?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'samsung-fixes-4' of git://github.com/kgene/linux-samsung
[linux-btrfs-devel.git] / fs / btrfs / inode.c
blobb2d004ad66a0159042c38f96691c92bb0c31537e
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
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
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/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include "compat.h"
42 #include "ctree.h"
43 #include "disk-io.h"
44 #include "transaction.h"
45 #include "btrfs_inode.h"
46 #include "ioctl.h"
47 #include "print-tree.h"
48 #include "volumes.h"
49 #include "ordered-data.h"
50 #include "xattr.h"
51 #include "tree-log.h"
52 #include "compression.h"
53 #include "locking.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
56
57 struct btrfs_iget_args {
58 u64 ino;
59 struct btrfs_root *root;
60 };
61
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct address_space_operations btrfs_symlink_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static struct extent_io_ops btrfs_extent_io_ops;
71
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_transaction_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
77
78 #define S_SHIFT 12
79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 };
88
89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
90 static int btrfs_truncate(struct inode *inode);
91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
92 static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
96
97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
98 struct inode *inode, struct inode *dir,
99 const struct qstr *qstr)
100 {
101 int err;
102
103 err = btrfs_init_acl(trans, inode, dir);
104 if (!err)
105 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
106 return err;
107 }
108
109 /*
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
113 */
114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
115 struct btrfs_root *root, struct inode *inode,
116 u64 start, size_t size, size_t compressed_size,
117 int compress_type,
118 struct page **compressed_pages)
119 {
120 struct btrfs_key key;
121 struct btrfs_path *path;
122 struct extent_buffer *leaf;
123 struct page *page = NULL;
124 char *kaddr;
125 unsigned long ptr;
126 struct btrfs_file_extent_item *ei;
127 int err = 0;
128 int ret;
129 size_t cur_size = size;
130 size_t datasize;
131 unsigned long offset;
132
133 if (compressed_size && compressed_pages)
134 cur_size = compressed_size;
135
136 path = btrfs_alloc_path();
137 if (!path)
138 return -ENOMEM;
139
140 path->leave_spinning = 1;
141
142 key.objectid = btrfs_ino(inode);
143 key.offset = start;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
149 datasize);
150 BUG_ON(ret);
151 if (ret) {
152 err = ret;
153 goto fail;
154 }
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
164
165 if (compress_type != BTRFS_COMPRESS_NONE) {
166 struct page *cpage;
167 int i = 0;
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
171 PAGE_CACHE_SIZE);
172
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
176
177 i++;
178 ptr += cur_size;
179 compressed_size -= cur_size;
180 }
181 btrfs_set_file_extent_compression(leaf, ei,
182 compress_type);
183 } else {
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
192 }
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
195
196 /*
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
200 *
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
204 */
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
207
208 return 0;
209 fail:
210 btrfs_free_path(path);
211 return err;
212 }
213
214
215 /*
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
219 */
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size, int compress_type,
224 struct page **compressed_pages)
225 {
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
231 u64 hint_byte;
232 u64 data_len = inline_len;
233 int ret;
234
235 if (compressed_size)
236 data_len = compressed_size;
237
238 if (start > 0 ||
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (!compressed_size &&
242 (actual_end & (root->sectorsize - 1)) == 0) ||
243 end + 1 < isize ||
244 data_len > root->fs_info->max_inline) {
245 return 1;
246 }
247
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 &hint_byte, 1);
250 BUG_ON(ret);
251
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
256 compress_type, compressed_pages);
257 BUG_ON(ret);
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
260 return 0;
261 }
262
263 struct async_extent {
264 u64 start;
265 u64 ram_size;
266 u64 compressed_size;
267 struct page **pages;
268 unsigned long nr_pages;
269 int compress_type;
270 struct list_head list;
271 };
272
273 struct async_cow {
274 struct inode *inode;
275 struct btrfs_root *root;
276 struct page *locked_page;
277 u64 start;
278 u64 end;
279 struct list_head extents;
280 struct btrfs_work work;
281 };
282
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
285 u64 compressed_size,
286 struct page **pages,
287 unsigned long nr_pages,
288 int compress_type)
289 {
290 struct async_extent *async_extent;
291
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 BUG_ON(!async_extent);
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 async_extent->compress_type = compress_type;
300 list_add_tail(&async_extent->list, &cow->extents);
301 return 0;
302 }
303
304 /*
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
308 *
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
314 *
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
319 */
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
322 u64 start, u64 end,
323 struct async_cow *async_cow,
324 int *num_added)
325 {
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
328 u64 num_bytes;
329 u64 blocksize = root->sectorsize;
330 u64 actual_end;
331 u64 isize = i_size_read(inode);
332 int ret = 0;
333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
339 unsigned long max_uncompressed = 128 * 1024;
340 int i;
341 int will_compress;
342 int compress_type = root->fs_info->compress_type;
343
344 /* if this is a small write inside eof, kick off a defragbot */
345 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL, inode);
347
348 actual_end = min_t(u64, isize, end + 1);
349 again:
350 will_compress = 0;
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353
354 /*
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
360 *
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
363 */
364 if (actual_end <= start)
365 goto cleanup_and_bail_uncompressed;
366
367 total_compressed = actual_end - start;
368
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
374 *
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
378 */
379 total_compressed = min(total_compressed, max_uncompressed);
380 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
381 num_bytes = max(blocksize, num_bytes);
382 total_in = 0;
383 ret = 0;
384
385 /*
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
389 */
390 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
391 (btrfs_test_opt(root, COMPRESS) ||
392 (BTRFS_I(inode)->force_compress) ||
393 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
394 WARN_ON(pages);
395 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
396 BUG_ON(!pages);
397
398 if (BTRFS_I(inode)->force_compress)
399 compress_type = BTRFS_I(inode)->force_compress;
400
401 ret = btrfs_compress_pages(compress_type,
402 inode->i_mapping, start,
403 total_compressed, pages,
404 nr_pages, &nr_pages_ret,
405 &total_in,
406 &total_compressed,
407 max_compressed);
408
409 if (!ret) {
410 unsigned long offset = total_compressed &
411 (PAGE_CACHE_SIZE - 1);
412 struct page *page = pages[nr_pages_ret - 1];
413 char *kaddr;
414
415 /* zero the tail end of the last page, we might be
416 * sending it down to disk
417 */
418 if (offset) {
419 kaddr = kmap_atomic(page, KM_USER0);
420 memset(kaddr + offset, 0,
421 PAGE_CACHE_SIZE - offset);
422 kunmap_atomic(kaddr, KM_USER0);
423 }
424 will_compress = 1;
425 }
426 }
427 if (start == 0) {
428 trans = btrfs_join_transaction(root);
429 BUG_ON(IS_ERR(trans));
430 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
431
432 /* lets try to make an inline extent */
433 if (ret || total_in < (actual_end - start)) {
434 /* we didn't compress the entire range, try
435 * to make an uncompressed inline extent.
436 */
437 ret = cow_file_range_inline(trans, root, inode,
438 start, end, 0, 0, NULL);
439 } else {
440 /* try making a compressed inline extent */
441 ret = cow_file_range_inline(trans, root, inode,
442 start, end,
443 total_compressed,
444 compress_type, pages);
445 }
446 if (ret == 0) {
447 /*
448 * inline extent creation worked, we don't need
449 * to create any more async work items. Unlock
450 * and free up our temp pages.
451 */
452 extent_clear_unlock_delalloc(inode,
453 &BTRFS_I(inode)->io_tree,
454 start, end, NULL,
455 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
456 EXTENT_CLEAR_DELALLOC |
457 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
458
459 btrfs_end_transaction(trans, root);
460 goto free_pages_out;
461 }
462 btrfs_end_transaction(trans, root);
463 }
464
465 if (will_compress) {
466 /*
467 * we aren't doing an inline extent round the compressed size
468 * up to a block size boundary so the allocator does sane
469 * things
470 */
471 total_compressed = (total_compressed + blocksize - 1) &
472 ~(blocksize - 1);
473
474 /*
475 * one last check to make sure the compression is really a
476 * win, compare the page count read with the blocks on disk
477 */
478 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
479 ~(PAGE_CACHE_SIZE - 1);
480 if (total_compressed >= total_in) {
481 will_compress = 0;
482 } else {
483 num_bytes = total_in;
484 }
485 }
486 if (!will_compress && pages) {
487 /*
488 * the compression code ran but failed to make things smaller,
489 * free any pages it allocated and our page pointer array
490 */
491 for (i = 0; i < nr_pages_ret; i++) {
492 WARN_ON(pages[i]->mapping);
493 page_cache_release(pages[i]);
494 }
495 kfree(pages);
496 pages = NULL;
497 total_compressed = 0;
498 nr_pages_ret = 0;
499
500 /* flag the file so we don't compress in the future */
501 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
502 !(BTRFS_I(inode)->force_compress)) {
503 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
504 }
505 }
506 if (will_compress) {
507 *num_added += 1;
508
509 /* the async work queues will take care of doing actual
510 * allocation on disk for these compressed pages,
511 * and will submit them to the elevator.
512 */
513 add_async_extent(async_cow, start, num_bytes,
514 total_compressed, pages, nr_pages_ret,
515 compress_type);
516
517 if (start + num_bytes < end) {
518 start += num_bytes;
519 pages = NULL;
520 cond_resched();
521 goto again;
522 }
523 } else {
524 cleanup_and_bail_uncompressed:
525 /*
526 * No compression, but we still need to write the pages in
527 * the file we've been given so far. redirty the locked
528 * page if it corresponds to our extent and set things up
529 * for the async work queue to run cow_file_range to do
530 * the normal delalloc dance
531 */
532 if (page_offset(locked_page) >= start &&
533 page_offset(locked_page) <= end) {
534 __set_page_dirty_nobuffers(locked_page);
535 /* unlocked later on in the async handlers */
536 }
537 add_async_extent(async_cow, start, end - start + 1,
538 0, NULL, 0, BTRFS_COMPRESS_NONE);
539 *num_added += 1;
540 }
541
542 out:
543 return 0;
544
545 free_pages_out:
546 for (i = 0; i < nr_pages_ret; i++) {
547 WARN_ON(pages[i]->mapping);
548 page_cache_release(pages[i]);
549 }
550 kfree(pages);
551
552 goto out;
553 }
554
555 /*
556 * phase two of compressed writeback. This is the ordered portion
557 * of the code, which only gets called in the order the work was
558 * queued. We walk all the async extents created by compress_file_range
559 * and send them down to the disk.
560 */
561 static noinline int submit_compressed_extents(struct inode *inode,
562 struct async_cow *async_cow)
563 {
564 struct async_extent *async_extent;
565 u64 alloc_hint = 0;
566 struct btrfs_trans_handle *trans;
567 struct btrfs_key ins;
568 struct extent_map *em;
569 struct btrfs_root *root = BTRFS_I(inode)->root;
570 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
571 struct extent_io_tree *io_tree;
572 int ret = 0;
573
574 if (list_empty(&async_cow->extents))
575 return 0;
576
577
578 while (!list_empty(&async_cow->extents)) {
579 async_extent = list_entry(async_cow->extents.next,
580 struct async_extent, list);
581 list_del(&async_extent->list);
582
583 io_tree = &BTRFS_I(inode)->io_tree;
584
585 retry:
586 /* did the compression code fall back to uncompressed IO? */
587 if (!async_extent->pages) {
588 int page_started = 0;
589 unsigned long nr_written = 0;
590
591 lock_extent(io_tree, async_extent->start,
592 async_extent->start +
593 async_extent->ram_size - 1, GFP_NOFS);
594
595 /* allocate blocks */
596 ret = cow_file_range(inode, async_cow->locked_page,
597 async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1,
600 &page_started, &nr_written, 0);
601
602 /*
603 * if page_started, cow_file_range inserted an
604 * inline extent and took care of all the unlocking
605 * and IO for us. Otherwise, we need to submit
606 * all those pages down to the drive.
607 */
608 if (!page_started && !ret)
609 extent_write_locked_range(io_tree,
610 inode, async_extent->start,
611 async_extent->start +
612 async_extent->ram_size - 1,
613 btrfs_get_extent,
614 WB_SYNC_ALL);
615 kfree(async_extent);
616 cond_resched();
617 continue;
618 }
619
620 lock_extent(io_tree, async_extent->start,
621 async_extent->start + async_extent->ram_size - 1,
622 GFP_NOFS);
623
624 trans = btrfs_join_transaction(root);
625 BUG_ON(IS_ERR(trans));
626 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
627 ret = btrfs_reserve_extent(trans, root,
628 async_extent->compressed_size,
629 async_extent->compressed_size,
630 0, alloc_hint,
631 (u64)-1, &ins, 1);
632 btrfs_end_transaction(trans, root);
633
634 if (ret) {
635 int i;
636 for (i = 0; i < async_extent->nr_pages; i++) {
637 WARN_ON(async_extent->pages[i]->mapping);
638 page_cache_release(async_extent->pages[i]);
639 }
640 kfree(async_extent->pages);
641 async_extent->nr_pages = 0;
642 async_extent->pages = NULL;
643 unlock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1, GFP_NOFS);
646 goto retry;
647 }
648
649 /*
650 * here we're doing allocation and writeback of the
651 * compressed pages
652 */
653 btrfs_drop_extent_cache(inode, async_extent->start,
654 async_extent->start +
655 async_extent->ram_size - 1, 0);
656
657 em = alloc_extent_map();
658 BUG_ON(!em);
659 em->start = async_extent->start;
660 em->len = async_extent->ram_size;
661 em->orig_start = em->start;
662
663 em->block_start = ins.objectid;
664 em->block_len = ins.offset;
665 em->bdev = root->fs_info->fs_devices->latest_bdev;
666 em->compress_type = async_extent->compress_type;
667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
668 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
669
670 while (1) {
671 write_lock(&em_tree->lock);
672 ret = add_extent_mapping(em_tree, em);
673 write_unlock(&em_tree->lock);
674 if (ret != -EEXIST) {
675 free_extent_map(em);
676 break;
677 }
678 btrfs_drop_extent_cache(inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1, 0);
681 }
682
683 ret = btrfs_add_ordered_extent_compress(inode,
684 async_extent->start,
685 ins.objectid,
686 async_extent->ram_size,
687 ins.offset,
688 BTRFS_ORDERED_COMPRESSED,
689 async_extent->compress_type);
690 BUG_ON(ret);
691
692 /*
693 * clear dirty, set writeback and unlock the pages.
694 */
695 extent_clear_unlock_delalloc(inode,
696 &BTRFS_I(inode)->io_tree,
697 async_extent->start,
698 async_extent->start +
699 async_extent->ram_size - 1,
700 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
701 EXTENT_CLEAR_UNLOCK |
702 EXTENT_CLEAR_DELALLOC |
703 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
704
705 ret = btrfs_submit_compressed_write(inode,
706 async_extent->start,
707 async_extent->ram_size,
708 ins.objectid,
709 ins.offset, async_extent->pages,
710 async_extent->nr_pages);
711
712 BUG_ON(ret);
713 alloc_hint = ins.objectid + ins.offset;
714 kfree(async_extent);
715 cond_resched();
716 }
717
718 return 0;
719 }
720
721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
722 u64 num_bytes)
723 {
724 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
725 struct extent_map *em;
726 u64 alloc_hint = 0;
727
728 read_lock(&em_tree->lock);
729 em = search_extent_mapping(em_tree, start, num_bytes);
730 if (em) {
731 /*
732 * if block start isn't an actual block number then find the
733 * first block in this inode and use that as a hint. If that
734 * block is also bogus then just don't worry about it.
735 */
736 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
737 free_extent_map(em);
738 em = search_extent_mapping(em_tree, 0, 0);
739 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
740 alloc_hint = em->block_start;
741 if (em)
742 free_extent_map(em);
743 } else {
744 alloc_hint = em->block_start;
745 free_extent_map(em);
746 }
747 }
748 read_unlock(&em_tree->lock);
749
750 return alloc_hint;
751 }
752
753 /*
754 * when extent_io.c finds a delayed allocation range in the file,
755 * the call backs end up in this code. The basic idea is to
756 * allocate extents on disk for the range, and create ordered data structs
757 * in ram to track those extents.
758 *
759 * locked_page is the page that writepage had locked already. We use
760 * it to make sure we don't do extra locks or unlocks.
761 *
762 * *page_started is set to one if we unlock locked_page and do everything
763 * required to start IO on it. It may be clean and already done with
764 * IO when we return.
765 */
766 static noinline int cow_file_range(struct inode *inode,
767 struct page *locked_page,
768 u64 start, u64 end, int *page_started,
769 unsigned long *nr_written,
770 int unlock)
771 {
772 struct btrfs_root *root = BTRFS_I(inode)->root;
773 struct btrfs_trans_handle *trans;
774 u64 alloc_hint = 0;
775 u64 num_bytes;
776 unsigned long ram_size;
777 u64 disk_num_bytes;
778 u64 cur_alloc_size;
779 u64 blocksize = root->sectorsize;
780 struct btrfs_key ins;
781 struct extent_map *em;
782 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
783 int ret = 0;
784
785 BUG_ON(btrfs_is_free_space_inode(root, inode));
786 trans = btrfs_join_transaction(root);
787 BUG_ON(IS_ERR(trans));
788 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
789
790 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
791 num_bytes = max(blocksize, num_bytes);
792 disk_num_bytes = num_bytes;
793 ret = 0;
794
795 /* if this is a small write inside eof, kick off defrag */
796 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
797 btrfs_add_inode_defrag(trans, inode);
798
799 if (start == 0) {
800 /* lets try to make an inline extent */
801 ret = cow_file_range_inline(trans, root, inode,
802 start, end, 0, 0, NULL);
803 if (ret == 0) {
804 extent_clear_unlock_delalloc(inode,
805 &BTRFS_I(inode)->io_tree,
806 start, end, NULL,
807 EXTENT_CLEAR_UNLOCK_PAGE |
808 EXTENT_CLEAR_UNLOCK |
809 EXTENT_CLEAR_DELALLOC |
810 EXTENT_CLEAR_DIRTY |
811 EXTENT_SET_WRITEBACK |
812 EXTENT_END_WRITEBACK);
813
814 *nr_written = *nr_written +
815 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
816 *page_started = 1;
817 ret = 0;
818 goto out;
819 }
820 }
821
822 BUG_ON(disk_num_bytes >
823 btrfs_super_total_bytes(&root->fs_info->super_copy));
824
825 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
826 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
827
828 while (disk_num_bytes > 0) {
829 unsigned long op;
830
831 cur_alloc_size = disk_num_bytes;
832 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
833 root->sectorsize, 0, alloc_hint,
834 (u64)-1, &ins, 1);
835 BUG_ON(ret);
836
837 em = alloc_extent_map();
838 BUG_ON(!em);
839 em->start = start;
840 em->orig_start = em->start;
841 ram_size = ins.offset;
842 em->len = ins.offset;
843
844 em->block_start = ins.objectid;
845 em->block_len = ins.offset;
846 em->bdev = root->fs_info->fs_devices->latest_bdev;
847 set_bit(EXTENT_FLAG_PINNED, &em->flags);
848
849 while (1) {
850 write_lock(&em_tree->lock);
851 ret = add_extent_mapping(em_tree, em);
852 write_unlock(&em_tree->lock);
853 if (ret != -EEXIST) {
854 free_extent_map(em);
855 break;
856 }
857 btrfs_drop_extent_cache(inode, start,
858 start + ram_size - 1, 0);
859 }
860
861 cur_alloc_size = ins.offset;
862 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
863 ram_size, cur_alloc_size, 0);
864 BUG_ON(ret);
865
866 if (root->root_key.objectid ==
867 BTRFS_DATA_RELOC_TREE_OBJECTID) {
868 ret = btrfs_reloc_clone_csums(inode, start,
869 cur_alloc_size);
870 BUG_ON(ret);
871 }
872
873 if (disk_num_bytes < cur_alloc_size)
874 break;
875
876 /* we're not doing compressed IO, don't unlock the first
877 * page (which the caller expects to stay locked), don't
878 * clear any dirty bits and don't set any writeback bits
879 *
880 * Do set the Private2 bit so we know this page was properly
881 * setup for writepage
882 */
883 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
884 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
885 EXTENT_SET_PRIVATE2;
886
887 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
888 start, start + ram_size - 1,
889 locked_page, op);
890 disk_num_bytes -= cur_alloc_size;
891 num_bytes -= cur_alloc_size;
892 alloc_hint = ins.objectid + ins.offset;
893 start += cur_alloc_size;
894 }
895 out:
896 ret = 0;
897 btrfs_end_transaction(trans, root);
898
899 return ret;
900 }
901
902 /*
903 * work queue call back to started compression on a file and pages
904 */
905 static noinline void async_cow_start(struct btrfs_work *work)
906 {
907 struct async_cow *async_cow;
908 int num_added = 0;
909 async_cow = container_of(work, struct async_cow, work);
910
911 compress_file_range(async_cow->inode, async_cow->locked_page,
912 async_cow->start, async_cow->end, async_cow,
913 &num_added);
914 if (num_added == 0)
915 async_cow->inode = NULL;
916 }
917
918 /*
919 * work queue call back to submit previously compressed pages
920 */
921 static noinline void async_cow_submit(struct btrfs_work *work)
922 {
923 struct async_cow *async_cow;
924 struct btrfs_root *root;
925 unsigned long nr_pages;
926
927 async_cow = container_of(work, struct async_cow, work);
928
929 root = async_cow->root;
930 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
931 PAGE_CACHE_SHIFT;
932
933 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
934
935 if (atomic_read(&root->fs_info->async_delalloc_pages) <
936 5 * 1042 * 1024 &&
937 waitqueue_active(&root->fs_info->async_submit_wait))
938 wake_up(&root->fs_info->async_submit_wait);
939
940 if (async_cow->inode)
941 submit_compressed_extents(async_cow->inode, async_cow);
942 }
943
944 static noinline void async_cow_free(struct btrfs_work *work)
945 {
946 struct async_cow *async_cow;
947 async_cow = container_of(work, struct async_cow, work);
948 kfree(async_cow);
949 }
950
951 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
952 u64 start, u64 end, int *page_started,
953 unsigned long *nr_written)
954 {
955 struct async_cow *async_cow;
956 struct btrfs_root *root = BTRFS_I(inode)->root;
957 unsigned long nr_pages;
958 u64 cur_end;
959 int limit = 10 * 1024 * 1042;
960
961 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
962 1, 0, NULL, GFP_NOFS);
963 while (start < end) {
964 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
965 BUG_ON(!async_cow);
966 async_cow->inode = inode;
967 async_cow->root = root;
968 async_cow->locked_page = locked_page;
969 async_cow->start = start;
970
971 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
972 cur_end = end;
973 else
974 cur_end = min(end, start + 512 * 1024 - 1);
975
976 async_cow->end = cur_end;
977 INIT_LIST_HEAD(&async_cow->extents);
978
979 async_cow->work.func = async_cow_start;
980 async_cow->work.ordered_func = async_cow_submit;
981 async_cow->work.ordered_free = async_cow_free;
982 async_cow->work.flags = 0;
983
984 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
985 PAGE_CACHE_SHIFT;
986 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
987
988 btrfs_queue_worker(&root->fs_info->delalloc_workers,
989 &async_cow->work);
990
991 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
992 wait_event(root->fs_info->async_submit_wait,
993 (atomic_read(&root->fs_info->async_delalloc_pages) <
994 limit));
995 }
996
997 while (atomic_read(&root->fs_info->async_submit_draining) &&
998 atomic_read(&root->fs_info->async_delalloc_pages)) {
999 wait_event(root->fs_info->async_submit_wait,
1000 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1001 0));
1002 }
1003
1004 *nr_written += nr_pages;
1005 start = cur_end + 1;
1006 }
1007 *page_started = 1;
1008 return 0;
1009 }
1010
1011 static noinline int csum_exist_in_range(struct btrfs_root *root,
1012 u64 bytenr, u64 num_bytes)
1013 {
1014 int ret;
1015 struct btrfs_ordered_sum *sums;
1016 LIST_HEAD(list);
1017
1018 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1019 bytenr + num_bytes - 1, &list, 0);
1020 if (ret == 0 && list_empty(&list))
1021 return 0;
1022
1023 while (!list_empty(&list)) {
1024 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025 list_del(&sums->list);
1026 kfree(sums);
1027 }
1028 return 1;
1029 }
1030
1031 /*
1032 * when nowcow writeback call back. This checks for snapshots or COW copies
1033 * of the extents that exist in the file, and COWs the file as required.
1034 *
1035 * If no cow copies or snapshots exist, we write directly to the existing
1036 * blocks on disk
1037 */
1038 static noinline int run_delalloc_nocow(struct inode *inode,
1039 struct page *locked_page,
1040 u64 start, u64 end, int *page_started, int force,
1041 unsigned long *nr_written)
1042 {
1043 struct btrfs_root *root = BTRFS_I(inode)->root;
1044 struct btrfs_trans_handle *trans;
1045 struct extent_buffer *leaf;
1046 struct btrfs_path *path;
1047 struct btrfs_file_extent_item *fi;
1048 struct btrfs_key found_key;
1049 u64 cow_start;
1050 u64 cur_offset;
1051 u64 extent_end;
1052 u64 extent_offset;
1053 u64 disk_bytenr;
1054 u64 num_bytes;
1055 int extent_type;
1056 int ret;
1057 int type;
1058 int nocow;
1059 int check_prev = 1;
1060 bool nolock;
1061 u64 ino = btrfs_ino(inode);
1062
1063 path = btrfs_alloc_path();
1064 if (!path)
1065 return -ENOMEM;
1066
1067 nolock = btrfs_is_free_space_inode(root, inode);
1068
1069 if (nolock)
1070 trans = btrfs_join_transaction_nolock(root);
1071 else
1072 trans = btrfs_join_transaction(root);
1073
1074 BUG_ON(IS_ERR(trans));
1075 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1076
1077 cow_start = (u64)-1;
1078 cur_offset = start;
1079 while (1) {
1080 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1081 cur_offset, 0);
1082 BUG_ON(ret < 0);
1083 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084 leaf = path->nodes[0];
1085 btrfs_item_key_to_cpu(leaf, &found_key,
1086 path->slots[0] - 1);
1087 if (found_key.objectid == ino &&
1088 found_key.type == BTRFS_EXTENT_DATA_KEY)
1089 path->slots[0]--;
1090 }
1091 check_prev = 0;
1092 next_slot:
1093 leaf = path->nodes[0];
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1096 if (ret < 0)
1097 BUG_ON(1);
1098 if (ret > 0)
1099 break;
1100 leaf = path->nodes[0];
1101 }
1102
1103 nocow = 0;
1104 disk_bytenr = 0;
1105 num_bytes = 0;
1106 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
1108 if (found_key.objectid > ino ||
1109 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110 found_key.offset > end)
1111 break;
1112
1113 if (found_key.offset > cur_offset) {
1114 extent_end = found_key.offset;
1115 extent_type = 0;
1116 goto out_check;
1117 }
1118
1119 fi = btrfs_item_ptr(leaf, path->slots[0],
1120 struct btrfs_file_extent_item);
1121 extent_type = btrfs_file_extent_type(leaf, fi);
1122
1123 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126 extent_offset = btrfs_file_extent_offset(leaf, fi);
1127 extent_end = found_key.offset +
1128 btrfs_file_extent_num_bytes(leaf, fi);
1129 if (extent_end <= start) {
1130 path->slots[0]++;
1131 goto next_slot;
1132 }
1133 if (disk_bytenr == 0)
1134 goto out_check;
1135 if (btrfs_file_extent_compression(leaf, fi) ||
1136 btrfs_file_extent_encryption(leaf, fi) ||
1137 btrfs_file_extent_other_encoding(leaf, fi))
1138 goto out_check;
1139 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140 goto out_check;
1141 if (btrfs_extent_readonly(root, disk_bytenr))
1142 goto out_check;
1143 if (btrfs_cross_ref_exist(trans, root, ino,
1144 found_key.offset -
1145 extent_offset, disk_bytenr))
1146 goto out_check;
1147 disk_bytenr += extent_offset;
1148 disk_bytenr += cur_offset - found_key.offset;
1149 num_bytes = min(end + 1, extent_end) - cur_offset;
1150 /*
1151 * force cow if csum exists in the range.
1152 * this ensure that csum for a given extent are
1153 * either valid or do not exist.
1154 */
1155 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156 goto out_check;
1157 nocow = 1;
1158 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159 extent_end = found_key.offset +
1160 btrfs_file_extent_inline_len(leaf, fi);
1161 extent_end = ALIGN(extent_end, root->sectorsize);
1162 } else {
1163 BUG_ON(1);
1164 }
1165 out_check:
1166 if (extent_end <= start) {
1167 path->slots[0]++;
1168 goto next_slot;
1169 }
1170 if (!nocow) {
1171 if (cow_start == (u64)-1)
1172 cow_start = cur_offset;
1173 cur_offset = extent_end;
1174 if (cur_offset > end)
1175 break;
1176 path->slots[0]++;
1177 goto next_slot;
1178 }
1179
1180 btrfs_release_path(path);
1181 if (cow_start != (u64)-1) {
1182 ret = cow_file_range(inode, locked_page, cow_start,
1183 found_key.offset - 1, page_started,
1184 nr_written, 1);
1185 BUG_ON(ret);
1186 cow_start = (u64)-1;
1187 }
1188
1189 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190 struct extent_map *em;
1191 struct extent_map_tree *em_tree;
1192 em_tree = &BTRFS_I(inode)->extent_tree;
1193 em = alloc_extent_map();
1194 BUG_ON(!em);
1195 em->start = cur_offset;
1196 em->orig_start = em->start;
1197 em->len = num_bytes;
1198 em->block_len = num_bytes;
1199 em->block_start = disk_bytenr;
1200 em->bdev = root->fs_info->fs_devices->latest_bdev;
1201 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202 while (1) {
1203 write_lock(&em_tree->lock);
1204 ret = add_extent_mapping(em_tree, em);
1205 write_unlock(&em_tree->lock);
1206 if (ret != -EEXIST) {
1207 free_extent_map(em);
1208 break;
1209 }
1210 btrfs_drop_extent_cache(inode, em->start,
1211 em->start + em->len - 1, 0);
1212 }
1213 type = BTRFS_ORDERED_PREALLOC;
1214 } else {
1215 type = BTRFS_ORDERED_NOCOW;
1216 }
1217
1218 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219 num_bytes, num_bytes, type);
1220 BUG_ON(ret);
1221
1222 if (root->root_key.objectid ==
1223 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225 num_bytes);
1226 BUG_ON(ret);
1227 }
1228
1229 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230 cur_offset, cur_offset + num_bytes - 1,
1231 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233 EXTENT_SET_PRIVATE2);
1234 cur_offset = extent_end;
1235 if (cur_offset > end)
1236 break;
1237 }
1238 btrfs_release_path(path);
1239
1240 if (cur_offset <= end && cow_start == (u64)-1)
1241 cow_start = cur_offset;
1242 if (cow_start != (u64)-1) {
1243 ret = cow_file_range(inode, locked_page, cow_start, end,
1244 page_started, nr_written, 1);
1245 BUG_ON(ret);
1246 }
1247
1248 if (nolock) {
1249 ret = btrfs_end_transaction_nolock(trans, root);
1250 BUG_ON(ret);
1251 } else {
1252 ret = btrfs_end_transaction(trans, root);
1253 BUG_ON(ret);
1254 }
1255 btrfs_free_path(path);
1256 return 0;
1257 }
1258
1259 /*
1260 * extent_io.c call back to do delayed allocation processing
1261 */
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263 u64 start, u64 end, int *page_started,
1264 unsigned long *nr_written)
1265 {
1266 int ret;
1267 struct btrfs_root *root = BTRFS_I(inode)->root;
1268
1269 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271 page_started, 1, nr_written);
1272 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274 page_started, 0, nr_written);
1275 else if (!btrfs_test_opt(root, COMPRESS) &&
1276 !(BTRFS_I(inode)->force_compress) &&
1277 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278 ret = cow_file_range(inode, locked_page, start, end,
1279 page_started, nr_written, 1);
1280 else
1281 ret = cow_file_range_async(inode, locked_page, start, end,
1282 page_started, nr_written);
1283 return ret;
1284 }
1285
1286 static void btrfs_split_extent_hook(struct inode *inode,
1287 struct extent_state *orig, u64 split)
1288 {
1289 /* not delalloc, ignore it */
1290 if (!(orig->state & EXTENT_DELALLOC))
1291 return;
1292
1293 spin_lock(&BTRFS_I(inode)->lock);
1294 BTRFS_I(inode)->outstanding_extents++;
1295 spin_unlock(&BTRFS_I(inode)->lock);
1296 }
1297
1298 /*
1299 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300 * extents so we can keep track of new extents that are just merged onto old
1301 * extents, such as when we are doing sequential writes, so we can properly
1302 * account for the metadata space we'll need.
1303 */
1304 static void btrfs_merge_extent_hook(struct inode *inode,
1305 struct extent_state *new,
1306 struct extent_state *other)
1307 {
1308 /* not delalloc, ignore it */
1309 if (!(other->state & EXTENT_DELALLOC))
1310 return;
1311
1312 spin_lock(&BTRFS_I(inode)->lock);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->lock);
1315 }
1316
1317 /*
1318 * extent_io.c set_bit_hook, used to track delayed allocation
1319 * bytes in this file, and to maintain the list of inodes that
1320 * have pending delalloc work to be done.
1321 */
1322 static void btrfs_set_bit_hook(struct inode *inode,
1323 struct extent_state *state, int *bits)
1324 {
1325
1326 /*
1327 * set_bit and clear bit hooks normally require _irqsave/restore
1328 * but in this case, we are only testing for the DELALLOC
1329 * bit, which is only set or cleared with irqs on
1330 */
1331 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1332 struct btrfs_root *root = BTRFS_I(inode)->root;
1333 u64 len = state->end + 1 - state->start;
1334 bool do_list = !btrfs_is_free_space_inode(root, inode);
1335
1336 if (*bits & EXTENT_FIRST_DELALLOC) {
1337 *bits &= ~EXTENT_FIRST_DELALLOC;
1338 } else {
1339 spin_lock(&BTRFS_I(inode)->lock);
1340 BTRFS_I(inode)->outstanding_extents++;
1341 spin_unlock(&BTRFS_I(inode)->lock);
1342 }
1343
1344 spin_lock(&root->fs_info->delalloc_lock);
1345 BTRFS_I(inode)->delalloc_bytes += len;
1346 root->fs_info->delalloc_bytes += len;
1347 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1348 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349 &root->fs_info->delalloc_inodes);
1350 }
1351 spin_unlock(&root->fs_info->delalloc_lock);
1352 }
1353 }
1354
1355 /*
1356 * extent_io.c clear_bit_hook, see set_bit_hook for why
1357 */
1358 static void btrfs_clear_bit_hook(struct inode *inode,
1359 struct extent_state *state, int *bits)
1360 {
1361 /*
1362 * set_bit and clear bit hooks normally require _irqsave/restore
1363 * but in this case, we are only testing for the DELALLOC
1364 * bit, which is only set or cleared with irqs on
1365 */
1366 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1367 struct btrfs_root *root = BTRFS_I(inode)->root;
1368 u64 len = state->end + 1 - state->start;
1369 bool do_list = !btrfs_is_free_space_inode(root, inode);
1370
1371 if (*bits & EXTENT_FIRST_DELALLOC) {
1372 *bits &= ~EXTENT_FIRST_DELALLOC;
1373 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374 spin_lock(&BTRFS_I(inode)->lock);
1375 BTRFS_I(inode)->outstanding_extents--;
1376 spin_unlock(&BTRFS_I(inode)->lock);
1377 }
1378
1379 if (*bits & EXTENT_DO_ACCOUNTING)
1380 btrfs_delalloc_release_metadata(inode, len);
1381
1382 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1383 && do_list)
1384 btrfs_free_reserved_data_space(inode, len);
1385
1386 spin_lock(&root->fs_info->delalloc_lock);
1387 root->fs_info->delalloc_bytes -= len;
1388 BTRFS_I(inode)->delalloc_bytes -= len;
1389
1390 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1391 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1393 }
1394 spin_unlock(&root->fs_info->delalloc_lock);
1395 }
1396 }
1397
1398 /*
1399 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400 * we don't create bios that span stripes or chunks
1401 */
1402 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1403 size_t size, struct bio *bio,
1404 unsigned long bio_flags)
1405 {
1406 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407 struct btrfs_mapping_tree *map_tree;
1408 u64 logical = (u64)bio->bi_sector << 9;
1409 u64 length = 0;
1410 u64 map_length;
1411 int ret;
1412
1413 if (bio_flags & EXTENT_BIO_COMPRESSED)
1414 return 0;
1415
1416 length = bio->bi_size;
1417 map_tree = &root->fs_info->mapping_tree;
1418 map_length = length;
1419 ret = btrfs_map_block(map_tree, READ, logical,
1420 &map_length, NULL, 0);
1421
1422 if (map_length < length + size)
1423 return 1;
1424 return ret;
1425 }
1426
1427 /*
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1431 *
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1434 */
1435 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436 struct bio *bio, int mirror_num,
1437 unsigned long bio_flags,
1438 u64 bio_offset)
1439 {
1440 struct btrfs_root *root = BTRFS_I(inode)->root;
1441 int ret = 0;
1442
1443 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1444 BUG_ON(ret);
1445 return 0;
1446 }
1447
1448 /*
1449 * in order to insert checksums into the metadata in large chunks,
1450 * we wait until bio submission time. All the pages in the bio are
1451 * checksummed and sums are attached onto the ordered extent record.
1452 *
1453 * At IO completion time the cums attached on the ordered extent record
1454 * are inserted into the btree
1455 */
1456 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1457 int mirror_num, unsigned long bio_flags,
1458 u64 bio_offset)
1459 {
1460 struct btrfs_root *root = BTRFS_I(inode)->root;
1461 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1462 }
1463
1464 /*
1465 * extent_io.c submission hook. This does the right thing for csum calculation
1466 * on write, or reading the csums from the tree before a read
1467 */
1468 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1469 int mirror_num, unsigned long bio_flags,
1470 u64 bio_offset)
1471 {
1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1473 int ret = 0;
1474 int skip_sum;
1475
1476 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1477
1478 if (btrfs_is_free_space_inode(root, inode))
1479 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1480 else
1481 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1482 BUG_ON(ret);
1483
1484 if (!(rw & REQ_WRITE)) {
1485 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1486 return btrfs_submit_compressed_read(inode, bio,
1487 mirror_num, bio_flags);
1488 } else if (!skip_sum) {
1489 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1490 if (ret)
1491 return ret;
1492 }
1493 goto mapit;
1494 } else if (!skip_sum) {
1495 /* csum items have already been cloned */
1496 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1497 goto mapit;
1498 /* we're doing a write, do the async checksumming */
1499 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1500 inode, rw, bio, mirror_num,
1501 bio_flags, bio_offset,
1502 __btrfs_submit_bio_start,
1503 __btrfs_submit_bio_done);
1504 }
1505
1506 mapit:
1507 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1508 }
1509
1510 /*
1511 * given a list of ordered sums record them in the inode. This happens
1512 * at IO completion time based on sums calculated at bio submission time.
1513 */
1514 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1515 struct inode *inode, u64 file_offset,
1516 struct list_head *list)
1517 {
1518 struct btrfs_ordered_sum *sum;
1519
1520 list_for_each_entry(sum, list, list) {
1521 btrfs_csum_file_blocks(trans,
1522 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1523 }
1524 return 0;
1525 }
1526
1527 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528 struct extent_state **cached_state)
1529 {
1530 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1531 WARN_ON(1);
1532 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1533 cached_state, GFP_NOFS);
1534 }
1535
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup {
1538 struct page *page;
1539 struct btrfs_work work;
1540 };
1541
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1543 {
1544 struct btrfs_writepage_fixup *fixup;
1545 struct btrfs_ordered_extent *ordered;
1546 struct extent_state *cached_state = NULL;
1547 struct page *page;
1548 struct inode *inode;
1549 u64 page_start;
1550 u64 page_end;
1551
1552 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1553 page = fixup->page;
1554 again:
1555 lock_page(page);
1556 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557 ClearPageChecked(page);
1558 goto out_page;
1559 }
1560
1561 inode = page->mapping->host;
1562 page_start = page_offset(page);
1563 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1564
1565 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566 &cached_state, GFP_NOFS);
1567
1568 /* already ordered? We're done */
1569 if (PagePrivate2(page))
1570 goto out;
1571
1572 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1573 if (ordered) {
1574 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575 page_end, &cached_state, GFP_NOFS);
1576 unlock_page(page);
1577 btrfs_start_ordered_extent(inode, ordered, 1);
1578 goto again;
1579 }
1580
1581 BUG();
1582 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1583 ClearPageChecked(page);
1584 out:
1585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586 &cached_state, GFP_NOFS);
1587 out_page:
1588 unlock_page(page);
1589 page_cache_release(page);
1590 kfree(fixup);
1591 }
1592
1593 /*
1594 * There are a few paths in the higher layers of the kernel that directly
1595 * set the page dirty bit without asking the filesystem if it is a
1596 * good idea. This causes problems because we want to make sure COW
1597 * properly happens and the data=ordered rules are followed.
1598 *
1599 * In our case any range that doesn't have the ORDERED bit set
1600 * hasn't been properly setup for IO. We kick off an async process
1601 * to fix it up. The async helper will wait for ordered extents, set
1602 * the delalloc bit and make it safe to write the page.
1603 */
1604 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1605 {
1606 struct inode *inode = page->mapping->host;
1607 struct btrfs_writepage_fixup *fixup;
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
1609
1610 /* this page is properly in the ordered list */
1611 if (TestClearPagePrivate2(page))
1612 return 0;
1613
1614 if (PageChecked(page))
1615 return -EAGAIN;
1616
1617 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1618 if (!fixup)
1619 return -EAGAIN;
1620
1621 SetPageChecked(page);
1622 page_cache_get(page);
1623 fixup->work.func = btrfs_writepage_fixup_worker;
1624 fixup->page = page;
1625 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1626 return -EAGAIN;
1627 }
1628
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630 struct inode *inode, u64 file_pos,
1631 u64 disk_bytenr, u64 disk_num_bytes,
1632 u64 num_bytes, u64 ram_bytes,
1633 u8 compression, u8 encryption,
1634 u16 other_encoding, int extent_type)
1635 {
1636 struct btrfs_root *root = BTRFS_I(inode)->root;
1637 struct btrfs_file_extent_item *fi;
1638 struct btrfs_path *path;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key ins;
1641 u64 hint;
1642 int ret;
1643
1644 path = btrfs_alloc_path();
1645 if (!path)
1646 return -ENOMEM;
1647
1648 path->leave_spinning = 1;
1649
1650 /*
1651 * we may be replacing one extent in the tree with another.
1652 * The new extent is pinned in the extent map, and we don't want
1653 * to drop it from the cache until it is completely in the btree.
1654 *
1655 * So, tell btrfs_drop_extents to leave this extent in the cache.
1656 * the caller is expected to unpin it and allow it to be merged
1657 * with the others.
1658 */
1659 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1660 &hint, 0);
1661 BUG_ON(ret);
1662
1663 ins.objectid = btrfs_ino(inode);
1664 ins.offset = file_pos;
1665 ins.type = BTRFS_EXTENT_DATA_KEY;
1666 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1667 BUG_ON(ret);
1668 leaf = path->nodes[0];
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672 btrfs_set_file_extent_type(leaf, fi, extent_type);
1673 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675 btrfs_set_file_extent_offset(leaf, fi, 0);
1676 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678 btrfs_set_file_extent_compression(leaf, fi, compression);
1679 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1681
1682 btrfs_unlock_up_safe(path, 1);
1683 btrfs_set_lock_blocking(leaf);
1684
1685 btrfs_mark_buffer_dirty(leaf);
1686
1687 inode_add_bytes(inode, num_bytes);
1688
1689 ins.objectid = disk_bytenr;
1690 ins.offset = disk_num_bytes;
1691 ins.type = BTRFS_EXTENT_ITEM_KEY;
1692 ret = btrfs_alloc_reserved_file_extent(trans, root,
1693 root->root_key.objectid,
1694 btrfs_ino(inode), file_pos, &ins);
1695 BUG_ON(ret);
1696 btrfs_free_path(path);
1697
1698 return 0;
1699 }
1700
1701 /*
1702 * helper function for btrfs_finish_ordered_io, this
1703 * just reads in some of the csum leaves to prime them into ram
1704 * before we start the transaction. It limits the amount of btree
1705 * reads required while inside the transaction.
1706 */
1707 /* as ordered data IO finishes, this gets called so we can finish
1708 * an ordered extent if the range of bytes in the file it covers are
1709 * fully written.
1710 */
1711 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1712 {
1713 struct btrfs_root *root = BTRFS_I(inode)->root;
1714 struct btrfs_trans_handle *trans = NULL;
1715 struct btrfs_ordered_extent *ordered_extent = NULL;
1716 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1717 struct extent_state *cached_state = NULL;
1718 int compress_type = 0;
1719 int ret;
1720 bool nolock;
1721
1722 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1723 end - start + 1);
1724 if (!ret)
1725 return 0;
1726 BUG_ON(!ordered_extent);
1727
1728 nolock = btrfs_is_free_space_inode(root, inode);
1729
1730 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731 BUG_ON(!list_empty(&ordered_extent->list));
1732 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1733 if (!ret) {
1734 if (nolock)
1735 trans = btrfs_join_transaction_nolock(root);
1736 else
1737 trans = btrfs_join_transaction(root);
1738 BUG_ON(IS_ERR(trans));
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740 ret = btrfs_update_inode(trans, root, inode);
1741 BUG_ON(ret);
1742 }
1743 goto out;
1744 }
1745
1746 lock_extent_bits(io_tree, ordered_extent->file_offset,
1747 ordered_extent->file_offset + ordered_extent->len - 1,
1748 0, &cached_state, GFP_NOFS);
1749
1750 if (nolock)
1751 trans = btrfs_join_transaction_nolock(root);
1752 else
1753 trans = btrfs_join_transaction(root);
1754 BUG_ON(IS_ERR(trans));
1755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756
1757 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1758 compress_type = ordered_extent->compress_type;
1759 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1760 BUG_ON(compress_type);
1761 ret = btrfs_mark_extent_written(trans, inode,
1762 ordered_extent->file_offset,
1763 ordered_extent->file_offset +
1764 ordered_extent->len);
1765 BUG_ON(ret);
1766 } else {
1767 BUG_ON(root == root->fs_info->tree_root);
1768 ret = insert_reserved_file_extent(trans, inode,
1769 ordered_extent->file_offset,
1770 ordered_extent->start,
1771 ordered_extent->disk_len,
1772 ordered_extent->len,
1773 ordered_extent->len,
1774 compress_type, 0, 0,
1775 BTRFS_FILE_EXTENT_REG);
1776 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777 ordered_extent->file_offset,
1778 ordered_extent->len);
1779 BUG_ON(ret);
1780 }
1781 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1784
1785 add_pending_csums(trans, inode, ordered_extent->file_offset,
1786 &ordered_extent->list);
1787
1788 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1789 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790 ret = btrfs_update_inode(trans, root, inode);
1791 BUG_ON(ret);
1792 }
1793 ret = 0;
1794 out:
1795 if (nolock) {
1796 if (trans)
1797 btrfs_end_transaction_nolock(trans, root);
1798 } else {
1799 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1800 if (trans)
1801 btrfs_end_transaction(trans, root);
1802 }
1803
1804 /* once for us */
1805 btrfs_put_ordered_extent(ordered_extent);
1806 /* once for the tree */
1807 btrfs_put_ordered_extent(ordered_extent);
1808
1809 return 0;
1810 }
1811
1812 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1813 struct extent_state *state, int uptodate)
1814 {
1815 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1816
1817 ClearPagePrivate2(page);
1818 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1819 }
1820
1821 /*
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1828 */
1829 struct io_failure_record {
1830 struct page *page;
1831 u64 start;
1832 u64 len;
1833 u64 logical;
1834 unsigned long bio_flags;
1835 int last_mirror;
1836 };
1837
1838 static int btrfs_io_failed_hook(struct bio *failed_bio,
1839 struct page *page, u64 start, u64 end,
1840 struct extent_state *state)
1841 {
1842 struct io_failure_record *failrec = NULL;
1843 u64 private;
1844 struct extent_map *em;
1845 struct inode *inode = page->mapping->host;
1846 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1848 struct bio *bio;
1849 int num_copies;
1850 int ret;
1851 int rw;
1852 u64 logical;
1853
1854 ret = get_state_private(failure_tree, start, &private);
1855 if (ret) {
1856 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1857 if (!failrec)
1858 return -ENOMEM;
1859 failrec->start = start;
1860 failrec->len = end - start + 1;
1861 failrec->last_mirror = 0;
1862 failrec->bio_flags = 0;
1863
1864 read_lock(&em_tree->lock);
1865 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866 if (em->start > start || em->start + em->len < start) {
1867 free_extent_map(em);
1868 em = NULL;
1869 }
1870 read_unlock(&em_tree->lock);
1871
1872 if (IS_ERR_OR_NULL(em)) {
1873 kfree(failrec);
1874 return -EIO;
1875 }
1876 logical = start - em->start;
1877 logical = em->block_start + logical;
1878 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879 logical = em->block_start;
1880 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881 extent_set_compress_type(&failrec->bio_flags,
1882 em->compress_type);
1883 }
1884 failrec->logical = logical;
1885 free_extent_map(em);
1886 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887 EXTENT_DIRTY, GFP_NOFS);
1888 set_state_private(failure_tree, start,
1889 (u64)(unsigned long)failrec);
1890 } else {
1891 failrec = (struct io_failure_record *)(unsigned long)private;
1892 }
1893 num_copies = btrfs_num_copies(
1894 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895 failrec->logical, failrec->len);
1896 failrec->last_mirror++;
1897 if (!state) {
1898 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900 failrec->start,
1901 EXTENT_LOCKED);
1902 if (state && state->start != failrec->start)
1903 state = NULL;
1904 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1905 }
1906 if (!state || failrec->last_mirror > num_copies) {
1907 set_state_private(failure_tree, failrec->start, 0);
1908 clear_extent_bits(failure_tree, failrec->start,
1909 failrec->start + failrec->len - 1,
1910 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911 kfree(failrec);
1912 return -EIO;
1913 }
1914 bio = bio_alloc(GFP_NOFS, 1);
1915 bio->bi_private = state;
1916 bio->bi_end_io = failed_bio->bi_end_io;
1917 bio->bi_sector = failrec->logical >> 9;
1918 bio->bi_bdev = failed_bio->bi_bdev;
1919 bio->bi_size = 0;
1920
1921 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922 if (failed_bio->bi_rw & REQ_WRITE)
1923 rw = WRITE;
1924 else
1925 rw = READ;
1926
1927 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928 failrec->last_mirror,
1929 failrec->bio_flags, 0);
1930 return ret;
1931 }
1932
1933 /*
1934 * each time an IO finishes, we do a fast check in the IO failure tree
1935 * to see if we need to process or clean up an io_failure_record
1936 */
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1938 {
1939 u64 private;
1940 u64 private_failure;
1941 struct io_failure_record *failure;
1942 int ret;
1943
1944 private = 0;
1945 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948 start, &private_failure);
1949 if (ret == 0) {
1950 failure = (struct io_failure_record *)(unsigned long)
1951 private_failure;
1952 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953 failure->start, 0);
1954 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955 failure->start,
1956 failure->start + failure->len - 1,
1957 EXTENT_DIRTY | EXTENT_LOCKED,
1958 GFP_NOFS);
1959 kfree(failure);
1960 }
1961 }
1962 return 0;
1963 }
1964
1965 /*
1966 * when reads are done, we need to check csums to verify the data is correct
1967 * if there's a match, we allow the bio to finish. If not, we go through
1968 * the io_failure_record routines to find good copies
1969 */
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971 struct extent_state *state)
1972 {
1973 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974 struct inode *inode = page->mapping->host;
1975 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1976 char *kaddr;
1977 u64 private = ~(u32)0;
1978 int ret;
1979 struct btrfs_root *root = BTRFS_I(inode)->root;
1980 u32 csum = ~(u32)0;
1981
1982 if (PageChecked(page)) {
1983 ClearPageChecked(page);
1984 goto good;
1985 }
1986
1987 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1988 goto good;
1989
1990 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993 GFP_NOFS);
1994 return 0;
1995 }
1996
1997 if (state && state->start == start) {
1998 private = state->private;
1999 ret = 0;
2000 } else {
2001 ret = get_state_private(io_tree, start, &private);
2002 }
2003 kaddr = kmap_atomic(page, KM_USER0);
2004 if (ret)
2005 goto zeroit;
2006
2007 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2008 btrfs_csum_final(csum, (char *)&csum);
2009 if (csum != private)
2010 goto zeroit;
2011
2012 kunmap_atomic(kaddr, KM_USER0);
2013 good:
2014 /* if the io failure tree for this inode is non-empty,
2015 * check to see if we've recovered from a failed IO
2016 */
2017 btrfs_clean_io_failures(inode, start);
2018 return 0;
2019
2020 zeroit:
2021 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2022 "private %llu\n",
2023 (unsigned long long)btrfs_ino(page->mapping->host),
2024 (unsigned long long)start, csum,
2025 (unsigned long long)private);
2026 memset(kaddr + offset, 1, end - start + 1);
2027 flush_dcache_page(page);
2028 kunmap_atomic(kaddr, KM_USER0);
2029 if (private == 0)
2030 return 0;
2031 return -EIO;
2032 }
2033
2034 struct delayed_iput {
2035 struct list_head list;
2036 struct inode *inode;
2037 };
2038
2039 void btrfs_add_delayed_iput(struct inode *inode)
2040 {
2041 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042 struct delayed_iput *delayed;
2043
2044 if (atomic_add_unless(&inode->i_count, -1, 1))
2045 return;
2046
2047 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048 delayed->inode = inode;
2049
2050 spin_lock(&fs_info->delayed_iput_lock);
2051 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052 spin_unlock(&fs_info->delayed_iput_lock);
2053 }
2054
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056 {
2057 LIST_HEAD(list);
2058 struct btrfs_fs_info *fs_info = root->fs_info;
2059 struct delayed_iput *delayed;
2060 int empty;
2061
2062 spin_lock(&fs_info->delayed_iput_lock);
2063 empty = list_empty(&fs_info->delayed_iputs);
2064 spin_unlock(&fs_info->delayed_iput_lock);
2065 if (empty)
2066 return;
2067
2068 down_read(&root->fs_info->cleanup_work_sem);
2069 spin_lock(&fs_info->delayed_iput_lock);
2070 list_splice_init(&fs_info->delayed_iputs, &list);
2071 spin_unlock(&fs_info->delayed_iput_lock);
2072
2073 while (!list_empty(&list)) {
2074 delayed = list_entry(list.next, struct delayed_iput, list);
2075 list_del(&delayed->list);
2076 iput(delayed->inode);
2077 kfree(delayed);
2078 }
2079 up_read(&root->fs_info->cleanup_work_sem);
2080 }
2081
2082 /*
2083 * calculate extra metadata reservation when snapshotting a subvolume
2084 * contains orphan files.
2085 */
2086 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2087 struct btrfs_pending_snapshot *pending,
2088 u64 *bytes_to_reserve)
2089 {
2090 struct btrfs_root *root;
2091 struct btrfs_block_rsv *block_rsv;
2092 u64 num_bytes;
2093 int index;
2094
2095 root = pending->root;
2096 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2097 return;
2098
2099 block_rsv = root->orphan_block_rsv;
2100
2101 /* orphan block reservation for the snapshot */
2102 num_bytes = block_rsv->size;
2103
2104 /*
2105 * after the snapshot is created, COWing tree blocks may use more
2106 * space than it frees. So we should make sure there is enough
2107 * reserved space.
2108 */
2109 index = trans->transid & 0x1;
2110 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2111 num_bytes += block_rsv->size -
2112 (block_rsv->reserved + block_rsv->freed[index]);
2113 }
2114
2115 *bytes_to_reserve += num_bytes;
2116 }
2117
2118 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2119 struct btrfs_pending_snapshot *pending)
2120 {
2121 struct btrfs_root *root = pending->root;
2122 struct btrfs_root *snap = pending->snap;
2123 struct btrfs_block_rsv *block_rsv;
2124 u64 num_bytes;
2125 int index;
2126 int ret;
2127
2128 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2129 return;
2130
2131 /* refill source subvolume's orphan block reservation */
2132 block_rsv = root->orphan_block_rsv;
2133 index = trans->transid & 0x1;
2134 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2135 num_bytes = block_rsv->size -
2136 (block_rsv->reserved + block_rsv->freed[index]);
2137 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138 root->orphan_block_rsv,
2139 num_bytes);
2140 BUG_ON(ret);
2141 }
2142
2143 /* setup orphan block reservation for the snapshot */
2144 block_rsv = btrfs_alloc_block_rsv(snap);
2145 BUG_ON(!block_rsv);
2146
2147 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2148 snap->orphan_block_rsv = block_rsv;
2149
2150 num_bytes = root->orphan_block_rsv->size;
2151 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2152 block_rsv, num_bytes);
2153 BUG_ON(ret);
2154
2155 #if 0
2156 /* insert orphan item for the snapshot */
2157 WARN_ON(!root->orphan_item_inserted);
2158 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2159 snap->root_key.objectid);
2160 BUG_ON(ret);
2161 snap->orphan_item_inserted = 1;
2162 #endif
2163 }
2164
2165 enum btrfs_orphan_cleanup_state {
2166 ORPHAN_CLEANUP_STARTED = 1,
2167 ORPHAN_CLEANUP_DONE = 2,
2168 };
2169
2170 /*
2171 * This is called in transaction commmit time. If there are no orphan
2172 * files in the subvolume, it removes orphan item and frees block_rsv
2173 * structure.
2174 */
2175 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2176 struct btrfs_root *root)
2177 {
2178 int ret;
2179
2180 if (!list_empty(&root->orphan_list) ||
2181 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2182 return;
2183
2184 if (root->orphan_item_inserted &&
2185 btrfs_root_refs(&root->root_item) > 0) {
2186 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187 root->root_key.objectid);
2188 BUG_ON(ret);
2189 root->orphan_item_inserted = 0;
2190 }
2191
2192 if (root->orphan_block_rsv) {
2193 WARN_ON(root->orphan_block_rsv->size > 0);
2194 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2195 root->orphan_block_rsv = NULL;
2196 }
2197 }
2198
2199 /*
2200 * This creates an orphan entry for the given inode in case something goes
2201 * wrong in the middle of an unlink/truncate.
2202 *
2203 * NOTE: caller of this function should reserve 5 units of metadata for
2204 * this function.
2205 */
2206 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2207 {
2208 struct btrfs_root *root = BTRFS_I(inode)->root;
2209 struct btrfs_block_rsv *block_rsv = NULL;
2210 int reserve = 0;
2211 int insert = 0;
2212 int ret;
2213
2214 if (!root->orphan_block_rsv) {
2215 block_rsv = btrfs_alloc_block_rsv(root);
2216 if (!block_rsv)
2217 return -ENOMEM;
2218 }
2219
2220 spin_lock(&root->orphan_lock);
2221 if (!root->orphan_block_rsv) {
2222 root->orphan_block_rsv = block_rsv;
2223 } else if (block_rsv) {
2224 btrfs_free_block_rsv(root, block_rsv);
2225 block_rsv = NULL;
2226 }
2227
2228 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2230 #if 0
2231 /*
2232 * For proper ENOSPC handling, we should do orphan
2233 * cleanup when mounting. But this introduces backward
2234 * compatibility issue.
2235 */
2236 if (!xchg(&root->orphan_item_inserted, 1))
2237 insert = 2;
2238 else
2239 insert = 1;
2240 #endif
2241 insert = 1;
2242 }
2243
2244 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245 BTRFS_I(inode)->orphan_meta_reserved = 1;
2246 reserve = 1;
2247 }
2248 spin_unlock(&root->orphan_lock);
2249
2250 if (block_rsv)
2251 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2252
2253 /* grab metadata reservation from transaction handle */
2254 if (reserve) {
2255 ret = btrfs_orphan_reserve_metadata(trans, inode);
2256 BUG_ON(ret);
2257 }
2258
2259 /* insert an orphan item to track this unlinked/truncated file */
2260 if (insert >= 1) {
2261 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2262 BUG_ON(ret);
2263 }
2264
2265 /* insert an orphan item to track subvolume contains orphan files */
2266 if (insert >= 2) {
2267 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268 root->root_key.objectid);
2269 BUG_ON(ret);
2270 }
2271 return 0;
2272 }
2273
2274 /*
2275 * We have done the truncate/delete so we can go ahead and remove the orphan
2276 * item for this particular inode.
2277 */
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2279 {
2280 struct btrfs_root *root = BTRFS_I(inode)->root;
2281 int delete_item = 0;
2282 int release_rsv = 0;
2283 int ret = 0;
2284
2285 spin_lock(&root->orphan_lock);
2286 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287 list_del_init(&BTRFS_I(inode)->i_orphan);
2288 delete_item = 1;
2289 }
2290
2291 if (BTRFS_I(inode)->orphan_meta_reserved) {
2292 BTRFS_I(inode)->orphan_meta_reserved = 0;
2293 release_rsv = 1;
2294 }
2295 spin_unlock(&root->orphan_lock);
2296
2297 if (trans && delete_item) {
2298 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2299 BUG_ON(ret);
2300 }
2301
2302 if (release_rsv)
2303 btrfs_orphan_release_metadata(inode);
2304
2305 return 0;
2306 }
2307
2308 /*
2309 * this cleans up any orphans that may be left on the list from the last use
2310 * of this root.
2311 */
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2313 {
2314 struct btrfs_path *path;
2315 struct extent_buffer *leaf;
2316 struct btrfs_key key, found_key;
2317 struct btrfs_trans_handle *trans;
2318 struct inode *inode;
2319 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2320
2321 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2322 return 0;
2323
2324 path = btrfs_alloc_path();
2325 if (!path) {
2326 ret = -ENOMEM;
2327 goto out;
2328 }
2329 path->reada = -1;
2330
2331 key.objectid = BTRFS_ORPHAN_OBJECTID;
2332 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333 key.offset = (u64)-1;
2334
2335 while (1) {
2336 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2337 if (ret < 0)
2338 goto out;
2339
2340 /*
2341 * if ret == 0 means we found what we were searching for, which
2342 * is weird, but possible, so only screw with path if we didn't
2343 * find the key and see if we have stuff that matches
2344 */
2345 if (ret > 0) {
2346 ret = 0;
2347 if (path->slots[0] == 0)
2348 break;
2349 path->slots[0]--;
2350 }
2351
2352 /* pull out the item */
2353 leaf = path->nodes[0];
2354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356 /* make sure the item matches what we want */
2357 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2358 break;
2359 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2360 break;
2361
2362 /* release the path since we're done with it */
2363 btrfs_release_path(path);
2364
2365 /*
2366 * this is where we are basically btrfs_lookup, without the
2367 * crossing root thing. we store the inode number in the
2368 * offset of the orphan item.
2369 */
2370 found_key.objectid = found_key.offset;
2371 found_key.type = BTRFS_INODE_ITEM_KEY;
2372 found_key.offset = 0;
2373 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374 if (IS_ERR(inode)) {
2375 ret = PTR_ERR(inode);
2376 goto out;
2377 }
2378
2379 /*
2380 * add this inode to the orphan list so btrfs_orphan_del does
2381 * the proper thing when we hit it
2382 */
2383 spin_lock(&root->orphan_lock);
2384 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385 spin_unlock(&root->orphan_lock);
2386
2387 /*
2388 * if this is a bad inode, means we actually succeeded in
2389 * removing the inode, but not the orphan record, which means
2390 * we need to manually delete the orphan since iput will just
2391 * do a destroy_inode
2392 */
2393 if (is_bad_inode(inode)) {
2394 trans = btrfs_start_transaction(root, 0);
2395 if (IS_ERR(trans)) {
2396 ret = PTR_ERR(trans);
2397 goto out;
2398 }
2399 btrfs_orphan_del(trans, inode);
2400 btrfs_end_transaction(trans, root);
2401 iput(inode);
2402 continue;
2403 }
2404
2405 /* if we have links, this was a truncate, lets do that */
2406 if (inode->i_nlink) {
2407 if (!S_ISREG(inode->i_mode)) {
2408 WARN_ON(1);
2409 iput(inode);
2410 continue;
2411 }
2412 nr_truncate++;
2413 ret = btrfs_truncate(inode);
2414 } else {
2415 nr_unlink++;
2416 }
2417
2418 /* this will do delete_inode and everything for us */
2419 iput(inode);
2420 if (ret)
2421 goto out;
2422 }
2423 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2424
2425 if (root->orphan_block_rsv)
2426 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2427 (u64)-1);
2428
2429 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430 trans = btrfs_join_transaction(root);
2431 if (!IS_ERR(trans))
2432 btrfs_end_transaction(trans, root);
2433 }
2434
2435 if (nr_unlink)
2436 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2437 if (nr_truncate)
2438 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2439
2440 out:
2441 if (ret)
2442 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443 btrfs_free_path(path);
2444 return ret;
2445 }
2446
2447 /*
2448 * very simple check to peek ahead in the leaf looking for xattrs. If we
2449 * don't find any xattrs, we know there can't be any acls.
2450 *
2451 * slot is the slot the inode is in, objectid is the objectid of the inode
2452 */
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454 int slot, u64 objectid)
2455 {
2456 u32 nritems = btrfs_header_nritems(leaf);
2457 struct btrfs_key found_key;
2458 int scanned = 0;
2459
2460 slot++;
2461 while (slot < nritems) {
2462 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2463
2464 /* we found a different objectid, there must not be acls */
2465 if (found_key.objectid != objectid)
2466 return 0;
2467
2468 /* we found an xattr, assume we've got an acl */
2469 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2470 return 1;
2471
2472 /*
2473 * we found a key greater than an xattr key, there can't
2474 * be any acls later on
2475 */
2476 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2477 return 0;
2478
2479 slot++;
2480 scanned++;
2481
2482 /*
2483 * it goes inode, inode backrefs, xattrs, extents,
2484 * so if there are a ton of hard links to an inode there can
2485 * be a lot of backrefs. Don't waste time searching too hard,
2486 * this is just an optimization
2487 */
2488 if (scanned >= 8)
2489 break;
2490 }
2491 /* we hit the end of the leaf before we found an xattr or
2492 * something larger than an xattr. We have to assume the inode
2493 * has acls
2494 */
2495 return 1;
2496 }
2497
2498 /*
2499 * read an inode from the btree into the in-memory inode
2500 */
2501 static void btrfs_read_locked_inode(struct inode *inode)
2502 {
2503 struct btrfs_path *path;
2504 struct extent_buffer *leaf;
2505 struct btrfs_inode_item *inode_item;
2506 struct btrfs_timespec *tspec;
2507 struct btrfs_root *root = BTRFS_I(inode)->root;
2508 struct btrfs_key location;
2509 int maybe_acls;
2510 u32 rdev;
2511 int ret;
2512 bool filled = false;
2513
2514 ret = btrfs_fill_inode(inode, &rdev);
2515 if (!ret)
2516 filled = true;
2517
2518 path = btrfs_alloc_path();
2519 if (!path)
2520 goto make_bad;
2521
2522 path->leave_spinning = 1;
2523 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2524
2525 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2526 if (ret)
2527 goto make_bad;
2528
2529 leaf = path->nodes[0];
2530
2531 if (filled)
2532 goto cache_acl;
2533
2534 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2535 struct btrfs_inode_item);
2536 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2537 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2538 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2539 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2540 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2541
2542 tspec = btrfs_inode_atime(inode_item);
2543 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2544 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2545
2546 tspec = btrfs_inode_mtime(inode_item);
2547 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2548 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2549
2550 tspec = btrfs_inode_ctime(inode_item);
2551 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2552 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2553
2554 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2555 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2556 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2557 inode->i_generation = BTRFS_I(inode)->generation;
2558 inode->i_rdev = 0;
2559 rdev = btrfs_inode_rdev(leaf, inode_item);
2560
2561 BTRFS_I(inode)->index_cnt = (u64)-1;
2562 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2563 cache_acl:
2564 /*
2565 * try to precache a NULL acl entry for files that don't have
2566 * any xattrs or acls
2567 */
2568 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2569 btrfs_ino(inode));
2570 if (!maybe_acls)
2571 cache_no_acl(inode);
2572
2573 btrfs_free_path(path);
2574
2575 switch (inode->i_mode & S_IFMT) {
2576 case S_IFREG:
2577 inode->i_mapping->a_ops = &btrfs_aops;
2578 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2579 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2580 inode->i_fop = &btrfs_file_operations;
2581 inode->i_op = &btrfs_file_inode_operations;
2582 break;
2583 case S_IFDIR:
2584 inode->i_fop = &btrfs_dir_file_operations;
2585 if (root == root->fs_info->tree_root)
2586 inode->i_op = &btrfs_dir_ro_inode_operations;
2587 else
2588 inode->i_op = &btrfs_dir_inode_operations;
2589 break;
2590 case S_IFLNK:
2591 inode->i_op = &btrfs_symlink_inode_operations;
2592 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2593 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2594 break;
2595 default:
2596 inode->i_op = &btrfs_special_inode_operations;
2597 init_special_inode(inode, inode->i_mode, rdev);
2598 break;
2599 }
2600
2601 btrfs_update_iflags(inode);
2602 return;
2603
2604 make_bad:
2605 btrfs_free_path(path);
2606 make_bad_inode(inode);
2607 }
2608
2609 /*
2610 * given a leaf and an inode, copy the inode fields into the leaf
2611 */
2612 static void fill_inode_item(struct btrfs_trans_handle *trans,
2613 struct extent_buffer *leaf,
2614 struct btrfs_inode_item *item,
2615 struct inode *inode)
2616 {
2617 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2618 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2619 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2620 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2621 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2622
2623 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2624 inode->i_atime.tv_sec);
2625 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2626 inode->i_atime.tv_nsec);
2627
2628 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2629 inode->i_mtime.tv_sec);
2630 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2631 inode->i_mtime.tv_nsec);
2632
2633 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2634 inode->i_ctime.tv_sec);
2635 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2636 inode->i_ctime.tv_nsec);
2637
2638 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2639 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2640 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2641 btrfs_set_inode_transid(leaf, item, trans->transid);
2642 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2643 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2644 btrfs_set_inode_block_group(leaf, item, 0);
2645 }
2646
2647 /*
2648 * copy everything in the in-memory inode into the btree.
2649 */
2650 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2651 struct btrfs_root *root, struct inode *inode)
2652 {
2653 struct btrfs_inode_item *inode_item;
2654 struct btrfs_path *path;
2655 struct extent_buffer *leaf;
2656 int ret;
2657
2658 /*
2659 * If the inode is a free space inode, we can deadlock during commit
2660 * if we put it into the delayed code.
2661 *
2662 * The data relocation inode should also be directly updated
2663 * without delay
2664 */
2665 if (!btrfs_is_free_space_inode(root, inode)
2666 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2667 ret = btrfs_delayed_update_inode(trans, root, inode);
2668 if (!ret)
2669 btrfs_set_inode_last_trans(trans, inode);
2670 return ret;
2671 }
2672
2673 path = btrfs_alloc_path();
2674 if (!path)
2675 return -ENOMEM;
2676
2677 path->leave_spinning = 1;
2678 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2679 1);
2680 if (ret) {
2681 if (ret > 0)
2682 ret = -ENOENT;
2683 goto failed;
2684 }
2685
2686 btrfs_unlock_up_safe(path, 1);
2687 leaf = path->nodes[0];
2688 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2689 struct btrfs_inode_item);
2690
2691 fill_inode_item(trans, leaf, inode_item, inode);
2692 btrfs_mark_buffer_dirty(leaf);
2693 btrfs_set_inode_last_trans(trans, inode);
2694 ret = 0;
2695 failed:
2696 btrfs_free_path(path);
2697 return ret;
2698 }
2699
2700 /*
2701 * unlink helper that gets used here in inode.c and in the tree logging
2702 * recovery code. It remove a link in a directory with a given name, and
2703 * also drops the back refs in the inode to the directory
2704 */
2705 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2706 struct btrfs_root *root,
2707 struct inode *dir, struct inode *inode,
2708 const char *name, int name_len)
2709 {
2710 struct btrfs_path *path;
2711 int ret = 0;
2712 struct extent_buffer *leaf;
2713 struct btrfs_dir_item *di;
2714 struct btrfs_key key;
2715 u64 index;
2716 u64 ino = btrfs_ino(inode);
2717 u64 dir_ino = btrfs_ino(dir);
2718
2719 path = btrfs_alloc_path();
2720 if (!path) {
2721 ret = -ENOMEM;
2722 goto out;
2723 }
2724
2725 path->leave_spinning = 1;
2726 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2727 name, name_len, -1);
2728 if (IS_ERR(di)) {
2729 ret = PTR_ERR(di);
2730 goto err;
2731 }
2732 if (!di) {
2733 ret = -ENOENT;
2734 goto err;
2735 }
2736 leaf = path->nodes[0];
2737 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2738 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2739 if (ret)
2740 goto err;
2741 btrfs_release_path(path);
2742
2743 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2744 dir_ino, &index);
2745 if (ret) {
2746 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2747 "inode %llu parent %llu\n", name_len, name,
2748 (unsigned long long)ino, (unsigned long long)dir_ino);
2749 goto err;
2750 }
2751
2752 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2753 if (ret)
2754 goto err;
2755
2756 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2757 inode, dir_ino);
2758 BUG_ON(ret != 0 && ret != -ENOENT);
2759
2760 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2761 dir, index);
2762 if (ret == -ENOENT)
2763 ret = 0;
2764 err:
2765 btrfs_free_path(path);
2766 if (ret)
2767 goto out;
2768
2769 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2770 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2771 btrfs_update_inode(trans, root, dir);
2772 out:
2773 return ret;
2774 }
2775
2776 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2777 struct btrfs_root *root,
2778 struct inode *dir, struct inode *inode,
2779 const char *name, int name_len)
2780 {
2781 int ret;
2782 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2783 if (!ret) {
2784 btrfs_drop_nlink(inode);
2785 ret = btrfs_update_inode(trans, root, inode);
2786 }
2787 return ret;
2788 }
2789
2790
2791 /* helper to check if there is any shared block in the path */
2792 static int check_path_shared(struct btrfs_root *root,
2793 struct btrfs_path *path)
2794 {
2795 struct extent_buffer *eb;
2796 int level;
2797 u64 refs = 1;
2798
2799 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2800 int ret;
2801
2802 if (!path->nodes[level])
2803 break;
2804 eb = path->nodes[level];
2805 if (!btrfs_block_can_be_shared(root, eb))
2806 continue;
2807 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2808 &refs, NULL);
2809 if (refs > 1)
2810 return 1;
2811 }
2812 return 0;
2813 }
2814
2815 /*
2816 * helper to start transaction for unlink and rmdir.
2817 *
2818 * unlink and rmdir are special in btrfs, they do not always free space.
2819 * so in enospc case, we should make sure they will free space before
2820 * allowing them to use the global metadata reservation.
2821 */
2822 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2823 struct dentry *dentry)
2824 {
2825 struct btrfs_trans_handle *trans;
2826 struct btrfs_root *root = BTRFS_I(dir)->root;
2827 struct btrfs_path *path;
2828 struct btrfs_inode_ref *ref;
2829 struct btrfs_dir_item *di;
2830 struct inode *inode = dentry->d_inode;
2831 u64 index;
2832 int check_link = 1;
2833 int err = -ENOSPC;
2834 int ret;
2835 u64 ino = btrfs_ino(inode);
2836 u64 dir_ino = btrfs_ino(dir);
2837
2838 trans = btrfs_start_transaction(root, 10);
2839 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2840 return trans;
2841
2842 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2843 return ERR_PTR(-ENOSPC);
2844
2845 /* check if there is someone else holds reference */
2846 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2847 return ERR_PTR(-ENOSPC);
2848
2849 if (atomic_read(&inode->i_count) > 2)
2850 return ERR_PTR(-ENOSPC);
2851
2852 if (xchg(&root->fs_info->enospc_unlink, 1))
2853 return ERR_PTR(-ENOSPC);
2854
2855 path = btrfs_alloc_path();
2856 if (!path) {
2857 root->fs_info->enospc_unlink = 0;
2858 return ERR_PTR(-ENOMEM);
2859 }
2860
2861 trans = btrfs_start_transaction(root, 0);
2862 if (IS_ERR(trans)) {
2863 btrfs_free_path(path);
2864 root->fs_info->enospc_unlink = 0;
2865 return trans;
2866 }
2867
2868 path->skip_locking = 1;
2869 path->search_commit_root = 1;
2870
2871 ret = btrfs_lookup_inode(trans, root, path,
2872 &BTRFS_I(dir)->location, 0);
2873 if (ret < 0) {
2874 err = ret;
2875 goto out;
2876 }
2877 if (ret == 0) {
2878 if (check_path_shared(root, path))
2879 goto out;
2880 } else {
2881 check_link = 0;
2882 }
2883 btrfs_release_path(path);
2884
2885 ret = btrfs_lookup_inode(trans, root, path,
2886 &BTRFS_I(inode)->location, 0);
2887 if (ret < 0) {
2888 err = ret;
2889 goto out;
2890 }
2891 if (ret == 0) {
2892 if (check_path_shared(root, path))
2893 goto out;
2894 } else {
2895 check_link = 0;
2896 }
2897 btrfs_release_path(path);
2898
2899 if (ret == 0 && S_ISREG(inode->i_mode)) {
2900 ret = btrfs_lookup_file_extent(trans, root, path,
2901 ino, (u64)-1, 0);
2902 if (ret < 0) {
2903 err = ret;
2904 goto out;
2905 }
2906 BUG_ON(ret == 0);
2907 if (check_path_shared(root, path))
2908 goto out;
2909 btrfs_release_path(path);
2910 }
2911
2912 if (!check_link) {
2913 err = 0;
2914 goto out;
2915 }
2916
2917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2918 dentry->d_name.name, dentry->d_name.len, 0);
2919 if (IS_ERR(di)) {
2920 err = PTR_ERR(di);
2921 goto out;
2922 }
2923 if (di) {
2924 if (check_path_shared(root, path))
2925 goto out;
2926 } else {
2927 err = 0;
2928 goto out;
2929 }
2930 btrfs_release_path(path);
2931
2932 ref = btrfs_lookup_inode_ref(trans, root, path,
2933 dentry->d_name.name, dentry->d_name.len,
2934 ino, dir_ino, 0);
2935 if (IS_ERR(ref)) {
2936 err = PTR_ERR(ref);
2937 goto out;
2938 }
2939 BUG_ON(!ref);
2940 if (check_path_shared(root, path))
2941 goto out;
2942 index = btrfs_inode_ref_index(path->nodes[0], ref);
2943 btrfs_release_path(path);
2944
2945 /*
2946 * This is a commit root search, if we can lookup inode item and other
2947 * relative items in the commit root, it means the transaction of
2948 * dir/file creation has been committed, and the dir index item that we
2949 * delay to insert has also been inserted into the commit root. So
2950 * we needn't worry about the delayed insertion of the dir index item
2951 * here.
2952 */
2953 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2954 dentry->d_name.name, dentry->d_name.len, 0);
2955 if (IS_ERR(di)) {
2956 err = PTR_ERR(di);
2957 goto out;
2958 }
2959 BUG_ON(ret == -ENOENT);
2960 if (check_path_shared(root, path))
2961 goto out;
2962
2963 err = 0;
2964 out:
2965 btrfs_free_path(path);
2966 if (err) {
2967 btrfs_end_transaction(trans, root);
2968 root->fs_info->enospc_unlink = 0;
2969 return ERR_PTR(err);
2970 }
2971
2972 trans->block_rsv = &root->fs_info->global_block_rsv;
2973 return trans;
2974 }
2975
2976 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2977 struct btrfs_root *root)
2978 {
2979 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2980 BUG_ON(!root->fs_info->enospc_unlink);
2981 root->fs_info->enospc_unlink = 0;
2982 }
2983 btrfs_end_transaction_throttle(trans, root);
2984 }
2985
2986 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2987 {
2988 struct btrfs_root *root = BTRFS_I(dir)->root;
2989 struct btrfs_trans_handle *trans;
2990 struct inode *inode = dentry->d_inode;
2991 int ret;
2992 unsigned long nr = 0;
2993
2994 trans = __unlink_start_trans(dir, dentry);
2995 if (IS_ERR(trans))
2996 return PTR_ERR(trans);
2997
2998 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2999
3000 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3001 dentry->d_name.name, dentry->d_name.len);
3002 if (ret)
3003 goto out;
3004
3005 if (inode->i_nlink == 0) {
3006 ret = btrfs_orphan_add(trans, inode);
3007 if (ret)
3008 goto out;
3009 }
3010
3011 out:
3012 nr = trans->blocks_used;
3013 __unlink_end_trans(trans, root);
3014 btrfs_btree_balance_dirty(root, nr);
3015 return ret;
3016 }
3017
3018 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3019 struct btrfs_root *root,
3020 struct inode *dir, u64 objectid,
3021 const char *name, int name_len)
3022 {
3023 struct btrfs_path *path;
3024 struct extent_buffer *leaf;
3025 struct btrfs_dir_item *di;
3026 struct btrfs_key key;
3027 u64 index;
3028 int ret;
3029 u64 dir_ino = btrfs_ino(dir);
3030
3031 path = btrfs_alloc_path();
3032 if (!path)
3033 return -ENOMEM;
3034
3035 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3036 name, name_len, -1);
3037 BUG_ON(IS_ERR_OR_NULL(di));
3038
3039 leaf = path->nodes[0];
3040 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3041 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3042 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3043 BUG_ON(ret);
3044 btrfs_release_path(path);
3045
3046 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3047 objectid, root->root_key.objectid,
3048 dir_ino, &index, name, name_len);
3049 if (ret < 0) {
3050 BUG_ON(ret != -ENOENT);
3051 di = btrfs_search_dir_index_item(root, path, dir_ino,
3052 name, name_len);
3053 BUG_ON(IS_ERR_OR_NULL(di));
3054
3055 leaf = path->nodes[0];
3056 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3057 btrfs_release_path(path);
3058 index = key.offset;
3059 }
3060 btrfs_release_path(path);
3061
3062 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3063 BUG_ON(ret);
3064
3065 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3066 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3067 ret = btrfs_update_inode(trans, root, dir);
3068 BUG_ON(ret);
3069
3070 btrfs_free_path(path);
3071 return 0;
3072 }
3073
3074 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3075 {
3076 struct inode *inode = dentry->d_inode;
3077 int err = 0;
3078 struct btrfs_root *root = BTRFS_I(dir)->root;
3079 struct btrfs_trans_handle *trans;
3080 unsigned long nr = 0;
3081
3082 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3083 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3084 return -ENOTEMPTY;
3085
3086 trans = __unlink_start_trans(dir, dentry);
3087 if (IS_ERR(trans))
3088 return PTR_ERR(trans);
3089
3090 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3091 err = btrfs_unlink_subvol(trans, root, dir,
3092 BTRFS_I(inode)->location.objectid,
3093 dentry->d_name.name,
3094 dentry->d_name.len);
3095 goto out;
3096 }
3097
3098 err = btrfs_orphan_add(trans, inode);
3099 if (err)
3100 goto out;
3101
3102 /* now the directory is empty */
3103 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3104 dentry->d_name.name, dentry->d_name.len);
3105 if (!err)
3106 btrfs_i_size_write(inode, 0);
3107 out:
3108 nr = trans->blocks_used;
3109 __unlink_end_trans(trans, root);
3110 btrfs_btree_balance_dirty(root, nr);
3111
3112 return err;
3113 }
3114
3115 /*
3116 * this can truncate away extent items, csum items and directory items.
3117 * It starts at a high offset and removes keys until it can't find
3118 * any higher than new_size
3119 *
3120 * csum items that cross the new i_size are truncated to the new size
3121 * as well.
3122 *
3123 * min_type is the minimum key type to truncate down to. If set to 0, this
3124 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3125 */
3126 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3127 struct btrfs_root *root,
3128 struct inode *inode,
3129 u64 new_size, u32 min_type)
3130 {
3131 struct btrfs_path *path;
3132 struct extent_buffer *leaf;
3133 struct btrfs_file_extent_item *fi;
3134 struct btrfs_key key;
3135 struct btrfs_key found_key;
3136 u64 extent_start = 0;
3137 u64 extent_num_bytes = 0;
3138 u64 extent_offset = 0;
3139 u64 item_end = 0;
3140 u64 mask = root->sectorsize - 1;
3141 u32 found_type = (u8)-1;
3142 int found_extent;
3143 int del_item;
3144 int pending_del_nr = 0;
3145 int pending_del_slot = 0;
3146 int extent_type = -1;
3147 int encoding;
3148 int ret;
3149 int err = 0;
3150 u64 ino = btrfs_ino(inode);
3151
3152 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3153
3154 path = btrfs_alloc_path();
3155 if (!path)
3156 return -ENOMEM;
3157 path->reada = -1;
3158
3159 if (root->ref_cows || root == root->fs_info->tree_root)
3160 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3161
3162 /*
3163 * This function is also used to drop the items in the log tree before
3164 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3165 * it is used to drop the loged items. So we shouldn't kill the delayed
3166 * items.
3167 */
3168 if (min_type == 0 && root == BTRFS_I(inode)->root)
3169 btrfs_kill_delayed_inode_items(inode);
3170
3171 key.objectid = ino;
3172 key.offset = (u64)-1;
3173 key.type = (u8)-1;
3174
3175 search_again:
3176 path->leave_spinning = 1;
3177 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3178 if (ret < 0) {
3179 err = ret;
3180 goto out;
3181 }
3182
3183 if (ret > 0) {
3184 /* there are no items in the tree for us to truncate, we're
3185 * done
3186 */
3187 if (path->slots[0] == 0)
3188 goto out;
3189 path->slots[0]--;
3190 }
3191
3192 while (1) {
3193 fi = NULL;
3194 leaf = path->nodes[0];
3195 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3196 found_type = btrfs_key_type(&found_key);
3197 encoding = 0;
3198
3199 if (found_key.objectid != ino)
3200 break;
3201
3202 if (found_type < min_type)
3203 break;
3204
3205 item_end = found_key.offset;
3206 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3207 fi = btrfs_item_ptr(leaf, path->slots[0],
3208 struct btrfs_file_extent_item);
3209 extent_type = btrfs_file_extent_type(leaf, fi);
3210 encoding = btrfs_file_extent_compression(leaf, fi);
3211 encoding |= btrfs_file_extent_encryption(leaf, fi);
3212 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3213
3214 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3215 item_end +=
3216 btrfs_file_extent_num_bytes(leaf, fi);
3217 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3218 item_end += btrfs_file_extent_inline_len(leaf,
3219 fi);
3220 }
3221 item_end--;
3222 }
3223 if (found_type > min_type) {
3224 del_item = 1;
3225 } else {
3226 if (item_end < new_size)
3227 break;
3228 if (found_key.offset >= new_size)
3229 del_item = 1;
3230 else
3231 del_item = 0;
3232 }
3233 found_extent = 0;
3234 /* FIXME, shrink the extent if the ref count is only 1 */
3235 if (found_type != BTRFS_EXTENT_DATA_KEY)
3236 goto delete;
3237
3238 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3239 u64 num_dec;
3240 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3241 if (!del_item && !encoding) {
3242 u64 orig_num_bytes =
3243 btrfs_file_extent_num_bytes(leaf, fi);
3244 extent_num_bytes = new_size -
3245 found_key.offset + root->sectorsize - 1;
3246 extent_num_bytes = extent_num_bytes &
3247 ~((u64)root->sectorsize - 1);
3248 btrfs_set_file_extent_num_bytes(leaf, fi,
3249 extent_num_bytes);
3250 num_dec = (orig_num_bytes -
3251 extent_num_bytes);
3252 if (root->ref_cows && extent_start != 0)
3253 inode_sub_bytes(inode, num_dec);
3254 btrfs_mark_buffer_dirty(leaf);
3255 } else {
3256 extent_num_bytes =
3257 btrfs_file_extent_disk_num_bytes(leaf,
3258 fi);
3259 extent_offset = found_key.offset -
3260 btrfs_file_extent_offset(leaf, fi);
3261
3262 /* FIXME blocksize != 4096 */
3263 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3264 if (extent_start != 0) {
3265 found_extent = 1;
3266 if (root->ref_cows)
3267 inode_sub_bytes(inode, num_dec);
3268 }
3269 }
3270 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3271 /*
3272 * we can't truncate inline items that have had
3273 * special encodings
3274 */
3275 if (!del_item &&
3276 btrfs_file_extent_compression(leaf, fi) == 0 &&
3277 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3278 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3279 u32 size = new_size - found_key.offset;
3280
3281 if (root->ref_cows) {
3282 inode_sub_bytes(inode, item_end + 1 -
3283 new_size);
3284 }
3285 size =
3286 btrfs_file_extent_calc_inline_size(size);
3287 ret = btrfs_truncate_item(trans, root, path,
3288 size, 1);
3289 } else if (root->ref_cows) {
3290 inode_sub_bytes(inode, item_end + 1 -
3291 found_key.offset);
3292 }
3293 }
3294 delete:
3295 if (del_item) {
3296 if (!pending_del_nr) {
3297 /* no pending yet, add ourselves */
3298 pending_del_slot = path->slots[0];
3299 pending_del_nr = 1;
3300 } else if (pending_del_nr &&
3301 path->slots[0] + 1 == pending_del_slot) {
3302 /* hop on the pending chunk */
3303 pending_del_nr++;
3304 pending_del_slot = path->slots[0];
3305 } else {
3306 BUG();
3307 }
3308 } else {
3309 break;
3310 }
3311 if (found_extent && (root->ref_cows ||
3312 root == root->fs_info->tree_root)) {
3313 btrfs_set_path_blocking(path);
3314 ret = btrfs_free_extent(trans, root, extent_start,
3315 extent_num_bytes, 0,
3316 btrfs_header_owner(leaf),
3317 ino, extent_offset);
3318 BUG_ON(ret);
3319 }
3320
3321 if (found_type == BTRFS_INODE_ITEM_KEY)
3322 break;
3323
3324 if (path->slots[0] == 0 ||
3325 path->slots[0] != pending_del_slot) {
3326 if (root->ref_cows &&
3327 BTRFS_I(inode)->location.objectid !=
3328 BTRFS_FREE_INO_OBJECTID) {
3329 err = -EAGAIN;
3330 goto out;
3331 }
3332 if (pending_del_nr) {
3333 ret = btrfs_del_items(trans, root, path,
3334 pending_del_slot,
3335 pending_del_nr);
3336 BUG_ON(ret);
3337 pending_del_nr = 0;
3338 }
3339 btrfs_release_path(path);
3340 goto search_again;
3341 } else {
3342 path->slots[0]--;
3343 }
3344 }
3345 out:
3346 if (pending_del_nr) {
3347 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3348 pending_del_nr);
3349 BUG_ON(ret);
3350 }
3351 btrfs_free_path(path);
3352 return err;
3353 }
3354
3355 /*
3356 * taken from block_truncate_page, but does cow as it zeros out
3357 * any bytes left in the last page in the file.
3358 */
3359 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3360 {
3361 struct inode *inode = mapping->host;
3362 struct btrfs_root *root = BTRFS_I(inode)->root;
3363 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3364 struct btrfs_ordered_extent *ordered;
3365 struct extent_state *cached_state = NULL;
3366 char *kaddr;
3367 u32 blocksize = root->sectorsize;
3368 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3369 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3370 struct page *page;
3371 int ret = 0;
3372 u64 page_start;
3373 u64 page_end;
3374
3375 if ((offset & (blocksize - 1)) == 0)
3376 goto out;
3377 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3378 if (ret)
3379 goto out;
3380
3381 ret = -ENOMEM;
3382 again:
3383 page = find_or_create_page(mapping, index, GFP_NOFS);
3384 if (!page) {
3385 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3386 goto out;
3387 }
3388
3389 page_start = page_offset(page);
3390 page_end = page_start + PAGE_CACHE_SIZE - 1;
3391
3392 if (!PageUptodate(page)) {
3393 ret = btrfs_readpage(NULL, page);
3394 lock_page(page);
3395 if (page->mapping != mapping) {
3396 unlock_page(page);
3397 page_cache_release(page);
3398 goto again;
3399 }
3400 if (!PageUptodate(page)) {
3401 ret = -EIO;
3402 goto out_unlock;
3403 }
3404 }
3405 wait_on_page_writeback(page);
3406
3407 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3408 GFP_NOFS);
3409 set_page_extent_mapped(page);
3410
3411 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3412 if (ordered) {
3413 unlock_extent_cached(io_tree, page_start, page_end,
3414 &cached_state, GFP_NOFS);
3415 unlock_page(page);
3416 page_cache_release(page);
3417 btrfs_start_ordered_extent(inode, ordered, 1);
3418 btrfs_put_ordered_extent(ordered);
3419 goto again;
3420 }
3421
3422 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3423 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3424 0, 0, &cached_state, GFP_NOFS);
3425
3426 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3427 &cached_state);
3428 if (ret) {
3429 unlock_extent_cached(io_tree, page_start, page_end,
3430 &cached_state, GFP_NOFS);
3431 goto out_unlock;
3432 }
3433
3434 ret = 0;
3435 if (offset != PAGE_CACHE_SIZE) {
3436 kaddr = kmap(page);
3437 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3438 flush_dcache_page(page);
3439 kunmap(page);
3440 }
3441 ClearPageChecked(page);
3442 set_page_dirty(page);
3443 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3444 GFP_NOFS);
3445
3446 out_unlock:
3447 if (ret)
3448 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3449 unlock_page(page);
3450 page_cache_release(page);
3451 out:
3452 return ret;
3453 }
3454
3455 /*
3456 * This function puts in dummy file extents for the area we're creating a hole
3457 * for. So if we are truncating this file to a larger size we need to insert
3458 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3459 * the range between oldsize and size
3460 */
3461 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3462 {
3463 struct btrfs_trans_handle *trans;
3464 struct btrfs_root *root = BTRFS_I(inode)->root;
3465 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3466 struct extent_map *em = NULL;
3467 struct extent_state *cached_state = NULL;
3468 u64 mask = root->sectorsize - 1;
3469 u64 hole_start = (oldsize + mask) & ~mask;
3470 u64 block_end = (size + mask) & ~mask;
3471 u64 last_byte;
3472 u64 cur_offset;
3473 u64 hole_size;
3474 int err = 0;
3475
3476 if (size <= hole_start)
3477 return 0;
3478
3479 while (1) {
3480 struct btrfs_ordered_extent *ordered;
3481 btrfs_wait_ordered_range(inode, hole_start,
3482 block_end - hole_start);
3483 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3484 &cached_state, GFP_NOFS);
3485 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3486 if (!ordered)
3487 break;
3488 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3489 &cached_state, GFP_NOFS);
3490 btrfs_put_ordered_extent(ordered);
3491 }
3492
3493 cur_offset = hole_start;
3494 while (1) {
3495 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3496 block_end - cur_offset, 0);
3497 BUG_ON(IS_ERR_OR_NULL(em));
3498 last_byte = min(extent_map_end(em), block_end);
3499 last_byte = (last_byte + mask) & ~mask;
3500 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3501 u64 hint_byte = 0;
3502 hole_size = last_byte - cur_offset;
3503
3504 trans = btrfs_start_transaction(root, 2);
3505 if (IS_ERR(trans)) {
3506 err = PTR_ERR(trans);
3507 break;
3508 }
3509
3510 err = btrfs_drop_extents(trans, inode, cur_offset,
3511 cur_offset + hole_size,
3512 &hint_byte, 1);
3513 if (err) {
3514 btrfs_end_transaction(trans, root);
3515 break;
3516 }
3517
3518 err = btrfs_insert_file_extent(trans, root,
3519 btrfs_ino(inode), cur_offset, 0,
3520 0, hole_size, 0, hole_size,
3521 0, 0, 0);
3522 if (err) {
3523 btrfs_end_transaction(trans, root);
3524 break;
3525 }
3526
3527 btrfs_drop_extent_cache(inode, hole_start,
3528 last_byte - 1, 0);
3529
3530 btrfs_end_transaction(trans, root);
3531 }
3532 free_extent_map(em);
3533 em = NULL;
3534 cur_offset = last_byte;
3535 if (cur_offset >= block_end)
3536 break;
3537 }
3538
3539 free_extent_map(em);
3540 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3541 GFP_NOFS);
3542 return err;
3543 }
3544
3545 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3546 {
3547 loff_t oldsize = i_size_read(inode);
3548 int ret;
3549
3550 if (newsize == oldsize)
3551 return 0;
3552
3553 if (newsize > oldsize) {
3554 i_size_write(inode, newsize);
3555 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3556 truncate_pagecache(inode, oldsize, newsize);
3557 ret = btrfs_cont_expand(inode, oldsize, newsize);
3558 if (ret) {
3559 btrfs_setsize(inode, oldsize);
3560 return ret;
3561 }
3562
3563 mark_inode_dirty(inode);
3564 } else {
3565
3566 /*
3567 * We're truncating a file that used to have good data down to
3568 * zero. Make sure it gets into the ordered flush list so that
3569 * any new writes get down to disk quickly.
3570 */
3571 if (newsize == 0)
3572 BTRFS_I(inode)->ordered_data_close = 1;
3573
3574 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3575 truncate_setsize(inode, newsize);
3576 ret = btrfs_truncate(inode);
3577 }
3578
3579 return ret;
3580 }
3581
3582 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3583 {
3584 struct inode *inode = dentry->d_inode;
3585 struct btrfs_root *root = BTRFS_I(inode)->root;
3586 int err;
3587
3588 if (btrfs_root_readonly(root))
3589 return -EROFS;
3590
3591 err = inode_change_ok(inode, attr);
3592 if (err)
3593 return err;
3594
3595 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3596 err = btrfs_setsize(inode, attr->ia_size);
3597 if (err)
3598 return err;
3599 }
3600
3601 if (attr->ia_valid) {
3602 setattr_copy(inode, attr);
3603 mark_inode_dirty(inode);
3604
3605 if (attr->ia_valid & ATTR_MODE)
3606 err = btrfs_acl_chmod(inode);
3607 }
3608
3609 return err;
3610 }
3611
3612 void btrfs_evict_inode(struct inode *inode)
3613 {
3614 struct btrfs_trans_handle *trans;
3615 struct btrfs_root *root = BTRFS_I(inode)->root;
3616 unsigned long nr;
3617 int ret;
3618
3619 trace_btrfs_inode_evict(inode);
3620
3621 truncate_inode_pages(&inode->i_data, 0);
3622 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3623 btrfs_is_free_space_inode(root, inode)))
3624 goto no_delete;
3625
3626 if (is_bad_inode(inode)) {
3627 btrfs_orphan_del(NULL, inode);
3628 goto no_delete;
3629 }
3630 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3631 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3632
3633 if (root->fs_info->log_root_recovering) {
3634 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3635 goto no_delete;
3636 }
3637
3638 if (inode->i_nlink > 0) {
3639 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3640 goto no_delete;
3641 }
3642
3643 btrfs_i_size_write(inode, 0);
3644
3645 while (1) {
3646 trans = btrfs_join_transaction(root);
3647 BUG_ON(IS_ERR(trans));
3648 trans->block_rsv = root->orphan_block_rsv;
3649
3650 ret = btrfs_block_rsv_check(trans, root,
3651 root->orphan_block_rsv, 0, 5);
3652 if (ret) {
3653 BUG_ON(ret != -EAGAIN);
3654 ret = btrfs_commit_transaction(trans, root);
3655 BUG_ON(ret);
3656 continue;
3657 }
3658
3659 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3660 if (ret != -EAGAIN)
3661 break;
3662
3663 nr = trans->blocks_used;
3664 btrfs_end_transaction(trans, root);
3665 trans = NULL;
3666 btrfs_btree_balance_dirty(root, nr);
3667
3668 }
3669
3670 if (ret == 0) {
3671 ret = btrfs_orphan_del(trans, inode);
3672 BUG_ON(ret);
3673 }
3674
3675 if (!(root == root->fs_info->tree_root ||
3676 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3677 btrfs_return_ino(root, btrfs_ino(inode));
3678
3679 nr = trans->blocks_used;
3680 btrfs_end_transaction(trans, root);
3681 btrfs_btree_balance_dirty(root, nr);
3682 no_delete:
3683 end_writeback(inode);
3684 return;
3685 }
3686
3687 /*
3688 * this returns the key found in the dir entry in the location pointer.
3689 * If no dir entries were found, location->objectid is 0.
3690 */
3691 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3692 struct btrfs_key *location)
3693 {
3694 const char *name = dentry->d_name.name;
3695 int namelen = dentry->d_name.len;
3696 struct btrfs_dir_item *di;
3697 struct btrfs_path *path;
3698 struct btrfs_root *root = BTRFS_I(dir)->root;
3699 int ret = 0;
3700
3701 path = btrfs_alloc_path();
3702 if (!path)
3703 return -ENOMEM;
3704
3705 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3706 namelen, 0);
3707 if (IS_ERR(di))
3708 ret = PTR_ERR(di);
3709
3710 if (IS_ERR_OR_NULL(di))
3711 goto out_err;
3712
3713 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3714 out:
3715 btrfs_free_path(path);
3716 return ret;
3717 out_err:
3718 location->objectid = 0;
3719 goto out;
3720 }
3721
3722 /*
3723 * when we hit a tree root in a directory, the btrfs part of the inode
3724 * needs to be changed to reflect the root directory of the tree root. This
3725 * is kind of like crossing a mount point.
3726 */
3727 static int fixup_tree_root_location(struct btrfs_root *root,
3728 struct inode *dir,
3729 struct dentry *dentry,
3730 struct btrfs_key *location,
3731 struct btrfs_root **sub_root)
3732 {
3733 struct btrfs_path *path;
3734 struct btrfs_root *new_root;
3735 struct btrfs_root_ref *ref;
3736 struct extent_buffer *leaf;
3737 int ret;
3738 int err = 0;
3739
3740 path = btrfs_alloc_path();
3741 if (!path) {
3742 err = -ENOMEM;
3743 goto out;
3744 }
3745
3746 err = -ENOENT;
3747 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3748 BTRFS_I(dir)->root->root_key.objectid,
3749 location->objectid);
3750 if (ret) {
3751 if (ret < 0)
3752 err = ret;
3753 goto out;
3754 }
3755
3756 leaf = path->nodes[0];
3757 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3758 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3759 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3760 goto out;
3761
3762 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3763 (unsigned long)(ref + 1),
3764 dentry->d_name.len);
3765 if (ret)
3766 goto out;
3767
3768 btrfs_release_path(path);
3769
3770 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3771 if (IS_ERR(new_root)) {
3772 err = PTR_ERR(new_root);
3773 goto out;
3774 }
3775
3776 if (btrfs_root_refs(&new_root->root_item) == 0) {
3777 err = -ENOENT;
3778 goto out;
3779 }
3780
3781 *sub_root = new_root;
3782 location->objectid = btrfs_root_dirid(&new_root->root_item);
3783 location->type = BTRFS_INODE_ITEM_KEY;
3784 location->offset = 0;
3785 err = 0;
3786 out:
3787 btrfs_free_path(path);
3788 return err;
3789 }
3790
3791 static void inode_tree_add(struct inode *inode)
3792 {
3793 struct btrfs_root *root = BTRFS_I(inode)->root;
3794 struct btrfs_inode *entry;
3795 struct rb_node **p;
3796 struct rb_node *parent;
3797 u64 ino = btrfs_ino(inode);
3798 again:
3799 p = &root->inode_tree.rb_node;
3800 parent = NULL;
3801
3802 if (inode_unhashed(inode))
3803 return;
3804
3805 spin_lock(&root->inode_lock);
3806 while (*p) {
3807 parent = *p;
3808 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3809
3810 if (ino < btrfs_ino(&entry->vfs_inode))
3811 p = &parent->rb_left;
3812 else if (ino > btrfs_ino(&entry->vfs_inode))
3813 p = &parent->rb_right;
3814 else {
3815 WARN_ON(!(entry->vfs_inode.i_state &
3816 (I_WILL_FREE | I_FREEING)));
3817 rb_erase(parent, &root->inode_tree);
3818 RB_CLEAR_NODE(parent);
3819 spin_unlock(&root->inode_lock);
3820 goto again;
3821 }
3822 }
3823 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3824 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3825 spin_unlock(&root->inode_lock);
3826 }
3827
3828 static void inode_tree_del(struct inode *inode)
3829 {
3830 struct btrfs_root *root = BTRFS_I(inode)->root;
3831 int empty = 0;
3832
3833 spin_lock(&root->inode_lock);
3834 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3835 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3836 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3837 empty = RB_EMPTY_ROOT(&root->inode_tree);
3838 }
3839 spin_unlock(&root->inode_lock);
3840
3841 /*
3842 * Free space cache has inodes in the tree root, but the tree root has a
3843 * root_refs of 0, so this could end up dropping the tree root as a
3844 * snapshot, so we need the extra !root->fs_info->tree_root check to
3845 * make sure we don't drop it.
3846 */
3847 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3848 root != root->fs_info->tree_root) {
3849 synchronize_srcu(&root->fs_info->subvol_srcu);
3850 spin_lock(&root->inode_lock);
3851 empty = RB_EMPTY_ROOT(&root->inode_tree);
3852 spin_unlock(&root->inode_lock);
3853 if (empty)
3854 btrfs_add_dead_root(root);
3855 }
3856 }
3857
3858 int btrfs_invalidate_inodes(struct btrfs_root *root)
3859 {
3860 struct rb_node *node;
3861 struct rb_node *prev;
3862 struct btrfs_inode *entry;
3863 struct inode *inode;
3864 u64 objectid = 0;
3865
3866 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3867
3868 spin_lock(&root->inode_lock);
3869 again:
3870 node = root->inode_tree.rb_node;
3871 prev = NULL;
3872 while (node) {
3873 prev = node;
3874 entry = rb_entry(node, struct btrfs_inode, rb_node);
3875
3876 if (objectid < btrfs_ino(&entry->vfs_inode))
3877 node = node->rb_left;
3878 else if (objectid > btrfs_ino(&entry->vfs_inode))
3879 node = node->rb_right;
3880 else
3881 break;
3882 }
3883 if (!node) {
3884 while (prev) {
3885 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3886 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3887 node = prev;
3888 break;
3889 }
3890 prev = rb_next(prev);
3891 }
3892 }
3893 while (node) {
3894 entry = rb_entry(node, struct btrfs_inode, rb_node);
3895 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3896 inode = igrab(&entry->vfs_inode);
3897 if (inode) {
3898 spin_unlock(&root->inode_lock);
3899 if (atomic_read(&inode->i_count) > 1)
3900 d_prune_aliases(inode);
3901 /*
3902 * btrfs_drop_inode will have it removed from
3903 * the inode cache when its usage count
3904 * hits zero.
3905 */
3906 iput(inode);
3907 cond_resched();
3908 spin_lock(&root->inode_lock);
3909 goto again;
3910 }
3911
3912 if (cond_resched_lock(&root->inode_lock))
3913 goto again;
3914
3915 node = rb_next(node);
3916 }
3917 spin_unlock(&root->inode_lock);
3918 return 0;
3919 }
3920
3921 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3922 {
3923 struct btrfs_iget_args *args = p;
3924 inode->i_ino = args->ino;
3925 BTRFS_I(inode)->root = args->root;
3926 btrfs_set_inode_space_info(args->root, inode);
3927 return 0;
3928 }
3929
3930 static int btrfs_find_actor(struct inode *inode, void *opaque)
3931 {
3932 struct btrfs_iget_args *args = opaque;
3933 return args->ino == btrfs_ino(inode) &&
3934 args->root == BTRFS_I(inode)->root;
3935 }
3936
3937 static struct inode *btrfs_iget_locked(struct super_block *s,
3938 u64 objectid,
3939 struct btrfs_root *root)
3940 {
3941 struct inode *inode;
3942 struct btrfs_iget_args args;
3943 args.ino = objectid;
3944 args.root = root;
3945
3946 inode = iget5_locked(s, objectid, btrfs_find_actor,
3947 btrfs_init_locked_inode,
3948 (void *)&args);
3949 return inode;
3950 }
3951
3952 /* Get an inode object given its location and corresponding root.
3953 * Returns in *is_new if the inode was read from disk
3954 */
3955 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3956 struct btrfs_root *root, int *new)
3957 {
3958 struct inode *inode;
3959
3960 inode = btrfs_iget_locked(s, location->objectid, root);
3961 if (!inode)
3962 return ERR_PTR(-ENOMEM);
3963
3964 if (inode->i_state & I_NEW) {
3965 BTRFS_I(inode)->root = root;
3966 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3967 btrfs_read_locked_inode(inode);
3968 if (!is_bad_inode(inode)) {
3969 inode_tree_add(inode);
3970 unlock_new_inode(inode);
3971 if (new)
3972 *new = 1;
3973 } else {
3974 unlock_new_inode(inode);
3975 iput(inode);
3976 inode = ERR_PTR(-ESTALE);
3977 }
3978 }
3979
3980 return inode;
3981 }
3982
3983 static struct inode *new_simple_dir(struct super_block *s,
3984 struct btrfs_key *key,
3985 struct btrfs_root *root)
3986 {
3987 struct inode *inode = new_inode(s);
3988
3989 if (!inode)
3990 return ERR_PTR(-ENOMEM);
3991
3992 BTRFS_I(inode)->root = root;
3993 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3994 BTRFS_I(inode)->dummy_inode = 1;
3995
3996 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3997 inode->i_op = &simple_dir_inode_operations;
3998 inode->i_fop = &simple_dir_operations;
3999 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4000 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4001
4002 return inode;
4003 }
4004
4005 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4006 {
4007 struct inode *inode;
4008 struct btrfs_root *root = BTRFS_I(dir)->root;
4009 struct btrfs_root *sub_root = root;
4010 struct btrfs_key location;
4011 int index;
4012 int ret = 0;
4013
4014 if (dentry->d_name.len > BTRFS_NAME_LEN)
4015 return ERR_PTR(-ENAMETOOLONG);
4016
4017 if (unlikely(d_need_lookup(dentry))) {
4018 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4019 kfree(dentry->d_fsdata);
4020 dentry->d_fsdata = NULL;
4021 /* This thing is hashed, drop it for now */
4022 d_drop(dentry);
4023 } else {
4024 ret = btrfs_inode_by_name(dir, dentry, &location);
4025 }
4026
4027 if (ret < 0)
4028 return ERR_PTR(ret);
4029
4030 if (location.objectid == 0)
4031 return NULL;
4032
4033 if (location.type == BTRFS_INODE_ITEM_KEY) {
4034 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4035 return inode;
4036 }
4037
4038 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4039
4040 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4041 ret = fixup_tree_root_location(root, dir, dentry,
4042 &location, &sub_root);
4043 if (ret < 0) {
4044 if (ret != -ENOENT)
4045 inode = ERR_PTR(ret);
4046 else
4047 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4048 } else {
4049 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4050 }
4051 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4052
4053 if (!IS_ERR(inode) && root != sub_root) {
4054 down_read(&root->fs_info->cleanup_work_sem);
4055 if (!(inode->i_sb->s_flags & MS_RDONLY))
4056 ret = btrfs_orphan_cleanup(sub_root);
4057 up_read(&root->fs_info->cleanup_work_sem);
4058 if (ret)
4059 inode = ERR_PTR(ret);
4060 }
4061
4062 return inode;
4063 }
4064
4065 static int btrfs_dentry_delete(const struct dentry *dentry)
4066 {
4067 struct btrfs_root *root;
4068
4069 if (!dentry->d_inode && !IS_ROOT(dentry))
4070 dentry = dentry->d_parent;
4071
4072 if (dentry->d_inode) {
4073 root = BTRFS_I(dentry->d_inode)->root;
4074 if (btrfs_root_refs(&root->root_item) == 0)
4075 return 1;
4076 }
4077 return 0;
4078 }
4079
4080 static void btrfs_dentry_release(struct dentry *dentry)
4081 {
4082 if (dentry->d_fsdata)
4083 kfree(dentry->d_fsdata);
4084 }
4085
4086 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4087 struct nameidata *nd)
4088 {
4089 struct dentry *ret;
4090
4091 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4092 if (unlikely(d_need_lookup(dentry))) {
4093 spin_lock(&dentry->d_lock);
4094 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4095 spin_unlock(&dentry->d_lock);
4096 }
4097 return ret;
4098 }
4099
4100 unsigned char btrfs_filetype_table[] = {
4101 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4102 };
4103
4104 static int btrfs_real_readdir(struct file *filp, void *dirent,
4105 filldir_t filldir)
4106 {
4107 struct inode *inode = filp->f_dentry->d_inode;
4108 struct btrfs_root *root = BTRFS_I(inode)->root;
4109 struct btrfs_item *item;
4110 struct btrfs_dir_item *di;
4111 struct btrfs_key key;
4112 struct btrfs_key found_key;
4113 struct btrfs_path *path;
4114 struct list_head ins_list;
4115 struct list_head del_list;
4116 struct qstr q;
4117 int ret;
4118 struct extent_buffer *leaf;
4119 int slot;
4120 unsigned char d_type;
4121 int over = 0;
4122 u32 di_cur;
4123 u32 di_total;
4124 u32 di_len;
4125 int key_type = BTRFS_DIR_INDEX_KEY;
4126 char tmp_name[32];
4127 char *name_ptr;
4128 int name_len;
4129 int is_curr = 0; /* filp->f_pos points to the current index? */
4130
4131 /* FIXME, use a real flag for deciding about the key type */
4132 if (root->fs_info->tree_root == root)
4133 key_type = BTRFS_DIR_ITEM_KEY;
4134
4135 /* special case for "." */
4136 if (filp->f_pos == 0) {
4137 over = filldir(dirent, ".", 1,
4138 filp->f_pos, btrfs_ino(inode), DT_DIR);
4139 if (over)
4140 return 0;
4141 filp->f_pos = 1;
4142 }
4143 /* special case for .., just use the back ref */
4144 if (filp->f_pos == 1) {
4145 u64 pino = parent_ino(filp->f_path.dentry);
4146 over = filldir(dirent, "..", 2,
4147 filp->f_pos, pino, DT_DIR);
4148 if (over)
4149 return 0;
4150 filp->f_pos = 2;
4151 }
4152 path = btrfs_alloc_path();
4153 if (!path)
4154 return -ENOMEM;
4155
4156 path->reada = 1;
4157
4158 if (key_type == BTRFS_DIR_INDEX_KEY) {
4159 INIT_LIST_HEAD(&ins_list);
4160 INIT_LIST_HEAD(&del_list);
4161 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4162 }
4163
4164 btrfs_set_key_type(&key, key_type);
4165 key.offset = filp->f_pos;
4166 key.objectid = btrfs_ino(inode);
4167
4168 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4169 if (ret < 0)
4170 goto err;
4171
4172 while (1) {
4173 leaf = path->nodes[0];
4174 slot = path->slots[0];
4175 if (slot >= btrfs_header_nritems(leaf)) {
4176 ret = btrfs_next_leaf(root, path);
4177 if (ret < 0)
4178 goto err;
4179 else if (ret > 0)
4180 break;
4181 continue;
4182 }
4183
4184 item = btrfs_item_nr(leaf, slot);
4185 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4186
4187 if (found_key.objectid != key.objectid)
4188 break;
4189 if (btrfs_key_type(&found_key) != key_type)
4190 break;
4191 if (found_key.offset < filp->f_pos)
4192 goto next;
4193 if (key_type == BTRFS_DIR_INDEX_KEY &&
4194 btrfs_should_delete_dir_index(&del_list,
4195 found_key.offset))
4196 goto next;
4197
4198 filp->f_pos = found_key.offset;
4199 is_curr = 1;
4200
4201 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4202 di_cur = 0;
4203 di_total = btrfs_item_size(leaf, item);
4204
4205 while (di_cur < di_total) {
4206 struct btrfs_key location;
4207 struct dentry *tmp;
4208
4209 if (verify_dir_item(root, leaf, di))
4210 break;
4211
4212 name_len = btrfs_dir_name_len(leaf, di);
4213 if (name_len <= sizeof(tmp_name)) {
4214 name_ptr = tmp_name;
4215 } else {
4216 name_ptr = kmalloc(name_len, GFP_NOFS);
4217 if (!name_ptr) {
4218 ret = -ENOMEM;
4219 goto err;
4220 }
4221 }
4222 read_extent_buffer(leaf, name_ptr,
4223 (unsigned long)(di + 1), name_len);
4224
4225 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4226 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4227
4228 q.name = name_ptr;
4229 q.len = name_len;
4230 q.hash = full_name_hash(q.name, q.len);
4231 tmp = d_lookup(filp->f_dentry, &q);
4232 if (!tmp) {
4233 struct btrfs_key *newkey;
4234
4235 newkey = kzalloc(sizeof(struct btrfs_key),
4236 GFP_NOFS);
4237 if (!newkey)
4238 goto no_dentry;
4239 tmp = d_alloc(filp->f_dentry, &q);
4240 if (!tmp) {
4241 kfree(newkey);
4242 dput(tmp);
4243 goto no_dentry;
4244 }
4245 memcpy(newkey, &location,
4246 sizeof(struct btrfs_key));
4247 tmp->d_fsdata = newkey;
4248 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4249 d_rehash(tmp);
4250 dput(tmp);
4251 } else {
4252 dput(tmp);
4253 }
4254 no_dentry:
4255 /* is this a reference to our own snapshot? If so
4256 * skip it
4257 */
4258 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4259 location.objectid == root->root_key.objectid) {
4260 over = 0;
4261 goto skip;
4262 }
4263 over = filldir(dirent, name_ptr, name_len,
4264 found_key.offset, location.objectid,
4265 d_type);
4266
4267 skip:
4268 if (name_ptr != tmp_name)
4269 kfree(name_ptr);
4270
4271 if (over)
4272 goto nopos;
4273 di_len = btrfs_dir_name_len(leaf, di) +
4274 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4275 di_cur += di_len;
4276 di = (struct btrfs_dir_item *)((char *)di + di_len);
4277 }
4278 next:
4279 path->slots[0]++;
4280 }
4281
4282 if (key_type == BTRFS_DIR_INDEX_KEY) {
4283 if (is_curr)
4284 filp->f_pos++;
4285 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4286 &ins_list);
4287 if (ret)
4288 goto nopos;
4289 }
4290
4291 /* Reached end of directory/root. Bump pos past the last item. */
4292 if (key_type == BTRFS_DIR_INDEX_KEY)
4293 /*
4294 * 32-bit glibc will use getdents64, but then strtol -
4295 * so the last number we can serve is this.
4296 */
4297 filp->f_pos = 0x7fffffff;
4298 else
4299 filp->f_pos++;
4300 nopos:
4301 ret = 0;
4302 err:
4303 if (key_type == BTRFS_DIR_INDEX_KEY)
4304 btrfs_put_delayed_items(&ins_list, &del_list);
4305 btrfs_free_path(path);
4306 return ret;
4307 }
4308
4309 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4310 {
4311 struct btrfs_root *root = BTRFS_I(inode)->root;
4312 struct btrfs_trans_handle *trans;
4313 int ret = 0;
4314 bool nolock = false;
4315
4316 if (BTRFS_I(inode)->dummy_inode)
4317 return 0;
4318
4319 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4320 nolock = true;
4321
4322 if (wbc->sync_mode == WB_SYNC_ALL) {
4323 if (nolock)
4324 trans = btrfs_join_transaction_nolock(root);
4325 else
4326 trans = btrfs_join_transaction(root);
4327 if (IS_ERR(trans))
4328 return PTR_ERR(trans);
4329 if (nolock)
4330 ret = btrfs_end_transaction_nolock(trans, root);
4331 else
4332 ret = btrfs_commit_transaction(trans, root);
4333 }
4334 return ret;
4335 }
4336
4337 /*
4338 * This is somewhat expensive, updating the tree every time the
4339 * inode changes. But, it is most likely to find the inode in cache.
4340 * FIXME, needs more benchmarking...there are no reasons other than performance
4341 * to keep or drop this code.
4342 */
4343 void btrfs_dirty_inode(struct inode *inode, int flags)
4344 {
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4346 struct btrfs_trans_handle *trans;
4347 int ret;
4348
4349 if (BTRFS_I(inode)->dummy_inode)
4350 return;
4351
4352 trans = btrfs_join_transaction(root);
4353 BUG_ON(IS_ERR(trans));
4354
4355 ret = btrfs_update_inode(trans, root, inode);
4356 if (ret && ret == -ENOSPC) {
4357 /* whoops, lets try again with the full transaction */
4358 btrfs_end_transaction(trans, root);
4359 trans = btrfs_start_transaction(root, 1);
4360 if (IS_ERR(trans)) {
4361 printk_ratelimited(KERN_ERR "btrfs: fail to "
4362 "dirty inode %llu error %ld\n",
4363 (unsigned long long)btrfs_ino(inode),
4364 PTR_ERR(trans));
4365 return;
4366 }
4367
4368 ret = btrfs_update_inode(trans, root, inode);
4369 if (ret) {
4370 printk_ratelimited(KERN_ERR "btrfs: fail to "
4371 "dirty inode %llu error %d\n",
4372 (unsigned long long)btrfs_ino(inode),
4373 ret);
4374 }
4375 }
4376 btrfs_end_transaction(trans, root);
4377 if (BTRFS_I(inode)->delayed_node)
4378 btrfs_balance_delayed_items(root);
4379 }
4380
4381 /*
4382 * find the highest existing sequence number in a directory
4383 * and then set the in-memory index_cnt variable to reflect
4384 * free sequence numbers
4385 */
4386 static int btrfs_set_inode_index_count(struct inode *inode)
4387 {
4388 struct btrfs_root *root = BTRFS_I(inode)->root;
4389 struct btrfs_key key, found_key;
4390 struct btrfs_path *path;
4391 struct extent_buffer *leaf;
4392 int ret;
4393
4394 key.objectid = btrfs_ino(inode);
4395 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4396 key.offset = (u64)-1;
4397
4398 path = btrfs_alloc_path();
4399 if (!path)
4400 return -ENOMEM;
4401
4402 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4403 if (ret < 0)
4404 goto out;
4405 /* FIXME: we should be able to handle this */
4406 if (ret == 0)
4407 goto out;
4408 ret = 0;
4409
4410 /*
4411 * MAGIC NUMBER EXPLANATION:
4412 * since we search a directory based on f_pos we have to start at 2
4413 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4414 * else has to start at 2
4415 */
4416 if (path->slots[0] == 0) {
4417 BTRFS_I(inode)->index_cnt = 2;
4418 goto out;
4419 }
4420
4421 path->slots[0]--;
4422
4423 leaf = path->nodes[0];
4424 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4425
4426 if (found_key.objectid != btrfs_ino(inode) ||
4427 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4428 BTRFS_I(inode)->index_cnt = 2;
4429 goto out;
4430 }
4431
4432 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4433 out:
4434 btrfs_free_path(path);
4435 return ret;
4436 }
4437
4438 /*
4439 * helper to find a free sequence number in a given directory. This current
4440 * code is very simple, later versions will do smarter things in the btree
4441 */
4442 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4443 {
4444 int ret = 0;
4445
4446 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4447 ret = btrfs_inode_delayed_dir_index_count(dir);
4448 if (ret) {
4449 ret = btrfs_set_inode_index_count(dir);
4450 if (ret)
4451 return ret;
4452 }
4453 }
4454
4455 *index = BTRFS_I(dir)->index_cnt;
4456 BTRFS_I(dir)->index_cnt++;
4457
4458 return ret;
4459 }
4460
4461 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4462 struct btrfs_root *root,
4463 struct inode *dir,
4464 const char *name, int name_len,
4465 u64 ref_objectid, u64 objectid, int mode,
4466 u64 *index)
4467 {
4468 struct inode *inode;
4469 struct btrfs_inode_item *inode_item;
4470 struct btrfs_key *location;
4471 struct btrfs_path *path;
4472 struct btrfs_inode_ref *ref;
4473 struct btrfs_key key[2];
4474 u32 sizes[2];
4475 unsigned long ptr;
4476 int ret;
4477 int owner;
4478
4479 path = btrfs_alloc_path();
4480 if (!path)
4481 return ERR_PTR(-ENOMEM);
4482
4483 inode = new_inode(root->fs_info->sb);
4484 if (!inode) {
4485 btrfs_free_path(path);
4486 return ERR_PTR(-ENOMEM);
4487 }
4488
4489 /*
4490 * we have to initialize this early, so we can reclaim the inode
4491 * number if we fail afterwards in this function.
4492 */
4493 inode->i_ino = objectid;
4494
4495 if (dir) {
4496 trace_btrfs_inode_request(dir);
4497
4498 ret = btrfs_set_inode_index(dir, index);
4499 if (ret) {
4500 btrfs_free_path(path);
4501 iput(inode);
4502 return ERR_PTR(ret);
4503 }
4504 }
4505 /*
4506 * index_cnt is ignored for everything but a dir,
4507 * btrfs_get_inode_index_count has an explanation for the magic
4508 * number
4509 */
4510 BTRFS_I(inode)->index_cnt = 2;
4511 BTRFS_I(inode)->root = root;
4512 BTRFS_I(inode)->generation = trans->transid;
4513 inode->i_generation = BTRFS_I(inode)->generation;
4514 btrfs_set_inode_space_info(root, inode);
4515
4516 if (S_ISDIR(mode))
4517 owner = 0;
4518 else
4519 owner = 1;
4520
4521 key[0].objectid = objectid;
4522 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4523 key[0].offset = 0;
4524
4525 key[1].objectid = objectid;
4526 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4527 key[1].offset = ref_objectid;
4528
4529 sizes[0] = sizeof(struct btrfs_inode_item);
4530 sizes[1] = name_len + sizeof(*ref);
4531
4532 path->leave_spinning = 1;
4533 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4534 if (ret != 0)
4535 goto fail;
4536
4537 inode_init_owner(inode, dir, mode);
4538 inode_set_bytes(inode, 0);
4539 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4540 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4541 struct btrfs_inode_item);
4542 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4543
4544 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4545 struct btrfs_inode_ref);
4546 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4547 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4548 ptr = (unsigned long)(ref + 1);
4549 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4550
4551 btrfs_mark_buffer_dirty(path->nodes[0]);
4552 btrfs_free_path(path);
4553
4554 location = &BTRFS_I(inode)->location;
4555 location->objectid = objectid;
4556 location->offset = 0;
4557 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4558
4559 btrfs_inherit_iflags(inode, dir);
4560
4561 if (S_ISREG(mode)) {
4562 if (btrfs_test_opt(root, NODATASUM))
4563 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4564 if (btrfs_test_opt(root, NODATACOW) ||
4565 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4566 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4567 }
4568
4569 insert_inode_hash(inode);
4570 inode_tree_add(inode);
4571
4572 trace_btrfs_inode_new(inode);
4573 btrfs_set_inode_last_trans(trans, inode);
4574
4575 return inode;
4576 fail:
4577 if (dir)
4578 BTRFS_I(dir)->index_cnt--;
4579 btrfs_free_path(path);
4580 iput(inode);
4581 return ERR_PTR(ret);
4582 }
4583
4584 static inline u8 btrfs_inode_type(struct inode *inode)
4585 {
4586 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4587 }
4588
4589 /*
4590 * utility function to add 'inode' into 'parent_inode' with
4591 * a give name and a given sequence number.
4592 * if 'add_backref' is true, also insert a backref from the
4593 * inode to the parent directory.
4594 */
4595 int btrfs_add_link(struct btrfs_trans_handle *trans,
4596 struct inode *parent_inode, struct inode *inode,
4597 const char *name, int name_len, int add_backref, u64 index)
4598 {
4599 int ret = 0;
4600 struct btrfs_key key;
4601 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4602 u64 ino = btrfs_ino(inode);
4603 u64 parent_ino = btrfs_ino(parent_inode);
4604
4605 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4606 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4607 } else {
4608 key.objectid = ino;
4609 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4610 key.offset = 0;
4611 }
4612
4613 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4614 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4615 key.objectid, root->root_key.objectid,
4616 parent_ino, index, name, name_len);
4617 } else if (add_backref) {
4618 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4619 parent_ino, index);
4620 }
4621
4622 if (ret == 0) {
4623 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4624 parent_inode, &key,
4625 btrfs_inode_type(inode), index);
4626 BUG_ON(ret);
4627
4628 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4629 name_len * 2);
4630 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4631 ret = btrfs_update_inode(trans, root, parent_inode);
4632 }
4633 return ret;
4634 }
4635
4636 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4637 struct inode *dir, struct dentry *dentry,
4638 struct inode *inode, int backref, u64 index)
4639 {
4640 int err = btrfs_add_link(trans, dir, inode,
4641 dentry->d_name.name, dentry->d_name.len,
4642 backref, index);
4643 if (!err) {
4644 d_instantiate(dentry, inode);
4645 return 0;
4646 }
4647 if (err > 0)
4648 err = -EEXIST;
4649 return err;
4650 }
4651
4652 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4653 int mode, dev_t rdev)
4654 {
4655 struct btrfs_trans_handle *trans;
4656 struct btrfs_root *root = BTRFS_I(dir)->root;
4657 struct inode *inode = NULL;
4658 int err;
4659 int drop_inode = 0;
4660 u64 objectid;
4661 unsigned long nr = 0;
4662 u64 index = 0;
4663
4664 if (!new_valid_dev(rdev))
4665 return -EINVAL;
4666
4667 /*
4668 * 2 for inode item and ref
4669 * 2 for dir items
4670 * 1 for xattr if selinux is on
4671 */
4672 trans = btrfs_start_transaction(root, 5);
4673 if (IS_ERR(trans))
4674 return PTR_ERR(trans);
4675
4676 err = btrfs_find_free_ino(root, &objectid);
4677 if (err)
4678 goto out_unlock;
4679
4680 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4681 dentry->d_name.len, btrfs_ino(dir), objectid,
4682 mode, &index);
4683 if (IS_ERR(inode)) {
4684 err = PTR_ERR(inode);
4685 goto out_unlock;
4686 }
4687
4688 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4689 if (err) {
4690 drop_inode = 1;
4691 goto out_unlock;
4692 }
4693
4694 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4695 if (err)
4696 drop_inode = 1;
4697 else {
4698 inode->i_op = &btrfs_special_inode_operations;
4699 init_special_inode(inode, inode->i_mode, rdev);
4700 btrfs_update_inode(trans, root, inode);
4701 }
4702 out_unlock:
4703 nr = trans->blocks_used;
4704 btrfs_end_transaction_throttle(trans, root);
4705 btrfs_btree_balance_dirty(root, nr);
4706 if (drop_inode) {
4707 inode_dec_link_count(inode);
4708 iput(inode);
4709 }
4710 return err;
4711 }
4712
4713 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4714 int mode, struct nameidata *nd)
4715 {
4716 struct btrfs_trans_handle *trans;
4717 struct btrfs_root *root = BTRFS_I(dir)->root;
4718 struct inode *inode = NULL;
4719 int drop_inode = 0;
4720 int err;
4721 unsigned long nr = 0;
4722 u64 objectid;
4723 u64 index = 0;
4724
4725 /*
4726 * 2 for inode item and ref
4727 * 2 for dir items
4728 * 1 for xattr if selinux is on
4729 */
4730 trans = btrfs_start_transaction(root, 5);
4731 if (IS_ERR(trans))
4732 return PTR_ERR(trans);
4733
4734 err = btrfs_find_free_ino(root, &objectid);
4735 if (err)
4736 goto out_unlock;
4737
4738 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4739 dentry->d_name.len, btrfs_ino(dir), objectid,
4740 mode, &index);
4741 if (IS_ERR(inode)) {
4742 err = PTR_ERR(inode);
4743 goto out_unlock;
4744 }
4745
4746 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4747 if (err) {
4748 drop_inode = 1;
4749 goto out_unlock;
4750 }
4751
4752 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4753 if (err)
4754 drop_inode = 1;
4755 else {
4756 inode->i_mapping->a_ops = &btrfs_aops;
4757 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4758 inode->i_fop = &btrfs_file_operations;
4759 inode->i_op = &btrfs_file_inode_operations;
4760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4761 }
4762 out_unlock:
4763 nr = trans->blocks_used;
4764 btrfs_end_transaction_throttle(trans, root);
4765 if (drop_inode) {
4766 inode_dec_link_count(inode);
4767 iput(inode);
4768 }
4769 btrfs_btree_balance_dirty(root, nr);
4770 return err;
4771 }
4772
4773 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4774 struct dentry *dentry)
4775 {
4776 struct btrfs_trans_handle *trans;
4777 struct btrfs_root *root = BTRFS_I(dir)->root;
4778 struct inode *inode = old_dentry->d_inode;
4779 u64 index;
4780 unsigned long nr = 0;
4781 int err;
4782 int drop_inode = 0;
4783
4784 /* do not allow sys_link's with other subvols of the same device */
4785 if (root->objectid != BTRFS_I(inode)->root->objectid)
4786 return -EXDEV;
4787
4788 if (inode->i_nlink == ~0U)
4789 return -EMLINK;
4790
4791 err = btrfs_set_inode_index(dir, &index);
4792 if (err)
4793 goto fail;
4794
4795 /*
4796 * 2 items for inode and inode ref
4797 * 2 items for dir items
4798 * 1 item for parent inode
4799 */
4800 trans = btrfs_start_transaction(root, 5);
4801 if (IS_ERR(trans)) {
4802 err = PTR_ERR(trans);
4803 goto fail;
4804 }
4805
4806 btrfs_inc_nlink(inode);
4807 inode->i_ctime = CURRENT_TIME;
4808 ihold(inode);
4809
4810 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4811
4812 if (err) {
4813 drop_inode = 1;
4814 } else {
4815 struct dentry *parent = dentry->d_parent;
4816 err = btrfs_update_inode(trans, root, inode);
4817 BUG_ON(err);
4818 btrfs_log_new_name(trans, inode, NULL, parent);
4819 }
4820
4821 nr = trans->blocks_used;
4822 btrfs_end_transaction_throttle(trans, root);
4823 fail:
4824 if (drop_inode) {
4825 inode_dec_link_count(inode);
4826 iput(inode);
4827 }
4828 btrfs_btree_balance_dirty(root, nr);
4829 return err;
4830 }
4831
4832 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4833 {
4834 struct inode *inode = NULL;
4835 struct btrfs_trans_handle *trans;
4836 struct btrfs_root *root = BTRFS_I(dir)->root;
4837 int err = 0;
4838 int drop_on_err = 0;
4839 u64 objectid = 0;
4840 u64 index = 0;
4841 unsigned long nr = 1;
4842
4843 /*
4844 * 2 items for inode and ref
4845 * 2 items for dir items
4846 * 1 for xattr if selinux is on
4847 */
4848 trans = btrfs_start_transaction(root, 5);
4849 if (IS_ERR(trans))
4850 return PTR_ERR(trans);
4851
4852 err = btrfs_find_free_ino(root, &objectid);
4853 if (err)
4854 goto out_fail;
4855
4856 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4857 dentry->d_name.len, btrfs_ino(dir), objectid,
4858 S_IFDIR | mode, &index);
4859 if (IS_ERR(inode)) {
4860 err = PTR_ERR(inode);
4861 goto out_fail;
4862 }
4863
4864 drop_on_err = 1;
4865
4866 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4867 if (err)
4868 goto out_fail;
4869
4870 inode->i_op = &btrfs_dir_inode_operations;
4871 inode->i_fop = &btrfs_dir_file_operations;
4872
4873 btrfs_i_size_write(inode, 0);
4874 err = btrfs_update_inode(trans, root, inode);
4875 if (err)
4876 goto out_fail;
4877
4878 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4879 dentry->d_name.len, 0, index);
4880 if (err)
4881 goto out_fail;
4882
4883 d_instantiate(dentry, inode);
4884 drop_on_err = 0;
4885
4886 out_fail:
4887 nr = trans->blocks_used;
4888 btrfs_end_transaction_throttle(trans, root);
4889 if (drop_on_err)
4890 iput(inode);
4891 btrfs_btree_balance_dirty(root, nr);
4892 return err;
4893 }
4894
4895 /* helper for btfs_get_extent. Given an existing extent in the tree,
4896 * and an extent that you want to insert, deal with overlap and insert
4897 * the new extent into the tree.
4898 */
4899 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4900 struct extent_map *existing,
4901 struct extent_map *em,
4902 u64 map_start, u64 map_len)
4903 {
4904 u64 start_diff;
4905
4906 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4907 start_diff = map_start - em->start;
4908 em->start = map_start;
4909 em->len = map_len;
4910 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4911 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4912 em->block_start += start_diff;
4913 em->block_len -= start_diff;
4914 }
4915 return add_extent_mapping(em_tree, em);
4916 }
4917
4918 static noinline int uncompress_inline(struct btrfs_path *path,
4919 struct inode *inode, struct page *page,
4920 size_t pg_offset, u64 extent_offset,
4921 struct btrfs_file_extent_item *item)
4922 {
4923 int ret;
4924 struct extent_buffer *leaf = path->nodes[0];
4925 char *tmp;
4926 size_t max_size;
4927 unsigned long inline_size;
4928 unsigned long ptr;
4929 int compress_type;
4930
4931 WARN_ON(pg_offset != 0);
4932 compress_type = btrfs_file_extent_compression(leaf, item);
4933 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4934 inline_size = btrfs_file_extent_inline_item_len(leaf,
4935 btrfs_item_nr(leaf, path->slots[0]));
4936 tmp = kmalloc(inline_size, GFP_NOFS);
4937 if (!tmp)
4938 return -ENOMEM;
4939 ptr = btrfs_file_extent_inline_start(item);
4940
4941 read_extent_buffer(leaf, tmp, ptr, inline_size);
4942
4943 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4944 ret = btrfs_decompress(compress_type, tmp, page,
4945 extent_offset, inline_size, max_size);
4946 if (ret) {
4947 char *kaddr = kmap_atomic(page, KM_USER0);
4948 unsigned long copy_size = min_t(u64,
4949 PAGE_CACHE_SIZE - pg_offset,
4950 max_size - extent_offset);
4951 memset(kaddr + pg_offset, 0, copy_size);
4952 kunmap_atomic(kaddr, KM_USER0);
4953 }
4954 kfree(tmp);
4955 return 0;
4956 }
4957
4958 /*
4959 * a bit scary, this does extent mapping from logical file offset to the disk.
4960 * the ugly parts come from merging extents from the disk with the in-ram
4961 * representation. This gets more complex because of the data=ordered code,
4962 * where the in-ram extents might be locked pending data=ordered completion.
4963 *
4964 * This also copies inline extents directly into the page.
4965 */
4966
4967 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4968 size_t pg_offset, u64 start, u64 len,
4969 int create)
4970 {
4971 int ret;
4972 int err = 0;
4973 u64 bytenr;
4974 u64 extent_start = 0;
4975 u64 extent_end = 0;
4976 u64 objectid = btrfs_ino(inode);
4977 u32 found_type;
4978 struct btrfs_path *path = NULL;
4979 struct btrfs_root *root = BTRFS_I(inode)->root;
4980 struct btrfs_file_extent_item *item;
4981 struct extent_buffer *leaf;
4982 struct btrfs_key found_key;
4983 struct extent_map *em = NULL;
4984 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4985 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4986 struct btrfs_trans_handle *trans = NULL;
4987 int compress_type;
4988
4989 again:
4990 read_lock(&em_tree->lock);
4991 em = lookup_extent_mapping(em_tree, start, len);
4992 if (em)
4993 em->bdev = root->fs_info->fs_devices->latest_bdev;
4994 read_unlock(&em_tree->lock);
4995
4996 if (em) {
4997 if (em->start > start || em->start + em->len <= start)
4998 free_extent_map(em);
4999 else if (em->block_start == EXTENT_MAP_INLINE && page)
5000 free_extent_map(em);
5001 else
5002 goto out;
5003 }
5004 em = alloc_extent_map();
5005 if (!em) {
5006 err = -ENOMEM;
5007 goto out;
5008 }
5009 em->bdev = root->fs_info->fs_devices->latest_bdev;
5010 em->start = EXTENT_MAP_HOLE;
5011 em->orig_start = EXTENT_MAP_HOLE;
5012 em->len = (u64)-1;
5013 em->block_len = (u64)-1;
5014
5015 if (!path) {
5016 path = btrfs_alloc_path();
5017 if (!path) {
5018 err = -ENOMEM;
5019 goto out;
5020 }
5021 /*
5022 * Chances are we'll be called again, so go ahead and do
5023 * readahead
5024 */
5025 path->reada = 1;
5026 }
5027
5028 ret = btrfs_lookup_file_extent(trans, root, path,
5029 objectid, start, trans != NULL);
5030 if (ret < 0) {
5031 err = ret;
5032 goto out;
5033 }
5034
5035 if (ret != 0) {
5036 if (path->slots[0] == 0)
5037 goto not_found;
5038 path->slots[0]--;
5039 }
5040
5041 leaf = path->nodes[0];
5042 item = btrfs_item_ptr(leaf, path->slots[0],
5043 struct btrfs_file_extent_item);
5044 /* are we inside the extent that was found? */
5045 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5046 found_type = btrfs_key_type(&found_key);
5047 if (found_key.objectid != objectid ||
5048 found_type != BTRFS_EXTENT_DATA_KEY) {
5049 goto not_found;
5050 }
5051
5052 found_type = btrfs_file_extent_type(leaf, item);
5053 extent_start = found_key.offset;
5054 compress_type = btrfs_file_extent_compression(leaf, item);
5055 if (found_type == BTRFS_FILE_EXTENT_REG ||
5056 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5057 extent_end = extent_start +
5058 btrfs_file_extent_num_bytes(leaf, item);
5059 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5060 size_t size;
5061 size = btrfs_file_extent_inline_len(leaf, item);
5062 extent_end = (extent_start + size + root->sectorsize - 1) &
5063 ~((u64)root->sectorsize - 1);
5064 }
5065
5066 if (start >= extent_end) {
5067 path->slots[0]++;
5068 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5069 ret = btrfs_next_leaf(root, path);
5070 if (ret < 0) {
5071 err = ret;
5072 goto out;
5073 }
5074 if (ret > 0)
5075 goto not_found;
5076 leaf = path->nodes[0];
5077 }
5078 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5079 if (found_key.objectid != objectid ||
5080 found_key.type != BTRFS_EXTENT_DATA_KEY)
5081 goto not_found;
5082 if (start + len <= found_key.offset)
5083 goto not_found;
5084 em->start = start;
5085 em->len = found_key.offset - start;
5086 goto not_found_em;
5087 }
5088
5089 if (found_type == BTRFS_FILE_EXTENT_REG ||
5090 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5091 em->start = extent_start;
5092 em->len = extent_end - extent_start;
5093 em->orig_start = extent_start -
5094 btrfs_file_extent_offset(leaf, item);
5095 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5096 if (bytenr == 0) {
5097 em->block_start = EXTENT_MAP_HOLE;
5098 goto insert;
5099 }
5100 if (compress_type != BTRFS_COMPRESS_NONE) {
5101 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5102 em->compress_type = compress_type;
5103 em->block_start = bytenr;
5104 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5105 item);
5106 } else {
5107 bytenr += btrfs_file_extent_offset(leaf, item);
5108 em->block_start = bytenr;
5109 em->block_len = em->len;
5110 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5111 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5112 }
5113 goto insert;
5114 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5115 unsigned long ptr;
5116 char *map;
5117 size_t size;
5118 size_t extent_offset;
5119 size_t copy_size;
5120
5121 em->block_start = EXTENT_MAP_INLINE;
5122 if (!page || create) {
5123 em->start = extent_start;
5124 em->len = extent_end - extent_start;
5125 goto out;
5126 }
5127
5128 size = btrfs_file_extent_inline_len(leaf, item);
5129 extent_offset = page_offset(page) + pg_offset - extent_start;
5130 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5131 size - extent_offset);
5132 em->start = extent_start + extent_offset;
5133 em->len = (copy_size + root->sectorsize - 1) &
5134 ~((u64)root->sectorsize - 1);
5135 em->orig_start = EXTENT_MAP_INLINE;
5136 if (compress_type) {
5137 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5138 em->compress_type = compress_type;
5139 }
5140 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5141 if (create == 0 && !PageUptodate(page)) {
5142 if (btrfs_file_extent_compression(leaf, item) !=
5143 BTRFS_COMPRESS_NONE) {
5144 ret = uncompress_inline(path, inode, page,
5145 pg_offset,
5146 extent_offset, item);
5147 BUG_ON(ret);
5148 } else {
5149 map = kmap(page);
5150 read_extent_buffer(leaf, map + pg_offset, ptr,
5151 copy_size);
5152 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5153 memset(map + pg_offset + copy_size, 0,
5154 PAGE_CACHE_SIZE - pg_offset -
5155 copy_size);
5156 }
5157 kunmap(page);
5158 }
5159 flush_dcache_page(page);
5160 } else if (create && PageUptodate(page)) {
5161 WARN_ON(1);
5162 if (!trans) {
5163 kunmap(page);
5164 free_extent_map(em);
5165 em = NULL;
5166
5167 btrfs_release_path(path);
5168 trans = btrfs_join_transaction(root);
5169
5170 if (IS_ERR(trans))
5171 return ERR_CAST(trans);
5172 goto again;
5173 }
5174 map = kmap(page);
5175 write_extent_buffer(leaf, map + pg_offset, ptr,
5176 copy_size);
5177 kunmap(page);
5178 btrfs_mark_buffer_dirty(leaf);
5179 }
5180 set_extent_uptodate(io_tree, em->start,
5181 extent_map_end(em) - 1, NULL, GFP_NOFS);
5182 goto insert;
5183 } else {
5184 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5185 WARN_ON(1);
5186 }
5187 not_found:
5188 em->start = start;
5189 em->len = len;
5190 not_found_em:
5191 em->block_start = EXTENT_MAP_HOLE;
5192 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5193 insert:
5194 btrfs_release_path(path);
5195 if (em->start > start || extent_map_end(em) <= start) {
5196 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5197 "[%llu %llu]\n", (unsigned long long)em->start,
5198 (unsigned long long)em->len,
5199 (unsigned long long)start,
5200 (unsigned long long)len);
5201 err = -EIO;
5202 goto out;
5203 }
5204
5205 err = 0;
5206 write_lock(&em_tree->lock);
5207 ret = add_extent_mapping(em_tree, em);
5208 /* it is possible that someone inserted the extent into the tree
5209 * while we had the lock dropped. It is also possible that
5210 * an overlapping map exists in the tree
5211 */
5212 if (ret == -EEXIST) {
5213 struct extent_map *existing;
5214
5215 ret = 0;
5216
5217 existing = lookup_extent_mapping(em_tree, start, len);
5218 if (existing && (existing->start > start ||
5219 existing->start + existing->len <= start)) {
5220 free_extent_map(existing);
5221 existing = NULL;
5222 }
5223 if (!existing) {
5224 existing = lookup_extent_mapping(em_tree, em->start,
5225 em->len);
5226 if (existing) {
5227 err = merge_extent_mapping(em_tree, existing,
5228 em, start,
5229 root->sectorsize);
5230 free_extent_map(existing);
5231 if (err) {
5232 free_extent_map(em);
5233 em = NULL;
5234 }
5235 } else {
5236 err = -EIO;
5237 free_extent_map(em);
5238 em = NULL;
5239 }
5240 } else {
5241 free_extent_map(em);
5242 em = existing;
5243 err = 0;
5244 }
5245 }
5246 write_unlock(&em_tree->lock);
5247 out:
5248
5249 trace_btrfs_get_extent(root, em);
5250
5251 if (path)
5252 btrfs_free_path(path);
5253 if (trans) {
5254 ret = btrfs_end_transaction(trans, root);
5255 if (!err)
5256 err = ret;
5257 }
5258 if (err) {
5259 free_extent_map(em);
5260 return ERR_PTR(err);
5261 }
5262 return em;
5263 }
5264
5265 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5266 size_t pg_offset, u64 start, u64 len,
5267 int create)
5268 {
5269 struct extent_map *em;
5270 struct extent_map *hole_em = NULL;
5271 u64 range_start = start;
5272 u64 end;
5273 u64 found;
5274 u64 found_end;
5275 int err = 0;
5276
5277 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5278 if (IS_ERR(em))
5279 return em;
5280 if (em) {
5281 /*
5282 * if our em maps to a hole, there might
5283 * actually be delalloc bytes behind it
5284 */
5285 if (em->block_start != EXTENT_MAP_HOLE)
5286 return em;
5287 else
5288 hole_em = em;
5289 }
5290
5291 /* check to see if we've wrapped (len == -1 or similar) */
5292 end = start + len;
5293 if (end < start)
5294 end = (u64)-1;
5295 else
5296 end -= 1;
5297
5298 em = NULL;
5299
5300 /* ok, we didn't find anything, lets look for delalloc */
5301 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5302 end, len, EXTENT_DELALLOC, 1);
5303 found_end = range_start + found;
5304 if (found_end < range_start)
5305 found_end = (u64)-1;
5306
5307 /*
5308 * we didn't find anything useful, return
5309 * the original results from get_extent()
5310 */
5311 if (range_start > end || found_end <= start) {
5312 em = hole_em;
5313 hole_em = NULL;
5314 goto out;
5315 }
5316
5317 /* adjust the range_start to make sure it doesn't
5318 * go backwards from the start they passed in
5319 */
5320 range_start = max(start,range_start);
5321 found = found_end - range_start;
5322
5323 if (found > 0) {
5324 u64 hole_start = start;
5325 u64 hole_len = len;
5326
5327 em = alloc_extent_map();
5328 if (!em) {
5329 err = -ENOMEM;
5330 goto out;
5331 }
5332 /*
5333 * when btrfs_get_extent can't find anything it
5334 * returns one huge hole
5335 *
5336 * make sure what it found really fits our range, and
5337 * adjust to make sure it is based on the start from
5338 * the caller
5339 */
5340 if (hole_em) {
5341 u64 calc_end = extent_map_end(hole_em);
5342
5343 if (calc_end <= start || (hole_em->start > end)) {
5344 free_extent_map(hole_em);
5345 hole_em = NULL;
5346 } else {
5347 hole_start = max(hole_em->start, start);
5348 hole_len = calc_end - hole_start;
5349 }
5350 }
5351 em->bdev = NULL;
5352 if (hole_em && range_start > hole_start) {
5353 /* our hole starts before our delalloc, so we
5354 * have to return just the parts of the hole
5355 * that go until the delalloc starts
5356 */
5357 em->len = min(hole_len,
5358 range_start - hole_start);
5359 em->start = hole_start;
5360 em->orig_start = hole_start;
5361 /*
5362 * don't adjust block start at all,
5363 * it is fixed at EXTENT_MAP_HOLE
5364 */
5365 em->block_start = hole_em->block_start;
5366 em->block_len = hole_len;
5367 } else {
5368 em->start = range_start;
5369 em->len = found;
5370 em->orig_start = range_start;
5371 em->block_start = EXTENT_MAP_DELALLOC;
5372 em->block_len = found;
5373 }
5374 } else if (hole_em) {
5375 return hole_em;
5376 }
5377 out:
5378
5379 free_extent_map(hole_em);
5380 if (err) {
5381 free_extent_map(em);
5382 return ERR_PTR(err);
5383 }
5384 return em;
5385 }
5386
5387 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5388 struct extent_map *em,
5389 u64 start, u64 len)
5390 {
5391 struct btrfs_root *root = BTRFS_I(inode)->root;
5392 struct btrfs_trans_handle *trans;
5393 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5394 struct btrfs_key ins;
5395 u64 alloc_hint;
5396 int ret;
5397 bool insert = false;
5398
5399 /*
5400 * Ok if the extent map we looked up is a hole and is for the exact
5401 * range we want, there is no reason to allocate a new one, however if
5402 * it is not right then we need to free this one and drop the cache for
5403 * our range.
5404 */
5405 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5406 em->len != len) {
5407 free_extent_map(em);
5408 em = NULL;
5409 insert = true;
5410 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5411 }
5412
5413 trans = btrfs_join_transaction(root);
5414 if (IS_ERR(trans))
5415 return ERR_CAST(trans);
5416
5417 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5418 btrfs_add_inode_defrag(trans, inode);
5419
5420 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5421
5422 alloc_hint = get_extent_allocation_hint(inode, start, len);
5423 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5424 alloc_hint, (u64)-1, &ins, 1);
5425 if (ret) {
5426 em = ERR_PTR(ret);
5427 goto out;
5428 }
5429
5430 if (!em) {
5431 em = alloc_extent_map();
5432 if (!em) {
5433 em = ERR_PTR(-ENOMEM);
5434 goto out;
5435 }
5436 }
5437
5438 em->start = start;
5439 em->orig_start = em->start;
5440 em->len = ins.offset;
5441
5442 em->block_start = ins.objectid;
5443 em->block_len = ins.offset;
5444 em->bdev = root->fs_info->fs_devices->latest_bdev;
5445
5446 /*
5447 * We need to do this because if we're using the original em we searched
5448 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5449 */
5450 em->flags = 0;
5451 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5452
5453 while (insert) {
5454 write_lock(&em_tree->lock);
5455 ret = add_extent_mapping(em_tree, em);
5456 write_unlock(&em_tree->lock);
5457 if (ret != -EEXIST)
5458 break;
5459 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5460 }
5461
5462 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5463 ins.offset, ins.offset, 0);
5464 if (ret) {
5465 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5466 em = ERR_PTR(ret);
5467 }
5468 out:
5469 btrfs_end_transaction(trans, root);
5470 return em;
5471 }
5472
5473 /*
5474 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5475 * block must be cow'd
5476 */
5477 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5478 struct inode *inode, u64 offset, u64 len)
5479 {
5480 struct btrfs_path *path;
5481 int ret;
5482 struct extent_buffer *leaf;
5483 struct btrfs_root *root = BTRFS_I(inode)->root;
5484 struct btrfs_file_extent_item *fi;
5485 struct btrfs_key key;
5486 u64 disk_bytenr;
5487 u64 backref_offset;
5488 u64 extent_end;
5489 u64 num_bytes;
5490 int slot;
5491 int found_type;
5492
5493 path = btrfs_alloc_path();
5494 if (!path)
5495 return -ENOMEM;
5496
5497 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5498 offset, 0);
5499 if (ret < 0)
5500 goto out;
5501
5502 slot = path->slots[0];
5503 if (ret == 1) {
5504 if (slot == 0) {
5505 /* can't find the item, must cow */
5506 ret = 0;
5507 goto out;
5508 }
5509 slot--;
5510 }
5511 ret = 0;
5512 leaf = path->nodes[0];
5513 btrfs_item_key_to_cpu(leaf, &key, slot);
5514 if (key.objectid != btrfs_ino(inode) ||
5515 key.type != BTRFS_EXTENT_DATA_KEY) {
5516 /* not our file or wrong item type, must cow */
5517 goto out;
5518 }
5519
5520 if (key.offset > offset) {
5521 /* Wrong offset, must cow */
5522 goto out;
5523 }
5524
5525 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5526 found_type = btrfs_file_extent_type(leaf, fi);
5527 if (found_type != BTRFS_FILE_EXTENT_REG &&
5528 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5529 /* not a regular extent, must cow */
5530 goto out;
5531 }
5532 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5533 backref_offset = btrfs_file_extent_offset(leaf, fi);
5534
5535 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5536 if (extent_end < offset + len) {
5537 /* extent doesn't include our full range, must cow */
5538 goto out;
5539 }
5540
5541 if (btrfs_extent_readonly(root, disk_bytenr))
5542 goto out;
5543
5544 /*
5545 * look for other files referencing this extent, if we
5546 * find any we must cow
5547 */
5548 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5549 key.offset - backref_offset, disk_bytenr))
5550 goto out;
5551
5552 /*
5553 * adjust disk_bytenr and num_bytes to cover just the bytes
5554 * in this extent we are about to write. If there
5555 * are any csums in that range we have to cow in order
5556 * to keep the csums correct
5557 */
5558 disk_bytenr += backref_offset;
5559 disk_bytenr += offset - key.offset;
5560 num_bytes = min(offset + len, extent_end) - offset;
5561 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5562 goto out;
5563 /*
5564 * all of the above have passed, it is safe to overwrite this extent
5565 * without cow
5566 */
5567 ret = 1;
5568 out:
5569 btrfs_free_path(path);
5570 return ret;
5571 }
5572
5573 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5574 struct buffer_head *bh_result, int create)
5575 {
5576 struct extent_map *em;
5577 struct btrfs_root *root = BTRFS_I(inode)->root;
5578 u64 start = iblock << inode->i_blkbits;
5579 u64 len = bh_result->b_size;
5580 struct btrfs_trans_handle *trans;
5581
5582 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5583 if (IS_ERR(em))
5584 return PTR_ERR(em);
5585
5586 /*
5587 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5588 * io. INLINE is special, and we could probably kludge it in here, but
5589 * it's still buffered so for safety lets just fall back to the generic
5590 * buffered path.
5591 *
5592 * For COMPRESSED we _have_ to read the entire extent in so we can
5593 * decompress it, so there will be buffering required no matter what we
5594 * do, so go ahead and fallback to buffered.
5595 *
5596 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5597 * to buffered IO. Don't blame me, this is the price we pay for using
5598 * the generic code.
5599 */
5600 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5601 em->block_start == EXTENT_MAP_INLINE) {
5602 free_extent_map(em);
5603 return -ENOTBLK;
5604 }
5605
5606 /* Just a good old fashioned hole, return */
5607 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5608 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5609 free_extent_map(em);
5610 /* DIO will do one hole at a time, so just unlock a sector */
5611 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5612 start + root->sectorsize - 1, GFP_NOFS);
5613 return 0;
5614 }
5615
5616 /*
5617 * We don't allocate a new extent in the following cases
5618 *
5619 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5620 * existing extent.
5621 * 2) The extent is marked as PREALLOC. We're good to go here and can
5622 * just use the extent.
5623 *
5624 */
5625 if (!create) {
5626 len = em->len - (start - em->start);
5627 goto map;
5628 }
5629
5630 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5631 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5632 em->block_start != EXTENT_MAP_HOLE)) {
5633 int type;
5634 int ret;
5635 u64 block_start;
5636
5637 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5638 type = BTRFS_ORDERED_PREALLOC;
5639 else
5640 type = BTRFS_ORDERED_NOCOW;
5641 len = min(len, em->len - (start - em->start));
5642 block_start = em->block_start + (start - em->start);
5643
5644 /*
5645 * we're not going to log anything, but we do need
5646 * to make sure the current transaction stays open
5647 * while we look for nocow cross refs
5648 */
5649 trans = btrfs_join_transaction(root);
5650 if (IS_ERR(trans))
5651 goto must_cow;
5652
5653 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5654 ret = btrfs_add_ordered_extent_dio(inode, start,
5655 block_start, len, len, type);
5656 btrfs_end_transaction(trans, root);
5657 if (ret) {
5658 free_extent_map(em);
5659 return ret;
5660 }
5661 goto unlock;
5662 }
5663 btrfs_end_transaction(trans, root);
5664 }
5665 must_cow:
5666 /*
5667 * this will cow the extent, reset the len in case we changed
5668 * it above
5669 */
5670 len = bh_result->b_size;
5671 em = btrfs_new_extent_direct(inode, em, start, len);
5672 if (IS_ERR(em))
5673 return PTR_ERR(em);
5674 len = min(len, em->len - (start - em->start));
5675 unlock:
5676 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5677 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5678 0, NULL, GFP_NOFS);
5679 map:
5680 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5681 inode->i_blkbits;
5682 bh_result->b_size = len;
5683 bh_result->b_bdev = em->bdev;
5684 set_buffer_mapped(bh_result);
5685 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5686 set_buffer_new(bh_result);
5687
5688 free_extent_map(em);
5689
5690 return 0;
5691 }
5692
5693 struct btrfs_dio_private {
5694 struct inode *inode;
5695 u64 logical_offset;
5696 u64 disk_bytenr;
5697 u64 bytes;
5698 u32 *csums;
5699 void *private;
5700
5701 /* number of bios pending for this dio */
5702 atomic_t pending_bios;
5703
5704 /* IO errors */
5705 int errors;
5706
5707 struct bio *orig_bio;
5708 };
5709
5710 static void btrfs_endio_direct_read(struct bio *bio, int err)
5711 {
5712 struct btrfs_dio_private *dip = bio->bi_private;
5713 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5714 struct bio_vec *bvec = bio->bi_io_vec;
5715 struct inode *inode = dip->inode;
5716 struct btrfs_root *root = BTRFS_I(inode)->root;
5717 u64 start;
5718 u32 *private = dip->csums;
5719
5720 start = dip->logical_offset;
5721 do {
5722 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5723 struct page *page = bvec->bv_page;
5724 char *kaddr;
5725 u32 csum = ~(u32)0;
5726 unsigned long flags;
5727
5728 local_irq_save(flags);
5729 kaddr = kmap_atomic(page, KM_IRQ0);
5730 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5731 csum, bvec->bv_len);
5732 btrfs_csum_final(csum, (char *)&csum);
5733 kunmap_atomic(kaddr, KM_IRQ0);
5734 local_irq_restore(flags);
5735
5736 flush_dcache_page(bvec->bv_page);
5737 if (csum != *private) {
5738 printk(KERN_ERR "btrfs csum failed ino %llu off"
5739 " %llu csum %u private %u\n",
5740 (unsigned long long)btrfs_ino(inode),
5741 (unsigned long long)start,
5742 csum, *private);
5743 err = -EIO;
5744 }
5745 }
5746
5747 start += bvec->bv_len;
5748 private++;
5749 bvec++;
5750 } while (bvec <= bvec_end);
5751
5752 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5753 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5754 bio->bi_private = dip->private;
5755
5756 kfree(dip->csums);
5757 kfree(dip);
5758
5759 /* If we had a csum failure make sure to clear the uptodate flag */
5760 if (err)
5761 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5762 dio_end_io(bio, err);
5763 }
5764
5765 static void btrfs_endio_direct_write(struct bio *bio, int err)
5766 {
5767 struct btrfs_dio_private *dip = bio->bi_private;
5768 struct inode *inode = dip->inode;
5769 struct btrfs_root *root = BTRFS_I(inode)->root;
5770 struct btrfs_trans_handle *trans;
5771 struct btrfs_ordered_extent *ordered = NULL;
5772 struct extent_state *cached_state = NULL;
5773 u64 ordered_offset = dip->logical_offset;
5774 u64 ordered_bytes = dip->bytes;
5775 int ret;
5776
5777 if (err)
5778 goto out_done;
5779 again:
5780 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5781 &ordered_offset,
5782 ordered_bytes);
5783 if (!ret)
5784 goto out_test;
5785
5786 BUG_ON(!ordered);
5787
5788 trans = btrfs_join_transaction(root);
5789 if (IS_ERR(trans)) {
5790 err = -ENOMEM;
5791 goto out;
5792 }
5793 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5794
5795 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5796 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5797 if (!ret)
5798 ret = btrfs_update_inode(trans, root, inode);
5799 err = ret;
5800 goto out;
5801 }
5802
5803 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5804 ordered->file_offset + ordered->len - 1, 0,
5805 &cached_state, GFP_NOFS);
5806
5807 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5808 ret = btrfs_mark_extent_written(trans, inode,
5809 ordered->file_offset,
5810 ordered->file_offset +
5811 ordered->len);
5812 if (ret) {
5813 err = ret;
5814 goto out_unlock;
5815 }
5816 } else {
5817 ret = insert_reserved_file_extent(trans, inode,
5818 ordered->file_offset,
5819 ordered->start,
5820 ordered->disk_len,
5821 ordered->len,
5822 ordered->len,
5823 0, 0, 0,
5824 BTRFS_FILE_EXTENT_REG);
5825 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5826 ordered->file_offset, ordered->len);
5827 if (ret) {
5828 err = ret;
5829 WARN_ON(1);
5830 goto out_unlock;
5831 }
5832 }
5833
5834 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5835 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5836 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5837 btrfs_update_inode(trans, root, inode);
5838 ret = 0;
5839 out_unlock:
5840 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5841 ordered->file_offset + ordered->len - 1,
5842 &cached_state, GFP_NOFS);
5843 out:
5844 btrfs_delalloc_release_metadata(inode, ordered->len);
5845 btrfs_end_transaction(trans, root);
5846 ordered_offset = ordered->file_offset + ordered->len;
5847 btrfs_put_ordered_extent(ordered);
5848 btrfs_put_ordered_extent(ordered);
5849
5850 out_test:
5851 /*
5852 * our bio might span multiple ordered extents. If we haven't
5853 * completed the accounting for the whole dio, go back and try again
5854 */
5855 if (ordered_offset < dip->logical_offset + dip->bytes) {
5856 ordered_bytes = dip->logical_offset + dip->bytes -
5857 ordered_offset;
5858 goto again;
5859 }
5860 out_done:
5861 bio->bi_private = dip->private;
5862
5863 kfree(dip->csums);
5864 kfree(dip);
5865
5866 /* If we had an error make sure to clear the uptodate flag */
5867 if (err)
5868 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5869 dio_end_io(bio, err);
5870 }
5871
5872 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5873 struct bio *bio, int mirror_num,
5874 unsigned long bio_flags, u64 offset)
5875 {
5876 int ret;
5877 struct btrfs_root *root = BTRFS_I(inode)->root;
5878 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5879 BUG_ON(ret);
5880 return 0;
5881 }
5882
5883 static void btrfs_end_dio_bio(struct bio *bio, int err)
5884 {
5885 struct btrfs_dio_private *dip = bio->bi_private;
5886
5887 if (err) {
5888 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5889 "sector %#Lx len %u err no %d\n",
5890 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5891 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5892 dip->errors = 1;
5893
5894 /*
5895 * before atomic variable goto zero, we must make sure
5896 * dip->errors is perceived to be set.
5897 */
5898 smp_mb__before_atomic_dec();
5899 }
5900
5901 /* if there are more bios still pending for this dio, just exit */
5902 if (!atomic_dec_and_test(&dip->pending_bios))
5903 goto out;
5904
5905 if (dip->errors)
5906 bio_io_error(dip->orig_bio);
5907 else {
5908 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5909 bio_endio(dip->orig_bio, 0);
5910 }
5911 out:
5912 bio_put(bio);
5913 }
5914
5915 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5916 u64 first_sector, gfp_t gfp_flags)
5917 {
5918 int nr_vecs = bio_get_nr_vecs(bdev);
5919 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5920 }
5921
5922 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5923 int rw, u64 file_offset, int skip_sum,
5924 u32 *csums, int async_submit)
5925 {
5926 int write = rw & REQ_WRITE;
5927 struct btrfs_root *root = BTRFS_I(inode)->root;
5928 int ret;
5929
5930 bio_get(bio);
5931 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5932 if (ret)
5933 goto err;
5934
5935 if (skip_sum)
5936 goto map;
5937
5938 if (write && async_submit) {
5939 ret = btrfs_wq_submit_bio(root->fs_info,
5940 inode, rw, bio, 0, 0,
5941 file_offset,
5942 __btrfs_submit_bio_start_direct_io,
5943 __btrfs_submit_bio_done);
5944 goto err;
5945 } else if (write) {
5946 /*
5947 * If we aren't doing async submit, calculate the csum of the
5948 * bio now.
5949 */
5950 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5951 if (ret)
5952 goto err;
5953 } else if (!skip_sum) {
5954 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5955 file_offset, csums);
5956 if (ret)
5957 goto err;
5958 }
5959
5960 map:
5961 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5962 err:
5963 bio_put(bio);
5964 return ret;
5965 }
5966
5967 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5968 int skip_sum)
5969 {
5970 struct inode *inode = dip->inode;
5971 struct btrfs_root *root = BTRFS_I(inode)->root;
5972 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5973 struct bio *bio;
5974 struct bio *orig_bio = dip->orig_bio;
5975 struct bio_vec *bvec = orig_bio->bi_io_vec;
5976 u64 start_sector = orig_bio->bi_sector;
5977 u64 file_offset = dip->logical_offset;
5978 u64 submit_len = 0;
5979 u64 map_length;
5980 int nr_pages = 0;
5981 u32 *csums = dip->csums;
5982 int ret = 0;
5983 int async_submit = 0;
5984 int write = rw & REQ_WRITE;
5985
5986 map_length = orig_bio->bi_size;
5987 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5988 &map_length, NULL, 0);
5989 if (ret) {
5990 bio_put(orig_bio);
5991 return -EIO;
5992 }
5993
5994 if (map_length >= orig_bio->bi_size) {
5995 bio = orig_bio;
5996 goto submit;
5997 }
5998
5999 async_submit = 1;
6000 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6001 if (!bio)
6002 return -ENOMEM;
6003 bio->bi_private = dip;
6004 bio->bi_end_io = btrfs_end_dio_bio;
6005 atomic_inc(&dip->pending_bios);
6006
6007 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6008 if (unlikely(map_length < submit_len + bvec->bv_len ||
6009 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6010 bvec->bv_offset) < bvec->bv_len)) {
6011 /*
6012 * inc the count before we submit the bio so
6013 * we know the end IO handler won't happen before
6014 * we inc the count. Otherwise, the dip might get freed
6015 * before we're done setting it up
6016 */
6017 atomic_inc(&dip->pending_bios);
6018 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6019 file_offset, skip_sum,
6020 csums, async_submit);
6021 if (ret) {
6022 bio_put(bio);
6023 atomic_dec(&dip->pending_bios);
6024 goto out_err;
6025 }
6026
6027 /* Write's use the ordered csums */
6028 if (!write && !skip_sum)
6029 csums = csums + nr_pages;
6030 start_sector += submit_len >> 9;
6031 file_offset += submit_len;
6032
6033 submit_len = 0;
6034 nr_pages = 0;
6035
6036 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6037 start_sector, GFP_NOFS);
6038 if (!bio)
6039 goto out_err;
6040 bio->bi_private = dip;
6041 bio->bi_end_io = btrfs_end_dio_bio;
6042
6043 map_length = orig_bio->bi_size;
6044 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6045 &map_length, NULL, 0);
6046 if (ret) {
6047 bio_put(bio);
6048 goto out_err;
6049 }
6050 } else {
6051 submit_len += bvec->bv_len;
6052 nr_pages ++;
6053 bvec++;
6054 }
6055 }
6056
6057 submit:
6058 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6059 csums, async_submit);
6060 if (!ret)
6061 return 0;
6062
6063 bio_put(bio);
6064 out_err:
6065 dip->errors = 1;
6066 /*
6067 * before atomic variable goto zero, we must
6068 * make sure dip->errors is perceived to be set.
6069 */
6070 smp_mb__before_atomic_dec();
6071 if (atomic_dec_and_test(&dip->pending_bios))
6072 bio_io_error(dip->orig_bio);
6073
6074 /* bio_end_io() will handle error, so we needn't return it */
6075 return 0;
6076 }
6077
6078 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6079 loff_t file_offset)
6080 {
6081 struct btrfs_root *root = BTRFS_I(inode)->root;
6082 struct btrfs_dio_private *dip;
6083 struct bio_vec *bvec = bio->bi_io_vec;
6084 int skip_sum;
6085 int write = rw & REQ_WRITE;
6086 int ret = 0;
6087
6088 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6089
6090 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6091 if (!dip) {
6092 ret = -ENOMEM;
6093 goto free_ordered;
6094 }
6095 dip->csums = NULL;
6096
6097 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6098 if (!write && !skip_sum) {
6099 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6100 if (!dip->csums) {
6101 kfree(dip);
6102 ret = -ENOMEM;
6103 goto free_ordered;
6104 }
6105 }
6106
6107 dip->private = bio->bi_private;
6108 dip->inode = inode;
6109 dip->logical_offset = file_offset;
6110
6111 dip->bytes = 0;
6112 do {
6113 dip->bytes += bvec->bv_len;
6114 bvec++;
6115 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6116
6117 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6118 bio->bi_private = dip;
6119 dip->errors = 0;
6120 dip->orig_bio = bio;
6121 atomic_set(&dip->pending_bios, 0);
6122
6123 if (write)
6124 bio->bi_end_io = btrfs_endio_direct_write;
6125 else
6126 bio->bi_end_io = btrfs_endio_direct_read;
6127
6128 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6129 if (!ret)
6130 return;
6131 free_ordered:
6132 /*
6133 * If this is a write, we need to clean up the reserved space and kill
6134 * the ordered extent.
6135 */
6136 if (write) {
6137 struct btrfs_ordered_extent *ordered;
6138 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6139 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6140 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6141 btrfs_free_reserved_extent(root, ordered->start,
6142 ordered->disk_len);
6143 btrfs_put_ordered_extent(ordered);
6144 btrfs_put_ordered_extent(ordered);
6145 }
6146 bio_endio(bio, ret);
6147 }
6148
6149 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6150 const struct iovec *iov, loff_t offset,
6151 unsigned long nr_segs)
6152 {
6153 int seg;
6154 int i;
6155 size_t size;
6156 unsigned long addr;
6157 unsigned blocksize_mask = root->sectorsize - 1;
6158 ssize_t retval = -EINVAL;
6159 loff_t end = offset;
6160
6161 if (offset & blocksize_mask)
6162 goto out;
6163
6164 /* Check the memory alignment. Blocks cannot straddle pages */
6165 for (seg = 0; seg < nr_segs; seg++) {
6166 addr = (unsigned long)iov[seg].iov_base;
6167 size = iov[seg].iov_len;
6168 end += size;
6169 if ((addr & blocksize_mask) || (size & blocksize_mask))
6170 goto out;
6171
6172 /* If this is a write we don't need to check anymore */
6173 if (rw & WRITE)
6174 continue;
6175
6176 /*
6177 * Check to make sure we don't have duplicate iov_base's in this
6178 * iovec, if so return EINVAL, otherwise we'll get csum errors
6179 * when reading back.
6180 */
6181 for (i = seg + 1; i < nr_segs; i++) {
6182 if (iov[seg].iov_base == iov[i].iov_base)
6183 goto out;
6184 }
6185 }
6186 retval = 0;
6187 out:
6188 return retval;
6189 }
6190 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6191 const struct iovec *iov, loff_t offset,
6192 unsigned long nr_segs)
6193 {
6194 struct file *file = iocb->ki_filp;
6195 struct inode *inode = file->f_mapping->host;
6196 struct btrfs_ordered_extent *ordered;
6197 struct extent_state *cached_state = NULL;
6198 u64 lockstart, lockend;
6199 ssize_t ret;
6200 int writing = rw & WRITE;
6201 int write_bits = 0;
6202 size_t count = iov_length(iov, nr_segs);
6203
6204 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6205 offset, nr_segs)) {
6206 return 0;
6207 }
6208
6209 lockstart = offset;
6210 lockend = offset + count - 1;
6211
6212 if (writing) {
6213 ret = btrfs_delalloc_reserve_space(inode, count);
6214 if (ret)
6215 goto out;
6216 }
6217
6218 while (1) {
6219 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6220 0, &cached_state, GFP_NOFS);
6221 /*
6222 * We're concerned with the entire range that we're going to be
6223 * doing DIO to, so we need to make sure theres no ordered
6224 * extents in this range.
6225 */
6226 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6227 lockend - lockstart + 1);
6228 if (!ordered)
6229 break;
6230 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6231 &cached_state, GFP_NOFS);
6232 btrfs_start_ordered_extent(inode, ordered, 1);
6233 btrfs_put_ordered_extent(ordered);
6234 cond_resched();
6235 }
6236
6237 /*
6238 * we don't use btrfs_set_extent_delalloc because we don't want
6239 * the dirty or uptodate bits
6240 */
6241 if (writing) {
6242 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6243 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6244 EXTENT_DELALLOC, 0, NULL, &cached_state,
6245 GFP_NOFS);
6246 if (ret) {
6247 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6248 lockend, EXTENT_LOCKED | write_bits,
6249 1, 0, &cached_state, GFP_NOFS);
6250 goto out;
6251 }
6252 }
6253
6254 free_extent_state(cached_state);
6255 cached_state = NULL;
6256
6257 ret = __blockdev_direct_IO(rw, iocb, inode,
6258 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6259 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6260 btrfs_submit_direct, 0);
6261
6262 if (ret < 0 && ret != -EIOCBQUEUED) {
6263 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6264 offset + iov_length(iov, nr_segs) - 1,
6265 EXTENT_LOCKED | write_bits, 1, 0,
6266 &cached_state, GFP_NOFS);
6267 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6268 /*
6269 * We're falling back to buffered, unlock the section we didn't
6270 * do IO on.
6271 */
6272 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6273 offset + iov_length(iov, nr_segs) - 1,
6274 EXTENT_LOCKED | write_bits, 1, 0,
6275 &cached_state, GFP_NOFS);
6276 }
6277 out:
6278 free_extent_state(cached_state);
6279 return ret;
6280 }
6281
6282 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6283 __u64 start, __u64 len)
6284 {
6285 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6286 }
6287
6288 int btrfs_readpage(struct file *file, struct page *page)
6289 {
6290 struct extent_io_tree *tree;
6291 tree = &BTRFS_I(page->mapping->host)->io_tree;
6292 return extent_read_full_page(tree, page, btrfs_get_extent);
6293 }
6294
6295 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6296 {
6297 struct extent_io_tree *tree;
6298
6299
6300 if (current->flags & PF_MEMALLOC) {
6301 redirty_page_for_writepage(wbc, page);
6302 unlock_page(page);
6303 return 0;
6304 }
6305 tree = &BTRFS_I(page->mapping->host)->io_tree;
6306 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6307 }
6308
6309 int btrfs_writepages(struct address_space *mapping,
6310 struct writeback_control *wbc)
6311 {
6312 struct extent_io_tree *tree;
6313
6314 tree = &BTRFS_I(mapping->host)->io_tree;
6315 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6316 }
6317
6318 static int
6319 btrfs_readpages(struct file *file, struct address_space *mapping,
6320 struct list_head *pages, unsigned nr_pages)
6321 {
6322 struct extent_io_tree *tree;
6323 tree = &BTRFS_I(mapping->host)->io_tree;
6324 return extent_readpages(tree, mapping, pages, nr_pages,
6325 btrfs_get_extent);
6326 }
6327 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6328 {
6329 struct extent_io_tree *tree;
6330 struct extent_map_tree *map;
6331 int ret;
6332
6333 tree = &BTRFS_I(page->mapping->host)->io_tree;
6334 map = &BTRFS_I(page->mapping->host)->extent_tree;
6335 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6336 if (ret == 1) {
6337 ClearPagePrivate(page);
6338 set_page_private(page, 0);
6339 page_cache_release(page);
6340 }
6341 return ret;
6342 }
6343
6344 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6345 {
6346 if (PageWriteback(page) || PageDirty(page))
6347 return 0;
6348 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6349 }
6350
6351 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6352 {
6353 struct extent_io_tree *tree;
6354 struct btrfs_ordered_extent *ordered;
6355 struct extent_state *cached_state = NULL;
6356 u64 page_start = page_offset(page);
6357 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6358
6359
6360 /*
6361 * we have the page locked, so new writeback can't start,
6362 * and the dirty bit won't be cleared while we are here.
6363 *
6364 * Wait for IO on this page so that we can safely clear
6365 * the PagePrivate2 bit and do ordered accounting
6366 */
6367 wait_on_page_writeback(page);
6368
6369 tree = &BTRFS_I(page->mapping->host)->io_tree;
6370 if (offset) {
6371 btrfs_releasepage(page, GFP_NOFS);
6372 return;
6373 }
6374 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6375 GFP_NOFS);
6376 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6377 page_offset(page));
6378 if (ordered) {
6379 /*
6380 * IO on this page will never be started, so we need
6381 * to account for any ordered extents now
6382 */
6383 clear_extent_bit(tree, page_start, page_end,
6384 EXTENT_DIRTY | EXTENT_DELALLOC |
6385 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6386 &cached_state, GFP_NOFS);
6387 /*
6388 * whoever cleared the private bit is responsible
6389 * for the finish_ordered_io
6390 */
6391 if (TestClearPagePrivate2(page)) {
6392 btrfs_finish_ordered_io(page->mapping->host,
6393 page_start, page_end);
6394 }
6395 btrfs_put_ordered_extent(ordered);
6396 cached_state = NULL;
6397 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6398 GFP_NOFS);
6399 }
6400 clear_extent_bit(tree, page_start, page_end,
6401 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6402 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6403 __btrfs_releasepage(page, GFP_NOFS);
6404
6405 ClearPageChecked(page);
6406 if (PagePrivate(page)) {
6407 ClearPagePrivate(page);
6408 set_page_private(page, 0);
6409 page_cache_release(page);
6410 }
6411 }
6412
6413 /*
6414 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6415 * called from a page fault handler when a page is first dirtied. Hence we must
6416 * be careful to check for EOF conditions here. We set the page up correctly
6417 * for a written page which means we get ENOSPC checking when writing into
6418 * holes and correct delalloc and unwritten extent mapping on filesystems that
6419 * support these features.
6420 *
6421 * We are not allowed to take the i_mutex here so we have to play games to
6422 * protect against truncate races as the page could now be beyond EOF. Because
6423 * vmtruncate() writes the inode size before removing pages, once we have the
6424 * page lock we can determine safely if the page is beyond EOF. If it is not
6425 * beyond EOF, then the page is guaranteed safe against truncation until we
6426 * unlock the page.
6427 */
6428 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6429 {
6430 struct page *page = vmf->page;
6431 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6432 struct btrfs_root *root = BTRFS_I(inode)->root;
6433 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6434 struct btrfs_ordered_extent *ordered;
6435 struct extent_state *cached_state = NULL;
6436 char *kaddr;
6437 unsigned long zero_start;
6438 loff_t size;
6439 int ret;
6440 u64 page_start;
6441 u64 page_end;
6442
6443 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6444 if (ret) {
6445 if (ret == -ENOMEM)
6446 ret = VM_FAULT_OOM;
6447 else /* -ENOSPC, -EIO, etc */
6448 ret = VM_FAULT_SIGBUS;
6449 goto out;
6450 }
6451
6452 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6453 again:
6454 lock_page(page);
6455 size = i_size_read(inode);
6456 page_start = page_offset(page);
6457 page_end = page_start + PAGE_CACHE_SIZE - 1;
6458
6459 if ((page->mapping != inode->i_mapping) ||
6460 (page_start >= size)) {
6461 /* page got truncated out from underneath us */
6462 goto out_unlock;
6463 }
6464 wait_on_page_writeback(page);
6465
6466 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6467 GFP_NOFS);
6468 set_page_extent_mapped(page);
6469
6470 /*
6471 * we can't set the delalloc bits if there are pending ordered
6472 * extents. Drop our locks and wait for them to finish
6473 */
6474 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6475 if (ordered) {
6476 unlock_extent_cached(io_tree, page_start, page_end,
6477 &cached_state, GFP_NOFS);
6478 unlock_page(page);
6479 btrfs_start_ordered_extent(inode, ordered, 1);
6480 btrfs_put_ordered_extent(ordered);
6481 goto again;
6482 }
6483
6484 /*
6485 * XXX - page_mkwrite gets called every time the page is dirtied, even
6486 * if it was already dirty, so for space accounting reasons we need to
6487 * clear any delalloc bits for the range we are fixing to save. There
6488 * is probably a better way to do this, but for now keep consistent with
6489 * prepare_pages in the normal write path.
6490 */
6491 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6492 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6493 0, 0, &cached_state, GFP_NOFS);
6494
6495 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6496 &cached_state);
6497 if (ret) {
6498 unlock_extent_cached(io_tree, page_start, page_end,
6499 &cached_state, GFP_NOFS);
6500 ret = VM_FAULT_SIGBUS;
6501 goto out_unlock;
6502 }
6503 ret = 0;
6504
6505 /* page is wholly or partially inside EOF */
6506 if (page_start + PAGE_CACHE_SIZE > size)
6507 zero_start = size & ~PAGE_CACHE_MASK;
6508 else
6509 zero_start = PAGE_CACHE_SIZE;
6510
6511 if (zero_start != PAGE_CACHE_SIZE) {
6512 kaddr = kmap(page);
6513 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6514 flush_dcache_page(page);
6515 kunmap(page);
6516 }
6517 ClearPageChecked(page);
6518 set_page_dirty(page);
6519 SetPageUptodate(page);
6520
6521 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6522 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6523
6524 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6525
6526 out_unlock:
6527 if (!ret)
6528 return VM_FAULT_LOCKED;
6529 unlock_page(page);
6530 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6531 out:
6532 return ret;
6533 }
6534
6535 static int btrfs_truncate(struct inode *inode)
6536 {
6537 struct btrfs_root *root = BTRFS_I(inode)->root;
6538 struct btrfs_block_rsv *rsv;
6539 int ret;
6540 int err = 0;
6541 struct btrfs_trans_handle *trans;
6542 unsigned long nr;
6543 u64 mask = root->sectorsize - 1;
6544
6545 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6546 if (ret)
6547 return ret;
6548
6549 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6550 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6551
6552 /*
6553 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6554 * 3 things going on here
6555 *
6556 * 1) We need to reserve space for our orphan item and the space to
6557 * delete our orphan item. Lord knows we don't want to have a dangling
6558 * orphan item because we didn't reserve space to remove it.
6559 *
6560 * 2) We need to reserve space to update our inode.
6561 *
6562 * 3) We need to have something to cache all the space that is going to
6563 * be free'd up by the truncate operation, but also have some slack
6564 * space reserved in case it uses space during the truncate (thank you
6565 * very much snapshotting).
6566 *
6567 * And we need these to all be seperate. The fact is we can use alot of
6568 * space doing the truncate, and we have no earthly idea how much space
6569 * we will use, so we need the truncate reservation to be seperate so it
6570 * doesn't end up using space reserved for updating the inode or
6571 * removing the orphan item. We also need to be able to stop the
6572 * transaction and start a new one, which means we need to be able to
6573 * update the inode several times, and we have no idea of knowing how
6574 * many times that will be, so we can't just reserve 1 item for the
6575 * entirety of the opration, so that has to be done seperately as well.
6576 * Then there is the orphan item, which does indeed need to be held on
6577 * to for the whole operation, and we need nobody to touch this reserved
6578 * space except the orphan code.
6579 *
6580 * So that leaves us with
6581 *
6582 * 1) root->orphan_block_rsv - for the orphan deletion.
6583 * 2) rsv - for the truncate reservation, which we will steal from the
6584 * transaction reservation.
6585 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6586 * updating the inode.
6587 */
6588 rsv = btrfs_alloc_block_rsv(root);
6589 if (!rsv)
6590 return -ENOMEM;
6591 btrfs_add_durable_block_rsv(root->fs_info, rsv);
6592
6593 trans = btrfs_start_transaction(root, 4);
6594 if (IS_ERR(trans)) {
6595 err = PTR_ERR(trans);
6596 goto out;
6597 }
6598
6599 /*
6600 * Reserve space for the truncate process. Truncate should be adding
6601 * space, but if there are snapshots it may end up using space.
6602 */
6603 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6604 BUG_ON(ret);
6605
6606 ret = btrfs_orphan_add(trans, inode);
6607 if (ret) {
6608 btrfs_end_transaction(trans, root);
6609 goto out;
6610 }
6611
6612 nr = trans->blocks_used;
6613 btrfs_end_transaction(trans, root);
6614 btrfs_btree_balance_dirty(root, nr);
6615
6616 /*
6617 * Ok so we've already migrated our bytes over for the truncate, so here
6618 * just reserve the one slot we need for updating the inode.
6619 */
6620 trans = btrfs_start_transaction(root, 1);
6621 if (IS_ERR(trans)) {
6622 err = PTR_ERR(trans);
6623 goto out;
6624 }
6625 trans->block_rsv = rsv;
6626
6627 /*
6628 * setattr is responsible for setting the ordered_data_close flag,
6629 * but that is only tested during the last file release. That
6630 * could happen well after the next commit, leaving a great big
6631 * window where new writes may get lost if someone chooses to write
6632 * to this file after truncating to zero
6633 *
6634 * The inode doesn't have any dirty data here, and so if we commit
6635 * this is a noop. If someone immediately starts writing to the inode
6636 * it is very likely we'll catch some of their writes in this
6637 * transaction, and the commit will find this file on the ordered
6638 * data list with good things to send down.
6639 *
6640 * This is a best effort solution, there is still a window where
6641 * using truncate to replace the contents of the file will
6642 * end up with a zero length file after a crash.
6643 */
6644 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6645 btrfs_add_ordered_operation(trans, root, inode);
6646
6647 while (1) {
6648 if (!trans) {
6649 trans = btrfs_start_transaction(root, 3);
6650 if (IS_ERR(trans)) {
6651 err = PTR_ERR(trans);
6652 goto out;
6653 }
6654
6655 ret = btrfs_truncate_reserve_metadata(trans, root,
6656 rsv);
6657 BUG_ON(ret);
6658
6659 trans->block_rsv = rsv;
6660 }
6661
6662 ret = btrfs_truncate_inode_items(trans, root, inode,
6663 inode->i_size,
6664 BTRFS_EXTENT_DATA_KEY);
6665 if (ret != -EAGAIN) {
6666 err = ret;
6667 break;
6668 }
6669
6670 trans->block_rsv = &root->fs_info->trans_block_rsv;
6671 ret = btrfs_update_inode(trans, root, inode);
6672 if (ret) {
6673 err = ret;
6674 break;
6675 }
6676
6677 nr = trans->blocks_used;
6678 btrfs_end_transaction(trans, root);
6679 trans = NULL;
6680 btrfs_btree_balance_dirty(root, nr);
6681 }
6682
6683 if (ret == 0 && inode->i_nlink > 0) {
6684 trans->block_rsv = root->orphan_block_rsv;
6685 ret = btrfs_orphan_del(trans, inode);
6686 if (ret)
6687 err = ret;
6688 } else if (ret && inode->i_nlink > 0) {
6689 /*
6690 * Failed to do the truncate, remove us from the in memory
6691 * orphan list.
6692 */
6693 ret = btrfs_orphan_del(NULL, inode);
6694 }
6695
6696 trans->block_rsv = &root->fs_info->trans_block_rsv;
6697 ret = btrfs_update_inode(trans, root, inode);
6698 if (ret && !err)
6699 err = ret;
6700
6701 nr = trans->blocks_used;
6702 ret = btrfs_end_transaction_throttle(trans, root);
6703 btrfs_btree_balance_dirty(root, nr);
6704
6705 out:
6706 btrfs_free_block_rsv(root, rsv);
6707
6708 if (ret && !err)
6709 err = ret;
6710
6711 return err;
6712 }
6713
6714 /*
6715 * create a new subvolume directory/inode (helper for the ioctl).
6716 */
6717 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6718 struct btrfs_root *new_root, u64 new_dirid)
6719 {
6720 struct inode *inode;
6721 int err;
6722 u64 index = 0;
6723
6724 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6725 new_dirid, S_IFDIR | 0700, &index);
6726 if (IS_ERR(inode))
6727 return PTR_ERR(inode);
6728 inode->i_op = &btrfs_dir_inode_operations;
6729 inode->i_fop = &btrfs_dir_file_operations;
6730
6731 inode->i_nlink = 1;
6732 btrfs_i_size_write(inode, 0);
6733
6734 err = btrfs_update_inode(trans, new_root, inode);
6735 BUG_ON(err);
6736
6737 iput(inode);
6738 return 0;
6739 }
6740
6741 struct inode *btrfs_alloc_inode(struct super_block *sb)
6742 {
6743 struct btrfs_inode *ei;
6744 struct inode *inode;
6745
6746 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6747 if (!ei)
6748 return NULL;
6749
6750 ei->root = NULL;
6751 ei->space_info = NULL;
6752 ei->generation = 0;
6753 ei->sequence = 0;
6754 ei->last_trans = 0;
6755 ei->last_sub_trans = 0;
6756 ei->logged_trans = 0;
6757 ei->delalloc_bytes = 0;
6758 ei->reserved_bytes = 0;
6759 ei->disk_i_size = 0;
6760 ei->flags = 0;
6761 ei->index_cnt = (u64)-1;
6762 ei->last_unlink_trans = 0;
6763
6764 spin_lock_init(&ei->lock);
6765 ei->outstanding_extents = 0;
6766 ei->reserved_extents = 0;
6767
6768 ei->ordered_data_close = 0;
6769 ei->orphan_meta_reserved = 0;
6770 ei->dummy_inode = 0;
6771 ei->in_defrag = 0;
6772 ei->force_compress = BTRFS_COMPRESS_NONE;
6773
6774 ei->delayed_node = NULL;
6775
6776 inode = &ei->vfs_inode;
6777 extent_map_tree_init(&ei->extent_tree);
6778 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6779 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6780 mutex_init(&ei->log_mutex);
6781 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6782 INIT_LIST_HEAD(&ei->i_orphan);
6783 INIT_LIST_HEAD(&ei->delalloc_inodes);
6784 INIT_LIST_HEAD(&ei->ordered_operations);
6785 RB_CLEAR_NODE(&ei->rb_node);
6786
6787 return inode;
6788 }
6789
6790 static void btrfs_i_callback(struct rcu_head *head)
6791 {
6792 struct inode *inode = container_of(head, struct inode, i_rcu);
6793 INIT_LIST_HEAD(&inode->i_dentry);
6794 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6795 }
6796
6797 void btrfs_destroy_inode(struct inode *inode)
6798 {
6799 struct btrfs_ordered_extent *ordered;
6800 struct btrfs_root *root = BTRFS_I(inode)->root;
6801
6802 WARN_ON(!list_empty(&inode->i_dentry));
6803 WARN_ON(inode->i_data.nrpages);
6804 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6805 WARN_ON(BTRFS_I(inode)->reserved_extents);
6806
6807 /*
6808 * This can happen where we create an inode, but somebody else also
6809 * created the same inode and we need to destroy the one we already
6810 * created.
6811 */
6812 if (!root)
6813 goto free;
6814
6815 /*
6816 * Make sure we're properly removed from the ordered operation
6817 * lists.
6818 */
6819 smp_mb();
6820 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6821 spin_lock(&root->fs_info->ordered_extent_lock);
6822 list_del_init(&BTRFS_I(inode)->ordered_operations);
6823 spin_unlock(&root->fs_info->ordered_extent_lock);
6824 }
6825
6826 spin_lock(&root->orphan_lock);
6827 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6828 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6829 (unsigned long long)btrfs_ino(inode));
6830 list_del_init(&BTRFS_I(inode)->i_orphan);
6831 }
6832 spin_unlock(&root->orphan_lock);
6833
6834 while (1) {
6835 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6836 if (!ordered)
6837 break;
6838 else {
6839 printk(KERN_ERR "btrfs found ordered "
6840 "extent %llu %llu on inode cleanup\n",
6841 (unsigned long long)ordered->file_offset,
6842 (unsigned long long)ordered->len);
6843 btrfs_remove_ordered_extent(inode, ordered);
6844 btrfs_put_ordered_extent(ordered);
6845 btrfs_put_ordered_extent(ordered);
6846 }
6847 }
6848 inode_tree_del(inode);
6849 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6850 free:
6851 btrfs_remove_delayed_node(inode);
6852 call_rcu(&inode->i_rcu, btrfs_i_callback);
6853 }
6854
6855 int btrfs_drop_inode(struct inode *inode)
6856 {
6857 struct btrfs_root *root = BTRFS_I(inode)->root;
6858
6859 if (btrfs_root_refs(&root->root_item) == 0 &&
6860 !btrfs_is_free_space_inode(root, inode))
6861 return 1;
6862 else
6863 return generic_drop_inode(inode);
6864 }
6865
6866 static void init_once(void *foo)
6867 {
6868 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6869
6870 inode_init_once(&ei->vfs_inode);
6871 }
6872
6873 void btrfs_destroy_cachep(void)
6874 {
6875 if (btrfs_inode_cachep)
6876 kmem_cache_destroy(btrfs_inode_cachep);
6877 if (btrfs_trans_handle_cachep)
6878 kmem_cache_destroy(btrfs_trans_handle_cachep);
6879 if (btrfs_transaction_cachep)
6880 kmem_cache_destroy(btrfs_transaction_cachep);
6881 if (btrfs_path_cachep)
6882 kmem_cache_destroy(btrfs_path_cachep);
6883 if (btrfs_free_space_cachep)
6884 kmem_cache_destroy(btrfs_free_space_cachep);
6885 }
6886
6887 int btrfs_init_cachep(void)
6888 {
6889 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6890 sizeof(struct btrfs_inode), 0,
6891 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6892 if (!btrfs_inode_cachep)
6893 goto fail;
6894
6895 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6896 sizeof(struct btrfs_trans_handle), 0,
6897 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6898 if (!btrfs_trans_handle_cachep)
6899 goto fail;
6900
6901 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6902 sizeof(struct btrfs_transaction), 0,
6903 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6904 if (!btrfs_transaction_cachep)
6905 goto fail;
6906
6907 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6908 sizeof(struct btrfs_path), 0,
6909 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6910 if (!btrfs_path_cachep)
6911 goto fail;
6912
6913 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6914 sizeof(struct btrfs_free_space), 0,
6915 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6916 if (!btrfs_free_space_cachep)
6917 goto fail;
6918
6919 return 0;
6920 fail:
6921 btrfs_destroy_cachep();
6922 return -ENOMEM;
6923 }
6924
6925 static int btrfs_getattr(struct vfsmount *mnt,
6926 struct dentry *dentry, struct kstat *stat)
6927 {
6928 struct inode *inode = dentry->d_inode;
6929 generic_fillattr(inode, stat);
6930 stat->dev = BTRFS_I(inode)->root->anon_dev;
6931 stat->blksize = PAGE_CACHE_SIZE;
6932 stat->blocks = (inode_get_bytes(inode) +
6933 BTRFS_I(inode)->delalloc_bytes) >> 9;
6934 return 0;
6935 }
6936
6937 /*
6938 * If a file is moved, it will inherit the cow and compression flags of the new
6939 * directory.
6940 */
6941 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6942 {
6943 struct btrfs_inode *b_dir = BTRFS_I(dir);
6944 struct btrfs_inode *b_inode = BTRFS_I(inode);
6945
6946 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6947 b_inode->flags |= BTRFS_INODE_NODATACOW;
6948 else
6949 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6950
6951 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6952 b_inode->flags |= BTRFS_INODE_COMPRESS;
6953 else
6954 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6955 }
6956
6957 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6958 struct inode *new_dir, struct dentry *new_dentry)
6959 {
6960 struct btrfs_trans_handle *trans;
6961 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6962 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6963 struct inode *new_inode = new_dentry->d_inode;
6964 struct inode *old_inode = old_dentry->d_inode;
6965 struct timespec ctime = CURRENT_TIME;
6966 u64 index = 0;
6967 u64 root_objectid;
6968 int ret;
6969 u64 old_ino = btrfs_ino(old_inode);
6970
6971 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6972 return -EPERM;
6973
6974 /* we only allow rename subvolume link between subvolumes */
6975 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6976 return -EXDEV;
6977
6978 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6979 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6980 return -ENOTEMPTY;
6981
6982 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6983 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6984 return -ENOTEMPTY;
6985 /*
6986 * we're using rename to replace one file with another.
6987 * and the replacement file is large. Start IO on it now so
6988 * we don't add too much work to the end of the transaction
6989 */
6990 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6991 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6992 filemap_flush(old_inode->i_mapping);
6993
6994 /* close the racy window with snapshot create/destroy ioctl */
6995 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6996 down_read(&root->fs_info->subvol_sem);
6997 /*
6998 * We want to reserve the absolute worst case amount of items. So if
6999 * both inodes are subvols and we need to unlink them then that would
7000 * require 4 item modifications, but if they are both normal inodes it
7001 * would require 5 item modifications, so we'll assume their normal
7002 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7003 * should cover the worst case number of items we'll modify.
7004 */
7005 trans = btrfs_start_transaction(root, 20);
7006 if (IS_ERR(trans)) {
7007 ret = PTR_ERR(trans);
7008 goto out_notrans;
7009 }
7010
7011 if (dest != root)
7012 btrfs_record_root_in_trans(trans, dest);
7013
7014 ret = btrfs_set_inode_index(new_dir, &index);
7015 if (ret)
7016 goto out_fail;
7017
7018 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7019 /* force full log commit if subvolume involved. */
7020 root->fs_info->last_trans_log_full_commit = trans->transid;
7021 } else {
7022 ret = btrfs_insert_inode_ref(trans, dest,
7023 new_dentry->d_name.name,
7024 new_dentry->d_name.len,
7025 old_ino,
7026 btrfs_ino(new_dir), index);
7027 if (ret)
7028 goto out_fail;
7029 /*
7030 * this is an ugly little race, but the rename is required
7031 * to make sure that if we crash, the inode is either at the
7032 * old name or the new one. pinning the log transaction lets
7033 * us make sure we don't allow a log commit to come in after
7034 * we unlink the name but before we add the new name back in.
7035 */
7036 btrfs_pin_log_trans(root);
7037 }
7038 /*
7039 * make sure the inode gets flushed if it is replacing
7040 * something.
7041 */
7042 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7043 btrfs_add_ordered_operation(trans, root, old_inode);
7044
7045 old_dir->i_ctime = old_dir->i_mtime = ctime;
7046 new_dir->i_ctime = new_dir->i_mtime = ctime;
7047 old_inode->i_ctime = ctime;
7048
7049 if (old_dentry->d_parent != new_dentry->d_parent)
7050 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7051
7052 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7053 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7054 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7055 old_dentry->d_name.name,
7056 old_dentry->d_name.len);
7057 } else {
7058 ret = __btrfs_unlink_inode(trans, root, old_dir,
7059 old_dentry->d_inode,
7060 old_dentry->d_name.name,
7061 old_dentry->d_name.len);
7062 if (!ret)
7063 ret = btrfs_update_inode(trans, root, old_inode);
7064 }
7065 BUG_ON(ret);
7066
7067 if (new_inode) {
7068 new_inode->i_ctime = CURRENT_TIME;
7069 if (unlikely(btrfs_ino(new_inode) ==
7070 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7071 root_objectid = BTRFS_I(new_inode)->location.objectid;
7072 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7073 root_objectid,
7074 new_dentry->d_name.name,
7075 new_dentry->d_name.len);
7076 BUG_ON(new_inode->i_nlink == 0);
7077 } else {
7078 ret = btrfs_unlink_inode(trans, dest, new_dir,
7079 new_dentry->d_inode,
7080 new_dentry->d_name.name,
7081 new_dentry->d_name.len);
7082 }
7083 BUG_ON(ret);
7084 if (new_inode->i_nlink == 0) {
7085 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7086 BUG_ON(ret);
7087 }
7088 }
7089
7090 fixup_inode_flags(new_dir, old_inode);
7091
7092 ret = btrfs_add_link(trans, new_dir, old_inode,
7093 new_dentry->d_name.name,
7094 new_dentry->d_name.len, 0, index);
7095 BUG_ON(ret);
7096
7097 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7098 struct dentry *parent = new_dentry->d_parent;
7099 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7100 btrfs_end_log_trans(root);
7101 }
7102 out_fail:
7103 btrfs_end_transaction_throttle(trans, root);
7104 out_notrans:
7105 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7106 up_read(&root->fs_info->subvol_sem);
7107
7108 return ret;
7109 }
7110
7111 /*
7112 * some fairly slow code that needs optimization. This walks the list
7113 * of all the inodes with pending delalloc and forces them to disk.
7114 */
7115 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7116 {
7117 struct list_head *head = &root->fs_info->delalloc_inodes;
7118 struct btrfs_inode *binode;
7119 struct inode *inode;
7120
7121 if (root->fs_info->sb->s_flags & MS_RDONLY)
7122 return -EROFS;
7123
7124 spin_lock(&root->fs_info->delalloc_lock);
7125 while (!list_empty(head)) {
7126 binode = list_entry(head->next, struct btrfs_inode,
7127 delalloc_inodes);
7128 inode = igrab(&binode->vfs_inode);
7129 if (!inode)
7130 list_del_init(&binode->delalloc_inodes);
7131 spin_unlock(&root->fs_info->delalloc_lock);
7132 if (inode) {
7133 filemap_flush(inode->i_mapping);
7134 if (delay_iput)
7135 btrfs_add_delayed_iput(inode);
7136 else
7137 iput(inode);
7138 }
7139 cond_resched();
7140 spin_lock(&root->fs_info->delalloc_lock);
7141 }
7142 spin_unlock(&root->fs_info->delalloc_lock);
7143
7144 /* the filemap_flush will queue IO into the worker threads, but
7145 * we have to make sure the IO is actually started and that
7146 * ordered extents get created before we return
7147 */
7148 atomic_inc(&root->fs_info->async_submit_draining);
7149 while (atomic_read(&root->fs_info->nr_async_submits) ||
7150 atomic_read(&root->fs_info->async_delalloc_pages)) {
7151 wait_event(root->fs_info->async_submit_wait,
7152 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7153 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7154 }
7155 atomic_dec(&root->fs_info->async_submit_draining);
7156 return 0;
7157 }
7158
7159 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7160 const char *symname)
7161 {
7162 struct btrfs_trans_handle *trans;
7163 struct btrfs_root *root = BTRFS_I(dir)->root;
7164 struct btrfs_path *path;
7165 struct btrfs_key key;
7166 struct inode *inode = NULL;
7167 int err;
7168 int drop_inode = 0;
7169 u64 objectid;
7170 u64 index = 0 ;
7171 int name_len;
7172 int datasize;
7173 unsigned long ptr;
7174 struct btrfs_file_extent_item *ei;
7175 struct extent_buffer *leaf;
7176 unsigned long nr = 0;
7177
7178 name_len = strlen(symname) + 1;
7179 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7180 return -ENAMETOOLONG;
7181
7182 /*
7183 * 2 items for inode item and ref
7184 * 2 items for dir items
7185 * 1 item for xattr if selinux is on
7186 */
7187 trans = btrfs_start_transaction(root, 5);
7188 if (IS_ERR(trans))
7189 return PTR_ERR(trans);
7190
7191 err = btrfs_find_free_ino(root, &objectid);
7192 if (err)
7193 goto out_unlock;
7194
7195 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7196 dentry->d_name.len, btrfs_ino(dir), objectid,
7197 S_IFLNK|S_IRWXUGO, &index);
7198 if (IS_ERR(inode)) {
7199 err = PTR_ERR(inode);
7200 goto out_unlock;
7201 }
7202
7203 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7204 if (err) {
7205 drop_inode = 1;
7206 goto out_unlock;
7207 }
7208
7209 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7210 if (err)
7211 drop_inode = 1;
7212 else {
7213 inode->i_mapping->a_ops = &btrfs_aops;
7214 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7215 inode->i_fop = &btrfs_file_operations;
7216 inode->i_op = &btrfs_file_inode_operations;
7217 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7218 }
7219 if (drop_inode)
7220 goto out_unlock;
7221
7222 path = btrfs_alloc_path();
7223 if (!path) {
7224 err = -ENOMEM;
7225 drop_inode = 1;
7226 goto out_unlock;
7227 }
7228 key.objectid = btrfs_ino(inode);
7229 key.offset = 0;
7230 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7231 datasize = btrfs_file_extent_calc_inline_size(name_len);
7232 err = btrfs_insert_empty_item(trans, root, path, &key,
7233 datasize);
7234 if (err) {
7235 drop_inode = 1;
7236 btrfs_free_path(path);
7237 goto out_unlock;
7238 }
7239 leaf = path->nodes[0];
7240 ei = btrfs_item_ptr(leaf, path->slots[0],
7241 struct btrfs_file_extent_item);
7242 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7243 btrfs_set_file_extent_type(leaf, ei,
7244 BTRFS_FILE_EXTENT_INLINE);
7245 btrfs_set_file_extent_encryption(leaf, ei, 0);
7246 btrfs_set_file_extent_compression(leaf, ei, 0);
7247 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7248 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7249
7250 ptr = btrfs_file_extent_inline_start(ei);
7251 write_extent_buffer(leaf, symname, ptr, name_len);
7252 btrfs_mark_buffer_dirty(leaf);
7253 btrfs_free_path(path);
7254
7255 inode->i_op = &btrfs_symlink_inode_operations;
7256 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7257 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7258 inode_set_bytes(inode, name_len);
7259 btrfs_i_size_write(inode, name_len - 1);
7260 err = btrfs_update_inode(trans, root, inode);
7261 if (err)
7262 drop_inode = 1;
7263
7264 out_unlock:
7265 nr = trans->blocks_used;
7266 btrfs_end_transaction_throttle(trans, root);
7267 if (drop_inode) {
7268 inode_dec_link_count(inode);
7269 iput(inode);
7270 }
7271 btrfs_btree_balance_dirty(root, nr);
7272 return err;
7273 }
7274
7275 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7276 u64 start, u64 num_bytes, u64 min_size,
7277 loff_t actual_len, u64 *alloc_hint,
7278 struct btrfs_trans_handle *trans)
7279 {
7280 struct btrfs_root *root = BTRFS_I(inode)->root;
7281 struct btrfs_key ins;
7282 u64 cur_offset = start;
7283 u64 i_size;
7284 int ret = 0;
7285 bool own_trans = true;
7286
7287 if (trans)
7288 own_trans = false;
7289 while (num_bytes > 0) {
7290 if (own_trans) {
7291 trans = btrfs_start_transaction(root, 3);
7292 if (IS_ERR(trans)) {
7293 ret = PTR_ERR(trans);
7294 break;
7295 }
7296 }
7297
7298 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7299 0, *alloc_hint, (u64)-1, &ins, 1);
7300 if (ret) {
7301 if (own_trans)
7302 btrfs_end_transaction(trans, root);
7303 break;
7304 }
7305
7306 ret = insert_reserved_file_extent(trans, inode,
7307 cur_offset, ins.objectid,
7308 ins.offset, ins.offset,
7309 ins.offset, 0, 0, 0,
7310 BTRFS_FILE_EXTENT_PREALLOC);
7311 BUG_ON(ret);
7312 btrfs_drop_extent_cache(inode, cur_offset,
7313 cur_offset + ins.offset -1, 0);
7314
7315 num_bytes -= ins.offset;
7316 cur_offset += ins.offset;
7317 *alloc_hint = ins.objectid + ins.offset;
7318
7319 inode->i_ctime = CURRENT_TIME;
7320 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7321 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7322 (actual_len > inode->i_size) &&
7323 (cur_offset > inode->i_size)) {
7324 if (cur_offset > actual_len)
7325 i_size = actual_len;
7326 else
7327 i_size = cur_offset;
7328 i_size_write(inode, i_size);
7329 btrfs_ordered_update_i_size(inode, i_size, NULL);
7330 }
7331
7332 ret = btrfs_update_inode(trans, root, inode);
7333 BUG_ON(ret);
7334
7335 if (own_trans)
7336 btrfs_end_transaction(trans, root);
7337 }
7338 return ret;
7339 }
7340
7341 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7342 u64 start, u64 num_bytes, u64 min_size,
7343 loff_t actual_len, u64 *alloc_hint)
7344 {
7345 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7346 min_size, actual_len, alloc_hint,
7347 NULL);
7348 }
7349
7350 int btrfs_prealloc_file_range_trans(struct inode *inode,
7351 struct btrfs_trans_handle *trans, int mode,
7352 u64 start, u64 num_bytes, u64 min_size,
7353 loff_t actual_len, u64 *alloc_hint)
7354 {
7355 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7356 min_size, actual_len, alloc_hint, trans);
7357 }
7358
7359 static int btrfs_set_page_dirty(struct page *page)
7360 {
7361 return __set_page_dirty_nobuffers(page);
7362 }
7363
7364 static int btrfs_permission(struct inode *inode, int mask)
7365 {
7366 struct btrfs_root *root = BTRFS_I(inode)->root;
7367 umode_t mode = inode->i_mode;
7368
7369 if (mask & MAY_WRITE &&
7370 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7371 if (btrfs_root_readonly(root))
7372 return -EROFS;
7373 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7374 return -EACCES;
7375 }
7376 return generic_permission(inode, mask);
7377 }
7378
7379 static const struct inode_operations btrfs_dir_inode_operations = {
7380 .getattr = btrfs_getattr,
7381 .lookup = btrfs_lookup,
7382 .create = btrfs_create,
7383 .unlink = btrfs_unlink,
7384 .link = btrfs_link,
7385 .mkdir = btrfs_mkdir,
7386 .rmdir = btrfs_rmdir,
7387 .rename = btrfs_rename,
7388 .symlink = btrfs_symlink,
7389 .setattr = btrfs_setattr,
7390 .mknod = btrfs_mknod,
7391 .setxattr = btrfs_setxattr,
7392 .getxattr = btrfs_getxattr,
7393 .listxattr = btrfs_listxattr,
7394 .removexattr = btrfs_removexattr,
7395 .permission = btrfs_permission,
7396 .get_acl = btrfs_get_acl,
7397 };
7398 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7399 .lookup = btrfs_lookup,
7400 .permission = btrfs_permission,
7401 .get_acl = btrfs_get_acl,
7402 };
7403
7404 static const struct file_operations btrfs_dir_file_operations = {
7405 .llseek = generic_file_llseek,
7406 .read = generic_read_dir,
7407 .readdir = btrfs_real_readdir,
7408 .unlocked_ioctl = btrfs_ioctl,
7409 #ifdef CONFIG_COMPAT
7410 .compat_ioctl = btrfs_ioctl,
7411 #endif
7412 .release = btrfs_release_file,
7413 .fsync = btrfs_sync_file,
7414 };
7415
7416 static struct extent_io_ops btrfs_extent_io_ops = {
7417 .fill_delalloc = run_delalloc_range,
7418 .submit_bio_hook = btrfs_submit_bio_hook,
7419 .merge_bio_hook = btrfs_merge_bio_hook,
7420 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7421 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7422 .writepage_start_hook = btrfs_writepage_start_hook,
7423 .readpage_io_failed_hook = btrfs_io_failed_hook,
7424 .set_bit_hook = btrfs_set_bit_hook,
7425 .clear_bit_hook = btrfs_clear_bit_hook,
7426 .merge_extent_hook = btrfs_merge_extent_hook,
7427 .split_extent_hook = btrfs_split_extent_hook,
7428 };
7429
7430 /*
7431 * btrfs doesn't support the bmap operation because swapfiles
7432 * use bmap to make a mapping of extents in the file. They assume
7433 * these extents won't change over the life of the file and they
7434 * use the bmap result to do IO directly to the drive.
7435 *
7436 * the btrfs bmap call would return logical addresses that aren't
7437 * suitable for IO and they also will change frequently as COW
7438 * operations happen. So, swapfile + btrfs == corruption.
7439 *
7440 * For now we're avoiding this by dropping bmap.
7441 */
7442 static const struct address_space_operations btrfs_aops = {
7443 .readpage = btrfs_readpage,
7444 .writepage = btrfs_writepage,
7445 .writepages = btrfs_writepages,
7446 .readpages = btrfs_readpages,
7447 .direct_IO = btrfs_direct_IO,
7448 .invalidatepage = btrfs_invalidatepage,
7449 .releasepage = btrfs_releasepage,
7450 .set_page_dirty = btrfs_set_page_dirty,
7451 .error_remove_page = generic_error_remove_page,
7452 };
7453
7454 static const struct address_space_operations btrfs_symlink_aops = {
7455 .readpage = btrfs_readpage,
7456 .writepage = btrfs_writepage,
7457 .invalidatepage = btrfs_invalidatepage,
7458 .releasepage = btrfs_releasepage,
7459 };
7460
7461 static const struct inode_operations btrfs_file_inode_operations = {
7462 .getattr = btrfs_getattr,
7463 .setattr = btrfs_setattr,
7464 .setxattr = btrfs_setxattr,
7465 .getxattr = btrfs_getxattr,
7466 .listxattr = btrfs_listxattr,
7467 .removexattr = btrfs_removexattr,
7468 .permission = btrfs_permission,
7469 .fiemap = btrfs_fiemap,
7470 .get_acl = btrfs_get_acl,
7471 };
7472 static const struct inode_operations btrfs_special_inode_operations = {
7473 .getattr = btrfs_getattr,
7474 .setattr = btrfs_setattr,
7475 .permission = btrfs_permission,
7476 .setxattr = btrfs_setxattr,
7477 .getxattr = btrfs_getxattr,
7478 .listxattr = btrfs_listxattr,
7479 .removexattr = btrfs_removexattr,
7480 .get_acl = btrfs_get_acl,
7481 };
7482 static const struct inode_operations btrfs_symlink_inode_operations = {
7483 .readlink = generic_readlink,
7484 .follow_link = page_follow_link_light,
7485 .put_link = page_put_link,
7486 .getattr = btrfs_getattr,
7487 .permission = btrfs_permission,
7488 .setxattr = btrfs_setxattr,
7489 .getxattr = btrfs_getxattr,
7490 .listxattr = btrfs_listxattr,
7491 .removexattr = btrfs_removexattr,
7492 .get_acl = btrfs_get_acl,
7493 };
7494
7495 const struct dentry_operations btrfs_dentry_operations = {
7496 .d_delete = btrfs_dentry_delete,
7497 .d_release = btrfs_dentry_release,
7498 };
Linux Btrfs Development