_alloc_path();
2100 BUG_ON(!path);
39279cc3
CM		 2101 ret = btrfs_lookup_inode(trans, root, path,
2102 &BTRFS_I(inode)->location, 1);
2103 if (ret) {
2104 if (ret > 0)
2105 ret = -ENOENT;
2106 goto failed;
2107 }
2108
5f39d397
CM		 2109 leaf = path->nodes[0];
2110 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2111 struct btrfs_inode_item);
2112
e02119d5 2113 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2114 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2115 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2116 ret = 0;
2117failed:
39279cc3
CM		 2118 btrfs_free_path(path);
2119 return ret;
2120}
2121
2122
d352ac68
CM		 2123/*
2124 * unlink helper that gets used here in inode.c and in the tree logging
2125 * recovery code. It remove a link in a directory with a given name, and
2126 * also drops the back refs in the inode to the directory
2127 */
e02119d5
CM		 2128int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2129 struct btrfs_root *root,
2130 struct inode *dir, struct inode *inode,
2131 const char *name, int name_len)
39279cc3
CM		 2132{
2133 struct btrfs_path *path;
39279cc3 2134 int ret = 0;
5f39d397 2135 struct extent_buffer *leaf;
39279cc3 2136 struct btrfs_dir_item *di;
5f39d397 2137 struct btrfs_key key;
aec7477b 2138 u64 index;
39279cc3
CM		 2139
2140 path = btrfs_alloc_path();
54aa1f4d
CM		 2141 if (!path) {
2142 ret = -ENOMEM;
2143 goto err;
2144 }
2145
39279cc3
CM		 2146 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2147 name, name_len, -1);
2148 if (IS_ERR(di)) {
2149 ret = PTR_ERR(di);
2150 goto err;
2151 }
2152 if (!di) {
2153 ret = -ENOENT;
2154 goto err;
2155 }
5f39d397
CM		 2156 leaf = path->nodes[0];
2157 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2158 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2159 if (ret)
2160 goto err;
39279cc3
CM		 2161 btrfs_release_path(root, path);
2162
aec7477b 2163 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2164 inode->i_ino,
2165 dir->i_ino, &index);
aec7477b 2166 if (ret) {
d397712b 2167 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
aec7477b 2168 "inode %lu parent %lu\n", name_len, name,
e02119d5 2169 inode->i_ino, dir->i_ino);
aec7477b
JB		 2170 goto err;
2171 }
2172
39279cc3 2173 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2174 index, name, name_len, -1);
39279cc3
CM		 2175 if (IS_ERR(di)) {
2176 ret = PTR_ERR(di);
2177 goto err;
2178 }
2179 if (!di) {
2180 ret = -ENOENT;
2181 goto err;
2182 }
2183 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2184 btrfs_release_path(root, path);
39279cc3 2185
e02119d5
CM		 2186 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2187 inode, dir->i_ino);
49eb7e46
CM		 2188 BUG_ON(ret != 0 && ret != -ENOENT);
2189 if (ret != -ENOENT)
2190 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2191
2192 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2193 dir, index);
2194 BUG_ON(ret);
39279cc3
CM		 2195err:
2196 btrfs_free_path(path);
e02119d5
CM		 2197 if (ret)
2198 goto out;
2199
2200 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2201 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2202 btrfs_update_inode(trans, root, dir);
2203 btrfs_drop_nlink(inode);
2204 ret = btrfs_update_inode(trans, root, inode);
2205 dir->i_sb->s_dirt = 1;
2206out:
39279cc3
CM		 2207 return ret;
2208}
2209
2210static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2211{
2212 struct btrfs_root *root;
2213 struct btrfs_trans_handle *trans;
7b128766 2214 struct inode *inode = dentry->d_inode;
39279cc3 2215 int ret;
1832a6d5 2216 unsigned long nr = 0;
39279cc3
CM		 2217
2218 root = BTRFS_I(dir)->root;
1832a6d5
CM		 2219
2220 ret = btrfs_check_free_space(root, 1, 1);
2221 if (ret)
2222 goto fail;
2223
39279cc3 2224 trans = btrfs_start_transaction(root, 1);
5f39d397 2225
39279cc3 2226 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2227 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2228 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2229
2230 if (inode->i_nlink == 0)
2231 ret = btrfs_orphan_add(trans, inode);
2232
d3c2fdcf 2233 nr = trans->blocks_used;
5f39d397 2234
89ce8a63 2235 btrfs_end_transaction_throttle(trans, root);
1832a6d5 2236fail:
d3c2fdcf 2237 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2238 return ret;
2239}
2240
2241static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2242{
2243 struct inode *inode = dentry->d_inode;
1832a6d5 2244 int err = 0;
39279cc3
CM		 2245 int ret;
2246 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2247 struct btrfs_trans_handle *trans;
1832a6d5 2248 unsigned long nr = 0;
39279cc3 2249
3394e160
CM		 2250 /*
2251 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2252 * the root of a subvolume or snapshot
2253 */
2254 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2255 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2256 return -ENOTEMPTY;
925baedd 2257 }
134d4512 2258
1832a6d5
CM		 2259 ret = btrfs_check_free_space(root, 1, 1);
2260 if (ret)
2261 goto fail;
2262
39279cc3
CM		 2263 trans = btrfs_start_transaction(root, 1);
2264 btrfs_set_trans_block_group(trans, dir);
39279cc3 2265
7b128766
JB		 2266 err = btrfs_orphan_add(trans, inode);
2267 if (err)
2268 goto fail_trans;
2269
39279cc3 2270 /* now the directory is empty */
e02119d5
CM		 2271 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2272 dentry->d_name.name, dentry->d_name.len);
d397712b 2273 if (!err)
dbe674a9 2274 btrfs_i_size_write(inode, 0);
3954401f 2275
7b128766 2276fail_trans:
d3c2fdcf 2277 nr = trans->blocks_used;
89ce8a63 2278 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 2279fail:
d3c2fdcf 2280 btrfs_btree_balance_dirty(root, nr);
3954401f 2281
39279cc3
CM		 2282 if (ret && !err)
2283 err = ret;
2284 return err;
2285}
2286
d20f7043 2287#if 0
323ac95b
CM		 2288/*
2289 * when truncating bytes in a file, it is possible to avoid reading
2290 * the leaves that contain only checksum items. This can be the
2291 * majority of the IO required to delete a large file, but it must
2292 * be done carefully.
2293 *
2294 * The keys in the level just above the leaves are checked to make sure
2295 * the lowest key in a given leaf is a csum key, and starts at an offset
2296 * after the new size.
2297 *
2298 * Then the key for the next leaf is checked to make sure it also has
2299 * a checksum item for the same file. If it does, we know our target leaf
2300 * contains only checksum items, and it can be safely freed without reading
2301 * it.
2302 *
2303 * This is just an optimization targeted at large files. It may do
2304 * nothing. It will return 0 unless things went badly.
2305 */
2306static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2307 struct btrfs_root *root,
2308 struct btrfs_path *path,
2309 struct inode *inode, u64 new_size)
2310{
2311 struct btrfs_key key;
2312 int ret;
2313 int nritems;
2314 struct btrfs_key found_key;
2315 struct btrfs_key other_key;
5b84e8d6
YZ		 2316 struct btrfs_leaf_ref *ref;
2317 u64 leaf_gen;
2318 u64 leaf_start;
323ac95b
CM		 2319
2320 path->lowest_level = 1;
2321 key.objectid = inode->i_ino;
2322 key.type = BTRFS_CSUM_ITEM_KEY;
2323 key.offset = new_size;
2324again:
2325 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2326 if (ret < 0)
2327 goto out;
2328
2329 if (path->nodes[1] == NULL) {
2330 ret = 0;
2331 goto out;
2332 }
2333 ret = 0;
2334 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2335 nritems = btrfs_header_nritems(path->nodes[1]);
2336
2337 if (!nritems)
2338 goto out;
2339
2340 if (path->slots[1] >= nritems)
2341 goto next_node;
2342
2343 /* did we find a key greater than anything we want to delete? */
2344 if (found_key.objectid > inode->i_ino ||
2345 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2346 goto out;
2347
2348 /* we check the next key in the node to make sure the leave contains
2349 * only checksum items. This comparison doesn't work if our
2350 * leaf is the last one in the node
2351 */
2352 if (path->slots[1] + 1 >= nritems) {
2353next_node:
2354 /* search forward from the last key in the node, this
2355 * will bring us into the next node in the tree
2356 */
2357 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2358
2359 /* unlikely, but we inc below, so check to be safe */
2360 if (found_key.offset == (u64)-1)
2361 goto out;
2362
2363 /* search_forward needs a path with locks held, do the
2364 * search again for the original key. It is possible
2365 * this will race with a balance and return a path that
2366 * we could modify, but this drop is just an optimization
2367 * and is allowed to miss some leaves.
2368 */
2369 btrfs_release_path(root, path);
2370 found_key.offset++;
2371
2372 /* setup a max key for search_forward */
2373 other_key.offset = (u64)-1;
2374 other_key.type = key.type;
2375 other_key.objectid = key.objectid;
2376
2377 path->keep_locks = 1;
2378 ret = btrfs_search_forward(root, &found_key, &other_key,
2379 path, 0, 0);
2380 path->keep_locks = 0;
2381 if (ret || found_key.objectid != key.objectid ||
2382 found_key.type != key.type) {
2383 ret = 0;
2384 goto out;
2385 }
2386
2387 key.offset = found_key.offset;
2388 btrfs_release_path(root, path);
2389 cond_resched();
2390 goto again;
2391 }
2392
2393 /* we know there's one more slot after us in the tree,
2394 * read that key so we can verify it is also a checksum item
2395 */
2396 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2397
2398 if (found_key.objectid < inode->i_ino)
2399 goto next_key;
2400
2401 if (found_key.type != key.type || found_key.offset < new_size)
2402 goto next_key;
2403
2404 /*
2405 * if the key for the next leaf isn't a csum key from this objectid,
2406 * we can't be sure there aren't good items inside this leaf.
2407 * Bail out
2408 */
2409 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2410 goto out;
2411
5b84e8d6
YZ		 2412 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2413 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2414 /*
2415 * it is safe to delete this leaf, it contains only
2416 * csum items from this inode at an offset >= new_size
2417 */
5b84e8d6 2418 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2419 BUG_ON(ret);
2420
5b84e8d6
YZ		 2421 if (root->ref_cows && leaf_gen < trans->transid) {
2422 ref = btrfs_alloc_leaf_ref(root, 0);
2423 if (ref) {
2424 ref->root_gen = root->root_key.offset;
2425 ref->bytenr = leaf_start;
2426 ref->owner = 0;
2427 ref->generation = leaf_gen;
2428 ref->nritems = 0;
2429
2430 ret = btrfs_add_leaf_ref(root, ref, 0);
2431 WARN_ON(ret);
2432 btrfs_free_leaf_ref(root, ref);
2433 } else {
2434 WARN_ON(1);
2435 }
2436 }
323ac95b
CM		 2437next_key:
2438 btrfs_release_path(root, path);
2439
2440 if (other_key.objectid == inode->i_ino &&
2441 other_key.type == key.type && other_key.offset > key.offset) {
2442 key.offset = other_key.offset;
2443 cond_resched();
2444 goto again;
2445 }
2446 ret = 0;
2447out:
2448 /* fixup any changes we've made to the path */
2449 path->lowest_level = 0;
2450 path->keep_locks = 0;
2451 btrfs_release_path(root, path);
2452 return ret;
2453}
2454
d20f7043
CM		 2455#endif
2456
39279cc3
CM		 2457/*
2458 * this can truncate away extent items, csum items and directory items.
2459 * It starts at a high offset and removes keys until it can't find
d352ac68 2460 * any higher than new_size
39279cc3
CM		 2461 *
2462 * csum items that cross the new i_size are truncated to the new size
2463 * as well.
7b128766
JB		 2464 *
2465 * min_type is the minimum key type to truncate down to. If set to 0, this
2466 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2467 */
e02119d5
CM		 2468noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2469 struct btrfs_root *root,
2470 struct inode *inode,
2471 u64 new_size, u32 min_type)
39279cc3
CM		 2472{
2473 int ret;
2474 struct btrfs_path *path;
2475 struct btrfs_key key;
5f39d397 2476 struct btrfs_key found_key;
39279cc3 2477 u32 found_type;
5f39d397 2478 struct extent_buffer *leaf;
39279cc3
CM		 2479 struct btrfs_file_extent_item *fi;
2480 u64 extent_start = 0;
db94535d 2481 u64 extent_num_bytes = 0;
39279cc3 2482 u64 item_end = 0;
7bb86316 2483 u64 root_gen = 0;
d8d5f3e1 2484 u64 root_owner = 0;
39279cc3
CM		 2485 int found_extent;
2486 int del_item;
85e21bac
CM		 2487 int pending_del_nr = 0;
2488 int pending_del_slot = 0;
179e29e4 2489 int extent_type = -1;
771ed689 2490 int encoding;
3b951516 2491 u64 mask = root->sectorsize - 1;
39279cc3 2492
e02119d5 2493 if (root->ref_cows)
5b21f2ed 2494 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2495 path = btrfs_alloc_path();
3c69faec 2496 path->reada = -1;
39279cc3 2497 BUG_ON(!path);
5f39d397 2498
39279cc3
CM		 2499 /* FIXME, add redo link to tree so we don't leak on crash */
2500 key.objectid = inode->i_ino;
2501 key.offset = (u64)-1;
5f39d397
CM		 2502 key.type = (u8)-1;
2503
85e21bac 2504 btrfs_init_path(path);
323ac95b 2505
85e21bac
CM		 2506search_again:
2507 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
d397712b 2508 if (ret < 0)
85e21bac 2509 goto error;
d397712b 2510
85e21bac 2511 if (ret > 0) {
e02119d5
CM		 2512 /* there are no items in the tree for us to truncate, we're
2513 * done
2514 */
2515 if (path->slots[0] == 0) {
2516 ret = 0;
2517 goto error;
2518 }
85e21bac
CM		 2519 path->slots[0]--;
2520 }
2521
d397712b 2522 while (1) {
39279cc3 2523 fi = NULL;
5f39d397
CM		 2524 leaf = path->nodes[0];
2525 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2526 found_type = btrfs_key_type(&found_key);
771ed689 2527 encoding = 0;
39279cc3 2528
5f39d397 2529 if (found_key.objectid != inode->i_ino)
39279cc3 2530 break;
5f39d397 2531
85e21bac 2532 if (found_type < min_type)
39279cc3
CM		 2533 break;
2534
5f39d397 2535 item_end = found_key.offset;
39279cc3 2536 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2537 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2538 struct btrfs_file_extent_item);
179e29e4 2539 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2540 encoding = btrfs_file_extent_compression(leaf, fi);
2541 encoding |= btrfs_file_extent_encryption(leaf, fi);
2542 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2543
179e29e4 2544 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2545 item_end +=
db94535d 2546 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2547 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2548 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2549 fi);
39279cc3 2550 }
008630c1 2551 item_end--;
39279cc3 2552 }
e02119d5 2553 if (item_end < new_size) {
d397712b 2554 if (found_type == BTRFS_DIR_ITEM_KEY)
b888db2b 2555 found_type = BTRFS_INODE_ITEM_KEY;
d397712b 2556 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
d20f7043 2557 found_type = BTRFS_EXTENT_DATA_KEY;
d397712b 2558 else if (found_type == BTRFS_EXTENT_DATA_KEY)
85e21bac 2559 found_type = BTRFS_XATTR_ITEM_KEY;
d397712b 2560 else if (found_type == BTRFS_XATTR_ITEM_KEY)
85e21bac 2561 found_type = BTRFS_INODE_REF_KEY;
d397712b 2562 else if (found_type)
b888db2b 2563 found_type--;
d397712b 2564 else
b888db2b 2565 break;
a61721d5 2566 btrfs_set_key_type(&key, found_type);
85e21bac 2567 goto next;
39279cc3 2568 }
e02119d5 2569 if (found_key.offset >= new_size)
39279cc3
CM		 2570 del_item = 1;
2571 else
2572 del_item = 0;
2573 found_extent = 0;
2574
2575 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2576 if (found_type != BTRFS_EXTENT_DATA_KEY)
2577 goto delete;
2578
2579 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2580 u64 num_dec;
db94535d 2581 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2582 if (!del_item && !encoding) {
db94535d
CM		 2583 u64 orig_num_bytes =
2584 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2585 extent_num_bytes = new_size -
5f39d397 2586 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2587 extent_num_bytes = extent_num_bytes &
2588 ~((u64)root->sectorsize - 1);
db94535d
CM		 2589 btrfs_set_file_extent_num_bytes(leaf, fi,
2590 extent_num_bytes);
2591 num_dec = (orig_num_bytes -
9069218d 2592 extent_num_bytes);
e02119d5 2593 if (root->ref_cows && extent_start != 0)
a76a3cd4 2594 inode_sub_bytes(inode, num_dec);
5f39d397 2595 btrfs_mark_buffer_dirty(leaf);
39279cc3 2596 } else {
db94535d
CM		 2597 extent_num_bytes =
2598 btrfs_file_extent_disk_num_bytes(leaf,
2599 fi);
39279cc3 2600 /* FIXME blocksize != 4096 */
9069218d 2601 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2602 if (extent_start != 0) {
2603 found_extent = 1;
e02119d5 2604 if (root->ref_cows)
a76a3cd4 2605 inode_sub_bytes(inode, num_dec);
e02119d5 2606 }
31840ae1 2607 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2608 root_owner = btrfs_header_owner(leaf);
39279cc3 2609 }
9069218d 2610 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2611 /*
2612 * we can't truncate inline items that have had
2613 * special encodings
2614 */
2615 if (!del_item &&
2616 btrfs_file_extent_compression(leaf, fi) == 0 &&
2617 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2618 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2619 u32 size = new_size - found_key.offset;
2620
2621 if (root->ref_cows) {
a76a3cd4
YZ		 2622 inode_sub_bytes(inode, item_end + 1 -
2623 new_size);
e02119d5
CM		 2624 }
2625 size =
2626 btrfs_file_extent_calc_inline_size(size);
9069218d 2627 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2628 size, 1);
9069218d 2629 BUG_ON(ret);
e02119d5 2630 } else if (root->ref_cows) {
a76a3cd4
YZ		 2631 inode_sub_bytes(inode, item_end + 1 -
2632 found_key.offset);
9069218d 2633 }
39279cc3 2634 }
179e29e4 2635delete:
39279cc3 2636 if (del_item) {
85e21bac
CM		 2637 if (!pending_del_nr) {
2638 /* no pending yet, add ourselves */
2639 pending_del_slot = path->slots[0];
2640 pending_del_nr = 1;
2641 } else if (pending_del_nr &&
2642 path->slots[0] + 1 == pending_del_slot) {
2643 /* hop on the pending chunk */
2644 pending_del_nr++;
2645 pending_del_slot = path->slots[0];
2646 } else {
d397712b 2647 BUG();
85e21bac 2648 }
39279cc3
CM		 2649 } else {
2650 break;
2651 }
39279cc3
CM		 2652 if (found_extent) {
2653 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2654 extent_num_bytes,
31840ae1 2655 leaf->start, root_owner,
3bb1a1bc 2656 root_gen, inode->i_ino, 0);
39279cc3
CM		 2657 BUG_ON(ret);
2658 }
85e21bac
CM		 2659next:
2660 if (path->slots[0] == 0) {
2661 if (pending_del_nr)
2662 goto del_pending;
2663 btrfs_release_path(root, path);
2664 goto search_again;
2665 }
2666
2667 path->slots[0]--;
2668 if (pending_del_nr &&
2669 path->slots[0] + 1 != pending_del_slot) {
2670 struct btrfs_key debug;
2671del_pending:
2672 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2673 pending_del_slot);
2674 ret = btrfs_del_items(trans, root, path,
2675 pending_del_slot,
2676 pending_del_nr);
2677 BUG_ON(ret);
2678 pending_del_nr = 0;
2679 btrfs_release_path(root, path);
2680 goto search_again;
2681 }
39279cc3
CM		 2682 }
2683 ret = 0;
2684error:
85e21bac
CM		 2685 if (pending_del_nr) {
2686 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2687 pending_del_nr);
2688 }
39279cc3
CM		 2689 btrfs_free_path(path);
2690 inode->i_sb->s_dirt = 1;
2691 return ret;
2692}
2693
2694/*
2695 * taken from block_truncate_page, but does cow as it zeros out
2696 * any bytes left in the last page in the file.
2697 */
2698static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2699{
2700 struct inode *inode = mapping->host;
db94535d 2701 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2702 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2703 struct btrfs_ordered_extent *ordered;
2704 char *kaddr;
db94535d 2705 u32 blocksize = root->sectorsize;
39279cc3
CM		 2706 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2707 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2708 struct page *page;
39279cc3 2709 int ret = 0;
a52d9a80 2710 u64 page_start;
e6dcd2dc 2711 u64 page_end;
39279cc3
CM		 2712
2713 if ((offset & (blocksize - 1)) == 0)
2714 goto out;
2715
2716 ret = -ENOMEM;
211c17f5 2717again:
39279cc3
CM		 2718 page = grab_cache_page(mapping, index);
2719 if (!page)
2720 goto out;
e6dcd2dc
CM		 2721
2722 page_start = page_offset(page);
2723 page_end = page_start + PAGE_CACHE_SIZE - 1;
2724
39279cc3 2725 if (!PageUptodate(page)) {
9ebefb18 2726 ret = btrfs_readpage(NULL, page);
39279cc3 2727 lock_page(page);
211c17f5
CM		 2728 if (page->mapping != mapping) {
2729 unlock_page(page);
2730 page_cache_release(page);
2731 goto again;
2732 }
39279cc3
CM		 2733 if (!PageUptodate(page)) {
2734 ret = -EIO;
89642229 2735 goto out_unlock;
39279cc3
CM		 2736 }
2737 }
211c17f5 2738 wait_on_page_writeback(page);
e6dcd2dc
CM		 2739
2740 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2741 set_page_extent_mapped(page);
2742
2743 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2744 if (ordered) {
2745 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2746 unlock_page(page);
2747 page_cache_release(page);
eb84ae03 2748 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2749 btrfs_put_ordered_extent(ordered);
2750 goto again;
2751 }
2752
ea8c2819 2753 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2754 ret = 0;
2755 if (offset != PAGE_CACHE_SIZE) {
2756 kaddr = kmap(page);
2757 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2758 flush_dcache_page(page);
2759 kunmap(page);
2760 }
247e743c 2761 ClearPageChecked(page);
e6dcd2dc
CM		 2762 set_page_dirty(page);
2763 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2764
89642229 2765out_unlock:
39279cc3
CM		 2766 unlock_page(page);
2767 page_cache_release(page);
2768out:
2769 return ret;
2770}
2771
9036c102 2772int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2773{
9036c102
YZ		 2774 struct btrfs_trans_handle *trans;
2775 struct btrfs_root *root = BTRFS_I(inode)->root;
2776 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2777 struct extent_map *em;
2778 u64 mask = root->sectorsize - 1;
2779 u64 hole_start = (inode->i_size + mask) & ~mask;
2780 u64 block_end = (size + mask) & ~mask;
2781 u64 last_byte;
2782 u64 cur_offset;
2783 u64 hole_size;
39279cc3
CM		 2784 int err;
2785
9036c102
YZ		 2786 if (size <= hole_start)
2787 return 0;
2788
2789 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2790 if (err)
2791 return err;
2792
9036c102 2793 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2794
9036c102
YZ		 2795 while (1) {
2796 struct btrfs_ordered_extent *ordered;
2797 btrfs_wait_ordered_range(inode, hole_start,
2798 block_end - hole_start);
2799 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2800 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2801 if (!ordered)
2802 break;
2803 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2804 btrfs_put_ordered_extent(ordered);
2805 }
39279cc3 2806
9036c102
YZ		 2807 trans = btrfs_start_transaction(root, 1);
2808 btrfs_set_trans_block_group(trans, inode);
39279cc3 2809
9036c102
YZ		 2810 cur_offset = hole_start;
2811 while (1) {
2812 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2813 block_end - cur_offset, 0);
2814 BUG_ON(IS_ERR(em) || !em);
2815 last_byte = min(extent_map_end(em), block_end);
2816 last_byte = (last_byte + mask) & ~mask;
2817 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2818 u64 hint_byte = 0;
9036c102 2819 hole_size = last_byte - cur_offset;
771ed689
CM		 2820 err = btrfs_drop_extents(trans, root, inode,
2821 cur_offset,
2822 cur_offset + hole_size,
2823 cur_offset, &hint_byte);
2824 if (err)
2825 break;
9036c102
YZ		 2826 err = btrfs_insert_file_extent(trans, root,
2827 inode->i_ino, cur_offset, 0,
2828 0, hole_size, 0, hole_size,
2829 0, 0, 0);
2830 btrfs_drop_extent_cache(inode, hole_start,
2831 last_byte - 1, 0);
2832 }
2833 free_extent_map(em);
2834 cur_offset = last_byte;
2835 if (err || cur_offset >= block_end)
2836 break;
2837 }
1832a6d5 2838
9036c102
YZ		 2839 btrfs_end_transaction(trans, root);
2840 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2841 return err;
2842}
39279cc3 2843
9036c102
YZ		 2844static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2845{
2846 struct inode *inode = dentry->d_inode;
2847 int err;
39279cc3 2848
9036c102
YZ		 2849 err = inode_change_ok(inode, attr);
2850 if (err)
2851 return err;
2bf5a725 2852
9036c102
YZ		 2853 if (S_ISREG(inode->i_mode) &&
2854 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2855 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2856 if (err)
2857 return err;
39279cc3 2858 }
9036c102 2859
39279cc3 2860 err = inode_setattr(inode, attr);
33268eaf
JB		 2861
2862 if (!err && ((attr->ia_valid & ATTR_MODE)))
2863 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2864 return err;
2865}
61295eb8 2866
39279cc3
CM		 2867void btrfs_delete_inode(struct inode *inode)
2868{
2869 struct btrfs_trans_handle *trans;
2870 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2871 unsigned long nr;
39279cc3
CM		 2872 int ret;
2873
2874 truncate_inode_pages(&inode->i_data, 0);
2875 if (is_bad_inode(inode)) {
7b128766 2876 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2877 goto no_delete;
2878 }
4a096752 2879 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2880
dbe674a9 2881 btrfs_i_size_write(inode, 0);
180591bc 2882 trans = btrfs_join_transaction(root, 1);
5f39d397 2883
39279cc3 2884 btrfs_set_trans_block_group(trans, inode);
e02119d5 2885 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2886 if (ret) {
2887 btrfs_orphan_del(NULL, inode);
54aa1f4d 2888 goto no_delete_lock;
7b128766
JB		 2889 }
2890
2891 btrfs_orphan_del(trans, inode);
85e21bac 2892
d3c2fdcf 2893 nr = trans->blocks_used;
85e21bac 2894 clear_inode(inode);
5f39d397 2895
39279cc3 2896 btrfs_end_transaction(trans, root);
d3c2fdcf 2897 btrfs_btree_balance_dirty(root, nr);
39279cc3 2898 return;
54aa1f4d
CM		 2899
2900no_delete_lock:
d3c2fdcf 2901 nr = trans->blocks_used;
54aa1f4d 2902 btrfs_end_transaction(trans, root);
d3c2fdcf 2903 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2904no_delete:
2905 clear_inode(inode);
2906}
2907
2908/*
2909 * this returns the key found in the dir entry in the location pointer.
2910 * If no dir entries were found, location->objectid is 0.
2911 */
2912static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2913 struct btrfs_key *location)
2914{
2915 const char *name = dentry->d_name.name;
2916 int namelen = dentry->d_name.len;
2917 struct btrfs_dir_item *di;
2918 struct btrfs_path *path;
2919 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2920 int ret = 0;
39279cc3
CM		 2921
2922 path = btrfs_alloc_path();
2923 BUG_ON(!path);
3954401f 2924
39279cc3
CM		 2925 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2926 namelen, 0);
0d9f7f3e
Y		 2927 if (IS_ERR(di))
2928 ret = PTR_ERR(di);
d397712b
CM		 2929
2930 if (!di || IS_ERR(di))
3954401f 2931 goto out_err;
d397712b 2932
5f39d397 2933 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2934out:
39279cc3
CM		 2935 btrfs_free_path(path);
2936 return ret;
3954401f
CM		 2937out_err:
2938 location->objectid = 0;
2939 goto out;
39279cc3
CM		 2940}
2941
2942/*
2943 * when we hit a tree root in a directory, the btrfs part of the inode
2944 * needs to be changed to reflect the root directory of the tree root. This
2945 * is kind of like crossing a mount point.
2946 */
2947static int fixup_tree_root_location(struct btrfs_root *root,
2948 struct btrfs_key *location,
58176a96
JB		 2949 struct btrfs_root **sub_root,
2950 struct dentry *dentry)
39279cc3 2951{
39279cc3
CM		 2952 struct btrfs_root_item *ri;
2953
2954 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2955 return 0;
2956 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2957 return 0;
2958
58176a96
JB		 2959 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2960 dentry->d_name.name,
2961 dentry->d_name.len);
39279cc3
CM		 2962 if (IS_ERR(*sub_root))
2963 return PTR_ERR(*sub_root);
2964
2965 ri = &(*sub_root)->root_item;
2966 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2967 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2968 location->offset = 0;
2969
39279cc3
CM		 2970 return 0;
2971}
2972
e02119d5 2973static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2974{
e02119d5
CM		 2975 struct btrfs_inode *bi = BTRFS_I(inode);
2976
2977 bi->i_acl = NULL;
2978 bi->i_default_acl = NULL;
2979
2980 bi->generation = 0;
c3027eb5 2981 bi->sequence = 0;
e02119d5
CM		 2982 bi->last_trans = 0;
2983 bi->logged_trans = 0;
2984 bi->delalloc_bytes = 0;
2985 bi->disk_i_size = 0;
2986 bi->flags = 0;
2987 bi->index_cnt = (u64)-1;
49eb7e46 2988 bi->log_dirty_trans = 0;
d1310b2e
CM		 2989 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2990 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2991 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2992 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2993 inode->i_mapping, GFP_NOFS);
ea8c2819 2994 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2995 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 2996 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2997 mutex_init(&BTRFS_I(inode)->log_mutex);
2998}
2999
3000static int btrfs_init_locked_inode(struct inode *inode, void *p)
3001{
3002 struct btrfs_iget_args *args = p;
3003 inode->i_ino = args->ino;
3004 init_btrfs_i(inode);
3005 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 3006 return 0;
3007}
3008
3009static int btrfs_find_actor(struct inode *inode, void *opaque)
3010{
3011 struct btrfs_iget_args *args = opaque;
d397712b
CM		 3012 return args->ino == inode->i_ino &&
3013 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3014}
3015
5b21f2ed
ZY		 3016struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3017 struct btrfs_root *root, int wait)
3018{
3019 struct inode *inode;
3020 struct btrfs_iget_args args;
3021 args.ino = objectid;
3022 args.root = root;
3023
3024 if (wait) {
3025 inode = ilookup5(s, objectid, btrfs_find_actor,
3026 (void *)&args);
3027 } else {
3028 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3029 (void *)&args);
3030 }
3031 return inode;
3032}
3033
39279cc3
CM		 3034struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3035 struct btrfs_root *root)
3036{
3037 struct inode *inode;
3038 struct btrfs_iget_args args;
3039 args.ino = objectid;
3040 args.root = root;
3041
3042 inode = iget5_locked(s, objectid, btrfs_find_actor,
3043 btrfs_init_locked_inode,
3044 (void *)&args);
3045 return inode;
3046}
3047
1a54ef8c
BR		 3048/* Get an inode object given its location and corresponding root.
3049 * Returns in *is_new if the inode was read from disk
3050 */
3051struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3052 struct btrfs_root *root, int *is_new)
3053{
3054 struct inode *inode;
3055
3056 inode = btrfs_iget_locked(s, location->objectid, root);
3057 if (!inode)
3058 return ERR_PTR(-EACCES);
3059
3060 if (inode->i_state & I_NEW) {
3061 BTRFS_I(inode)->root = root;
3062 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3063 btrfs_read_locked_inode(inode);
3064 unlock_new_inode(inode);
3065 if (is_new)
3066 *is_new = 1;
3067 } else {
3068 if (is_new)
3069 *is_new = 0;
3070 }
3071
3072 return inode;
3073}
3074
3de4586c 3075struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3076{
d397712b 3077 struct inode *inode;
39279cc3
CM		 3078 struct btrfs_inode *bi = BTRFS_I(dir);
3079 struct btrfs_root *root = bi->root;
3080 struct btrfs_root *sub_root = root;
3081 struct btrfs_key location;
c146afad 3082 int ret, new;
39279cc3
CM		 3083
3084 if (dentry->d_name.len > BTRFS_NAME_LEN)
3085 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3086
39279cc3 3087 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3088
39279cc3
CM		 3089 if (ret < 0)
3090 return ERR_PTR(ret);
5f39d397 3091
39279cc3
CM		 3092 inode = NULL;
3093 if (location.objectid) {
58176a96
JB		 3094 ret = fixup_tree_root_location(root, &location, &sub_root,
3095 dentry);
39279cc3
CM		 3096 if (ret < 0)
3097 return ERR_PTR(ret);
3098 if (ret > 0)
3099 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3100 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3101 if (IS_ERR(inode))
3102 return ERR_CAST(inode);
39279cc3 3103 }
3de4586c
CM		 3104 return inode;
3105}
3106
3107static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3108 struct nameidata *nd)
3109{
3110 struct inode *inode;
3111
3112 if (dentry->d_name.len > BTRFS_NAME_LEN)
3113 return ERR_PTR(-ENAMETOOLONG);
3114
3115 inode = btrfs_lookup_dentry(dir, dentry);
3116 if (IS_ERR(inode))
3117 return ERR_CAST(inode);
7b128766 3118
39279cc3
CM		 3119 return d_splice_alias(inode, dentry);
3120}
3121
39279cc3
CM		 3122static unsigned char btrfs_filetype_table[] = {
3123 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3124};
3125
cbdf5a24
DW		 3126static int btrfs_real_readdir(struct file *filp, void *dirent,
3127 filldir_t filldir)
39279cc3 3128{
6da6abae 3129 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3130 struct btrfs_root *root = BTRFS_I(inode)->root;
3131 struct btrfs_item *item;
3132 struct btrfs_dir_item *di;
3133 struct btrfs_key key;
5f39d397 3134 struct btrfs_key found_key;
39279cc3
CM		 3135 struct btrfs_path *path;
3136 int ret;
3137 u32 nritems;
5f39d397 3138 struct extent_buffer *leaf;
39279cc3
CM		 3139 int slot;
3140 int advance;
3141 unsigned char d_type;
3142 int over = 0;
3143 u32 di_cur;
3144 u32 di_total;
3145 u32 di_len;
3146 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3147 char tmp_name[32];
3148 char *name_ptr;
3149 int name_len;
39279cc3
CM		 3150
3151 /* FIXME, use a real flag for deciding about the key type */
3152 if (root->fs_info->tree_root == root)
3153 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3154
3954401f
CM		 3155 /* special case for "." */
3156 if (filp->f_pos == 0) {
3157 over = filldir(dirent, ".", 1,
3158 1, inode->i_ino,
3159 DT_DIR);
3160 if (over)
3161 return 0;
3162 filp->f_pos = 1;
3163 }
3954401f
CM		 3164 /* special case for .., just use the back ref */
3165 if (filp->f_pos == 1) {
5ecc7e5d 3166 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3167 over = filldir(dirent, "..", 2,
5ecc7e5d 3168 2, pino, DT_DIR);
3954401f 3169 if (over)
49593bfa 3170 return 0;
3954401f
CM		 3171 filp->f_pos = 2;
3172 }
49593bfa
DW		 3173 path = btrfs_alloc_path();
3174 path->reada = 2;
3175
39279cc3
CM		 3176 btrfs_set_key_type(&key, key_type);
3177 key.offset = filp->f_pos;
49593bfa 3178 key.objectid = inode->i_ino;
5f39d397 3179
39279cc3
CM		 3180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3181 if (ret < 0)
3182 goto err;
3183 advance = 0;
49593bfa
DW		 3184
3185 while (1) {
5f39d397
CM		 3186 leaf = path->nodes[0];
3187 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3188 slot = path->slots[0];
3189 if (advance || slot >= nritems) {
49593bfa 3190 if (slot >= nritems - 1) {
39279cc3
CM		 3191 ret = btrfs_next_leaf(root, path);
3192 if (ret)
3193 break;
5f39d397
CM		 3194 leaf = path->nodes[0];
3195 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3196 slot = path->slots[0];
3197 } else {
3198 slot++;
3199 path->slots[0]++;
3200 }
3201 }
3de4586c 3202
39279cc3 3203 advance = 1;
5f39d397
CM		 3204 item = btrfs_item_nr(leaf, slot);
3205 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3206
3207 if (found_key.objectid != key.objectid)
39279cc3 3208 break;
5f39d397 3209 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3210 break;
5f39d397 3211 if (found_key.offset < filp->f_pos)
39279cc3 3212 continue;
5f39d397
CM		 3213
3214 filp->f_pos = found_key.offset;
49593bfa 3215
39279cc3
CM		 3216 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3217 di_cur = 0;
5f39d397 3218 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3219
3220 while (di_cur < di_total) {
5f39d397
CM		 3221 struct btrfs_key location;
3222
3223 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3224 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3225 name_ptr = tmp_name;
3226 } else {
3227 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3228 if (!name_ptr) {
3229 ret = -ENOMEM;
3230 goto err;
3231 }
5f39d397
CM		 3232 }
3233 read_extent_buffer(leaf, name_ptr,
3234 (unsigned long)(di + 1), name_len);
3235
3236 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3237 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3238
3239 /* is this a reference to our own snapshot? If so
3240 * skip it
3241 */
3242 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3243 location.objectid == root->root_key.objectid) {
3244 over = 0;
3245 goto skip;
3246 }
5f39d397 3247 over = filldir(dirent, name_ptr, name_len,
49593bfa 3248 found_key.offset, location.objectid,
39279cc3 3249 d_type);
5f39d397 3250
3de4586c 3251skip:
5f39d397
CM		 3252 if (name_ptr != tmp_name)
3253 kfree(name_ptr);
3254
39279cc3
CM		 3255 if (over)
3256 goto nopos;
5103e947 3257 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3258 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3259 di_cur += di_len;
3260 di = (struct btrfs_dir_item *)((char *)di + di_len);
3261 }
3262 }
49593bfa
DW		 3263
3264 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 3265 if (key_type == BTRFS_DIR_INDEX_KEY)
3266 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3267 else
3268 filp->f_pos++;
39279cc3
CM		 3269nopos:
3270 ret = 0;
3271err:
39279cc3 3272 btrfs_free_path(path);
39279cc3
CM		 3273 return ret;
3274}
3275
3276int btrfs_write_inode(struct inode *inode, int wait)
3277{
3278 struct btrfs_root *root = BTRFS_I(inode)->root;
3279 struct btrfs_trans_handle *trans;
3280 int ret = 0;
3281
c146afad 3282 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3283 return 0;
3284
39279cc3 3285 if (wait) {
f9295749 3286 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3287 btrfs_set_trans_block_group(trans, inode);
3288 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3289 }
3290 return ret;
3291}
3292
3293/*
54aa1f4d 3294 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3295 * inode changes. But, it is most likely to find the inode in cache.
3296 * FIXME, needs more benchmarking...there are no reasons other than performance
3297 * to keep or drop this code.
3298 */
3299void btrfs_dirty_inode(struct inode *inode)
3300{
3301 struct btrfs_root *root = BTRFS_I(inode)->root;
3302 struct btrfs_trans_handle *trans;
3303
f9295749 3304 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3305 btrfs_set_trans_block_group(trans, inode);
3306 btrfs_update_inode(trans, root, inode);
3307 btrfs_end_transaction(trans, root);
39279cc3
CM		 3308}
3309
d352ac68
CM		 3310/*
3311 * find the highest existing sequence number in a directory
3312 * and then set the in-memory index_cnt variable to reflect
3313 * free sequence numbers
3314 */
aec7477b
JB		 3315static int btrfs_set_inode_index_count(struct inode *inode)
3316{
3317 struct btrfs_root *root = BTRFS_I(inode)->root;
3318 struct btrfs_key key, found_key;
3319 struct btrfs_path *path;
3320 struct extent_buffer *leaf;
3321 int ret;
3322
3323 key.objectid = inode->i_ino;
3324 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3325 key.offset = (u64)-1;
3326
3327 path = btrfs_alloc_path();
3328 if (!path)
3329 return -ENOMEM;
3330
3331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3332 if (ret < 0)
3333 goto out;
3334 /* FIXME: we should be able to handle this */
3335 if (ret == 0)
3336 goto out;
3337 ret = 0;
3338
3339 /*
3340 * MAGIC NUMBER EXPLANATION:
3341 * since we search a directory based on f_pos we have to start at 2
3342 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3343 * else has to start at 2
3344 */
3345 if (path->slots[0] == 0) {
3346 BTRFS_I(inode)->index_cnt = 2;
3347 goto out;
3348 }
3349
3350 path->slots[0]--;
3351
3352 leaf = path->nodes[0];
3353 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3354
3355 if (found_key.objectid != inode->i_ino ||
3356 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3357 BTRFS_I(inode)->index_cnt = 2;
3358 goto out;
3359 }
3360
3361 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3362out:
3363 btrfs_free_path(path);
3364 return ret;
3365}
3366
d352ac68
CM		 3367/*
3368 * helper to find a free sequence number in a given directory. This current
3369 * code is very simple, later versions will do smarter things in the btree
3370 */
3de4586c 3371int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3372{
3373 int ret = 0;
3374
3375 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3376 ret = btrfs_set_inode_index_count(dir);
d397712b 3377 if (ret)
aec7477b
JB		 3378 return ret;
3379 }
3380
00e4e6b3 3381 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3382 BTRFS_I(dir)->index_cnt++;
3383
3384 return ret;
3385}
3386
39279cc3
CM		 3387static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3388 struct btrfs_root *root,
aec7477b 3389 struct inode *dir,
9c58309d 3390 const char *name, int name_len,
d2fb3437
YZ		 3391 u64 ref_objectid, u64 objectid,
3392 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3393{
3394 struct inode *inode;
5f39d397 3395 struct btrfs_inode_item *inode_item;
39279cc3 3396 struct btrfs_key *location;
5f39d397 3397 struct btrfs_path *path;
9c58309d
CM		 3398 struct btrfs_inode_ref *ref;
3399 struct btrfs_key key[2];
3400 u32 sizes[2];
3401 unsigned long ptr;
39279cc3
CM		 3402 int ret;
3403 int owner;
3404
5f39d397
CM		 3405 path = btrfs_alloc_path();
3406 BUG_ON(!path);
3407
39279cc3
CM		 3408 inode = new_inode(root->fs_info->sb);
3409 if (!inode)
3410 return ERR_PTR(-ENOMEM);
3411
aec7477b 3412 if (dir) {
3de4586c 3413 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB		 3414 if (ret)
3415 return ERR_PTR(ret);
aec7477b
JB		 3416 }
3417 /*
3418 * index_cnt is ignored for everything but a dir,
3419 * btrfs_get_inode_index_count has an explanation for the magic
3420 * number
3421 */
e02119d5 3422 init_btrfs_i(inode);
aec7477b 3423 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3424 BTRFS_I(inode)->root = root;
e02119d5 3425 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3426
39279cc3
CM		 3427 if (mode & S_IFDIR)
3428 owner = 0;
3429 else
3430 owner = 1;
d2fb3437
YZ		 3431 BTRFS_I(inode)->block_group =
3432 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ		 3433 if ((mode & S_IFREG)) {
3434 if (btrfs_test_opt(root, NODATASUM))
3435 btrfs_set_flag(inode, NODATASUM);
3436 if (btrfs_test_opt(root, NODATACOW))
3437 btrfs_set_flag(inode, NODATACOW);
3438 }
9c58309d
CM		 3439
3440 key[0].objectid = objectid;
3441 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3442 key[0].offset = 0;
3443
3444 key[1].objectid = objectid;
3445 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3446 key[1].offset = ref_objectid;
3447
3448 sizes[0] = sizeof(struct btrfs_inode_item);
3449 sizes[1] = name_len + sizeof(*ref);
3450
3451 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3452 if (ret != 0)
5f39d397
CM		 3453 goto fail;
3454
9c58309d
CM		 3455 if (objectid > root->highest_inode)
3456 root->highest_inode = objectid;
3457
79683f2d
CM		 3458 inode->i_uid = current_fsuid();
3459 inode->i_gid = current_fsgid();
39279cc3
CM		 3460 inode->i_mode = mode;
3461 inode->i_ino = objectid;
a76a3cd4 3462 inode_set_bytes(inode, 0);
39279cc3 3463 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3464 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3465 struct btrfs_inode_item);
e02119d5 3466 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3467
3468 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3469 struct btrfs_inode_ref);
3470 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3471 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3472 ptr = (unsigned long)(ref + 1);
3473 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3474
5f39d397
CM		 3475 btrfs_mark_buffer_dirty(path->nodes[0]);
3476 btrfs_free_path(path);
3477
39279cc3
CM		 3478 location = &BTRFS_I(inode)->location;
3479 location->objectid = objectid;
39279cc3
CM		 3480 location->offset = 0;
3481 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3482
39279cc3
CM		 3483 insert_inode_hash(inode);
3484 return inode;
5f39d397 3485fail:
aec7477b
JB		 3486 if (dir)
3487 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3488 btrfs_free_path(path);
3489 return ERR_PTR(ret);
39279cc3
CM		 3490}
3491
3492static inline u8 btrfs_inode_type(struct inode *inode)
3493{
3494 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3495}
3496
d352ac68
CM		 3497/*
3498 * utility function to add 'inode' into 'parent_inode' with
3499 * a give name and a given sequence number.
3500 * if 'add_backref' is true, also insert a backref from the
3501 * inode to the parent directory.
3502 */
e02119d5
CM		 3503int btrfs_add_link(struct btrfs_trans_handle *trans,
3504 struct inode *parent_inode, struct inode *inode,
3505 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3506{
3507 int ret;
3508 struct btrfs_key key;
e02119d5 3509 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3510
39279cc3 3511 key.objectid = inode->i_ino;
39279cc3
CM		 3512 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3513 key.offset = 0;
3514
e02119d5
CM		 3515 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3516 parent_inode->i_ino,
aec7477b 3517 &key, btrfs_inode_type(inode),
00e4e6b3 3518 index);
39279cc3 3519 if (ret == 0) {
9c58309d
CM		 3520 if (add_backref) {
3521 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3522 name, name_len,
3523 inode->i_ino,
3524 parent_inode->i_ino,
3525 index);
9c58309d 3526 }
dbe674a9 3527 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3528 name_len * 2);
79c44584 3529 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3530 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3531 }
3532 return ret;
3533}
3534
3535static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3536 struct dentry *dentry, struct inode *inode,
00e4e6b3 3537 int backref, u64 index)
39279cc3 3538{
e02119d5
CM		 3539 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3540 inode, dentry->d_name.name,
3541 dentry->d_name.len, backref, index);
39279cc3
CM		 3542 if (!err) {
3543 d_instantiate(dentry, inode);
3544 return 0;
3545 }
3546 if (err > 0)
3547 err = -EEXIST;
3548 return err;
3549}
3550
618e21d5
JB		 3551static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3552 int mode, dev_t rdev)
3553{
3554 struct btrfs_trans_handle *trans;
3555 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3556 struct inode *inode = NULL;
618e21d5
JB		 3557 int err;
3558 int drop_inode = 0;
3559 u64 objectid;
1832a6d5 3560 unsigned long nr = 0;
00e4e6b3 3561 u64 index = 0;
618e21d5
JB		 3562
3563 if (!new_valid_dev(rdev))
3564 return -EINVAL;
3565
1832a6d5
CM		 3566 err = btrfs_check_free_space(root, 1, 0);
3567 if (err)
3568 goto fail;
3569
618e21d5
JB		 3570 trans = btrfs_start_transaction(root, 1);
3571 btrfs_set_trans_block_group(trans, dir);
3572
3573 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3574 if (err) {
3575 err = -ENOSPC;
3576 goto out_unlock;
3577 }
3578
aec7477b 3579 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3580 dentry->d_name.len,
3581 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3582 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3583 err = PTR_ERR(inode);
3584 if (IS_ERR(inode))
3585 goto out_unlock;
3586
33268eaf
JB		 3587 err = btrfs_init_acl(inode, dir);
3588 if (err) {
3589 drop_inode = 1;
3590 goto out_unlock;
3591 }
3592
618e21d5 3593 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3594 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3595 if (err)
3596 drop_inode = 1;
3597 else {
3598 inode->i_op = &btrfs_special_inode_operations;
3599 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3600 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3601 }
3602 dir->i_sb->s_dirt = 1;
3603 btrfs_update_inode_block_group(trans, inode);
3604 btrfs_update_inode_block_group(trans, dir);
3605out_unlock:
d3c2fdcf 3606 nr = trans->blocks_used;
89ce8a63 3607 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3608fail:
618e21d5
JB		 3609 if (drop_inode) {
3610 inode_dec_link_count(inode);
3611 iput(inode);
3612 }
d3c2fdcf 3613 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3614 return err;
3615}
3616
39279cc3
CM		 3617static int btrfs_create(struct inode *dir, struct dentry *dentry,
3618 int mode, struct nameidata *nd)
3619{
3620 struct btrfs_trans_handle *trans;
3621 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3622 struct inode *inode = NULL;
39279cc3
CM		 3623 int err;
3624 int drop_inode = 0;
1832a6d5 3625 unsigned long nr = 0;
39279cc3 3626 u64 objectid;
00e4e6b3 3627 u64 index = 0;
39279cc3 3628
1832a6d5
CM		 3629 err = btrfs_check_free_space(root, 1, 0);
3630 if (err)
3631 goto fail;
39279cc3
CM		 3632 trans = btrfs_start_transaction(root, 1);
3633 btrfs_set_trans_block_group(trans, dir);
3634
3635 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3636 if (err) {
3637 err = -ENOSPC;
3638 goto out_unlock;
3639 }
3640
aec7477b 3641 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3642 dentry->d_name.len,
3643 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3644 objectid, BTRFS_I(dir)->block_group, mode,
3645 &index);
39279cc3
CM		 3646 err = PTR_ERR(inode);
3647 if (IS_ERR(inode))
3648 goto out_unlock;
3649
33268eaf
JB		 3650 err = btrfs_init_acl(inode, dir);
3651 if (err) {
3652 drop_inode = 1;
3653 goto out_unlock;
3654 }
3655
39279cc3 3656 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3657 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3658 if (err)
3659 drop_inode = 1;
3660 else {
3661 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3662 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3663 inode->i_fop = &btrfs_file_operations;
3664 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3665 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3666 }
3667 dir->i_sb->s_dirt = 1;
3668 btrfs_update_inode_block_group(trans, inode);
3669 btrfs_update_inode_block_group(trans, dir);
3670out_unlock:
d3c2fdcf 3671 nr = trans->blocks_used;
ab78c84d 3672 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3673fail:
39279cc3
CM		 3674 if (drop_inode) {
3675 inode_dec_link_count(inode);
3676 iput(inode);
3677 }
d3c2fdcf 3678 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3679 return err;
3680}
3681
3682static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3683 struct dentry *dentry)
3684{
3685 struct btrfs_trans_handle *trans;
3686 struct btrfs_root *root = BTRFS_I(dir)->root;
3687 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3688 u64 index;
1832a6d5 3689 unsigned long nr = 0;
39279cc3
CM		 3690 int err;
3691 int drop_inode = 0;
3692
3693 if (inode->i_nlink == 0)
3694 return -ENOENT;
3695
e02119d5 3696 btrfs_inc_nlink(inode);
1832a6d5
CM		 3697 err = btrfs_check_free_space(root, 1, 0);
3698 if (err)
3699 goto fail;
3de4586c 3700 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3701 if (err)
3702 goto fail;
3703
39279cc3 3704 trans = btrfs_start_transaction(root, 1);
5f39d397 3705
39279cc3
CM		 3706 btrfs_set_trans_block_group(trans, dir);
3707 atomic_inc(&inode->i_count);
aec7477b 3708
00e4e6b3 3709 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3710
39279cc3
CM		 3711 if (err)
3712 drop_inode = 1;
5f39d397 3713
39279cc3
CM		 3714 dir->i_sb->s_dirt = 1;
3715 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3716 err = btrfs_update_inode(trans, root, inode);
5f39d397 3717
54aa1f4d
CM		 3718 if (err)
3719 drop_inode = 1;
39279cc3 3720
d3c2fdcf 3721 nr = trans->blocks_used;
ab78c84d 3722 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3723fail:
39279cc3
CM		 3724 if (drop_inode) {
3725 inode_dec_link_count(inode);
3726 iput(inode);
3727 }
d3c2fdcf 3728 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3729 return err;
3730}
3731
39279cc3
CM		 3732static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3733{
b9d86667 3734 struct inode *inode = NULL;
39279cc3
CM		 3735 struct btrfs_trans_handle *trans;
3736 struct btrfs_root *root = BTRFS_I(dir)->root;
3737 int err = 0;
3738 int drop_on_err = 0;
b9d86667 3739 u64 objectid = 0;
00e4e6b3 3740 u64 index = 0;
d3c2fdcf 3741 unsigned long nr = 1;
39279cc3 3742
1832a6d5
CM		 3743 err = btrfs_check_free_space(root, 1, 0);
3744 if (err)
3745 goto out_unlock;
3746
39279cc3
CM		 3747 trans = btrfs_start_transaction(root, 1);
3748 btrfs_set_trans_block_group(trans, dir);
5f39d397 3749
39279cc3
CM		 3750 if (IS_ERR(trans)) {
3751 err = PTR_ERR(trans);
3752 goto out_unlock;
3753 }
3754
3755 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3756 if (err) {
3757 err = -ENOSPC;
3758 goto out_unlock;
3759 }
3760
aec7477b 3761 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3762 dentry->d_name.len,
3763 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3764 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3765 &index);
39279cc3
CM		 3766 if (IS_ERR(inode)) {
3767 err = PTR_ERR(inode);
3768 goto out_fail;
3769 }
5f39d397 3770
39279cc3 3771 drop_on_err = 1;
33268eaf
JB		 3772
3773 err = btrfs_init_acl(inode, dir);
3774 if (err)
3775 goto out_fail;
3776
39279cc3
CM		 3777 inode->i_op = &btrfs_dir_inode_operations;
3778 inode->i_fop = &btrfs_dir_file_operations;
3779 btrfs_set_trans_block_group(trans, inode);
3780
dbe674a9 3781 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3782 err = btrfs_update_inode(trans, root, inode);
3783 if (err)
3784 goto out_fail;
5f39d397 3785
e02119d5
CM		 3786 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3787 inode, dentry->d_name.name,
3788 dentry->d_name.len, 0, index);
39279cc3
CM		 3789 if (err)
3790 goto out_fail;
5f39d397 3791
39279cc3
CM		 3792 d_instantiate(dentry, inode);
3793 drop_on_err = 0;
3794 dir->i_sb->s_dirt = 1;
3795 btrfs_update_inode_block_group(trans, inode);
3796 btrfs_update_inode_block_group(trans, dir);
3797
3798out_fail:
d3c2fdcf 3799 nr = trans->blocks_used;
ab78c84d 3800 btrfs_end_transaction_throttle(trans, root);
5f39d397 3801
39279cc3 3802out_unlock:
39279cc3
CM		 3803 if (drop_on_err)
3804 iput(inode);
d3c2fdcf 3805 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3806 return err;
3807}
3808
d352ac68
CM		 3809/* helper for btfs_get_extent. Given an existing extent in the tree,
3810 * and an extent that you want to insert, deal with overlap and insert
3811 * the new extent into the tree.
3812 */
3b951516
CM		 3813static int merge_extent_mapping(struct extent_map_tree *em_tree,
3814 struct extent_map *existing,
e6dcd2dc
CM		 3815 struct extent_map *em,
3816 u64 map_start, u64 map_len)
3b951516
CM		 3817{
3818 u64 start_diff;
3b951516 3819
e6dcd2dc
CM		 3820 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3821 start_diff = map_start - em->start;
3822 em->start = map_start;
3823 em->len = map_len;
c8b97818
CM		 3824 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3825 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3826 em->block_start += start_diff;
c8b97818
CM		 3827 em->block_len -= start_diff;
3828 }
e6dcd2dc 3829 return add_extent_mapping(em_tree, em);
3b951516
CM		 3830}
3831
c8b97818
CM		 3832static noinline int uncompress_inline(struct btrfs_path *path,
3833 struct inode *inode, struct page *page,
3834 size_t pg_offset, u64 extent_offset,
3835 struct btrfs_file_extent_item *item)
3836{
3837 int ret;
3838 struct extent_buffer *leaf = path->nodes[0];
3839 char *tmp;
3840 size_t max_size;
3841 unsigned long inline_size;
3842 unsigned long ptr;
3843
3844 WARN_ON(pg_offset != 0);
3845 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3846 inline_size = btrfs_file_extent_inline_item_len(leaf,
3847 btrfs_item_nr(leaf, path->slots[0]));
3848 tmp = kmalloc(inline_size, GFP_NOFS);
3849 ptr = btrfs_file_extent_inline_start(item);
3850
3851 read_extent_buffer(leaf, tmp, ptr, inline_size);
3852
5b050f04 3853 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3854 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3855 inline_size, max_size);
3856 if (ret) {
3857 char *kaddr = kmap_atomic(page, KM_USER0);
3858 unsigned long copy_size = min_t(u64,
3859 PAGE_CACHE_SIZE - pg_offset,
3860 max_size - extent_offset);
3861 memset(kaddr + pg_offset, 0, copy_size);
3862 kunmap_atomic(kaddr, KM_USER0);
3863 }
3864 kfree(tmp);
3865 return 0;
3866}
3867
d352ac68
CM		 3868/*
3869 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 3870 * the ugly parts come from merging extents from the disk with the in-ram
3871 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 3872 * where the in-ram extents might be locked pending data=ordered completion.
3873 *
3874 * This also copies inline extents directly into the page.
3875 */
d397712b 3876
a52d9a80 3877struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3878 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3879 int create)
3880{
3881 int ret;
3882 int err = 0;
db94535d 3883 u64 bytenr;
a52d9a80
CM		 3884 u64 extent_start = 0;
3885 u64 extent_end = 0;
3886 u64 objectid = inode->i_ino;
3887 u32 found_type;
f421950f 3888 struct btrfs_path *path = NULL;
a52d9a80
CM		 3889 struct btrfs_root *root = BTRFS_I(inode)->root;
3890 struct btrfs_file_extent_item *item;
5f39d397
CM		 3891 struct extent_buffer *leaf;
3892 struct btrfs_key found_key;
a52d9a80
CM		 3893 struct extent_map *em = NULL;
3894 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3895 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3896 struct btrfs_trans_handle *trans = NULL;
c8b97818 3897 int compressed;
a52d9a80 3898
a52d9a80 3899again:
d1310b2e
CM		 3900 spin_lock(&em_tree->lock);
3901 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3902 if (em)
3903 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3904 spin_unlock(&em_tree->lock);
3905
a52d9a80 3906 if (em) {
e1c4b745
CM		 3907 if (em->start > start || em->start + em->len <= start)
3908 free_extent_map(em);
3909 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3910 free_extent_map(em);
3911 else
3912 goto out;
a52d9a80 3913 }
d1310b2e 3914 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3915 if (!em) {
d1310b2e
CM		 3916 err = -ENOMEM;
3917 goto out;
a52d9a80 3918 }
e6dcd2dc 3919 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3920 em->start = EXTENT_MAP_HOLE;
445a6944 3921 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3922 em->len = (u64)-1;
c8b97818 3923 em->block_len = (u64)-1;
f421950f
CM		 3924
3925 if (!path) {
3926 path = btrfs_alloc_path();
3927 BUG_ON(!path);
3928 }
3929
179e29e4
CM		 3930 ret = btrfs_lookup_file_extent(trans, root, path,
3931 objectid, start, trans != NULL);
a52d9a80
CM		 3932 if (ret < 0) {
3933 err = ret;
3934 goto out;
3935 }
3936
3937 if (ret != 0) {
3938 if (path->slots[0] == 0)
3939 goto not_found;
3940 path->slots[0]--;
3941 }
3942
5f39d397
CM		 3943 leaf = path->nodes[0];
3944 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3945 struct btrfs_file_extent_item);
a52d9a80 3946 /* are we inside the extent that was found? */
5f39d397
CM		 3947 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3948 found_type = btrfs_key_type(&found_key);
3949 if (found_key.objectid != objectid ||
a52d9a80
CM		 3950 found_type != BTRFS_EXTENT_DATA_KEY) {
3951 goto not_found;
3952 }
3953
5f39d397
CM		 3954 found_type = btrfs_file_extent_type(leaf, item);
3955 extent_start = found_key.offset;
c8b97818 3956 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3957 if (found_type == BTRFS_FILE_EXTENT_REG ||
3958 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3959 extent_end = extent_start +
db94535d 3960 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3961 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3962 size_t size;
3963 size = btrfs_file_extent_inline_len(leaf, item);
3964 extent_end = (extent_start + size + root->sectorsize - 1) &
3965 ~((u64)root->sectorsize - 1);
3966 }
3967
3968 if (start >= extent_end) {
3969 path->slots[0]++;
3970 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3971 ret = btrfs_next_leaf(root, path);
3972 if (ret < 0) {
3973 err = ret;
3974 goto out;
a52d9a80 3975 }
9036c102
YZ		 3976 if (ret > 0)
3977 goto not_found;
3978 leaf = path->nodes[0];
a52d9a80 3979 }
9036c102
YZ		 3980 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3981 if (found_key.objectid != objectid ||
3982 found_key.type != BTRFS_EXTENT_DATA_KEY)
3983 goto not_found;
3984 if (start + len <= found_key.offset)
3985 goto not_found;
3986 em->start = start;
3987 em->len = found_key.offset - start;
3988 goto not_found_em;
3989 }
3990
d899e052
YZ		 3991 if (found_type == BTRFS_FILE_EXTENT_REG ||
3992 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 3993 em->start = extent_start;
3994 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 3995 em->orig_start = extent_start -
3996 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 3997 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3998 if (bytenr == 0) {
5f39d397 3999 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4000 goto insert;
4001 }
c8b97818
CM		 4002 if (compressed) {
4003 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4004 em->block_start = bytenr;
4005 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4006 item);
4007 } else {
4008 bytenr += btrfs_file_extent_offset(leaf, item);
4009 em->block_start = bytenr;
4010 em->block_len = em->len;
d899e052
YZ		 4011 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4012 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4013 }
a52d9a80
CM		 4014 goto insert;
4015 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4016 unsigned long ptr;
a52d9a80 4017 char *map;
3326d1b0
CM		 4018 size_t size;
4019 size_t extent_offset;
4020 size_t copy_size;
a52d9a80 4021
689f9346 4022 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4023 if (!page || create) {
689f9346 4024 em->start = extent_start;
9036c102 4025 em->len = extent_end - extent_start;
689f9346
Y		 4026 goto out;
4027 }
5f39d397 4028
9036c102
YZ		 4029 size = btrfs_file_extent_inline_len(leaf, item);
4030 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4031 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4032 size - extent_offset);
3326d1b0 4033 em->start = extent_start + extent_offset;
70dec807
CM		 4034 em->len = (copy_size + root->sectorsize - 1) &
4035 ~((u64)root->sectorsize - 1);
ff5b7ee3 4036 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 4037 if (compressed)
4038 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4039 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4040 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 4041 if (btrfs_file_extent_compression(leaf, item) ==
4042 BTRFS_COMPRESS_ZLIB) {
4043 ret = uncompress_inline(path, inode, page,
4044 pg_offset,
4045 extent_offset, item);
4046 BUG_ON(ret);
4047 } else {
4048 map = kmap(page);
4049 read_extent_buffer(leaf, map + pg_offset, ptr,
4050 copy_size);
4051 kunmap(page);
4052 }
179e29e4
CM		 4053 flush_dcache_page(page);
4054 } else if (create && PageUptodate(page)) {
4055 if (!trans) {
4056 kunmap(page);
4057 free_extent_map(em);
4058 em = NULL;
4059 btrfs_release_path(root, path);
f9295749 4060 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4061 goto again;
4062 }
c8b97818 4063 map = kmap(page);
70dec807 4064 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4065 copy_size);
c8b97818 4066 kunmap(page);
179e29e4 4067 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4068 }
d1310b2e
CM		 4069 set_extent_uptodate(io_tree, em->start,
4070 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4071 goto insert;
4072 } else {
d397712b 4073 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 4074 WARN_ON(1);
4075 }
4076not_found:
4077 em->start = start;
d1310b2e 4078 em->len = len;
a52d9a80 4079not_found_em:
5f39d397 4080 em->block_start = EXTENT_MAP_HOLE;
9036c102 4081 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4082insert:
4083 btrfs_release_path(root, path);
d1310b2e 4084 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 4085 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4086 "[%llu %llu]\n", (unsigned long long)em->start,
4087 (unsigned long long)em->len,
4088 (unsigned long long)start,
4089 (unsigned long long)len);
a52d9a80
CM		 4090 err = -EIO;
4091 goto out;
4092 }
d1310b2e
CM		 4093
4094 err = 0;
4095 spin_lock(&em_tree->lock);
a52d9a80 4096 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4097 /* it is possible that someone inserted the extent into the tree
4098 * while we had the lock dropped. It is also possible that
4099 * an overlapping map exists in the tree
4100 */
a52d9a80 4101 if (ret == -EEXIST) {
3b951516 4102 struct extent_map *existing;
e6dcd2dc
CM		 4103
4104 ret = 0;
4105
3b951516 4106 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4107 if (existing && (existing->start > start ||
4108 existing->start + existing->len <= start)) {
4109 free_extent_map(existing);
4110 existing = NULL;
4111 }
3b951516
CM		 4112 if (!existing) {
4113 existing = lookup_extent_mapping(em_tree, em->start,
4114 em->len);
4115 if (existing) {
4116 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4117 em, start,
4118 root->sectorsize);
3b951516
CM		 4119 free_extent_map(existing);
4120 if (err) {
4121 free_extent_map(em);
4122 em = NULL;
4123 }
4124 } else {
4125 err = -EIO;
3b951516
CM		 4126 free_extent_map(em);
4127 em = NULL;
4128 }
4129 } else {
4130 free_extent_map(em);
4131 em = existing;
e6dcd2dc 4132 err = 0;
a52d9a80 4133 }
a52d9a80 4134 }
d1310b2e 4135 spin_unlock(&em_tree->lock);
a52d9a80 4136out:
f421950f
CM		 4137 if (path)
4138 btrfs_free_path(path);
a52d9a80
CM		 4139 if (trans) {
4140 ret = btrfs_end_transaction(trans, root);
d397712b 4141 if (!err)
a52d9a80
CM		 4142 err = ret;
4143 }
a52d9a80
CM		 4144 if (err) {
4145 free_extent_map(em);
4146 WARN_ON(1);
4147 return ERR_PTR(err);
4148 }
4149 return em;
4150}
4151
16432985
CM		 4152static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4153 const struct iovec *iov, loff_t offset,
4154 unsigned long nr_segs)
4155{
e1c4b745 4156 return -EINVAL;
16432985
CM		 4157}
4158
d396c6f5 4159static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 4160{
d396c6f5 4161 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 4162}
4163
a52d9a80 4164int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4165{
d1310b2e
CM		 4166 struct extent_io_tree *tree;
4167 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4168 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4169}
1832a6d5 4170
a52d9a80 4171static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4172{
d1310b2e 4173 struct extent_io_tree *tree;
b888db2b
CM		 4174
4175
4176 if (current->flags & PF_MEMALLOC) {
4177 redirty_page_for_writepage(wbc, page);
4178 unlock_page(page);
4179 return 0;
4180 }
d1310b2e 4181 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4182 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4183}
4184
f421950f
CM		 4185int btrfs_writepages(struct address_space *mapping,
4186 struct writeback_control *wbc)
b293f02e 4187{
d1310b2e 4188 struct extent_io_tree *tree;
771ed689 4189
d1310b2e 4190 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4191 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4192}
4193
3ab2fb5a
CM		 4194static int
4195btrfs_readpages(struct file *file, struct address_space *mapping,
4196 struct list_head *pages, unsigned nr_pages)
4197{
d1310b2e
CM		 4198 struct extent_io_tree *tree;
4199 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4200 return extent_readpages(tree, mapping, pages, nr_pages,
4201 btrfs_get_extent);
4202}
e6dcd2dc 4203static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4204{
d1310b2e
CM		 4205 struct extent_io_tree *tree;
4206 struct extent_map_tree *map;
a52d9a80 4207 int ret;
8c2383c3 4208
d1310b2e
CM		 4209 tree = &BTRFS_I(page->mapping->host)->io_tree;
4210 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4211 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4212 if (ret == 1) {
4213 ClearPagePrivate(page);
4214 set_page_private(page, 0);
4215 page_cache_release(page);
39279cc3 4216 }
a52d9a80 4217 return ret;
39279cc3
CM		 4218}
4219
e6dcd2dc
CM		 4220static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4221{
98509cfc
CM		 4222 if (PageWriteback(page) || PageDirty(page))
4223 return 0;
e6dcd2dc
CM		 4224 return __btrfs_releasepage(page, gfp_flags);
4225}
4226
a52d9a80 4227static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4228{
d1310b2e 4229 struct extent_io_tree *tree;
e6dcd2dc
CM		 4230 struct btrfs_ordered_extent *ordered;
4231 u64 page_start = page_offset(page);
4232 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4233
e6dcd2dc 4234 wait_on_page_writeback(page);
d1310b2e 4235 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4236 if (offset) {
4237 btrfs_releasepage(page, GFP_NOFS);
4238 return;
4239 }
4240
4241 lock_extent(tree, page_start, page_end, GFP_NOFS);
4242 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4243 page_offset(page));
4244 if (ordered) {
eb84ae03
CM		 4245 /*
4246 * IO on this page will never be started, so we need
4247 * to account for any ordered extents now
4248 */
e6dcd2dc
CM		 4249 clear_extent_bit(tree, page_start, page_end,
4250 EXTENT_DIRTY | EXTENT_DELALLOC |
4251 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4252 btrfs_finish_ordered_io(page->mapping->host,
4253 page_start, page_end);
e6dcd2dc
CM		 4254 btrfs_put_ordered_extent(ordered);
4255 lock_extent(tree, page_start, page_end, GFP_NOFS);
4256 }
4257 clear_extent_bit(tree, page_start, page_end,
4258 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4259 EXTENT_ORDERED,
4260 1, 1, GFP_NOFS);
4261 __btrfs_releasepage(page, GFP_NOFS);
4262
4a096752 4263 ClearPageChecked(page);
9ad6b7bc 4264 if (PagePrivate(page)) {
9ad6b7bc
CM		 4265 ClearPagePrivate(page);
4266 set_page_private(page, 0);
4267 page_cache_release(page);
4268 }
39279cc3
CM		 4269}
4270
9ebefb18
CM		 4271/*
4272 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4273 * called from a page fault handler when a page is first dirtied. Hence we must
4274 * be careful to check for EOF conditions here. We set the page up correctly
4275 * for a written page which means we get ENOSPC checking when writing into
4276 * holes and correct delalloc and unwritten extent mapping on filesystems that
4277 * support these features.
4278 *
4279 * We are not allowed to take the i_mutex here so we have to play games to
4280 * protect against truncate races as the page could now be beyond EOF. Because
4281 * vmtruncate() writes the inode size before removing pages, once we have the
4282 * page lock we can determine safely if the page is beyond EOF. If it is not
4283 * beyond EOF, then the page is guaranteed safe against truncation until we
4284 * unlock the page.
4285 */
4286int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4287{
6da6abae 4288 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4289 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4290 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4291 struct btrfs_ordered_extent *ordered;
4292 char *kaddr;
4293 unsigned long zero_start;
9ebefb18 4294 loff_t size;
1832a6d5 4295 int ret;
a52d9a80 4296 u64 page_start;
e6dcd2dc 4297 u64 page_end;
9ebefb18 4298
1832a6d5 4299 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 4300 if (ret)
4301 goto out;
4302
4303 ret = -EINVAL;
e6dcd2dc 4304again:
9ebefb18 4305 lock_page(page);
9ebefb18 4306 size = i_size_read(inode);
e6dcd2dc
CM		 4307 page_start = page_offset(page);
4308 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4309
9ebefb18 4310 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4311 (page_start >= size)) {
9ebefb18
CM		 4312 /* page got truncated out from underneath us */
4313 goto out_unlock;
4314 }
e6dcd2dc
CM		 4315 wait_on_page_writeback(page);
4316
4317 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4318 set_page_extent_mapped(page);
4319
eb84ae03
CM		 4320 /*
4321 * we can't set the delalloc bits if there are pending ordered
4322 * extents. Drop our locks and wait for them to finish
4323 */
e6dcd2dc
CM		 4324 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4325 if (ordered) {
4326 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4327 unlock_page(page);
eb84ae03 4328 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4329 btrfs_put_ordered_extent(ordered);
4330 goto again;
4331 }
4332
ea8c2819 4333 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4334 ret = 0;
9ebefb18
CM		 4335
4336 /* page is wholly or partially inside EOF */
a52d9a80 4337 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4338 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4339 else
e6dcd2dc 4340 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4341
e6dcd2dc
CM		 4342 if (zero_start != PAGE_CACHE_SIZE) {
4343 kaddr = kmap(page);
4344 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4345 flush_dcache_page(page);
4346 kunmap(page);
4347 }
247e743c 4348 ClearPageChecked(page);
e6dcd2dc
CM		 4349 set_page_dirty(page);
4350 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4351
4352out_unlock:
4353 unlock_page(page);
1832a6d5 4354out:
9ebefb18
CM		 4355 return ret;
4356}
4357
39279cc3
CM		 4358static void btrfs_truncate(struct inode *inode)
4359{
4360 struct btrfs_root *root = BTRFS_I(inode)->root;
4361 int ret;
4362 struct btrfs_trans_handle *trans;
d3c2fdcf 4363 unsigned long nr;
dbe674a9 4364 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4365
4366 if (!S_ISREG(inode->i_mode))
4367 return;
4368 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4369 return;
4370
4371 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4372 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4373
39279cc3
CM		 4374 trans = btrfs_start_transaction(root, 1);
4375 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4376 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4377
7b128766
JB		 4378 ret = btrfs_orphan_add(trans, inode);
4379 if (ret)
4380 goto out;
39279cc3 4381 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4382 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4383 BTRFS_EXTENT_DATA_KEY);
39279cc3 4384 btrfs_update_inode(trans, root, inode);
5f39d397 4385
7b128766
JB		 4386 ret = btrfs_orphan_del(trans, inode);
4387 BUG_ON(ret);
4388
4389out:
4390 nr = trans->blocks_used;
89ce8a63 4391 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4392 BUG_ON(ret);
d3c2fdcf 4393 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4394}
4395
d352ac68
CM		 4396/*
4397 * create a new subvolume directory/inode (helper for the ioctl).
4398 */
d2fb3437
YZ		 4399int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4400 struct btrfs_root *new_root, struct dentry *dentry,
4401 u64 new_dirid, u64 alloc_hint)
39279cc3 4402{
39279cc3 4403 struct inode *inode;
cb8e7090 4404 int error;
00e4e6b3 4405 u64 index = 0;
39279cc3 4406
aec7477b 4407 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4408 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4409 if (IS_ERR(inode))
f46b5a66 4410 return PTR_ERR(inode);
39279cc3
CM		 4411 inode->i_op = &btrfs_dir_inode_operations;
4412 inode->i_fop = &btrfs_dir_file_operations;
4413
39279cc3 4414 inode->i_nlink = 1;
dbe674a9 4415 btrfs_i_size_write(inode, 0);
3b96362c 4416
cb8e7090
CH		 4417 error = btrfs_update_inode(trans, new_root, inode);
4418 if (error)
4419 return error;
4420
4421 d_instantiate(dentry, inode);
4422 return 0;
39279cc3
CM		 4423}
4424
d352ac68
CM		 4425/* helper function for file defrag and space balancing. This
4426 * forces readahead on a given range of bytes in an inode
4427 */
edbd8d4e 4428unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4429 struct file_ra_state *ra, struct file *file,
4430 pgoff_t offset, pgoff_t last_index)
4431{
8e7bf94f 4432 pgoff_t req_size = last_index - offset + 1;
86479a04 4433
86479a04
CM		 4434 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4435 return offset + req_size;
86479a04
CM		 4436}
4437
39279cc3
CM		 4438struct inode *btrfs_alloc_inode(struct super_block *sb)
4439{
4440 struct btrfs_inode *ei;
4441
4442 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4443 if (!ei)
4444 return NULL;
15ee9bc7 4445 ei->last_trans = 0;
e02119d5 4446 ei->logged_trans = 0;
e6dcd2dc 4447 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4448 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4449 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4450 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4451 return &ei->vfs_inode;
4452}
4453
4454void btrfs_destroy_inode(struct inode *inode)
4455{
e6dcd2dc 4456 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4457 WARN_ON(!list_empty(&inode->i_dentry));
4458 WARN_ON(inode->i_data.nrpages);
4459
33268eaf
JB		 4460 if (BTRFS_I(inode)->i_acl &&
4461 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4462 posix_acl_release(BTRFS_I(inode)->i_acl);
4463 if (BTRFS_I(inode)->i_default_acl &&
4464 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4465 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4466
bcc63abb 4467 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4468 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4469 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4470 " list\n", inode->i_ino);
4471 dump_stack();
4472 }
bcc63abb 4473 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4474
d397712b 4475 while (1) {
e6dcd2dc
CM		 4476 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4477 if (!ordered)
4478 break;
4479 else {
d397712b
CM		 4480 printk(KERN_ERR "btrfs found ordered "
4481 "extent %llu %llu on inode cleanup\n",
4482 (unsigned long long)ordered->file_offset,
4483 (unsigned long long)ordered->len);
e6dcd2dc
CM		 4484 btrfs_remove_ordered_extent(inode, ordered);
4485 btrfs_put_ordered_extent(ordered);
4486 btrfs_put_ordered_extent(ordered);
4487 }
4488 }
5b21f2ed 4489 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4490 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4491}
4492
0ee0fda0 4493static void init_once(void *foo)
39279cc3
CM		 4494{
4495 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4496
4497 inode_init_once(&ei->vfs_inode);
4498}
4499
4500void btrfs_destroy_cachep(void)
4501{
4502 if (btrfs_inode_cachep)
4503 kmem_cache_destroy(btrfs_inode_cachep);
4504 if (btrfs_trans_handle_cachep)
4505 kmem_cache_destroy(btrfs_trans_handle_cachep);
4506 if (btrfs_transaction_cachep)
4507 kmem_cache_destroy(btrfs_transaction_cachep);
4508 if (btrfs_bit_radix_cachep)
4509 kmem_cache_destroy(btrfs_bit_radix_cachep);
4510 if (btrfs_path_cachep)
4511 kmem_cache_destroy(btrfs_path_cachep);
4512}
4513
86479a04 4514struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4515 unsigned long extra_flags,
2b1f55b0 4516 void (*ctor)(void *))
92fee66d
CM		 4517{
4518 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4519 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4520}
4521
39279cc3
CM		 4522int btrfs_init_cachep(void)
4523{
86479a04 4524 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4525 sizeof(struct btrfs_inode),
4526 0, init_once);
39279cc3
CM		 4527 if (!btrfs_inode_cachep)
4528 goto fail;
86479a04
CM		 4529 btrfs_trans_handle_cachep =
4530 btrfs_cache_create("btrfs_trans_handle_cache",
4531 sizeof(struct btrfs_trans_handle),
4532 0, NULL);
39279cc3
CM		 4533 if (!btrfs_trans_handle_cachep)
4534 goto fail;
86479a04 4535 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4536 sizeof(struct btrfs_transaction),
92fee66d 4537 0, NULL);
39279cc3
CM		 4538 if (!btrfs_transaction_cachep)
4539 goto fail;
86479a04 4540 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4541 sizeof(struct btrfs_path),
92fee66d 4542 0, NULL);
39279cc3
CM		 4543 if (!btrfs_path_cachep)
4544 goto fail;
86479a04 4545 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4546 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4547 if (!btrfs_bit_radix_cachep)
4548 goto fail;
4549 return 0;
4550fail:
4551 btrfs_destroy_cachep();
4552 return -ENOMEM;
4553}
4554
4555static int btrfs_getattr(struct vfsmount *mnt,
4556 struct dentry *dentry, struct kstat *stat)
4557{
4558 struct inode *inode = dentry->d_inode;
4559 generic_fillattr(inode, stat);
3394e160 4560 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4561 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4562 stat->blocks = (inode_get_bytes(inode) +
4563 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4564 return 0;
4565}
4566
d397712b
CM		 4567static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4568 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 4569{
4570 struct btrfs_trans_handle *trans;
4571 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4572 struct inode *new_inode = new_dentry->d_inode;
4573 struct inode *old_inode = old_dentry->d_inode;
4574 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4575 u64 index = 0;
39279cc3
CM		 4576 int ret;
4577
3394e160
CM		 4578 /* we're not allowed to rename between subvolumes */
4579 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4580 BTRFS_I(new_dir)->root->root_key.objectid)
4581 return -EXDEV;
4582
39279cc3
CM		 4583 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4584 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4585 return -ENOTEMPTY;
4586 }
5f39d397 4587
0660b5af
CM		 4588 /* to rename a snapshot or subvolume, we need to juggle the
4589 * backrefs. This isn't coded yet
4590 */
4591 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4592 return -EXDEV;
4593
1832a6d5
CM		 4594 ret = btrfs_check_free_space(root, 1, 0);
4595 if (ret)
4596 goto out_unlock;
4597
39279cc3 4598 trans = btrfs_start_transaction(root, 1);
5f39d397 4599
39279cc3 4600 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4601
e02119d5 4602 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4603 old_dir->i_ctime = old_dir->i_mtime = ctime;
4604 new_dir->i_ctime = new_dir->i_mtime = ctime;
4605 old_inode->i_ctime = ctime;
5f39d397 4606
e02119d5
CM		 4607 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4608 old_dentry->d_name.name,
4609 old_dentry->d_name.len);
39279cc3
CM		 4610 if (ret)
4611 goto out_fail;
4612
4613 if (new_inode) {
4614 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4615 ret = btrfs_unlink_inode(trans, root, new_dir,
4616 new_dentry->d_inode,
4617 new_dentry->d_name.name,
4618 new_dentry->d_name.len);
39279cc3
CM		 4619 if (ret)
4620 goto out_fail;
7b128766 4621 if (new_inode->i_nlink == 0) {
e02119d5 4622 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4623 if (ret)
4624 goto out_fail;
4625 }
e02119d5 4626
39279cc3 4627 }
3de4586c 4628 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4629 if (ret)
4630 goto out_fail;
4631
e02119d5
CM		 4632 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4633 old_inode, new_dentry->d_name.name,
4634 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4635 if (ret)
4636 goto out_fail;
4637
4638out_fail:
ab78c84d 4639 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4640out_unlock:
39279cc3
CM		 4641 return ret;
4642}
4643
d352ac68
CM		 4644/*
4645 * some fairly slow code that needs optimization. This walks the list
4646 * of all the inodes with pending delalloc and forces them to disk.
4647 */
ea8c2819
CM		 4648int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4649{
4650 struct list_head *head = &root->fs_info->delalloc_inodes;
4651 struct btrfs_inode *binode;
5b21f2ed 4652 struct inode *inode;
ea8c2819 4653
c146afad
YZ		 4654 if (root->fs_info->sb->s_flags & MS_RDONLY)
4655 return -EROFS;
4656
75eff68e 4657 spin_lock(&root->fs_info->delalloc_lock);
d397712b 4658 while (!list_empty(head)) {
ea8c2819
CM		 4659 binode = list_entry(head->next, struct btrfs_inode,
4660 delalloc_inodes);
5b21f2ed
ZY		 4661 inode = igrab(&binode->vfs_inode);
4662 if (!inode)
4663 list_del_init(&binode->delalloc_inodes);
75eff68e 4664 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 4665 if (inode) {
8c8bee1d 4666 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4667 iput(inode);
4668 }
4669 cond_resched();
75eff68e 4670 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 4671 }
75eff68e 4672 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 4673
4674 /* the filemap_flush will queue IO into the worker threads, but
4675 * we have to make sure the IO is actually started and that
4676 * ordered extents get created before we return
4677 */
4678 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 4679 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 4680 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4681 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4682 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4683 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4684 }
4685 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4686 return 0;
4687}
4688
39279cc3
CM		 4689static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4690 const char *symname)
4691{
4692 struct btrfs_trans_handle *trans;
4693 struct btrfs_root *root = BTRFS_I(dir)->root;
4694 struct btrfs_path *path;
4695 struct btrfs_key key;
1832a6d5 4696 struct inode *inode = NULL;
39279cc3
CM		 4697 int err;
4698 int drop_inode = 0;
4699 u64 objectid;
00e4e6b3 4700 u64 index = 0 ;
39279cc3
CM		 4701 int name_len;
4702 int datasize;
5f39d397 4703 unsigned long ptr;
39279cc3 4704 struct btrfs_file_extent_item *ei;
5f39d397 4705 struct extent_buffer *leaf;
1832a6d5 4706 unsigned long nr = 0;
39279cc3
CM		 4707
4708 name_len = strlen(symname) + 1;
4709 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4710 return -ENAMETOOLONG;
1832a6d5 4711
1832a6d5
CM		 4712 err = btrfs_check_free_space(root, 1, 0);
4713 if (err)
4714 goto out_fail;
4715
39279cc3
CM		 4716 trans = btrfs_start_transaction(root, 1);
4717 btrfs_set_trans_block_group(trans, dir);
4718
4719 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4720 if (err) {
4721 err = -ENOSPC;
4722 goto out_unlock;
4723 }
4724
aec7477b 4725 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4726 dentry->d_name.len,
4727 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4728 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4729 &index);
39279cc3
CM		 4730 err = PTR_ERR(inode);
4731 if (IS_ERR(inode))
4732 goto out_unlock;
4733
33268eaf
JB		 4734 err = btrfs_init_acl(inode, dir);
4735 if (err) {
4736 drop_inode = 1;
4737 goto out_unlock;
4738 }
4739
39279cc3 4740 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4741 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4742 if (err)
4743 drop_inode = 1;
4744 else {
4745 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4746 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4747 inode->i_fop = &btrfs_file_operations;
4748 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4749 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4750 }
4751 dir->i_sb->s_dirt = 1;
4752 btrfs_update_inode_block_group(trans, inode);
4753 btrfs_update_inode_block_group(trans, dir);
4754 if (drop_inode)
4755 goto out_unlock;
4756
4757 path = btrfs_alloc_path();
4758 BUG_ON(!path);
4759 key.objectid = inode->i_ino;
4760 key.offset = 0;
39279cc3
CM		 4761 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4762 datasize = btrfs_file_extent_calc_inline_size(name_len);
4763 err = btrfs_insert_empty_item(trans, root, path, &key,
4764 datasize);
54aa1f4d
CM		 4765 if (err) {
4766 drop_inode = 1;
4767 goto out_unlock;
4768 }
5f39d397
CM		 4769 leaf = path->nodes[0];
4770 ei = btrfs_item_ptr(leaf, path->slots[0],
4771 struct btrfs_file_extent_item);
4772 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4773 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4774 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4775 btrfs_set_file_extent_encryption(leaf, ei, 0);
4776 btrfs_set_file_extent_compression(leaf, ei, 0);
4777 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4778 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4779
39279cc3 4780 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4781 write_extent_buffer(leaf, symname, ptr, name_len);
4782 btrfs_mark_buffer_dirty(leaf);
39279cc3 4783 btrfs_free_path(path);
5f39d397 4784
39279cc3
CM		 4785 inode->i_op = &btrfs_symlink_inode_operations;
4786 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4787 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4788 inode_set_bytes(inode, name_len);
dbe674a9 4789 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4790 err = btrfs_update_inode(trans, root, inode);
4791 if (err)
4792 drop_inode = 1;
39279cc3
CM		 4793
4794out_unlock:
d3c2fdcf 4795 nr = trans->blocks_used;
ab78c84d 4796 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4797out_fail:
39279cc3
CM		 4798 if (drop_inode) {
4799 inode_dec_link_count(inode);
4800 iput(inode);
4801 }
d3c2fdcf 4802 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4803 return err;
4804}
16432985 4805
d899e052
YZ		 4806static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4807 u64 alloc_hint, int mode)
4808{
4809 struct btrfs_trans_handle *trans;
4810 struct btrfs_root *root = BTRFS_I(inode)->root;
4811 struct btrfs_key ins;
4812 u64 alloc_size;
4813 u64 cur_offset = start;
4814 u64 num_bytes = end - start;
4815 int ret = 0;
4816
4817 trans = btrfs_join_transaction(root, 1);
4818 BUG_ON(!trans);
4819 btrfs_set_trans_block_group(trans, inode);
4820
4821 while (num_bytes > 0) {
4822 alloc_size = min(num_bytes, root->fs_info->max_extent);
4823 ret = btrfs_reserve_extent(trans, root, alloc_size,
4824 root->sectorsize, 0, alloc_hint,
4825 (u64)-1, &ins, 1);
4826 if (ret) {
4827 WARN_ON(1);
4828 goto out;
4829 }
4830 ret = insert_reserved_file_extent(trans, inode,
4831 cur_offset, ins.objectid,
4832 ins.offset, ins.offset,
4833 ins.offset, 0, 0, 0,
4834 BTRFS_FILE_EXTENT_PREALLOC);
4835 BUG_ON(ret);
4836 num_bytes -= ins.offset;
4837 cur_offset += ins.offset;
4838 alloc_hint = ins.objectid + ins.offset;
4839 }
4840out:
4841 if (cur_offset > start) {
4842 inode->i_ctime = CURRENT_TIME;
4843 btrfs_set_flag(inode, PREALLOC);
4844 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4845 cur_offset > i_size_read(inode))
4846 btrfs_i_size_write(inode, cur_offset);
4847 ret = btrfs_update_inode(trans, root, inode);
4848 BUG_ON(ret);
4849 }
4850
4851 btrfs_end_transaction(trans, root);
4852 return ret;
4853}
4854
4855static long btrfs_fallocate(struct inode *inode, int mode,
4856 loff_t offset, loff_t len)
4857{
4858 u64 cur_offset;
4859 u64 last_byte;
4860 u64 alloc_start;
4861 u64 alloc_end;
4862 u64 alloc_hint = 0;
4863 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4864 struct extent_map *em;
4865 int ret;
4866
4867 alloc_start = offset & ~mask;
4868 alloc_end = (offset + len + mask) & ~mask;
4869
4870 mutex_lock(&inode->i_mutex);
4871 if (alloc_start > inode->i_size) {
4872 ret = btrfs_cont_expand(inode, alloc_start);
4873 if (ret)
4874 goto out;
4875 }
4876
4877 while (1) {
4878 struct btrfs_ordered_extent *ordered;
4879 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4880 alloc_end - 1, GFP_NOFS);
4881 ordered = btrfs_lookup_first_ordered_extent(inode,
4882 alloc_end - 1);
4883 if (ordered &&
4884 ordered->file_offset + ordered->len > alloc_start &&
4885 ordered->file_offset < alloc_end) {
4886 btrfs_put_ordered_extent(ordered);
4887 unlock_extent(&BTRFS_I(inode)->io_tree,
4888 alloc_start, alloc_end - 1, GFP_NOFS);
4889 btrfs_wait_ordered_range(inode, alloc_start,
4890 alloc_end - alloc_start);
4891 } else {
4892 if (ordered)
4893 btrfs_put_ordered_extent(ordered);
4894 break;
4895 }
4896 }
4897
4898 cur_offset = alloc_start;
4899 while (1) {
4900 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4901 alloc_end - cur_offset, 0);
4902 BUG_ON(IS_ERR(em) || !em);
4903 last_byte = min(extent_map_end(em), alloc_end);
4904 last_byte = (last_byte + mask) & ~mask;
4905 if (em->block_start == EXTENT_MAP_HOLE) {
4906 ret = prealloc_file_range(inode, cur_offset,
4907 last_byte, alloc_hint, mode);
4908 if (ret < 0) {
4909 free_extent_map(em);
4910 break;
4911 }
4912 }
4913 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4914 alloc_hint = em->block_start;
4915 free_extent_map(em);
4916
4917 cur_offset = last_byte;
4918 if (cur_offset >= alloc_end) {
4919 ret = 0;
4920 break;
4921 }
4922 }
4923 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4924 GFP_NOFS);
4925out:
4926 mutex_unlock(&inode->i_mutex);
4927 return ret;
4928}
4929
e6dcd2dc
CM		 4930static int btrfs_set_page_dirty(struct page *page)
4931{
e6dcd2dc
CM		 4932 return __set_page_dirty_nobuffers(page);
4933}
4934
0ee0fda0 4935static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4936{
4937 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4938 return -EACCES;
33268eaf 4939 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4940}
39279cc3
CM		 4941
4942static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4943 .getattr = btrfs_getattr,
39279cc3
CM		 4944 .lookup = btrfs_lookup,
4945 .create = btrfs_create,
4946 .unlink = btrfs_unlink,
4947 .link = btrfs_link,
4948 .mkdir = btrfs_mkdir,
4949 .rmdir = btrfs_rmdir,
4950 .rename = btrfs_rename,
4951 .symlink = btrfs_symlink,
4952 .setattr = btrfs_setattr,
618e21d5 4953 .mknod = btrfs_mknod,
95819c05
CH		 4954 .setxattr = btrfs_setxattr,
4955 .getxattr = btrfs_getxattr,
5103e947 4956 .listxattr = btrfs_listxattr,
95819c05 4957 .removexattr = btrfs_removexattr,
fdebe2bd 4958 .permission = btrfs_permission,
39279cc3 4959};
39279cc3
CM		 4960static struct inode_operations btrfs_dir_ro_inode_operations = {
4961 .lookup = btrfs_lookup,
fdebe2bd 4962 .permission = btrfs_permission,
39279cc3 4963};
39279cc3
CM		 4964static struct file_operations btrfs_dir_file_operations = {
4965 .llseek = generic_file_llseek,
4966 .read = generic_read_dir,
cbdf5a24 4967 .readdir = btrfs_real_readdir,
34287aa3 4968 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4969#ifdef CONFIG_COMPAT
34287aa3 4970 .compat_ioctl = btrfs_ioctl,
39279cc3 4971#endif
6bf13c0c 4972 .release = btrfs_release_file,
e02119d5 4973 .fsync = btrfs_sync_file,
39279cc3
CM		 4974};
4975
d1310b2e 4976static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4977 .fill_delalloc = run_delalloc_range,
065631f6 4978 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4979 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4980 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4981 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4982 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4983 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4984 .set_bit_hook = btrfs_set_bit_hook,
4985 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4986};
4987
39279cc3
CM		 4988static struct address_space_operations btrfs_aops = {
4989 .readpage = btrfs_readpage,
4990 .writepage = btrfs_writepage,
b293f02e 4991 .writepages = btrfs_writepages,
3ab2fb5a 4992 .readpages = btrfs_readpages,
39279cc3 4993 .sync_page = block_sync_page,
39279cc3 4994 .bmap = btrfs_bmap,
16432985 4995 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4996 .invalidatepage = btrfs_invalidatepage,
4997 .releasepage = btrfs_releasepage,
e6dcd2dc 4998 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 4999};
5000
5001static struct address_space_operations btrfs_symlink_aops = {
5002 .readpage = btrfs_readpage,
5003 .writepage = btrfs_writepage,
2bf5a725
CM		 5004 .invalidatepage = btrfs_invalidatepage,
5005 .releasepage = btrfs_releasepage,
39279cc3
CM		 5006};
5007
5008static struct inode_operations btrfs_file_inode_operations = {
5009 .truncate = btrfs_truncate,
5010 .getattr = btrfs_getattr,
5011 .setattr = btrfs_setattr,
95819c05
CH		 5012 .setxattr = btrfs_setxattr,
5013 .getxattr = btrfs_getxattr,
5103e947 5014 .listxattr = btrfs_listxattr,
95819c05 5015 .removexattr = btrfs_removexattr,
fdebe2bd 5016 .permission = btrfs_permission,
d899e052 5017 .fallocate = btrfs_fallocate,
39279cc3 5018};
618e21d5
JB		 5019static struct inode_operations btrfs_special_inode_operations = {
5020 .getattr = btrfs_getattr,
5021 .setattr = btrfs_setattr,
fdebe2bd 5022 .permission = btrfs_permission,
95819c05
CH		 5023 .setxattr = btrfs_setxattr,
5024 .getxattr = btrfs_getxattr,
33268eaf 5025 .listxattr = btrfs_listxattr,
95819c05 5026 .removexattr = btrfs_removexattr,
618e21d5 5027};
39279cc3
CM		 5028static struct inode_operations btrfs_symlink_inode_operations = {
5029 .readlink = generic_readlink,
5030 .follow_link = page_follow_link_light,
5031 .put_link = page_put_link,
fdebe2bd 5032 .permission = btrfs_permission,
39279cc3 5033};
Linux 6.x block layer and io_uring tree(s)