Merge commit 'jwb/next' into next
[deliverable/linux.git] / fs / btrfs / inode.c
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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "ref-cache.h"
52 #include "compression.h"
53 #include "locking.h"
54
55 struct btrfs_iget_args {
56 u64 ino;
57 struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 };
86
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
93
94 static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
95 {
96 int err;
97
98 err = btrfs_init_acl(inode, dir);
99 if (!err)
100 err = btrfs_xattr_security_init(inode, dir);
101 return err;
102 }
103
104 /*
105 * a very lame attempt at stopping writes when the FS is 85% full. There
106 * are countless ways this is incorrect, but it is better than nothing.
107 */
108 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
109 int for_del)
110 {
111 u64 total;
112 u64 used;
113 u64 thresh;
114 int ret = 0;
115
116 spin_lock(&root->fs_info->delalloc_lock);
117 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
118 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
119 if (for_del)
120 thresh = total * 90;
121 else
122 thresh = total * 85;
123
124 do_div(thresh, 100);
125
126 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
127 ret = -ENOSPC;
128 spin_unlock(&root->fs_info->delalloc_lock);
129 return ret;
130 }
131
132 /*
133 * this does all the hard work for inserting an inline extent into
134 * the btree. The caller should have done a btrfs_drop_extents so that
135 * no overlapping inline items exist in the btree
136 */
137 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
140 struct page **compressed_pages)
141 {
142 struct btrfs_key key;
143 struct btrfs_path *path;
144 struct extent_buffer *leaf;
145 struct page *page = NULL;
146 char *kaddr;
147 unsigned long ptr;
148 struct btrfs_file_extent_item *ei;
149 int err = 0;
150 int ret;
151 size_t cur_size = size;
152 size_t datasize;
153 unsigned long offset;
154 int use_compress = 0;
155
156 if (compressed_size && compressed_pages) {
157 use_compress = 1;
158 cur_size = compressed_size;
159 }
160
161 path = btrfs_alloc_path();
162 if (!path)
163 return -ENOMEM;
164
165 btrfs_set_trans_block_group(trans, inode);
166
167 key.objectid = inode->i_ino;
168 key.offset = start;
169 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
170 datasize = btrfs_file_extent_calc_inline_size(cur_size);
171
172 inode_add_bytes(inode, size);
173 ret = btrfs_insert_empty_item(trans, root, path, &key,
174 datasize);
175 BUG_ON(ret);
176 if (ret) {
177 err = ret;
178 goto fail;
179 }
180 leaf = path->nodes[0];
181 ei = btrfs_item_ptr(leaf, path->slots[0],
182 struct btrfs_file_extent_item);
183 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
184 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
185 btrfs_set_file_extent_encryption(leaf, ei, 0);
186 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
187 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
188 ptr = btrfs_file_extent_inline_start(ei);
189
190 if (use_compress) {
191 struct page *cpage;
192 int i = 0;
193 while (compressed_size > 0) {
194 cpage = compressed_pages[i];
195 cur_size = min_t(unsigned long, compressed_size,
196 PAGE_CACHE_SIZE);
197
198 kaddr = kmap(cpage);
199 write_extent_buffer(leaf, kaddr, ptr, cur_size);
200 kunmap(cpage);
201
202 i++;
203 ptr += cur_size;
204 compressed_size -= cur_size;
205 }
206 btrfs_set_file_extent_compression(leaf, ei,
207 BTRFS_COMPRESS_ZLIB);
208 } else {
209 page = find_get_page(inode->i_mapping,
210 start >> PAGE_CACHE_SHIFT);
211 btrfs_set_file_extent_compression(leaf, ei, 0);
212 kaddr = kmap_atomic(page, KM_USER0);
213 offset = start & (PAGE_CACHE_SIZE - 1);
214 write_extent_buffer(leaf, kaddr + offset, ptr, size);
215 kunmap_atomic(kaddr, KM_USER0);
216 page_cache_release(page);
217 }
218 btrfs_mark_buffer_dirty(leaf);
219 btrfs_free_path(path);
220
221 BTRFS_I(inode)->disk_i_size = inode->i_size;
222 btrfs_update_inode(trans, root, inode);
223 return 0;
224 fail:
225 btrfs_free_path(path);
226 return err;
227 }
228
229
230 /*
231 * conditionally insert an inline extent into the file. This
232 * does the checks required to make sure the data is small enough
233 * to fit as an inline extent.
234 */
235 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
236 struct btrfs_root *root,
237 struct inode *inode, u64 start, u64 end,
238 size_t compressed_size,
239 struct page **compressed_pages)
240 {
241 u64 isize = i_size_read(inode);
242 u64 actual_end = min(end + 1, isize);
243 u64 inline_len = actual_end - start;
244 u64 aligned_end = (end + root->sectorsize - 1) &
245 ~((u64)root->sectorsize - 1);
246 u64 hint_byte;
247 u64 data_len = inline_len;
248 int ret;
249
250 if (compressed_size)
251 data_len = compressed_size;
252
253 if (start > 0 ||
254 actual_end >= PAGE_CACHE_SIZE ||
255 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
256 (!compressed_size &&
257 (actual_end & (root->sectorsize - 1)) == 0) ||
258 end + 1 < isize ||
259 data_len > root->fs_info->max_inline) {
260 return 1;
261 }
262
263 ret = btrfs_drop_extents(trans, root, inode, start,
264 aligned_end, start, &hint_byte);
265 BUG_ON(ret);
266
267 if (isize > actual_end)
268 inline_len = min_t(u64, isize, actual_end);
269 ret = insert_inline_extent(trans, root, inode, start,
270 inline_len, compressed_size,
271 compressed_pages);
272 BUG_ON(ret);
273 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
274 return 0;
275 }
276
277 struct async_extent {
278 u64 start;
279 u64 ram_size;
280 u64 compressed_size;
281 struct page **pages;
282 unsigned long nr_pages;
283 struct list_head list;
284 };
285
286 struct async_cow {
287 struct inode *inode;
288 struct btrfs_root *root;
289 struct page *locked_page;
290 u64 start;
291 u64 end;
292 struct list_head extents;
293 struct btrfs_work work;
294 };
295
296 static noinline int add_async_extent(struct async_cow *cow,
297 u64 start, u64 ram_size,
298 u64 compressed_size,
299 struct page **pages,
300 unsigned long nr_pages)
301 {
302 struct async_extent *async_extent;
303
304 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
305 async_extent->start = start;
306 async_extent->ram_size = ram_size;
307 async_extent->compressed_size = compressed_size;
308 async_extent->pages = pages;
309 async_extent->nr_pages = nr_pages;
310 list_add_tail(&async_extent->list, &cow->extents);
311 return 0;
312 }
313
314 /*
315 * we create compressed extents in two phases. The first
316 * phase compresses a range of pages that have already been
317 * locked (both pages and state bits are locked).
318 *
319 * This is done inside an ordered work queue, and the compression
320 * is spread across many cpus. The actual IO submission is step
321 * two, and the ordered work queue takes care of making sure that
322 * happens in the same order things were put onto the queue by
323 * writepages and friends.
324 *
325 * If this code finds it can't get good compression, it puts an
326 * entry onto the work queue to write the uncompressed bytes. This
327 * makes sure that both compressed inodes and uncompressed inodes
328 * are written in the same order that pdflush sent them down.
329 */
330 static noinline int compress_file_range(struct inode *inode,
331 struct page *locked_page,
332 u64 start, u64 end,
333 struct async_cow *async_cow,
334 int *num_added)
335 {
336 struct btrfs_root *root = BTRFS_I(inode)->root;
337 struct btrfs_trans_handle *trans;
338 u64 num_bytes;
339 u64 orig_start;
340 u64 disk_num_bytes;
341 u64 blocksize = root->sectorsize;
342 u64 actual_end;
343 u64 isize = i_size_read(inode);
344 int ret = 0;
345 struct page **pages = NULL;
346 unsigned long nr_pages;
347 unsigned long nr_pages_ret = 0;
348 unsigned long total_compressed = 0;
349 unsigned long total_in = 0;
350 unsigned long max_compressed = 128 * 1024;
351 unsigned long max_uncompressed = 128 * 1024;
352 int i;
353 int will_compress;
354
355 orig_start = start;
356
357 actual_end = min_t(u64, isize, end + 1);
358 again:
359 will_compress = 0;
360 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
361 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
362
363 /*
364 * we don't want to send crud past the end of i_size through
365 * compression, that's just a waste of CPU time. So, if the
366 * end of the file is before the start of our current
367 * requested range of bytes, we bail out to the uncompressed
368 * cleanup code that can deal with all of this.
369 *
370 * It isn't really the fastest way to fix things, but this is a
371 * very uncommon corner.
372 */
373 if (actual_end <= start)
374 goto cleanup_and_bail_uncompressed;
375
376 total_compressed = actual_end - start;
377
378 /* we want to make sure that amount of ram required to uncompress
379 * an extent is reasonable, so we limit the total size in ram
380 * of a compressed extent to 128k. This is a crucial number
381 * because it also controls how easily we can spread reads across
382 * cpus for decompression.
383 *
384 * We also want to make sure the amount of IO required to do
385 * a random read is reasonably small, so we limit the size of
386 * a compressed extent to 128k.
387 */
388 total_compressed = min(total_compressed, max_uncompressed);
389 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
390 num_bytes = max(blocksize, num_bytes);
391 disk_num_bytes = num_bytes;
392 total_in = 0;
393 ret = 0;
394
395 /*
396 * we do compression for mount -o compress and when the
397 * inode has not been flagged as nocompress. This flag can
398 * change at any time if we discover bad compression ratios.
399 */
400 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
401 btrfs_test_opt(root, COMPRESS)) {
402 WARN_ON(pages);
403 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
404
405 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
406 total_compressed, pages,
407 nr_pages, &nr_pages_ret,
408 &total_in,
409 &total_compressed,
410 max_compressed);
411
412 if (!ret) {
413 unsigned long offset = total_compressed &
414 (PAGE_CACHE_SIZE - 1);
415 struct page *page = pages[nr_pages_ret - 1];
416 char *kaddr;
417
418 /* zero the tail end of the last page, we might be
419 * sending it down to disk
420 */
421 if (offset) {
422 kaddr = kmap_atomic(page, KM_USER0);
423 memset(kaddr + offset, 0,
424 PAGE_CACHE_SIZE - offset);
425 kunmap_atomic(kaddr, KM_USER0);
426 }
427 will_compress = 1;
428 }
429 }
430 if (start == 0) {
431 trans = btrfs_join_transaction(root, 1);
432 BUG_ON(!trans);
433 btrfs_set_trans_block_group(trans, inode);
434
435 /* lets try to make an inline extent */
436 if (ret || total_in < (actual_end - start)) {
437 /* we didn't compress the entire range, try
438 * to make an uncompressed inline extent.
439 */
440 ret = cow_file_range_inline(trans, root, inode,
441 start, end, 0, NULL);
442 } else {
443 /* try making a compressed inline extent */
444 ret = cow_file_range_inline(trans, root, inode,
445 start, end,
446 total_compressed, pages);
447 }
448 btrfs_end_transaction(trans, root);
449 if (ret == 0) {
450 /*
451 * inline extent creation worked, we don't need
452 * to create any more async work items. Unlock
453 * and free up our temp pages.
454 */
455 extent_clear_unlock_delalloc(inode,
456 &BTRFS_I(inode)->io_tree,
457 start, end, NULL, 1, 0,
458 0, 1, 1, 1);
459 ret = 0;
460 goto free_pages_out;
461 }
462 }
463
464 if (will_compress) {
465 /*
466 * we aren't doing an inline extent round the compressed size
467 * up to a block size boundary so the allocator does sane
468 * things
469 */
470 total_compressed = (total_compressed + blocksize - 1) &
471 ~(blocksize - 1);
472
473 /*
474 * one last check to make sure the compression is really a
475 * win, compare the page count read with the blocks on disk
476 */
477 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
478 ~(PAGE_CACHE_SIZE - 1);
479 if (total_compressed >= total_in) {
480 will_compress = 0;
481 } else {
482 disk_num_bytes = total_compressed;
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 btrfs_set_flag(inode, NOCOMPRESS);
502 }
503 if (will_compress) {
504 *num_added += 1;
505
506 /* the async work queues will take care of doing actual
507 * allocation on disk for these compressed pages,
508 * and will submit them to the elevator.
509 */
510 add_async_extent(async_cow, start, num_bytes,
511 total_compressed, pages, nr_pages_ret);
512
513 if (start + num_bytes < end && start + num_bytes < actual_end) {
514 start += num_bytes;
515 pages = NULL;
516 cond_resched();
517 goto again;
518 }
519 } else {
520 cleanup_and_bail_uncompressed:
521 /*
522 * No compression, but we still need to write the pages in
523 * the file we've been given so far. redirty the locked
524 * page if it corresponds to our extent and set things up
525 * for the async work queue to run cow_file_range to do
526 * the normal delalloc dance
527 */
528 if (page_offset(locked_page) >= start &&
529 page_offset(locked_page) <= end) {
530 __set_page_dirty_nobuffers(locked_page);
531 /* unlocked later on in the async handlers */
532 }
533 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
534 *num_added += 1;
535 }
536
537 out:
538 return 0;
539
540 free_pages_out:
541 for (i = 0; i < nr_pages_ret; i++) {
542 WARN_ON(pages[i]->mapping);
543 page_cache_release(pages[i]);
544 }
545 kfree(pages);
546
547 goto out;
548 }
549
550 /*
551 * phase two of compressed writeback. This is the ordered portion
552 * of the code, which only gets called in the order the work was
553 * queued. We walk all the async extents created by compress_file_range
554 * and send them down to the disk.
555 */
556 static noinline int submit_compressed_extents(struct inode *inode,
557 struct async_cow *async_cow)
558 {
559 struct async_extent *async_extent;
560 u64 alloc_hint = 0;
561 struct btrfs_trans_handle *trans;
562 struct btrfs_key ins;
563 struct extent_map *em;
564 struct btrfs_root *root = BTRFS_I(inode)->root;
565 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
566 struct extent_io_tree *io_tree;
567 int ret;
568
569 if (list_empty(&async_cow->extents))
570 return 0;
571
572 trans = btrfs_join_transaction(root, 1);
573
574 while (!list_empty(&async_cow->extents)) {
575 async_extent = list_entry(async_cow->extents.next,
576 struct async_extent, list);
577 list_del(&async_extent->list);
578
579 io_tree = &BTRFS_I(inode)->io_tree;
580
581 /* did the compression code fall back to uncompressed IO? */
582 if (!async_extent->pages) {
583 int page_started = 0;
584 unsigned long nr_written = 0;
585
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start +
588 async_extent->ram_size - 1, GFP_NOFS);
589
590 /* allocate blocks */
591 cow_file_range(inode, async_cow->locked_page,
592 async_extent->start,
593 async_extent->start +
594 async_extent->ram_size - 1,
595 &page_started, &nr_written, 0);
596
597 /*
598 * if page_started, cow_file_range inserted an
599 * inline extent and took care of all the unlocking
600 * and IO for us. Otherwise, we need to submit
601 * all those pages down to the drive.
602 */
603 if (!page_started)
604 extent_write_locked_range(io_tree,
605 inode, async_extent->start,
606 async_extent->start +
607 async_extent->ram_size - 1,
608 btrfs_get_extent,
609 WB_SYNC_ALL);
610 kfree(async_extent);
611 cond_resched();
612 continue;
613 }
614
615 lock_extent(io_tree, async_extent->start,
616 async_extent->start + async_extent->ram_size - 1,
617 GFP_NOFS);
618 /*
619 * here we're doing allocation and writeback of the
620 * compressed pages
621 */
622 btrfs_drop_extent_cache(inode, async_extent->start,
623 async_extent->start +
624 async_extent->ram_size - 1, 0);
625
626 ret = btrfs_reserve_extent(trans, root,
627 async_extent->compressed_size,
628 async_extent->compressed_size,
629 0, alloc_hint,
630 (u64)-1, &ins, 1);
631 BUG_ON(ret);
632 em = alloc_extent_map(GFP_NOFS);
633 em->start = async_extent->start;
634 em->len = async_extent->ram_size;
635 em->orig_start = em->start;
636
637 em->block_start = ins.objectid;
638 em->block_len = ins.offset;
639 em->bdev = root->fs_info->fs_devices->latest_bdev;
640 set_bit(EXTENT_FLAG_PINNED, &em->flags);
641 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
642
643 while (1) {
644 spin_lock(&em_tree->lock);
645 ret = add_extent_mapping(em_tree, em);
646 spin_unlock(&em_tree->lock);
647 if (ret != -EEXIST) {
648 free_extent_map(em);
649 break;
650 }
651 btrfs_drop_extent_cache(inode, async_extent->start,
652 async_extent->start +
653 async_extent->ram_size - 1, 0);
654 }
655
656 ret = btrfs_add_ordered_extent(inode, async_extent->start,
657 ins.objectid,
658 async_extent->ram_size,
659 ins.offset,
660 BTRFS_ORDERED_COMPRESSED);
661 BUG_ON(ret);
662
663 btrfs_end_transaction(trans, root);
664
665 /*
666 * clear dirty, set writeback and unlock the pages.
667 */
668 extent_clear_unlock_delalloc(inode,
669 &BTRFS_I(inode)->io_tree,
670 async_extent->start,
671 async_extent->start +
672 async_extent->ram_size - 1,
673 NULL, 1, 1, 0, 1, 1, 0);
674
675 ret = btrfs_submit_compressed_write(inode,
676 async_extent->start,
677 async_extent->ram_size,
678 ins.objectid,
679 ins.offset, async_extent->pages,
680 async_extent->nr_pages);
681
682 BUG_ON(ret);
683 trans = btrfs_join_transaction(root, 1);
684 alloc_hint = ins.objectid + ins.offset;
685 kfree(async_extent);
686 cond_resched();
687 }
688
689 btrfs_end_transaction(trans, root);
690 return 0;
691 }
692
693 /*
694 * when extent_io.c finds a delayed allocation range in the file,
695 * the call backs end up in this code. The basic idea is to
696 * allocate extents on disk for the range, and create ordered data structs
697 * in ram to track those extents.
698 *
699 * locked_page is the page that writepage had locked already. We use
700 * it to make sure we don't do extra locks or unlocks.
701 *
702 * *page_started is set to one if we unlock locked_page and do everything
703 * required to start IO on it. It may be clean and already done with
704 * IO when we return.
705 */
706 static noinline int cow_file_range(struct inode *inode,
707 struct page *locked_page,
708 u64 start, u64 end, int *page_started,
709 unsigned long *nr_written,
710 int unlock)
711 {
712 struct btrfs_root *root = BTRFS_I(inode)->root;
713 struct btrfs_trans_handle *trans;
714 u64 alloc_hint = 0;
715 u64 num_bytes;
716 unsigned long ram_size;
717 u64 disk_num_bytes;
718 u64 cur_alloc_size;
719 u64 blocksize = root->sectorsize;
720 u64 actual_end;
721 u64 isize = i_size_read(inode);
722 struct btrfs_key ins;
723 struct extent_map *em;
724 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
725 int ret = 0;
726
727 trans = btrfs_join_transaction(root, 1);
728 BUG_ON(!trans);
729 btrfs_set_trans_block_group(trans, inode);
730
731 actual_end = min_t(u64, isize, end + 1);
732
733 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
734 num_bytes = max(blocksize, num_bytes);
735 disk_num_bytes = num_bytes;
736 ret = 0;
737
738 if (start == 0) {
739 /* lets try to make an inline extent */
740 ret = cow_file_range_inline(trans, root, inode,
741 start, end, 0, NULL);
742 if (ret == 0) {
743 extent_clear_unlock_delalloc(inode,
744 &BTRFS_I(inode)->io_tree,
745 start, end, NULL, 1, 1,
746 1, 1, 1, 1);
747 *nr_written = *nr_written +
748 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
749 *page_started = 1;
750 ret = 0;
751 goto out;
752 }
753 }
754
755 BUG_ON(disk_num_bytes >
756 btrfs_super_total_bytes(&root->fs_info->super_copy));
757
758 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
759
760 while (disk_num_bytes > 0) {
761 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
762 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
763 root->sectorsize, 0, alloc_hint,
764 (u64)-1, &ins, 1);
765 BUG_ON(ret);
766
767 em = alloc_extent_map(GFP_NOFS);
768 em->start = start;
769 em->orig_start = em->start;
770
771 ram_size = ins.offset;
772 em->len = ins.offset;
773
774 em->block_start = ins.objectid;
775 em->block_len = ins.offset;
776 em->bdev = root->fs_info->fs_devices->latest_bdev;
777 set_bit(EXTENT_FLAG_PINNED, &em->flags);
778
779 while (1) {
780 spin_lock(&em_tree->lock);
781 ret = add_extent_mapping(em_tree, em);
782 spin_unlock(&em_tree->lock);
783 if (ret != -EEXIST) {
784 free_extent_map(em);
785 break;
786 }
787 btrfs_drop_extent_cache(inode, start,
788 start + ram_size - 1, 0);
789 }
790
791 cur_alloc_size = ins.offset;
792 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
793 ram_size, cur_alloc_size, 0);
794 BUG_ON(ret);
795
796 if (root->root_key.objectid ==
797 BTRFS_DATA_RELOC_TREE_OBJECTID) {
798 ret = btrfs_reloc_clone_csums(inode, start,
799 cur_alloc_size);
800 BUG_ON(ret);
801 }
802
803 if (disk_num_bytes < cur_alloc_size)
804 break;
805
806 /* we're not doing compressed IO, don't unlock the first
807 * page (which the caller expects to stay locked), don't
808 * clear any dirty bits and don't set any writeback bits
809 */
810 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
811 start, start + ram_size - 1,
812 locked_page, unlock, 1,
813 1, 0, 0, 0);
814 disk_num_bytes -= cur_alloc_size;
815 num_bytes -= cur_alloc_size;
816 alloc_hint = ins.objectid + ins.offset;
817 start += cur_alloc_size;
818 }
819 out:
820 ret = 0;
821 btrfs_end_transaction(trans, root);
822
823 return ret;
824 }
825
826 /*
827 * work queue call back to started compression on a file and pages
828 */
829 static noinline void async_cow_start(struct btrfs_work *work)
830 {
831 struct async_cow *async_cow;
832 int num_added = 0;
833 async_cow = container_of(work, struct async_cow, work);
834
835 compress_file_range(async_cow->inode, async_cow->locked_page,
836 async_cow->start, async_cow->end, async_cow,
837 &num_added);
838 if (num_added == 0)
839 async_cow->inode = NULL;
840 }
841
842 /*
843 * work queue call back to submit previously compressed pages
844 */
845 static noinline void async_cow_submit(struct btrfs_work *work)
846 {
847 struct async_cow *async_cow;
848 struct btrfs_root *root;
849 unsigned long nr_pages;
850
851 async_cow = container_of(work, struct async_cow, work);
852
853 root = async_cow->root;
854 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
855 PAGE_CACHE_SHIFT;
856
857 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
858
859 if (atomic_read(&root->fs_info->async_delalloc_pages) <
860 5 * 1042 * 1024 &&
861 waitqueue_active(&root->fs_info->async_submit_wait))
862 wake_up(&root->fs_info->async_submit_wait);
863
864 if (async_cow->inode)
865 submit_compressed_extents(async_cow->inode, async_cow);
866 }
867
868 static noinline void async_cow_free(struct btrfs_work *work)
869 {
870 struct async_cow *async_cow;
871 async_cow = container_of(work, struct async_cow, work);
872 kfree(async_cow);
873 }
874
875 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
876 u64 start, u64 end, int *page_started,
877 unsigned long *nr_written)
878 {
879 struct async_cow *async_cow;
880 struct btrfs_root *root = BTRFS_I(inode)->root;
881 unsigned long nr_pages;
882 u64 cur_end;
883 int limit = 10 * 1024 * 1042;
884
885 if (!btrfs_test_opt(root, COMPRESS)) {
886 return cow_file_range(inode, locked_page, start, end,
887 page_started, nr_written, 1);
888 }
889
890 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
891 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
892 while (start < end) {
893 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
894 async_cow->inode = inode;
895 async_cow->root = root;
896 async_cow->locked_page = locked_page;
897 async_cow->start = start;
898
899 if (btrfs_test_flag(inode, NOCOMPRESS))
900 cur_end = end;
901 else
902 cur_end = min(end, start + 512 * 1024 - 1);
903
904 async_cow->end = cur_end;
905 INIT_LIST_HEAD(&async_cow->extents);
906
907 async_cow->work.func = async_cow_start;
908 async_cow->work.ordered_func = async_cow_submit;
909 async_cow->work.ordered_free = async_cow_free;
910 async_cow->work.flags = 0;
911
912 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
913 PAGE_CACHE_SHIFT;
914 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
915
916 btrfs_queue_worker(&root->fs_info->delalloc_workers,
917 &async_cow->work);
918
919 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
920 wait_event(root->fs_info->async_submit_wait,
921 (atomic_read(&root->fs_info->async_delalloc_pages) <
922 limit));
923 }
924
925 while (atomic_read(&root->fs_info->async_submit_draining) &&
926 atomic_read(&root->fs_info->async_delalloc_pages)) {
927 wait_event(root->fs_info->async_submit_wait,
928 (atomic_read(&root->fs_info->async_delalloc_pages) ==
929 0));
930 }
931
932 *nr_written += nr_pages;
933 start = cur_end + 1;
934 }
935 *page_started = 1;
936 return 0;
937 }
938
939 static noinline int csum_exist_in_range(struct btrfs_root *root,
940 u64 bytenr, u64 num_bytes)
941 {
942 int ret;
943 struct btrfs_ordered_sum *sums;
944 LIST_HEAD(list);
945
946 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
947 bytenr + num_bytes - 1, &list);
948 if (ret == 0 && list_empty(&list))
949 return 0;
950
951 while (!list_empty(&list)) {
952 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
953 list_del(&sums->list);
954 kfree(sums);
955 }
956 return 1;
957 }
958
959 /*
960 * when nowcow writeback call back. This checks for snapshots or COW copies
961 * of the extents that exist in the file, and COWs the file as required.
962 *
963 * If no cow copies or snapshots exist, we write directly to the existing
964 * blocks on disk
965 */
966 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
967 u64 start, u64 end, int *page_started, int force,
968 unsigned long *nr_written)
969 {
970 struct btrfs_root *root = BTRFS_I(inode)->root;
971 struct btrfs_trans_handle *trans;
972 struct extent_buffer *leaf;
973 struct btrfs_path *path;
974 struct btrfs_file_extent_item *fi;
975 struct btrfs_key found_key;
976 u64 cow_start;
977 u64 cur_offset;
978 u64 extent_end;
979 u64 disk_bytenr;
980 u64 num_bytes;
981 int extent_type;
982 int ret;
983 int type;
984 int nocow;
985 int check_prev = 1;
986
987 path = btrfs_alloc_path();
988 BUG_ON(!path);
989 trans = btrfs_join_transaction(root, 1);
990 BUG_ON(!trans);
991
992 cow_start = (u64)-1;
993 cur_offset = start;
994 while (1) {
995 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
996 cur_offset, 0);
997 BUG_ON(ret < 0);
998 if (ret > 0 && path->slots[0] > 0 && check_prev) {
999 leaf = path->nodes[0];
1000 btrfs_item_key_to_cpu(leaf, &found_key,
1001 path->slots[0] - 1);
1002 if (found_key.objectid == inode->i_ino &&
1003 found_key.type == BTRFS_EXTENT_DATA_KEY)
1004 path->slots[0]--;
1005 }
1006 check_prev = 0;
1007 next_slot:
1008 leaf = path->nodes[0];
1009 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1010 ret = btrfs_next_leaf(root, path);
1011 if (ret < 0)
1012 BUG_ON(1);
1013 if (ret > 0)
1014 break;
1015 leaf = path->nodes[0];
1016 }
1017
1018 nocow = 0;
1019 disk_bytenr = 0;
1020 num_bytes = 0;
1021 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1022
1023 if (found_key.objectid > inode->i_ino ||
1024 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1025 found_key.offset > end)
1026 break;
1027
1028 if (found_key.offset > cur_offset) {
1029 extent_end = found_key.offset;
1030 goto out_check;
1031 }
1032
1033 fi = btrfs_item_ptr(leaf, path->slots[0],
1034 struct btrfs_file_extent_item);
1035 extent_type = btrfs_file_extent_type(leaf, fi);
1036
1037 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1038 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1039 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1040 extent_end = found_key.offset +
1041 btrfs_file_extent_num_bytes(leaf, fi);
1042 if (extent_end <= start) {
1043 path->slots[0]++;
1044 goto next_slot;
1045 }
1046 if (disk_bytenr == 0)
1047 goto out_check;
1048 if (btrfs_file_extent_compression(leaf, fi) ||
1049 btrfs_file_extent_encryption(leaf, fi) ||
1050 btrfs_file_extent_other_encoding(leaf, fi))
1051 goto out_check;
1052 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1053 goto out_check;
1054 if (btrfs_extent_readonly(root, disk_bytenr))
1055 goto out_check;
1056 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1057 disk_bytenr))
1058 goto out_check;
1059 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1060 disk_bytenr += cur_offset - found_key.offset;
1061 num_bytes = min(end + 1, extent_end) - cur_offset;
1062 /*
1063 * force cow if csum exists in the range.
1064 * this ensure that csum for a given extent are
1065 * either valid or do not exist.
1066 */
1067 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1068 goto out_check;
1069 nocow = 1;
1070 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1071 extent_end = found_key.offset +
1072 btrfs_file_extent_inline_len(leaf, fi);
1073 extent_end = ALIGN(extent_end, root->sectorsize);
1074 } else {
1075 BUG_ON(1);
1076 }
1077 out_check:
1078 if (extent_end <= start) {
1079 path->slots[0]++;
1080 goto next_slot;
1081 }
1082 if (!nocow) {
1083 if (cow_start == (u64)-1)
1084 cow_start = cur_offset;
1085 cur_offset = extent_end;
1086 if (cur_offset > end)
1087 break;
1088 path->slots[0]++;
1089 goto next_slot;
1090 }
1091
1092 btrfs_release_path(root, path);
1093 if (cow_start != (u64)-1) {
1094 ret = cow_file_range(inode, locked_page, cow_start,
1095 found_key.offset - 1, page_started,
1096 nr_written, 1);
1097 BUG_ON(ret);
1098 cow_start = (u64)-1;
1099 }
1100
1101 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1102 struct extent_map *em;
1103 struct extent_map_tree *em_tree;
1104 em_tree = &BTRFS_I(inode)->extent_tree;
1105 em = alloc_extent_map(GFP_NOFS);
1106 em->start = cur_offset;
1107 em->orig_start = em->start;
1108 em->len = num_bytes;
1109 em->block_len = num_bytes;
1110 em->block_start = disk_bytenr;
1111 em->bdev = root->fs_info->fs_devices->latest_bdev;
1112 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1113 while (1) {
1114 spin_lock(&em_tree->lock);
1115 ret = add_extent_mapping(em_tree, em);
1116 spin_unlock(&em_tree->lock);
1117 if (ret != -EEXIST) {
1118 free_extent_map(em);
1119 break;
1120 }
1121 btrfs_drop_extent_cache(inode, em->start,
1122 em->start + em->len - 1, 0);
1123 }
1124 type = BTRFS_ORDERED_PREALLOC;
1125 } else {
1126 type = BTRFS_ORDERED_NOCOW;
1127 }
1128
1129 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1130 num_bytes, num_bytes, type);
1131 BUG_ON(ret);
1132
1133 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1134 cur_offset, cur_offset + num_bytes - 1,
1135 locked_page, 1, 1, 1, 0, 0, 0);
1136 cur_offset = extent_end;
1137 if (cur_offset > end)
1138 break;
1139 }
1140 btrfs_release_path(root, path);
1141
1142 if (cur_offset <= end && cow_start == (u64)-1)
1143 cow_start = cur_offset;
1144 if (cow_start != (u64)-1) {
1145 ret = cow_file_range(inode, locked_page, cow_start, end,
1146 page_started, nr_written, 1);
1147 BUG_ON(ret);
1148 }
1149
1150 ret = btrfs_end_transaction(trans, root);
1151 BUG_ON(ret);
1152 btrfs_free_path(path);
1153 return 0;
1154 }
1155
1156 /*
1157 * extent_io.c call back to do delayed allocation processing
1158 */
1159 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1160 u64 start, u64 end, int *page_started,
1161 unsigned long *nr_written)
1162 {
1163 int ret;
1164
1165 if (btrfs_test_flag(inode, NODATACOW))
1166 ret = run_delalloc_nocow(inode, locked_page, start, end,
1167 page_started, 1, nr_written);
1168 else if (btrfs_test_flag(inode, PREALLOC))
1169 ret = run_delalloc_nocow(inode, locked_page, start, end,
1170 page_started, 0, nr_written);
1171 else
1172 ret = cow_file_range_async(inode, locked_page, start, end,
1173 page_started, nr_written);
1174
1175 return ret;
1176 }
1177
1178 /*
1179 * extent_io.c set_bit_hook, used to track delayed allocation
1180 * bytes in this file, and to maintain the list of inodes that
1181 * have pending delalloc work to be done.
1182 */
1183 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1184 unsigned long old, unsigned long bits)
1185 {
1186 /*
1187 * set_bit and clear bit hooks normally require _irqsave/restore
1188 * but in this case, we are only testeing for the DELALLOC
1189 * bit, which is only set or cleared with irqs on
1190 */
1191 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1192 struct btrfs_root *root = BTRFS_I(inode)->root;
1193 spin_lock(&root->fs_info->delalloc_lock);
1194 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1195 root->fs_info->delalloc_bytes += end - start + 1;
1196 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1197 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1198 &root->fs_info->delalloc_inodes);
1199 }
1200 spin_unlock(&root->fs_info->delalloc_lock);
1201 }
1202 return 0;
1203 }
1204
1205 /*
1206 * extent_io.c clear_bit_hook, see set_bit_hook for why
1207 */
1208 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1209 unsigned long old, unsigned long bits)
1210 {
1211 /*
1212 * set_bit and clear bit hooks normally require _irqsave/restore
1213 * but in this case, we are only testeing for the DELALLOC
1214 * bit, which is only set or cleared with irqs on
1215 */
1216 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1217 struct btrfs_root *root = BTRFS_I(inode)->root;
1218
1219 spin_lock(&root->fs_info->delalloc_lock);
1220 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1221 printk(KERN_INFO "btrfs warning: delalloc account "
1222 "%llu %llu\n",
1223 (unsigned long long)end - start + 1,
1224 (unsigned long long)
1225 root->fs_info->delalloc_bytes);
1226 root->fs_info->delalloc_bytes = 0;
1227 BTRFS_I(inode)->delalloc_bytes = 0;
1228 } else {
1229 root->fs_info->delalloc_bytes -= end - start + 1;
1230 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1231 }
1232 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1233 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1234 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1235 }
1236 spin_unlock(&root->fs_info->delalloc_lock);
1237 }
1238 return 0;
1239 }
1240
1241 /*
1242 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1243 * we don't create bios that span stripes or chunks
1244 */
1245 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1246 size_t size, struct bio *bio,
1247 unsigned long bio_flags)
1248 {
1249 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1250 struct btrfs_mapping_tree *map_tree;
1251 u64 logical = (u64)bio->bi_sector << 9;
1252 u64 length = 0;
1253 u64 map_length;
1254 int ret;
1255
1256 if (bio_flags & EXTENT_BIO_COMPRESSED)
1257 return 0;
1258
1259 length = bio->bi_size;
1260 map_tree = &root->fs_info->mapping_tree;
1261 map_length = length;
1262 ret = btrfs_map_block(map_tree, READ, logical,
1263 &map_length, NULL, 0);
1264
1265 if (map_length < length + size)
1266 return 1;
1267 return 0;
1268 }
1269
1270 /*
1271 * in order to insert checksums into the metadata in large chunks,
1272 * we wait until bio submission time. All the pages in the bio are
1273 * checksummed and sums are attached onto the ordered extent record.
1274 *
1275 * At IO completion time the cums attached on the ordered extent record
1276 * are inserted into the btree
1277 */
1278 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1279 struct bio *bio, int mirror_num,
1280 unsigned long bio_flags)
1281 {
1282 struct btrfs_root *root = BTRFS_I(inode)->root;
1283 int ret = 0;
1284
1285 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1286 BUG_ON(ret);
1287 return 0;
1288 }
1289
1290 /*
1291 * in order to insert checksums into the metadata in large chunks,
1292 * we wait until bio submission time. All the pages in the bio are
1293 * checksummed and sums are attached onto the ordered extent record.
1294 *
1295 * At IO completion time the cums attached on the ordered extent record
1296 * are inserted into the btree
1297 */
1298 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1299 int mirror_num, unsigned long bio_flags)
1300 {
1301 struct btrfs_root *root = BTRFS_I(inode)->root;
1302 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1303 }
1304
1305 /*
1306 * extent_io.c submission hook. This does the right thing for csum calculation
1307 * on write, or reading the csums from the tree before a read
1308 */
1309 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1310 int mirror_num, unsigned long bio_flags)
1311 {
1312 struct btrfs_root *root = BTRFS_I(inode)->root;
1313 int ret = 0;
1314 int skip_sum;
1315
1316 skip_sum = btrfs_test_flag(inode, NODATASUM);
1317
1318 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1319 BUG_ON(ret);
1320
1321 if (!(rw & (1 << BIO_RW))) {
1322 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1323 return btrfs_submit_compressed_read(inode, bio,
1324 mirror_num, bio_flags);
1325 } else if (!skip_sum)
1326 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1327 goto mapit;
1328 } else if (!skip_sum) {
1329 /* csum items have already been cloned */
1330 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1331 goto mapit;
1332 /* we're doing a write, do the async checksumming */
1333 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1334 inode, rw, bio, mirror_num,
1335 bio_flags, __btrfs_submit_bio_start,
1336 __btrfs_submit_bio_done);
1337 }
1338
1339 mapit:
1340 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1341 }
1342
1343 /*
1344 * given a list of ordered sums record them in the inode. This happens
1345 * at IO completion time based on sums calculated at bio submission time.
1346 */
1347 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1348 struct inode *inode, u64 file_offset,
1349 struct list_head *list)
1350 {
1351 struct btrfs_ordered_sum *sum;
1352
1353 btrfs_set_trans_block_group(trans, inode);
1354
1355 list_for_each_entry(sum, list, list) {
1356 btrfs_csum_file_blocks(trans,
1357 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1358 }
1359 return 0;
1360 }
1361
1362 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1363 {
1364 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1365 WARN_ON(1);
1366 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1367 GFP_NOFS);
1368 }
1369
1370 /* see btrfs_writepage_start_hook for details on why this is required */
1371 struct btrfs_writepage_fixup {
1372 struct page *page;
1373 struct btrfs_work work;
1374 };
1375
1376 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1377 {
1378 struct btrfs_writepage_fixup *fixup;
1379 struct btrfs_ordered_extent *ordered;
1380 struct page *page;
1381 struct inode *inode;
1382 u64 page_start;
1383 u64 page_end;
1384
1385 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1386 page = fixup->page;
1387 again:
1388 lock_page(page);
1389 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1390 ClearPageChecked(page);
1391 goto out_page;
1392 }
1393
1394 inode = page->mapping->host;
1395 page_start = page_offset(page);
1396 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1397
1398 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1399
1400 /* already ordered? We're done */
1401 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1402 EXTENT_ORDERED, 0)) {
1403 goto out;
1404 }
1405
1406 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1407 if (ordered) {
1408 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1409 page_end, GFP_NOFS);
1410 unlock_page(page);
1411 btrfs_start_ordered_extent(inode, ordered, 1);
1412 goto again;
1413 }
1414
1415 btrfs_set_extent_delalloc(inode, page_start, page_end);
1416 ClearPageChecked(page);
1417 out:
1418 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1419 out_page:
1420 unlock_page(page);
1421 page_cache_release(page);
1422 }
1423
1424 /*
1425 * There are a few paths in the higher layers of the kernel that directly
1426 * set the page dirty bit without asking the filesystem if it is a
1427 * good idea. This causes problems because we want to make sure COW
1428 * properly happens and the data=ordered rules are followed.
1429 *
1430 * In our case any range that doesn't have the ORDERED bit set
1431 * hasn't been properly setup for IO. We kick off an async process
1432 * to fix it up. The async helper will wait for ordered extents, set
1433 * the delalloc bit and make it safe to write the page.
1434 */
1435 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1436 {
1437 struct inode *inode = page->mapping->host;
1438 struct btrfs_writepage_fixup *fixup;
1439 struct btrfs_root *root = BTRFS_I(inode)->root;
1440 int ret;
1441
1442 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1443 EXTENT_ORDERED, 0);
1444 if (ret)
1445 return 0;
1446
1447 if (PageChecked(page))
1448 return -EAGAIN;
1449
1450 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1451 if (!fixup)
1452 return -EAGAIN;
1453
1454 SetPageChecked(page);
1455 page_cache_get(page);
1456 fixup->work.func = btrfs_writepage_fixup_worker;
1457 fixup->page = page;
1458 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1459 return -EAGAIN;
1460 }
1461
1462 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1463 struct inode *inode, u64 file_pos,
1464 u64 disk_bytenr, u64 disk_num_bytes,
1465 u64 num_bytes, u64 ram_bytes,
1466 u8 compression, u8 encryption,
1467 u16 other_encoding, int extent_type)
1468 {
1469 struct btrfs_root *root = BTRFS_I(inode)->root;
1470 struct btrfs_file_extent_item *fi;
1471 struct btrfs_path *path;
1472 struct extent_buffer *leaf;
1473 struct btrfs_key ins;
1474 u64 hint;
1475 int ret;
1476
1477 path = btrfs_alloc_path();
1478 BUG_ON(!path);
1479
1480 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1481 file_pos + num_bytes, file_pos, &hint);
1482 BUG_ON(ret);
1483
1484 ins.objectid = inode->i_ino;
1485 ins.offset = file_pos;
1486 ins.type = BTRFS_EXTENT_DATA_KEY;
1487 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1488 BUG_ON(ret);
1489 leaf = path->nodes[0];
1490 fi = btrfs_item_ptr(leaf, path->slots[0],
1491 struct btrfs_file_extent_item);
1492 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1493 btrfs_set_file_extent_type(leaf, fi, extent_type);
1494 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1495 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1496 btrfs_set_file_extent_offset(leaf, fi, 0);
1497 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1498 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1499 btrfs_set_file_extent_compression(leaf, fi, compression);
1500 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1501 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1502 btrfs_mark_buffer_dirty(leaf);
1503
1504 inode_add_bytes(inode, num_bytes);
1505 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1506
1507 ins.objectid = disk_bytenr;
1508 ins.offset = disk_num_bytes;
1509 ins.type = BTRFS_EXTENT_ITEM_KEY;
1510 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1511 root->root_key.objectid,
1512 trans->transid, inode->i_ino, &ins);
1513 BUG_ON(ret);
1514
1515 btrfs_free_path(path);
1516 return 0;
1517 }
1518
1519 /* as ordered data IO finishes, this gets called so we can finish
1520 * an ordered extent if the range of bytes in the file it covers are
1521 * fully written.
1522 */
1523 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1524 {
1525 struct btrfs_root *root = BTRFS_I(inode)->root;
1526 struct btrfs_trans_handle *trans;
1527 struct btrfs_ordered_extent *ordered_extent;
1528 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1529 int compressed = 0;
1530 int ret;
1531
1532 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1533 if (!ret)
1534 return 0;
1535
1536 trans = btrfs_join_transaction(root, 1);
1537
1538 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1539 BUG_ON(!ordered_extent);
1540 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1541 goto nocow;
1542
1543 lock_extent(io_tree, ordered_extent->file_offset,
1544 ordered_extent->file_offset + ordered_extent->len - 1,
1545 GFP_NOFS);
1546
1547 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1548 compressed = 1;
1549 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1550 BUG_ON(compressed);
1551 ret = btrfs_mark_extent_written(trans, root, inode,
1552 ordered_extent->file_offset,
1553 ordered_extent->file_offset +
1554 ordered_extent->len);
1555 BUG_ON(ret);
1556 } else {
1557 ret = insert_reserved_file_extent(trans, inode,
1558 ordered_extent->file_offset,
1559 ordered_extent->start,
1560 ordered_extent->disk_len,
1561 ordered_extent->len,
1562 ordered_extent->len,
1563 compressed, 0, 0,
1564 BTRFS_FILE_EXTENT_REG);
1565 BUG_ON(ret);
1566 }
1567 unlock_extent(io_tree, ordered_extent->file_offset,
1568 ordered_extent->file_offset + ordered_extent->len - 1,
1569 GFP_NOFS);
1570 nocow:
1571 add_pending_csums(trans, inode, ordered_extent->file_offset,
1572 &ordered_extent->list);
1573
1574 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1575 btrfs_ordered_update_i_size(inode, ordered_extent);
1576 btrfs_update_inode(trans, root, inode);
1577 btrfs_remove_ordered_extent(inode, ordered_extent);
1578 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1579
1580 /* once for us */
1581 btrfs_put_ordered_extent(ordered_extent);
1582 /* once for the tree */
1583 btrfs_put_ordered_extent(ordered_extent);
1584
1585 btrfs_end_transaction(trans, root);
1586 return 0;
1587 }
1588
1589 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1590 struct extent_state *state, int uptodate)
1591 {
1592 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1593 }
1594
1595 /*
1596 * When IO fails, either with EIO or csum verification fails, we
1597 * try other mirrors that might have a good copy of the data. This
1598 * io_failure_record is used to record state as we go through all the
1599 * mirrors. If another mirror has good data, the page is set up to date
1600 * and things continue. If a good mirror can't be found, the original
1601 * bio end_io callback is called to indicate things have failed.
1602 */
1603 struct io_failure_record {
1604 struct page *page;
1605 u64 start;
1606 u64 len;
1607 u64 logical;
1608 unsigned long bio_flags;
1609 int last_mirror;
1610 };
1611
1612 static int btrfs_io_failed_hook(struct bio *failed_bio,
1613 struct page *page, u64 start, u64 end,
1614 struct extent_state *state)
1615 {
1616 struct io_failure_record *failrec = NULL;
1617 u64 private;
1618 struct extent_map *em;
1619 struct inode *inode = page->mapping->host;
1620 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1621 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1622 struct bio *bio;
1623 int num_copies;
1624 int ret;
1625 int rw;
1626 u64 logical;
1627
1628 ret = get_state_private(failure_tree, start, &private);
1629 if (ret) {
1630 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1631 if (!failrec)
1632 return -ENOMEM;
1633 failrec->start = start;
1634 failrec->len = end - start + 1;
1635 failrec->last_mirror = 0;
1636 failrec->bio_flags = 0;
1637
1638 spin_lock(&em_tree->lock);
1639 em = lookup_extent_mapping(em_tree, start, failrec->len);
1640 if (em->start > start || em->start + em->len < start) {
1641 free_extent_map(em);
1642 em = NULL;
1643 }
1644 spin_unlock(&em_tree->lock);
1645
1646 if (!em || IS_ERR(em)) {
1647 kfree(failrec);
1648 return -EIO;
1649 }
1650 logical = start - em->start;
1651 logical = em->block_start + logical;
1652 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1653 logical = em->block_start;
1654 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1655 }
1656 failrec->logical = logical;
1657 free_extent_map(em);
1658 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1659 EXTENT_DIRTY, GFP_NOFS);
1660 set_state_private(failure_tree, start,
1661 (u64)(unsigned long)failrec);
1662 } else {
1663 failrec = (struct io_failure_record *)(unsigned long)private;
1664 }
1665 num_copies = btrfs_num_copies(
1666 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1667 failrec->logical, failrec->len);
1668 failrec->last_mirror++;
1669 if (!state) {
1670 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1671 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1672 failrec->start,
1673 EXTENT_LOCKED);
1674 if (state && state->start != failrec->start)
1675 state = NULL;
1676 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1677 }
1678 if (!state || failrec->last_mirror > num_copies) {
1679 set_state_private(failure_tree, failrec->start, 0);
1680 clear_extent_bits(failure_tree, failrec->start,
1681 failrec->start + failrec->len - 1,
1682 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1683 kfree(failrec);
1684 return -EIO;
1685 }
1686 bio = bio_alloc(GFP_NOFS, 1);
1687 bio->bi_private = state;
1688 bio->bi_end_io = failed_bio->bi_end_io;
1689 bio->bi_sector = failrec->logical >> 9;
1690 bio->bi_bdev = failed_bio->bi_bdev;
1691 bio->bi_size = 0;
1692
1693 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1694 if (failed_bio->bi_rw & (1 << BIO_RW))
1695 rw = WRITE;
1696 else
1697 rw = READ;
1698
1699 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1700 failrec->last_mirror,
1701 failrec->bio_flags);
1702 return 0;
1703 }
1704
1705 /*
1706 * each time an IO finishes, we do a fast check in the IO failure tree
1707 * to see if we need to process or clean up an io_failure_record
1708 */
1709 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1710 {
1711 u64 private;
1712 u64 private_failure;
1713 struct io_failure_record *failure;
1714 int ret;
1715
1716 private = 0;
1717 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1718 (u64)-1, 1, EXTENT_DIRTY)) {
1719 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1720 start, &private_failure);
1721 if (ret == 0) {
1722 failure = (struct io_failure_record *)(unsigned long)
1723 private_failure;
1724 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1725 failure->start, 0);
1726 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1727 failure->start,
1728 failure->start + failure->len - 1,
1729 EXTENT_DIRTY | EXTENT_LOCKED,
1730 GFP_NOFS);
1731 kfree(failure);
1732 }
1733 }
1734 return 0;
1735 }
1736
1737 /*
1738 * when reads are done, we need to check csums to verify the data is correct
1739 * if there's a match, we allow the bio to finish. If not, we go through
1740 * the io_failure_record routines to find good copies
1741 */
1742 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1743 struct extent_state *state)
1744 {
1745 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1746 struct inode *inode = page->mapping->host;
1747 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1748 char *kaddr;
1749 u64 private = ~(u32)0;
1750 int ret;
1751 struct btrfs_root *root = BTRFS_I(inode)->root;
1752 u32 csum = ~(u32)0;
1753
1754 if (PageChecked(page)) {
1755 ClearPageChecked(page);
1756 goto good;
1757 }
1758 if (btrfs_test_flag(inode, NODATASUM))
1759 return 0;
1760
1761 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1762 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1763 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1764 GFP_NOFS);
1765 return 0;
1766 }
1767
1768 if (state && state->start == start) {
1769 private = state->private;
1770 ret = 0;
1771 } else {
1772 ret = get_state_private(io_tree, start, &private);
1773 }
1774 kaddr = kmap_atomic(page, KM_USER0);
1775 if (ret)
1776 goto zeroit;
1777
1778 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1779 btrfs_csum_final(csum, (char *)&csum);
1780 if (csum != private)
1781 goto zeroit;
1782
1783 kunmap_atomic(kaddr, KM_USER0);
1784 good:
1785 /* if the io failure tree for this inode is non-empty,
1786 * check to see if we've recovered from a failed IO
1787 */
1788 btrfs_clean_io_failures(inode, start);
1789 return 0;
1790
1791 zeroit:
1792 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1793 "private %llu\n", page->mapping->host->i_ino,
1794 (unsigned long long)start, csum,
1795 (unsigned long long)private);
1796 memset(kaddr + offset, 1, end - start + 1);
1797 flush_dcache_page(page);
1798 kunmap_atomic(kaddr, KM_USER0);
1799 if (private == 0)
1800 return 0;
1801 return -EIO;
1802 }
1803
1804 /*
1805 * This creates an orphan entry for the given inode in case something goes
1806 * wrong in the middle of an unlink/truncate.
1807 */
1808 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1809 {
1810 struct btrfs_root *root = BTRFS_I(inode)->root;
1811 int ret = 0;
1812
1813 spin_lock(&root->list_lock);
1814
1815 /* already on the orphan list, we're good */
1816 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1817 spin_unlock(&root->list_lock);
1818 return 0;
1819 }
1820
1821 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1822
1823 spin_unlock(&root->list_lock);
1824
1825 /*
1826 * insert an orphan item to track this unlinked/truncated file
1827 */
1828 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1829
1830 return ret;
1831 }
1832
1833 /*
1834 * We have done the truncate/delete so we can go ahead and remove the orphan
1835 * item for this particular inode.
1836 */
1837 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1838 {
1839 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 int ret = 0;
1841
1842 spin_lock(&root->list_lock);
1843
1844 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1845 spin_unlock(&root->list_lock);
1846 return 0;
1847 }
1848
1849 list_del_init(&BTRFS_I(inode)->i_orphan);
1850 if (!trans) {
1851 spin_unlock(&root->list_lock);
1852 return 0;
1853 }
1854
1855 spin_unlock(&root->list_lock);
1856
1857 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1858
1859 return ret;
1860 }
1861
1862 /*
1863 * this cleans up any orphans that may be left on the list from the last use
1864 * of this root.
1865 */
1866 void btrfs_orphan_cleanup(struct btrfs_root *root)
1867 {
1868 struct btrfs_path *path;
1869 struct extent_buffer *leaf;
1870 struct btrfs_item *item;
1871 struct btrfs_key key, found_key;
1872 struct btrfs_trans_handle *trans;
1873 struct inode *inode;
1874 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1875
1876 path = btrfs_alloc_path();
1877 if (!path)
1878 return;
1879 path->reada = -1;
1880
1881 key.objectid = BTRFS_ORPHAN_OBJECTID;
1882 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1883 key.offset = (u64)-1;
1884
1885
1886 while (1) {
1887 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1888 if (ret < 0) {
1889 printk(KERN_ERR "Error searching slot for orphan: %d"
1890 "\n", ret);
1891 break;
1892 }
1893
1894 /*
1895 * if ret == 0 means we found what we were searching for, which
1896 * is weird, but possible, so only screw with path if we didnt
1897 * find the key and see if we have stuff that matches
1898 */
1899 if (ret > 0) {
1900 if (path->slots[0] == 0)
1901 break;
1902 path->slots[0]--;
1903 }
1904
1905 /* pull out the item */
1906 leaf = path->nodes[0];
1907 item = btrfs_item_nr(leaf, path->slots[0]);
1908 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1909
1910 /* make sure the item matches what we want */
1911 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1912 break;
1913 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1914 break;
1915
1916 /* release the path since we're done with it */
1917 btrfs_release_path(root, path);
1918
1919 /*
1920 * this is where we are basically btrfs_lookup, without the
1921 * crossing root thing. we store the inode number in the
1922 * offset of the orphan item.
1923 */
1924 inode = btrfs_iget_locked(root->fs_info->sb,
1925 found_key.offset, root);
1926 if (!inode)
1927 break;
1928
1929 if (inode->i_state & I_NEW) {
1930 BTRFS_I(inode)->root = root;
1931
1932 /* have to set the location manually */
1933 BTRFS_I(inode)->location.objectid = inode->i_ino;
1934 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1935 BTRFS_I(inode)->location.offset = 0;
1936
1937 btrfs_read_locked_inode(inode);
1938 unlock_new_inode(inode);
1939 }
1940
1941 /*
1942 * add this inode to the orphan list so btrfs_orphan_del does
1943 * the proper thing when we hit it
1944 */
1945 spin_lock(&root->list_lock);
1946 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1947 spin_unlock(&root->list_lock);
1948
1949 /*
1950 * if this is a bad inode, means we actually succeeded in
1951 * removing the inode, but not the orphan record, which means
1952 * we need to manually delete the orphan since iput will just
1953 * do a destroy_inode
1954 */
1955 if (is_bad_inode(inode)) {
1956 trans = btrfs_start_transaction(root, 1);
1957 btrfs_orphan_del(trans, inode);
1958 btrfs_end_transaction(trans, root);
1959 iput(inode);
1960 continue;
1961 }
1962
1963 /* if we have links, this was a truncate, lets do that */
1964 if (inode->i_nlink) {
1965 nr_truncate++;
1966 btrfs_truncate(inode);
1967 } else {
1968 nr_unlink++;
1969 }
1970
1971 /* this will do delete_inode and everything for us */
1972 iput(inode);
1973 }
1974
1975 if (nr_unlink)
1976 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1977 if (nr_truncate)
1978 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1979
1980 btrfs_free_path(path);
1981 }
1982
1983 /*
1984 * read an inode from the btree into the in-memory inode
1985 */
1986 void btrfs_read_locked_inode(struct inode *inode)
1987 {
1988 struct btrfs_path *path;
1989 struct extent_buffer *leaf;
1990 struct btrfs_inode_item *inode_item;
1991 struct btrfs_timespec *tspec;
1992 struct btrfs_root *root = BTRFS_I(inode)->root;
1993 struct btrfs_key location;
1994 u64 alloc_group_block;
1995 u32 rdev;
1996 int ret;
1997
1998 path = btrfs_alloc_path();
1999 BUG_ON(!path);
2000 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2001
2002 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2003 if (ret)
2004 goto make_bad;
2005
2006 leaf = path->nodes[0];
2007 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2008 struct btrfs_inode_item);
2009
2010 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2011 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2012 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2013 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2014 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2015
2016 tspec = btrfs_inode_atime(inode_item);
2017 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2018 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2019
2020 tspec = btrfs_inode_mtime(inode_item);
2021 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2022 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2023
2024 tspec = btrfs_inode_ctime(inode_item);
2025 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2026 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2027
2028 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2029 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2030 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2031 inode->i_generation = BTRFS_I(inode)->generation;
2032 inode->i_rdev = 0;
2033 rdev = btrfs_inode_rdev(leaf, inode_item);
2034
2035 BTRFS_I(inode)->index_cnt = (u64)-1;
2036 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2037
2038 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2039
2040 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2041 alloc_group_block, 0);
2042 btrfs_free_path(path);
2043 inode_item = NULL;
2044
2045 switch (inode->i_mode & S_IFMT) {
2046 case S_IFREG:
2047 inode->i_mapping->a_ops = &btrfs_aops;
2048 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2049 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2050 inode->i_fop = &btrfs_file_operations;
2051 inode->i_op = &btrfs_file_inode_operations;
2052 break;
2053 case S_IFDIR:
2054 inode->i_fop = &btrfs_dir_file_operations;
2055 if (root == root->fs_info->tree_root)
2056 inode->i_op = &btrfs_dir_ro_inode_operations;
2057 else
2058 inode->i_op = &btrfs_dir_inode_operations;
2059 break;
2060 case S_IFLNK:
2061 inode->i_op = &btrfs_symlink_inode_operations;
2062 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2063 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2064 break;
2065 default:
2066 inode->i_op = &btrfs_special_inode_operations;
2067 init_special_inode(inode, inode->i_mode, rdev);
2068 break;
2069 }
2070 return;
2071
2072 make_bad:
2073 btrfs_free_path(path);
2074 make_bad_inode(inode);
2075 }
2076
2077 /*
2078 * given a leaf and an inode, copy the inode fields into the leaf
2079 */
2080 static void fill_inode_item(struct btrfs_trans_handle *trans,
2081 struct extent_buffer *leaf,
2082 struct btrfs_inode_item *item,
2083 struct inode *inode)
2084 {
2085 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2086 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2087 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2088 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2089 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2090
2091 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2092 inode->i_atime.tv_sec);
2093 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2094 inode->i_atime.tv_nsec);
2095
2096 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2097 inode->i_mtime.tv_sec);
2098 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2099 inode->i_mtime.tv_nsec);
2100
2101 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2102 inode->i_ctime.tv_sec);
2103 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2104 inode->i_ctime.tv_nsec);
2105
2106 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2107 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2108 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2109 btrfs_set_inode_transid(leaf, item, trans->transid);
2110 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2111 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2112 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2113 }
2114
2115 /*
2116 * copy everything in the in-memory inode into the btree.
2117 */
2118 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2119 struct btrfs_root *root, struct inode *inode)
2120 {
2121 struct btrfs_inode_item *inode_item;
2122 struct btrfs_path *path;
2123 struct extent_buffer *leaf;
2124 int ret;
2125
2126 path = btrfs_alloc_path();
2127 BUG_ON(!path);
2128 ret = btrfs_lookup_inode(trans, root, path,
2129 &BTRFS_I(inode)->location, 1);
2130 if (ret) {
2131 if (ret > 0)
2132 ret = -ENOENT;
2133 goto failed;
2134 }
2135
2136 btrfs_unlock_up_safe(path, 1);
2137 leaf = path->nodes[0];
2138 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2139 struct btrfs_inode_item);
2140
2141 fill_inode_item(trans, leaf, inode_item, inode);
2142 btrfs_mark_buffer_dirty(leaf);
2143 btrfs_set_inode_last_trans(trans, inode);
2144 ret = 0;
2145 failed:
2146 btrfs_free_path(path);
2147 return ret;
2148 }
2149
2150
2151 /*
2152 * unlink helper that gets used here in inode.c and in the tree logging
2153 * recovery code. It remove a link in a directory with a given name, and
2154 * also drops the back refs in the inode to the directory
2155 */
2156 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2157 struct btrfs_root *root,
2158 struct inode *dir, struct inode *inode,
2159 const char *name, int name_len)
2160 {
2161 struct btrfs_path *path;
2162 int ret = 0;
2163 struct extent_buffer *leaf;
2164 struct btrfs_dir_item *di;
2165 struct btrfs_key key;
2166 u64 index;
2167
2168 path = btrfs_alloc_path();
2169 if (!path) {
2170 ret = -ENOMEM;
2171 goto err;
2172 }
2173
2174 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2175 name, name_len, -1);
2176 if (IS_ERR(di)) {
2177 ret = PTR_ERR(di);
2178 goto err;
2179 }
2180 if (!di) {
2181 ret = -ENOENT;
2182 goto err;
2183 }
2184 leaf = path->nodes[0];
2185 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2186 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2187 if (ret)
2188 goto err;
2189 btrfs_release_path(root, path);
2190
2191 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2192 inode->i_ino,
2193 dir->i_ino, &index);
2194 if (ret) {
2195 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2196 "inode %lu parent %lu\n", name_len, name,
2197 inode->i_ino, dir->i_ino);
2198 goto err;
2199 }
2200
2201 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2202 index, name, name_len, -1);
2203 if (IS_ERR(di)) {
2204 ret = PTR_ERR(di);
2205 goto err;
2206 }
2207 if (!di) {
2208 ret = -ENOENT;
2209 goto err;
2210 }
2211 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2212 btrfs_release_path(root, path);
2213
2214 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2215 inode, dir->i_ino);
2216 BUG_ON(ret != 0 && ret != -ENOENT);
2217 if (ret != -ENOENT)
2218 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2219
2220 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2221 dir, index);
2222 BUG_ON(ret);
2223 err:
2224 btrfs_free_path(path);
2225 if (ret)
2226 goto out;
2227
2228 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2229 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2230 btrfs_update_inode(trans, root, dir);
2231 btrfs_drop_nlink(inode);
2232 ret = btrfs_update_inode(trans, root, inode);
2233 dir->i_sb->s_dirt = 1;
2234 out:
2235 return ret;
2236 }
2237
2238 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2239 {
2240 struct btrfs_root *root;
2241 struct btrfs_trans_handle *trans;
2242 struct inode *inode = dentry->d_inode;
2243 int ret;
2244 unsigned long nr = 0;
2245
2246 root = BTRFS_I(dir)->root;
2247
2248 ret = btrfs_check_free_space(root, 1, 1);
2249 if (ret)
2250 goto fail;
2251
2252 trans = btrfs_start_transaction(root, 1);
2253
2254 btrfs_set_trans_block_group(trans, dir);
2255 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2256 dentry->d_name.name, dentry->d_name.len);
2257
2258 if (inode->i_nlink == 0)
2259 ret = btrfs_orphan_add(trans, inode);
2260
2261 nr = trans->blocks_used;
2262
2263 btrfs_end_transaction_throttle(trans, root);
2264 fail:
2265 btrfs_btree_balance_dirty(root, nr);
2266 return ret;
2267 }
2268
2269 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2270 {
2271 struct inode *inode = dentry->d_inode;
2272 int err = 0;
2273 int ret;
2274 struct btrfs_root *root = BTRFS_I(dir)->root;
2275 struct btrfs_trans_handle *trans;
2276 unsigned long nr = 0;
2277
2278 /*
2279 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2280 * the root of a subvolume or snapshot
2281 */
2282 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2283 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2284 return -ENOTEMPTY;
2285 }
2286
2287 ret = btrfs_check_free_space(root, 1, 1);
2288 if (ret)
2289 goto fail;
2290
2291 trans = btrfs_start_transaction(root, 1);
2292 btrfs_set_trans_block_group(trans, dir);
2293
2294 err = btrfs_orphan_add(trans, inode);
2295 if (err)
2296 goto fail_trans;
2297
2298 /* now the directory is empty */
2299 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2300 dentry->d_name.name, dentry->d_name.len);
2301 if (!err)
2302 btrfs_i_size_write(inode, 0);
2303
2304 fail_trans:
2305 nr = trans->blocks_used;
2306 ret = btrfs_end_transaction_throttle(trans, root);
2307 fail:
2308 btrfs_btree_balance_dirty(root, nr);
2309
2310 if (ret && !err)
2311 err = ret;
2312 return err;
2313 }
2314
2315 #if 0
2316 /*
2317 * when truncating bytes in a file, it is possible to avoid reading
2318 * the leaves that contain only checksum items. This can be the
2319 * majority of the IO required to delete a large file, but it must
2320 * be done carefully.
2321 *
2322 * The keys in the level just above the leaves are checked to make sure
2323 * the lowest key in a given leaf is a csum key, and starts at an offset
2324 * after the new size.
2325 *
2326 * Then the key for the next leaf is checked to make sure it also has
2327 * a checksum item for the same file. If it does, we know our target leaf
2328 * contains only checksum items, and it can be safely freed without reading
2329 * it.
2330 *
2331 * This is just an optimization targeted at large files. It may do
2332 * nothing. It will return 0 unless things went badly.
2333 */
2334 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2335 struct btrfs_root *root,
2336 struct btrfs_path *path,
2337 struct inode *inode, u64 new_size)
2338 {
2339 struct btrfs_key key;
2340 int ret;
2341 int nritems;
2342 struct btrfs_key found_key;
2343 struct btrfs_key other_key;
2344 struct btrfs_leaf_ref *ref;
2345 u64 leaf_gen;
2346 u64 leaf_start;
2347
2348 path->lowest_level = 1;
2349 key.objectid = inode->i_ino;
2350 key.type = BTRFS_CSUM_ITEM_KEY;
2351 key.offset = new_size;
2352 again:
2353 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2354 if (ret < 0)
2355 goto out;
2356
2357 if (path->nodes[1] == NULL) {
2358 ret = 0;
2359 goto out;
2360 }
2361 ret = 0;
2362 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2363 nritems = btrfs_header_nritems(path->nodes[1]);
2364
2365 if (!nritems)
2366 goto out;
2367
2368 if (path->slots[1] >= nritems)
2369 goto next_node;
2370
2371 /* did we find a key greater than anything we want to delete? */
2372 if (found_key.objectid > inode->i_ino ||
2373 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2374 goto out;
2375
2376 /* we check the next key in the node to make sure the leave contains
2377 * only checksum items. This comparison doesn't work if our
2378 * leaf is the last one in the node
2379 */
2380 if (path->slots[1] + 1 >= nritems) {
2381 next_node:
2382 /* search forward from the last key in the node, this
2383 * will bring us into the next node in the tree
2384 */
2385 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2386
2387 /* unlikely, but we inc below, so check to be safe */
2388 if (found_key.offset == (u64)-1)
2389 goto out;
2390
2391 /* search_forward needs a path with locks held, do the
2392 * search again for the original key. It is possible
2393 * this will race with a balance and return a path that
2394 * we could modify, but this drop is just an optimization
2395 * and is allowed to miss some leaves.
2396 */
2397 btrfs_release_path(root, path);
2398 found_key.offset++;
2399
2400 /* setup a max key for search_forward */
2401 other_key.offset = (u64)-1;
2402 other_key.type = key.type;
2403 other_key.objectid = key.objectid;
2404
2405 path->keep_locks = 1;
2406 ret = btrfs_search_forward(root, &found_key, &other_key,
2407 path, 0, 0);
2408 path->keep_locks = 0;
2409 if (ret || found_key.objectid != key.objectid ||
2410 found_key.type != key.type) {
2411 ret = 0;
2412 goto out;
2413 }
2414
2415 key.offset = found_key.offset;
2416 btrfs_release_path(root, path);
2417 cond_resched();
2418 goto again;
2419 }
2420
2421 /* we know there's one more slot after us in the tree,
2422 * read that key so we can verify it is also a checksum item
2423 */
2424 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2425
2426 if (found_key.objectid < inode->i_ino)
2427 goto next_key;
2428
2429 if (found_key.type != key.type || found_key.offset < new_size)
2430 goto next_key;
2431
2432 /*
2433 * if the key for the next leaf isn't a csum key from this objectid,
2434 * we can't be sure there aren't good items inside this leaf.
2435 * Bail out
2436 */
2437 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2438 goto out;
2439
2440 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2441 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2442 /*
2443 * it is safe to delete this leaf, it contains only
2444 * csum items from this inode at an offset >= new_size
2445 */
2446 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2447 BUG_ON(ret);
2448
2449 if (root->ref_cows && leaf_gen < trans->transid) {
2450 ref = btrfs_alloc_leaf_ref(root, 0);
2451 if (ref) {
2452 ref->root_gen = root->root_key.offset;
2453 ref->bytenr = leaf_start;
2454 ref->owner = 0;
2455 ref->generation = leaf_gen;
2456 ref->nritems = 0;
2457
2458 btrfs_sort_leaf_ref(ref);
2459
2460 ret = btrfs_add_leaf_ref(root, ref, 0);
2461 WARN_ON(ret);
2462 btrfs_free_leaf_ref(root, ref);
2463 } else {
2464 WARN_ON(1);
2465 }
2466 }
2467 next_key:
2468 btrfs_release_path(root, path);
2469
2470 if (other_key.objectid == inode->i_ino &&
2471 other_key.type == key.type && other_key.offset > key.offset) {
2472 key.offset = other_key.offset;
2473 cond_resched();
2474 goto again;
2475 }
2476 ret = 0;
2477 out:
2478 /* fixup any changes we've made to the path */
2479 path->lowest_level = 0;
2480 path->keep_locks = 0;
2481 btrfs_release_path(root, path);
2482 return ret;
2483 }
2484
2485 #endif
2486
2487 /*
2488 * this can truncate away extent items, csum items and directory items.
2489 * It starts at a high offset and removes keys until it can't find
2490 * any higher than new_size
2491 *
2492 * csum items that cross the new i_size are truncated to the new size
2493 * as well.
2494 *
2495 * min_type is the minimum key type to truncate down to. If set to 0, this
2496 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2497 */
2498 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2499 struct btrfs_root *root,
2500 struct inode *inode,
2501 u64 new_size, u32 min_type)
2502 {
2503 int ret;
2504 struct btrfs_path *path;
2505 struct btrfs_key key;
2506 struct btrfs_key found_key;
2507 u32 found_type = (u8)-1;
2508 struct extent_buffer *leaf;
2509 struct btrfs_file_extent_item *fi;
2510 u64 extent_start = 0;
2511 u64 extent_num_bytes = 0;
2512 u64 item_end = 0;
2513 u64 root_gen = 0;
2514 u64 root_owner = 0;
2515 int found_extent;
2516 int del_item;
2517 int pending_del_nr = 0;
2518 int pending_del_slot = 0;
2519 int extent_type = -1;
2520 int encoding;
2521 u64 mask = root->sectorsize - 1;
2522
2523 if (root->ref_cows)
2524 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2525 path = btrfs_alloc_path();
2526 path->reada = -1;
2527 BUG_ON(!path);
2528
2529 /* FIXME, add redo link to tree so we don't leak on crash */
2530 key.objectid = inode->i_ino;
2531 key.offset = (u64)-1;
2532 key.type = (u8)-1;
2533
2534 search_again:
2535 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2536 if (ret < 0)
2537 goto error;
2538
2539 if (ret > 0) {
2540 /* there are no items in the tree for us to truncate, we're
2541 * done
2542 */
2543 if (path->slots[0] == 0) {
2544 ret = 0;
2545 goto error;
2546 }
2547 path->slots[0]--;
2548 }
2549
2550 while (1) {
2551 fi = NULL;
2552 leaf = path->nodes[0];
2553 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2554 found_type = btrfs_key_type(&found_key);
2555 encoding = 0;
2556
2557 if (found_key.objectid != inode->i_ino)
2558 break;
2559
2560 if (found_type < min_type)
2561 break;
2562
2563 item_end = found_key.offset;
2564 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2565 fi = btrfs_item_ptr(leaf, path->slots[0],
2566 struct btrfs_file_extent_item);
2567 extent_type = btrfs_file_extent_type(leaf, fi);
2568 encoding = btrfs_file_extent_compression(leaf, fi);
2569 encoding |= btrfs_file_extent_encryption(leaf, fi);
2570 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2571
2572 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2573 item_end +=
2574 btrfs_file_extent_num_bytes(leaf, fi);
2575 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2576 item_end += btrfs_file_extent_inline_len(leaf,
2577 fi);
2578 }
2579 item_end--;
2580 }
2581 if (item_end < new_size) {
2582 if (found_type == BTRFS_DIR_ITEM_KEY)
2583 found_type = BTRFS_INODE_ITEM_KEY;
2584 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2585 found_type = BTRFS_EXTENT_DATA_KEY;
2586 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2587 found_type = BTRFS_XATTR_ITEM_KEY;
2588 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2589 found_type = BTRFS_INODE_REF_KEY;
2590 else if (found_type)
2591 found_type--;
2592 else
2593 break;
2594 btrfs_set_key_type(&key, found_type);
2595 goto next;
2596 }
2597 if (found_key.offset >= new_size)
2598 del_item = 1;
2599 else
2600 del_item = 0;
2601 found_extent = 0;
2602
2603 /* FIXME, shrink the extent if the ref count is only 1 */
2604 if (found_type != BTRFS_EXTENT_DATA_KEY)
2605 goto delete;
2606
2607 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2608 u64 num_dec;
2609 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2610 if (!del_item && !encoding) {
2611 u64 orig_num_bytes =
2612 btrfs_file_extent_num_bytes(leaf, fi);
2613 extent_num_bytes = new_size -
2614 found_key.offset + root->sectorsize - 1;
2615 extent_num_bytes = extent_num_bytes &
2616 ~((u64)root->sectorsize - 1);
2617 btrfs_set_file_extent_num_bytes(leaf, fi,
2618 extent_num_bytes);
2619 num_dec = (orig_num_bytes -
2620 extent_num_bytes);
2621 if (root->ref_cows && extent_start != 0)
2622 inode_sub_bytes(inode, num_dec);
2623 btrfs_mark_buffer_dirty(leaf);
2624 } else {
2625 extent_num_bytes =
2626 btrfs_file_extent_disk_num_bytes(leaf,
2627 fi);
2628 /* FIXME blocksize != 4096 */
2629 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2630 if (extent_start != 0) {
2631 found_extent = 1;
2632 if (root->ref_cows)
2633 inode_sub_bytes(inode, num_dec);
2634 }
2635 root_gen = btrfs_header_generation(leaf);
2636 root_owner = btrfs_header_owner(leaf);
2637 }
2638 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2639 /*
2640 * we can't truncate inline items that have had
2641 * special encodings
2642 */
2643 if (!del_item &&
2644 btrfs_file_extent_compression(leaf, fi) == 0 &&
2645 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2646 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2647 u32 size = new_size - found_key.offset;
2648
2649 if (root->ref_cows) {
2650 inode_sub_bytes(inode, item_end + 1 -
2651 new_size);
2652 }
2653 size =
2654 btrfs_file_extent_calc_inline_size(size);
2655 ret = btrfs_truncate_item(trans, root, path,
2656 size, 1);
2657 BUG_ON(ret);
2658 } else if (root->ref_cows) {
2659 inode_sub_bytes(inode, item_end + 1 -
2660 found_key.offset);
2661 }
2662 }
2663 delete:
2664 if (del_item) {
2665 if (!pending_del_nr) {
2666 /* no pending yet, add ourselves */
2667 pending_del_slot = path->slots[0];
2668 pending_del_nr = 1;
2669 } else if (pending_del_nr &&
2670 path->slots[0] + 1 == pending_del_slot) {
2671 /* hop on the pending chunk */
2672 pending_del_nr++;
2673 pending_del_slot = path->slots[0];
2674 } else {
2675 BUG();
2676 }
2677 } else {
2678 break;
2679 }
2680 if (found_extent) {
2681 ret = btrfs_free_extent(trans, root, extent_start,
2682 extent_num_bytes,
2683 leaf->start, root_owner,
2684 root_gen, inode->i_ino, 0);
2685 BUG_ON(ret);
2686 }
2687 next:
2688 if (path->slots[0] == 0) {
2689 if (pending_del_nr)
2690 goto del_pending;
2691 btrfs_release_path(root, path);
2692 if (found_type == BTRFS_INODE_ITEM_KEY)
2693 break;
2694 goto search_again;
2695 }
2696
2697 path->slots[0]--;
2698 if (pending_del_nr &&
2699 path->slots[0] + 1 != pending_del_slot) {
2700 struct btrfs_key debug;
2701 del_pending:
2702 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2703 pending_del_slot);
2704 ret = btrfs_del_items(trans, root, path,
2705 pending_del_slot,
2706 pending_del_nr);
2707 BUG_ON(ret);
2708 pending_del_nr = 0;
2709 btrfs_release_path(root, path);
2710 if (found_type == BTRFS_INODE_ITEM_KEY)
2711 break;
2712 goto search_again;
2713 }
2714 }
2715 ret = 0;
2716 error:
2717 if (pending_del_nr) {
2718 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2719 pending_del_nr);
2720 }
2721 btrfs_free_path(path);
2722 inode->i_sb->s_dirt = 1;
2723 return ret;
2724 }
2725
2726 /*
2727 * taken from block_truncate_page, but does cow as it zeros out
2728 * any bytes left in the last page in the file.
2729 */
2730 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2731 {
2732 struct inode *inode = mapping->host;
2733 struct btrfs_root *root = BTRFS_I(inode)->root;
2734 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2735 struct btrfs_ordered_extent *ordered;
2736 char *kaddr;
2737 u32 blocksize = root->sectorsize;
2738 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2739 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2740 struct page *page;
2741 int ret = 0;
2742 u64 page_start;
2743 u64 page_end;
2744
2745 if ((offset & (blocksize - 1)) == 0)
2746 goto out;
2747
2748 ret = -ENOMEM;
2749 again:
2750 page = grab_cache_page(mapping, index);
2751 if (!page)
2752 goto out;
2753
2754 page_start = page_offset(page);
2755 page_end = page_start + PAGE_CACHE_SIZE - 1;
2756
2757 if (!PageUptodate(page)) {
2758 ret = btrfs_readpage(NULL, page);
2759 lock_page(page);
2760 if (page->mapping != mapping) {
2761 unlock_page(page);
2762 page_cache_release(page);
2763 goto again;
2764 }
2765 if (!PageUptodate(page)) {
2766 ret = -EIO;
2767 goto out_unlock;
2768 }
2769 }
2770 wait_on_page_writeback(page);
2771
2772 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2773 set_page_extent_mapped(page);
2774
2775 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2776 if (ordered) {
2777 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2778 unlock_page(page);
2779 page_cache_release(page);
2780 btrfs_start_ordered_extent(inode, ordered, 1);
2781 btrfs_put_ordered_extent(ordered);
2782 goto again;
2783 }
2784
2785 btrfs_set_extent_delalloc(inode, page_start, page_end);
2786 ret = 0;
2787 if (offset != PAGE_CACHE_SIZE) {
2788 kaddr = kmap(page);
2789 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2790 flush_dcache_page(page);
2791 kunmap(page);
2792 }
2793 ClearPageChecked(page);
2794 set_page_dirty(page);
2795 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2796
2797 out_unlock:
2798 unlock_page(page);
2799 page_cache_release(page);
2800 out:
2801 return ret;
2802 }
2803
2804 int btrfs_cont_expand(struct inode *inode, loff_t size)
2805 {
2806 struct btrfs_trans_handle *trans;
2807 struct btrfs_root *root = BTRFS_I(inode)->root;
2808 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2809 struct extent_map *em;
2810 u64 mask = root->sectorsize - 1;
2811 u64 hole_start = (inode->i_size + mask) & ~mask;
2812 u64 block_end = (size + mask) & ~mask;
2813 u64 last_byte;
2814 u64 cur_offset;
2815 u64 hole_size;
2816 int err;
2817
2818 if (size <= hole_start)
2819 return 0;
2820
2821 err = btrfs_check_free_space(root, 1, 0);
2822 if (err)
2823 return err;
2824
2825 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2826
2827 while (1) {
2828 struct btrfs_ordered_extent *ordered;
2829 btrfs_wait_ordered_range(inode, hole_start,
2830 block_end - hole_start);
2831 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2832 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2833 if (!ordered)
2834 break;
2835 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2836 btrfs_put_ordered_extent(ordered);
2837 }
2838
2839 trans = btrfs_start_transaction(root, 1);
2840 btrfs_set_trans_block_group(trans, inode);
2841
2842 cur_offset = hole_start;
2843 while (1) {
2844 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2845 block_end - cur_offset, 0);
2846 BUG_ON(IS_ERR(em) || !em);
2847 last_byte = min(extent_map_end(em), block_end);
2848 last_byte = (last_byte + mask) & ~mask;
2849 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2850 u64 hint_byte = 0;
2851 hole_size = last_byte - cur_offset;
2852 err = btrfs_drop_extents(trans, root, inode,
2853 cur_offset,
2854 cur_offset + hole_size,
2855 cur_offset, &hint_byte);
2856 if (err)
2857 break;
2858 err = btrfs_insert_file_extent(trans, root,
2859 inode->i_ino, cur_offset, 0,
2860 0, hole_size, 0, hole_size,
2861 0, 0, 0);
2862 btrfs_drop_extent_cache(inode, hole_start,
2863 last_byte - 1, 0);
2864 }
2865 free_extent_map(em);
2866 cur_offset = last_byte;
2867 if (err || cur_offset >= block_end)
2868 break;
2869 }
2870
2871 btrfs_end_transaction(trans, root);
2872 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2873 return err;
2874 }
2875
2876 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2877 {
2878 struct inode *inode = dentry->d_inode;
2879 int err;
2880
2881 err = inode_change_ok(inode, attr);
2882 if (err)
2883 return err;
2884
2885 if (S_ISREG(inode->i_mode) &&
2886 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2887 err = btrfs_cont_expand(inode, attr->ia_size);
2888 if (err)
2889 return err;
2890 }
2891
2892 err = inode_setattr(inode, attr);
2893
2894 if (!err && ((attr->ia_valid & ATTR_MODE)))
2895 err = btrfs_acl_chmod(inode);
2896 return err;
2897 }
2898
2899 void btrfs_delete_inode(struct inode *inode)
2900 {
2901 struct btrfs_trans_handle *trans;
2902 struct btrfs_root *root = BTRFS_I(inode)->root;
2903 unsigned long nr;
2904 int ret;
2905
2906 truncate_inode_pages(&inode->i_data, 0);
2907 if (is_bad_inode(inode)) {
2908 btrfs_orphan_del(NULL, inode);
2909 goto no_delete;
2910 }
2911 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2912
2913 btrfs_i_size_write(inode, 0);
2914 trans = btrfs_join_transaction(root, 1);
2915
2916 btrfs_set_trans_block_group(trans, inode);
2917 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2918 if (ret) {
2919 btrfs_orphan_del(NULL, inode);
2920 goto no_delete_lock;
2921 }
2922
2923 btrfs_orphan_del(trans, inode);
2924
2925 nr = trans->blocks_used;
2926 clear_inode(inode);
2927
2928 btrfs_end_transaction(trans, root);
2929 btrfs_btree_balance_dirty(root, nr);
2930 return;
2931
2932 no_delete_lock:
2933 nr = trans->blocks_used;
2934 btrfs_end_transaction(trans, root);
2935 btrfs_btree_balance_dirty(root, nr);
2936 no_delete:
2937 clear_inode(inode);
2938 }
2939
2940 /*
2941 * this returns the key found in the dir entry in the location pointer.
2942 * If no dir entries were found, location->objectid is 0.
2943 */
2944 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2945 struct btrfs_key *location)
2946 {
2947 const char *name = dentry->d_name.name;
2948 int namelen = dentry->d_name.len;
2949 struct btrfs_dir_item *di;
2950 struct btrfs_path *path;
2951 struct btrfs_root *root = BTRFS_I(dir)->root;
2952 int ret = 0;
2953
2954 path = btrfs_alloc_path();
2955 BUG_ON(!path);
2956
2957 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2958 namelen, 0);
2959 if (IS_ERR(di))
2960 ret = PTR_ERR(di);
2961
2962 if (!di || IS_ERR(di))
2963 goto out_err;
2964
2965 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2966 out:
2967 btrfs_free_path(path);
2968 return ret;
2969 out_err:
2970 location->objectid = 0;
2971 goto out;
2972 }
2973
2974 /*
2975 * when we hit a tree root in a directory, the btrfs part of the inode
2976 * needs to be changed to reflect the root directory of the tree root. This
2977 * is kind of like crossing a mount point.
2978 */
2979 static int fixup_tree_root_location(struct btrfs_root *root,
2980 struct btrfs_key *location,
2981 struct btrfs_root **sub_root,
2982 struct dentry *dentry)
2983 {
2984 struct btrfs_root_item *ri;
2985
2986 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2987 return 0;
2988 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2989 return 0;
2990
2991 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2992 dentry->d_name.name,
2993 dentry->d_name.len);
2994 if (IS_ERR(*sub_root))
2995 return PTR_ERR(*sub_root);
2996
2997 ri = &(*sub_root)->root_item;
2998 location->objectid = btrfs_root_dirid(ri);
2999 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3000 location->offset = 0;
3001
3002 return 0;
3003 }
3004
3005 static noinline void init_btrfs_i(struct inode *inode)
3006 {
3007 struct btrfs_inode *bi = BTRFS_I(inode);
3008
3009 bi->i_acl = NULL;
3010 bi->i_default_acl = NULL;
3011
3012 bi->generation = 0;
3013 bi->sequence = 0;
3014 bi->last_trans = 0;
3015 bi->logged_trans = 0;
3016 bi->delalloc_bytes = 0;
3017 bi->disk_i_size = 0;
3018 bi->flags = 0;
3019 bi->index_cnt = (u64)-1;
3020 bi->log_dirty_trans = 0;
3021 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3022 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3023 inode->i_mapping, GFP_NOFS);
3024 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3025 inode->i_mapping, GFP_NOFS);
3026 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3027 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3028 mutex_init(&BTRFS_I(inode)->extent_mutex);
3029 mutex_init(&BTRFS_I(inode)->log_mutex);
3030 }
3031
3032 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3033 {
3034 struct btrfs_iget_args *args = p;
3035 inode->i_ino = args->ino;
3036 init_btrfs_i(inode);
3037 BTRFS_I(inode)->root = args->root;
3038 return 0;
3039 }
3040
3041 static int btrfs_find_actor(struct inode *inode, void *opaque)
3042 {
3043 struct btrfs_iget_args *args = opaque;
3044 return args->ino == inode->i_ino &&
3045 args->root == BTRFS_I(inode)->root;
3046 }
3047
3048 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3049 struct btrfs_root *root, int wait)
3050 {
3051 struct inode *inode;
3052 struct btrfs_iget_args args;
3053 args.ino = objectid;
3054 args.root = root;
3055
3056 if (wait) {
3057 inode = ilookup5(s, objectid, btrfs_find_actor,
3058 (void *)&args);
3059 } else {
3060 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3061 (void *)&args);
3062 }
3063 return inode;
3064 }
3065
3066 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3067 struct btrfs_root *root)
3068 {
3069 struct inode *inode;
3070 struct btrfs_iget_args args;
3071 args.ino = objectid;
3072 args.root = root;
3073
3074 inode = iget5_locked(s, objectid, btrfs_find_actor,
3075 btrfs_init_locked_inode,
3076 (void *)&args);
3077 return inode;
3078 }
3079
3080 /* Get an inode object given its location and corresponding root.
3081 * Returns in *is_new if the inode was read from disk
3082 */
3083 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3084 struct btrfs_root *root, int *is_new)
3085 {
3086 struct inode *inode;
3087
3088 inode = btrfs_iget_locked(s, location->objectid, root);
3089 if (!inode)
3090 return ERR_PTR(-EACCES);
3091
3092 if (inode->i_state & I_NEW) {
3093 BTRFS_I(inode)->root = root;
3094 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3095 btrfs_read_locked_inode(inode);
3096 unlock_new_inode(inode);
3097 if (is_new)
3098 *is_new = 1;
3099 } else {
3100 if (is_new)
3101 *is_new = 0;
3102 }
3103
3104 return inode;
3105 }
3106
3107 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3108 {
3109 struct inode *inode;
3110 struct btrfs_inode *bi = BTRFS_I(dir);
3111 struct btrfs_root *root = bi->root;
3112 struct btrfs_root *sub_root = root;
3113 struct btrfs_key location;
3114 int ret, new;
3115
3116 if (dentry->d_name.len > BTRFS_NAME_LEN)
3117 return ERR_PTR(-ENAMETOOLONG);
3118
3119 ret = btrfs_inode_by_name(dir, dentry, &location);
3120
3121 if (ret < 0)
3122 return ERR_PTR(ret);
3123
3124 inode = NULL;
3125 if (location.objectid) {
3126 ret = fixup_tree_root_location(root, &location, &sub_root,
3127 dentry);
3128 if (ret < 0)
3129 return ERR_PTR(ret);
3130 if (ret > 0)
3131 return ERR_PTR(-ENOENT);
3132 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3133 if (IS_ERR(inode))
3134 return ERR_CAST(inode);
3135 }
3136 return inode;
3137 }
3138
3139 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3140 struct nameidata *nd)
3141 {
3142 struct inode *inode;
3143
3144 if (dentry->d_name.len > BTRFS_NAME_LEN)
3145 return ERR_PTR(-ENAMETOOLONG);
3146
3147 inode = btrfs_lookup_dentry(dir, dentry);
3148 if (IS_ERR(inode))
3149 return ERR_CAST(inode);
3150
3151 return d_splice_alias(inode, dentry);
3152 }
3153
3154 static unsigned char btrfs_filetype_table[] = {
3155 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3156 };
3157
3158 static int btrfs_real_readdir(struct file *filp, void *dirent,
3159 filldir_t filldir)
3160 {
3161 struct inode *inode = filp->f_dentry->d_inode;
3162 struct btrfs_root *root = BTRFS_I(inode)->root;
3163 struct btrfs_item *item;
3164 struct btrfs_dir_item *di;
3165 struct btrfs_key key;
3166 struct btrfs_key found_key;
3167 struct btrfs_path *path;
3168 int ret;
3169 u32 nritems;
3170 struct extent_buffer *leaf;
3171 int slot;
3172 int advance;
3173 unsigned char d_type;
3174 int over = 0;
3175 u32 di_cur;
3176 u32 di_total;
3177 u32 di_len;
3178 int key_type = BTRFS_DIR_INDEX_KEY;
3179 char tmp_name[32];
3180 char *name_ptr;
3181 int name_len;
3182
3183 /* FIXME, use a real flag for deciding about the key type */
3184 if (root->fs_info->tree_root == root)
3185 key_type = BTRFS_DIR_ITEM_KEY;
3186
3187 /* special case for "." */
3188 if (filp->f_pos == 0) {
3189 over = filldir(dirent, ".", 1,
3190 1, inode->i_ino,
3191 DT_DIR);
3192 if (over)
3193 return 0;
3194 filp->f_pos = 1;
3195 }
3196 /* special case for .., just use the back ref */
3197 if (filp->f_pos == 1) {
3198 u64 pino = parent_ino(filp->f_path.dentry);
3199 over = filldir(dirent, "..", 2,
3200 2, pino, DT_DIR);
3201 if (over)
3202 return 0;
3203 filp->f_pos = 2;
3204 }
3205 path = btrfs_alloc_path();
3206 path->reada = 2;
3207
3208 btrfs_set_key_type(&key, key_type);
3209 key.offset = filp->f_pos;
3210 key.objectid = inode->i_ino;
3211
3212 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3213 if (ret < 0)
3214 goto err;
3215 advance = 0;
3216
3217 while (1) {
3218 leaf = path->nodes[0];
3219 nritems = btrfs_header_nritems(leaf);
3220 slot = path->slots[0];
3221 if (advance || slot >= nritems) {
3222 if (slot >= nritems - 1) {
3223 ret = btrfs_next_leaf(root, path);
3224 if (ret)
3225 break;
3226 leaf = path->nodes[0];
3227 nritems = btrfs_header_nritems(leaf);
3228 slot = path->slots[0];
3229 } else {
3230 slot++;
3231 path->slots[0]++;
3232 }
3233 }
3234
3235 advance = 1;
3236 item = btrfs_item_nr(leaf, slot);
3237 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3238
3239 if (found_key.objectid != key.objectid)
3240 break;
3241 if (btrfs_key_type(&found_key) != key_type)
3242 break;
3243 if (found_key.offset < filp->f_pos)
3244 continue;
3245
3246 filp->f_pos = found_key.offset;
3247
3248 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3249 di_cur = 0;
3250 di_total = btrfs_item_size(leaf, item);
3251
3252 while (di_cur < di_total) {
3253 struct btrfs_key location;
3254
3255 name_len = btrfs_dir_name_len(leaf, di);
3256 if (name_len <= sizeof(tmp_name)) {
3257 name_ptr = tmp_name;
3258 } else {
3259 name_ptr = kmalloc(name_len, GFP_NOFS);
3260 if (!name_ptr) {
3261 ret = -ENOMEM;
3262 goto err;
3263 }
3264 }
3265 read_extent_buffer(leaf, name_ptr,
3266 (unsigned long)(di + 1), name_len);
3267
3268 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3269 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3270
3271 /* is this a reference to our own snapshot? If so
3272 * skip it
3273 */
3274 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3275 location.objectid == root->root_key.objectid) {
3276 over = 0;
3277 goto skip;
3278 }
3279 over = filldir(dirent, name_ptr, name_len,
3280 found_key.offset, location.objectid,
3281 d_type);
3282
3283 skip:
3284 if (name_ptr != tmp_name)
3285 kfree(name_ptr);
3286
3287 if (over)
3288 goto nopos;
3289 di_len = btrfs_dir_name_len(leaf, di) +
3290 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3291 di_cur += di_len;
3292 di = (struct btrfs_dir_item *)((char *)di + di_len);
3293 }
3294 }
3295
3296 /* Reached end of directory/root. Bump pos past the last item. */
3297 if (key_type == BTRFS_DIR_INDEX_KEY)
3298 filp->f_pos = INT_LIMIT(off_t);
3299 else
3300 filp->f_pos++;
3301 nopos:
3302 ret = 0;
3303 err:
3304 btrfs_free_path(path);
3305 return ret;
3306 }
3307
3308 int btrfs_write_inode(struct inode *inode, int wait)
3309 {
3310 struct btrfs_root *root = BTRFS_I(inode)->root;
3311 struct btrfs_trans_handle *trans;
3312 int ret = 0;
3313
3314 if (root->fs_info->btree_inode == inode)
3315 return 0;
3316
3317 if (wait) {
3318 trans = btrfs_join_transaction(root, 1);
3319 btrfs_set_trans_block_group(trans, inode);
3320 ret = btrfs_commit_transaction(trans, root);
3321 }
3322 return ret;
3323 }
3324
3325 /*
3326 * This is somewhat expensive, updating the tree every time the
3327 * inode changes. But, it is most likely to find the inode in cache.
3328 * FIXME, needs more benchmarking...there are no reasons other than performance
3329 * to keep or drop this code.
3330 */
3331 void btrfs_dirty_inode(struct inode *inode)
3332 {
3333 struct btrfs_root *root = BTRFS_I(inode)->root;
3334 struct btrfs_trans_handle *trans;
3335
3336 trans = btrfs_join_transaction(root, 1);
3337 btrfs_set_trans_block_group(trans, inode);
3338 btrfs_update_inode(trans, root, inode);
3339 btrfs_end_transaction(trans, root);
3340 }
3341
3342 /*
3343 * find the highest existing sequence number in a directory
3344 * and then set the in-memory index_cnt variable to reflect
3345 * free sequence numbers
3346 */
3347 static int btrfs_set_inode_index_count(struct inode *inode)
3348 {
3349 struct btrfs_root *root = BTRFS_I(inode)->root;
3350 struct btrfs_key key, found_key;
3351 struct btrfs_path *path;
3352 struct extent_buffer *leaf;
3353 int ret;
3354
3355 key.objectid = inode->i_ino;
3356 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3357 key.offset = (u64)-1;
3358
3359 path = btrfs_alloc_path();
3360 if (!path)
3361 return -ENOMEM;
3362
3363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3364 if (ret < 0)
3365 goto out;
3366 /* FIXME: we should be able to handle this */
3367 if (ret == 0)
3368 goto out;
3369 ret = 0;
3370
3371 /*
3372 * MAGIC NUMBER EXPLANATION:
3373 * since we search a directory based on f_pos we have to start at 2
3374 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3375 * else has to start at 2
3376 */
3377 if (path->slots[0] == 0) {
3378 BTRFS_I(inode)->index_cnt = 2;
3379 goto out;
3380 }
3381
3382 path->slots[0]--;
3383
3384 leaf = path->nodes[0];
3385 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3386
3387 if (found_key.objectid != inode->i_ino ||
3388 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3389 BTRFS_I(inode)->index_cnt = 2;
3390 goto out;
3391 }
3392
3393 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3394 out:
3395 btrfs_free_path(path);
3396 return ret;
3397 }
3398
3399 /*
3400 * helper to find a free sequence number in a given directory. This current
3401 * code is very simple, later versions will do smarter things in the btree
3402 */
3403 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3404 {
3405 int ret = 0;
3406
3407 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3408 ret = btrfs_set_inode_index_count(dir);
3409 if (ret)
3410 return ret;
3411 }
3412
3413 *index = BTRFS_I(dir)->index_cnt;
3414 BTRFS_I(dir)->index_cnt++;
3415
3416 return ret;
3417 }
3418
3419 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3420 struct btrfs_root *root,
3421 struct inode *dir,
3422 const char *name, int name_len,
3423 u64 ref_objectid, u64 objectid,
3424 u64 alloc_hint, int mode, u64 *index)
3425 {
3426 struct inode *inode;
3427 struct btrfs_inode_item *inode_item;
3428 struct btrfs_key *location;
3429 struct btrfs_path *path;
3430 struct btrfs_inode_ref *ref;
3431 struct btrfs_key key[2];
3432 u32 sizes[2];
3433 unsigned long ptr;
3434 int ret;
3435 int owner;
3436
3437 path = btrfs_alloc_path();
3438 BUG_ON(!path);
3439
3440 inode = new_inode(root->fs_info->sb);
3441 if (!inode)
3442 return ERR_PTR(-ENOMEM);
3443
3444 if (dir) {
3445 ret = btrfs_set_inode_index(dir, index);
3446 if (ret)
3447 return ERR_PTR(ret);
3448 }
3449 /*
3450 * index_cnt is ignored for everything but a dir,
3451 * btrfs_get_inode_index_count has an explanation for the magic
3452 * number
3453 */
3454 init_btrfs_i(inode);
3455 BTRFS_I(inode)->index_cnt = 2;
3456 BTRFS_I(inode)->root = root;
3457 BTRFS_I(inode)->generation = trans->transid;
3458
3459 if (mode & S_IFDIR)
3460 owner = 0;
3461 else
3462 owner = 1;
3463 BTRFS_I(inode)->block_group =
3464 btrfs_find_block_group(root, 0, alloc_hint, owner);
3465 if ((mode & S_IFREG)) {
3466 if (btrfs_test_opt(root, NODATASUM))
3467 btrfs_set_flag(inode, NODATASUM);
3468 if (btrfs_test_opt(root, NODATACOW))
3469 btrfs_set_flag(inode, NODATACOW);
3470 }
3471
3472 key[0].objectid = objectid;
3473 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3474 key[0].offset = 0;
3475
3476 key[1].objectid = objectid;
3477 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3478 key[1].offset = ref_objectid;
3479
3480 sizes[0] = sizeof(struct btrfs_inode_item);
3481 sizes[1] = name_len + sizeof(*ref);
3482
3483 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3484 if (ret != 0)
3485 goto fail;
3486
3487 if (objectid > root->highest_inode)
3488 root->highest_inode = objectid;
3489
3490 inode->i_uid = current_fsuid();
3491
3492 if (dir && (dir->i_mode & S_ISGID)) {
3493 inode->i_gid = dir->i_gid;
3494 if (S_ISDIR(mode))
3495 mode |= S_ISGID;
3496 } else
3497 inode->i_gid = current_fsgid();
3498
3499 inode->i_mode = mode;
3500 inode->i_ino = objectid;
3501 inode_set_bytes(inode, 0);
3502 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3503 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3504 struct btrfs_inode_item);
3505 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3506
3507 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3508 struct btrfs_inode_ref);
3509 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3510 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3511 ptr = (unsigned long)(ref + 1);
3512 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3513
3514 btrfs_mark_buffer_dirty(path->nodes[0]);
3515 btrfs_free_path(path);
3516
3517 location = &BTRFS_I(inode)->location;
3518 location->objectid = objectid;
3519 location->offset = 0;
3520 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3521
3522 insert_inode_hash(inode);
3523 return inode;
3524 fail:
3525 if (dir)
3526 BTRFS_I(dir)->index_cnt--;
3527 btrfs_free_path(path);
3528 return ERR_PTR(ret);
3529 }
3530
3531 static inline u8 btrfs_inode_type(struct inode *inode)
3532 {
3533 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3534 }
3535
3536 /*
3537 * utility function to add 'inode' into 'parent_inode' with
3538 * a give name and a given sequence number.
3539 * if 'add_backref' is true, also insert a backref from the
3540 * inode to the parent directory.
3541 */
3542 int btrfs_add_link(struct btrfs_trans_handle *trans,
3543 struct inode *parent_inode, struct inode *inode,
3544 const char *name, int name_len, int add_backref, u64 index)
3545 {
3546 int ret;
3547 struct btrfs_key key;
3548 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3549
3550 key.objectid = inode->i_ino;
3551 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3552 key.offset = 0;
3553
3554 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3555 parent_inode->i_ino,
3556 &key, btrfs_inode_type(inode),
3557 index);
3558 if (ret == 0) {
3559 if (add_backref) {
3560 ret = btrfs_insert_inode_ref(trans, root,
3561 name, name_len,
3562 inode->i_ino,
3563 parent_inode->i_ino,
3564 index);
3565 }
3566 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3567 name_len * 2);
3568 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3569 ret = btrfs_update_inode(trans, root, parent_inode);
3570 }
3571 return ret;
3572 }
3573
3574 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3575 struct dentry *dentry, struct inode *inode,
3576 int backref, u64 index)
3577 {
3578 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3579 inode, dentry->d_name.name,
3580 dentry->d_name.len, backref, index);
3581 if (!err) {
3582 d_instantiate(dentry, inode);
3583 return 0;
3584 }
3585 if (err > 0)
3586 err = -EEXIST;
3587 return err;
3588 }
3589
3590 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3591 int mode, dev_t rdev)
3592 {
3593 struct btrfs_trans_handle *trans;
3594 struct btrfs_root *root = BTRFS_I(dir)->root;
3595 struct inode *inode = NULL;
3596 int err;
3597 int drop_inode = 0;
3598 u64 objectid;
3599 unsigned long nr = 0;
3600 u64 index = 0;
3601
3602 if (!new_valid_dev(rdev))
3603 return -EINVAL;
3604
3605 err = btrfs_check_free_space(root, 1, 0);
3606 if (err)
3607 goto fail;
3608
3609 trans = btrfs_start_transaction(root, 1);
3610 btrfs_set_trans_block_group(trans, dir);
3611
3612 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3613 if (err) {
3614 err = -ENOSPC;
3615 goto out_unlock;
3616 }
3617
3618 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3619 dentry->d_name.len,
3620 dentry->d_parent->d_inode->i_ino, objectid,
3621 BTRFS_I(dir)->block_group, mode, &index);
3622 err = PTR_ERR(inode);
3623 if (IS_ERR(inode))
3624 goto out_unlock;
3625
3626 err = btrfs_init_inode_security(inode, dir);
3627 if (err) {
3628 drop_inode = 1;
3629 goto out_unlock;
3630 }
3631
3632 btrfs_set_trans_block_group(trans, inode);
3633 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3634 if (err)
3635 drop_inode = 1;
3636 else {
3637 inode->i_op = &btrfs_special_inode_operations;
3638 init_special_inode(inode, inode->i_mode, rdev);
3639 btrfs_update_inode(trans, root, inode);
3640 }
3641 dir->i_sb->s_dirt = 1;
3642 btrfs_update_inode_block_group(trans, inode);
3643 btrfs_update_inode_block_group(trans, dir);
3644 out_unlock:
3645 nr = trans->blocks_used;
3646 btrfs_end_transaction_throttle(trans, root);
3647 fail:
3648 if (drop_inode) {
3649 inode_dec_link_count(inode);
3650 iput(inode);
3651 }
3652 btrfs_btree_balance_dirty(root, nr);
3653 return err;
3654 }
3655
3656 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3657 int mode, struct nameidata *nd)
3658 {
3659 struct btrfs_trans_handle *trans;
3660 struct btrfs_root *root = BTRFS_I(dir)->root;
3661 struct inode *inode = NULL;
3662 int err;
3663 int drop_inode = 0;
3664 unsigned long nr = 0;
3665 u64 objectid;
3666 u64 index = 0;
3667
3668 err = btrfs_check_free_space(root, 1, 0);
3669 if (err)
3670 goto fail;
3671 trans = btrfs_start_transaction(root, 1);
3672 btrfs_set_trans_block_group(trans, dir);
3673
3674 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3675 if (err) {
3676 err = -ENOSPC;
3677 goto out_unlock;
3678 }
3679
3680 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3681 dentry->d_name.len,
3682 dentry->d_parent->d_inode->i_ino,
3683 objectid, BTRFS_I(dir)->block_group, mode,
3684 &index);
3685 err = PTR_ERR(inode);
3686 if (IS_ERR(inode))
3687 goto out_unlock;
3688
3689 err = btrfs_init_inode_security(inode, dir);
3690 if (err) {
3691 drop_inode = 1;
3692 goto out_unlock;
3693 }
3694
3695 btrfs_set_trans_block_group(trans, inode);
3696 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3697 if (err)
3698 drop_inode = 1;
3699 else {
3700 inode->i_mapping->a_ops = &btrfs_aops;
3701 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3702 inode->i_fop = &btrfs_file_operations;
3703 inode->i_op = &btrfs_file_inode_operations;
3704 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3705 }
3706 dir->i_sb->s_dirt = 1;
3707 btrfs_update_inode_block_group(trans, inode);
3708 btrfs_update_inode_block_group(trans, dir);
3709 out_unlock:
3710 nr = trans->blocks_used;
3711 btrfs_end_transaction_throttle(trans, root);
3712 fail:
3713 if (drop_inode) {
3714 inode_dec_link_count(inode);
3715 iput(inode);
3716 }
3717 btrfs_btree_balance_dirty(root, nr);
3718 return err;
3719 }
3720
3721 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3722 struct dentry *dentry)
3723 {
3724 struct btrfs_trans_handle *trans;
3725 struct btrfs_root *root = BTRFS_I(dir)->root;
3726 struct inode *inode = old_dentry->d_inode;
3727 u64 index;
3728 unsigned long nr = 0;
3729 int err;
3730 int drop_inode = 0;
3731
3732 if (inode->i_nlink == 0)
3733 return -ENOENT;
3734
3735 btrfs_inc_nlink(inode);
3736 err = btrfs_check_free_space(root, 1, 0);
3737 if (err)
3738 goto fail;
3739 err = btrfs_set_inode_index(dir, &index);
3740 if (err)
3741 goto fail;
3742
3743 trans = btrfs_start_transaction(root, 1);
3744
3745 btrfs_set_trans_block_group(trans, dir);
3746 atomic_inc(&inode->i_count);
3747
3748 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3749
3750 if (err)
3751 drop_inode = 1;
3752
3753 dir->i_sb->s_dirt = 1;
3754 btrfs_update_inode_block_group(trans, dir);
3755 err = btrfs_update_inode(trans, root, inode);
3756
3757 if (err)
3758 drop_inode = 1;
3759
3760 nr = trans->blocks_used;
3761 btrfs_end_transaction_throttle(trans, root);
3762 fail:
3763 if (drop_inode) {
3764 inode_dec_link_count(inode);
3765 iput(inode);
3766 }
3767 btrfs_btree_balance_dirty(root, nr);
3768 return err;
3769 }
3770
3771 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3772 {
3773 struct inode *inode = NULL;
3774 struct btrfs_trans_handle *trans;
3775 struct btrfs_root *root = BTRFS_I(dir)->root;
3776 int err = 0;
3777 int drop_on_err = 0;
3778 u64 objectid = 0;
3779 u64 index = 0;
3780 unsigned long nr = 1;
3781
3782 err = btrfs_check_free_space(root, 1, 0);
3783 if (err)
3784 goto out_unlock;
3785
3786 trans = btrfs_start_transaction(root, 1);
3787 btrfs_set_trans_block_group(trans, dir);
3788
3789 if (IS_ERR(trans)) {
3790 err = PTR_ERR(trans);
3791 goto out_unlock;
3792 }
3793
3794 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3795 if (err) {
3796 err = -ENOSPC;
3797 goto out_unlock;
3798 }
3799
3800 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3801 dentry->d_name.len,
3802 dentry->d_parent->d_inode->i_ino, objectid,
3803 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3804 &index);
3805 if (IS_ERR(inode)) {
3806 err = PTR_ERR(inode);
3807 goto out_fail;
3808 }
3809
3810 drop_on_err = 1;
3811
3812 err = btrfs_init_inode_security(inode, dir);
3813 if (err)
3814 goto out_fail;
3815
3816 inode->i_op = &btrfs_dir_inode_operations;
3817 inode->i_fop = &btrfs_dir_file_operations;
3818 btrfs_set_trans_block_group(trans, inode);
3819
3820 btrfs_i_size_write(inode, 0);
3821 err = btrfs_update_inode(trans, root, inode);
3822 if (err)
3823 goto out_fail;
3824
3825 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3826 inode, dentry->d_name.name,
3827 dentry->d_name.len, 0, index);
3828 if (err)
3829 goto out_fail;
3830
3831 d_instantiate(dentry, inode);
3832 drop_on_err = 0;
3833 dir->i_sb->s_dirt = 1;
3834 btrfs_update_inode_block_group(trans, inode);
3835 btrfs_update_inode_block_group(trans, dir);
3836
3837 out_fail:
3838 nr = trans->blocks_used;
3839 btrfs_end_transaction_throttle(trans, root);
3840
3841 out_unlock:
3842 if (drop_on_err)
3843 iput(inode);
3844 btrfs_btree_balance_dirty(root, nr);
3845 return err;
3846 }
3847
3848 /* helper for btfs_get_extent. Given an existing extent in the tree,
3849 * and an extent that you want to insert, deal with overlap and insert
3850 * the new extent into the tree.
3851 */
3852 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3853 struct extent_map *existing,
3854 struct extent_map *em,
3855 u64 map_start, u64 map_len)
3856 {
3857 u64 start_diff;
3858
3859 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3860 start_diff = map_start - em->start;
3861 em->start = map_start;
3862 em->len = map_len;
3863 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3864 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3865 em->block_start += start_diff;
3866 em->block_len -= start_diff;
3867 }
3868 return add_extent_mapping(em_tree, em);
3869 }
3870
3871 static noinline int uncompress_inline(struct btrfs_path *path,
3872 struct inode *inode, struct page *page,
3873 size_t pg_offset, u64 extent_offset,
3874 struct btrfs_file_extent_item *item)
3875 {
3876 int ret;
3877 struct extent_buffer *leaf = path->nodes[0];
3878 char *tmp;
3879 size_t max_size;
3880 unsigned long inline_size;
3881 unsigned long ptr;
3882
3883 WARN_ON(pg_offset != 0);
3884 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3885 inline_size = btrfs_file_extent_inline_item_len(leaf,
3886 btrfs_item_nr(leaf, path->slots[0]));
3887 tmp = kmalloc(inline_size, GFP_NOFS);
3888 ptr = btrfs_file_extent_inline_start(item);
3889
3890 read_extent_buffer(leaf, tmp, ptr, inline_size);
3891
3892 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3893 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3894 inline_size, max_size);
3895 if (ret) {
3896 char *kaddr = kmap_atomic(page, KM_USER0);
3897 unsigned long copy_size = min_t(u64,
3898 PAGE_CACHE_SIZE - pg_offset,
3899 max_size - extent_offset);
3900 memset(kaddr + pg_offset, 0, copy_size);
3901 kunmap_atomic(kaddr, KM_USER0);
3902 }
3903 kfree(tmp);
3904 return 0;
3905 }
3906
3907 /*
3908 * a bit scary, this does extent mapping from logical file offset to the disk.
3909 * the ugly parts come from merging extents from the disk with the in-ram
3910 * representation. This gets more complex because of the data=ordered code,
3911 * where the in-ram extents might be locked pending data=ordered completion.
3912 *
3913 * This also copies inline extents directly into the page.
3914 */
3915
3916 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3917 size_t pg_offset, u64 start, u64 len,
3918 int create)
3919 {
3920 int ret;
3921 int err = 0;
3922 u64 bytenr;
3923 u64 extent_start = 0;
3924 u64 extent_end = 0;
3925 u64 objectid = inode->i_ino;
3926 u32 found_type;
3927 struct btrfs_path *path = NULL;
3928 struct btrfs_root *root = BTRFS_I(inode)->root;
3929 struct btrfs_file_extent_item *item;
3930 struct extent_buffer *leaf;
3931 struct btrfs_key found_key;
3932 struct extent_map *em = NULL;
3933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3934 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3935 struct btrfs_trans_handle *trans = NULL;
3936 int compressed;
3937
3938 again:
3939 spin_lock(&em_tree->lock);
3940 em = lookup_extent_mapping(em_tree, start, len);
3941 if (em)
3942 em->bdev = root->fs_info->fs_devices->latest_bdev;
3943 spin_unlock(&em_tree->lock);
3944
3945 if (em) {
3946 if (em->start > start || em->start + em->len <= start)
3947 free_extent_map(em);
3948 else if (em->block_start == EXTENT_MAP_INLINE && page)
3949 free_extent_map(em);
3950 else
3951 goto out;
3952 }
3953 em = alloc_extent_map(GFP_NOFS);
3954 if (!em) {
3955 err = -ENOMEM;
3956 goto out;
3957 }
3958 em->bdev = root->fs_info->fs_devices->latest_bdev;
3959 em->start = EXTENT_MAP_HOLE;
3960 em->orig_start = EXTENT_MAP_HOLE;
3961 em->len = (u64)-1;
3962 em->block_len = (u64)-1;
3963
3964 if (!path) {
3965 path = btrfs_alloc_path();
3966 BUG_ON(!path);
3967 }
3968
3969 ret = btrfs_lookup_file_extent(trans, root, path,
3970 objectid, start, trans != NULL);
3971 if (ret < 0) {
3972 err = ret;
3973 goto out;
3974 }
3975
3976 if (ret != 0) {
3977 if (path->slots[0] == 0)
3978 goto not_found;
3979 path->slots[0]--;
3980 }
3981
3982 leaf = path->nodes[0];
3983 item = btrfs_item_ptr(leaf, path->slots[0],
3984 struct btrfs_file_extent_item);
3985 /* are we inside the extent that was found? */
3986 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3987 found_type = btrfs_key_type(&found_key);
3988 if (found_key.objectid != objectid ||
3989 found_type != BTRFS_EXTENT_DATA_KEY) {
3990 goto not_found;
3991 }
3992
3993 found_type = btrfs_file_extent_type(leaf, item);
3994 extent_start = found_key.offset;
3995 compressed = btrfs_file_extent_compression(leaf, item);
3996 if (found_type == BTRFS_FILE_EXTENT_REG ||
3997 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3998 extent_end = extent_start +
3999 btrfs_file_extent_num_bytes(leaf, item);
4000 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4001 size_t size;
4002 size = btrfs_file_extent_inline_len(leaf, item);
4003 extent_end = (extent_start + size + root->sectorsize - 1) &
4004 ~((u64)root->sectorsize - 1);
4005 }
4006
4007 if (start >= extent_end) {
4008 path->slots[0]++;
4009 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4010 ret = btrfs_next_leaf(root, path);
4011 if (ret < 0) {
4012 err = ret;
4013 goto out;
4014 }
4015 if (ret > 0)
4016 goto not_found;
4017 leaf = path->nodes[0];
4018 }
4019 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4020 if (found_key.objectid != objectid ||
4021 found_key.type != BTRFS_EXTENT_DATA_KEY)
4022 goto not_found;
4023 if (start + len <= found_key.offset)
4024 goto not_found;
4025 em->start = start;
4026 em->len = found_key.offset - start;
4027 goto not_found_em;
4028 }
4029
4030 if (found_type == BTRFS_FILE_EXTENT_REG ||
4031 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4032 em->start = extent_start;
4033 em->len = extent_end - extent_start;
4034 em->orig_start = extent_start -
4035 btrfs_file_extent_offset(leaf, item);
4036 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4037 if (bytenr == 0) {
4038 em->block_start = EXTENT_MAP_HOLE;
4039 goto insert;
4040 }
4041 if (compressed) {
4042 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4043 em->block_start = bytenr;
4044 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4045 item);
4046 } else {
4047 bytenr += btrfs_file_extent_offset(leaf, item);
4048 em->block_start = bytenr;
4049 em->block_len = em->len;
4050 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4051 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4052 }
4053 goto insert;
4054 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4055 unsigned long ptr;
4056 char *map;
4057 size_t size;
4058 size_t extent_offset;
4059 size_t copy_size;
4060
4061 em->block_start = EXTENT_MAP_INLINE;
4062 if (!page || create) {
4063 em->start = extent_start;
4064 em->len = extent_end - extent_start;
4065 goto out;
4066 }
4067
4068 size = btrfs_file_extent_inline_len(leaf, item);
4069 extent_offset = page_offset(page) + pg_offset - extent_start;
4070 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4071 size - extent_offset);
4072 em->start = extent_start + extent_offset;
4073 em->len = (copy_size + root->sectorsize - 1) &
4074 ~((u64)root->sectorsize - 1);
4075 em->orig_start = EXTENT_MAP_INLINE;
4076 if (compressed)
4077 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4078 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4079 if (create == 0 && !PageUptodate(page)) {
4080 if (btrfs_file_extent_compression(leaf, item) ==
4081 BTRFS_COMPRESS_ZLIB) {
4082 ret = uncompress_inline(path, inode, page,
4083 pg_offset,
4084 extent_offset, item);
4085 BUG_ON(ret);
4086 } else {
4087 map = kmap(page);
4088 read_extent_buffer(leaf, map + pg_offset, ptr,
4089 copy_size);
4090 kunmap(page);
4091 }
4092 flush_dcache_page(page);
4093 } else if (create && PageUptodate(page)) {
4094 if (!trans) {
4095 kunmap(page);
4096 free_extent_map(em);
4097 em = NULL;
4098 btrfs_release_path(root, path);
4099 trans = btrfs_join_transaction(root, 1);
4100 goto again;
4101 }
4102 map = kmap(page);
4103 write_extent_buffer(leaf, map + pg_offset, ptr,
4104 copy_size);
4105 kunmap(page);
4106 btrfs_mark_buffer_dirty(leaf);
4107 }
4108 set_extent_uptodate(io_tree, em->start,
4109 extent_map_end(em) - 1, GFP_NOFS);
4110 goto insert;
4111 } else {
4112 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4113 WARN_ON(1);
4114 }
4115 not_found:
4116 em->start = start;
4117 em->len = len;
4118 not_found_em:
4119 em->block_start = EXTENT_MAP_HOLE;
4120 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4121 insert:
4122 btrfs_release_path(root, path);
4123 if (em->start > start || extent_map_end(em) <= start) {
4124 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4125 "[%llu %llu]\n", (unsigned long long)em->start,
4126 (unsigned long long)em->len,
4127 (unsigned long long)start,
4128 (unsigned long long)len);
4129 err = -EIO;
4130 goto out;
4131 }
4132
4133 err = 0;
4134 spin_lock(&em_tree->lock);
4135 ret = add_extent_mapping(em_tree, em);
4136 /* it is possible that someone inserted the extent into the tree
4137 * while we had the lock dropped. It is also possible that
4138 * an overlapping map exists in the tree
4139 */
4140 if (ret == -EEXIST) {
4141 struct extent_map *existing;
4142
4143 ret = 0;
4144
4145 existing = lookup_extent_mapping(em_tree, start, len);
4146 if (existing && (existing->start > start ||
4147 existing->start + existing->len <= start)) {
4148 free_extent_map(existing);
4149 existing = NULL;
4150 }
4151 if (!existing) {
4152 existing = lookup_extent_mapping(em_tree, em->start,
4153 em->len);
4154 if (existing) {
4155 err = merge_extent_mapping(em_tree, existing,
4156 em, start,
4157 root->sectorsize);
4158 free_extent_map(existing);
4159 if (err) {
4160 free_extent_map(em);
4161 em = NULL;
4162 }
4163 } else {
4164 err = -EIO;
4165 free_extent_map(em);
4166 em = NULL;
4167 }
4168 } else {
4169 free_extent_map(em);
4170 em = existing;
4171 err = 0;
4172 }
4173 }
4174 spin_unlock(&em_tree->lock);
4175 out:
4176 if (path)
4177 btrfs_free_path(path);
4178 if (trans) {
4179 ret = btrfs_end_transaction(trans, root);
4180 if (!err)
4181 err = ret;
4182 }
4183 if (err) {
4184 free_extent_map(em);
4185 WARN_ON(1);
4186 return ERR_PTR(err);
4187 }
4188 return em;
4189 }
4190
4191 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4192 const struct iovec *iov, loff_t offset,
4193 unsigned long nr_segs)
4194 {
4195 return -EINVAL;
4196 }
4197
4198 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4199 __u64 start, __u64 len)
4200 {
4201 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4202 }
4203
4204 int btrfs_readpage(struct file *file, struct page *page)
4205 {
4206 struct extent_io_tree *tree;
4207 tree = &BTRFS_I(page->mapping->host)->io_tree;
4208 return extent_read_full_page(tree, page, btrfs_get_extent);
4209 }
4210
4211 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4212 {
4213 struct extent_io_tree *tree;
4214
4215
4216 if (current->flags & PF_MEMALLOC) {
4217 redirty_page_for_writepage(wbc, page);
4218 unlock_page(page);
4219 return 0;
4220 }
4221 tree = &BTRFS_I(page->mapping->host)->io_tree;
4222 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4223 }
4224
4225 int btrfs_writepages(struct address_space *mapping,
4226 struct writeback_control *wbc)
4227 {
4228 struct extent_io_tree *tree;
4229
4230 tree = &BTRFS_I(mapping->host)->io_tree;
4231 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4232 }
4233
4234 static int
4235 btrfs_readpages(struct file *file, struct address_space *mapping,
4236 struct list_head *pages, unsigned nr_pages)
4237 {
4238 struct extent_io_tree *tree;
4239 tree = &BTRFS_I(mapping->host)->io_tree;
4240 return extent_readpages(tree, mapping, pages, nr_pages,
4241 btrfs_get_extent);
4242 }
4243 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4244 {
4245 struct extent_io_tree *tree;
4246 struct extent_map_tree *map;
4247 int ret;
4248
4249 tree = &BTRFS_I(page->mapping->host)->io_tree;
4250 map = &BTRFS_I(page->mapping->host)->extent_tree;
4251 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4252 if (ret == 1) {
4253 ClearPagePrivate(page);
4254 set_page_private(page, 0);
4255 page_cache_release(page);
4256 }
4257 return ret;
4258 }
4259
4260 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4261 {
4262 if (PageWriteback(page) || PageDirty(page))
4263 return 0;
4264 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4265 }
4266
4267 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4268 {
4269 struct extent_io_tree *tree;
4270 struct btrfs_ordered_extent *ordered;
4271 u64 page_start = page_offset(page);
4272 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4273
4274 wait_on_page_writeback(page);
4275 tree = &BTRFS_I(page->mapping->host)->io_tree;
4276 if (offset) {
4277 btrfs_releasepage(page, GFP_NOFS);
4278 return;
4279 }
4280
4281 lock_extent(tree, page_start, page_end, GFP_NOFS);
4282 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4283 page_offset(page));
4284 if (ordered) {
4285 /*
4286 * IO on this page will never be started, so we need
4287 * to account for any ordered extents now
4288 */
4289 clear_extent_bit(tree, page_start, page_end,
4290 EXTENT_DIRTY | EXTENT_DELALLOC |
4291 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4292 btrfs_finish_ordered_io(page->mapping->host,
4293 page_start, page_end);
4294 btrfs_put_ordered_extent(ordered);
4295 lock_extent(tree, page_start, page_end, GFP_NOFS);
4296 }
4297 clear_extent_bit(tree, page_start, page_end,
4298 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4299 EXTENT_ORDERED,
4300 1, 1, GFP_NOFS);
4301 __btrfs_releasepage(page, GFP_NOFS);
4302
4303 ClearPageChecked(page);
4304 if (PagePrivate(page)) {
4305 ClearPagePrivate(page);
4306 set_page_private(page, 0);
4307 page_cache_release(page);
4308 }
4309 }
4310
4311 /*
4312 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4313 * called from a page fault handler when a page is first dirtied. Hence we must
4314 * be careful to check for EOF conditions here. We set the page up correctly
4315 * for a written page which means we get ENOSPC checking when writing into
4316 * holes and correct delalloc and unwritten extent mapping on filesystems that
4317 * support these features.
4318 *
4319 * We are not allowed to take the i_mutex here so we have to play games to
4320 * protect against truncate races as the page could now be beyond EOF. Because
4321 * vmtruncate() writes the inode size before removing pages, once we have the
4322 * page lock we can determine safely if the page is beyond EOF. If it is not
4323 * beyond EOF, then the page is guaranteed safe against truncation until we
4324 * unlock the page.
4325 */
4326 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4327 {
4328 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4329 struct btrfs_root *root = BTRFS_I(inode)->root;
4330 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4331 struct btrfs_ordered_extent *ordered;
4332 char *kaddr;
4333 unsigned long zero_start;
4334 loff_t size;
4335 int ret;
4336 u64 page_start;
4337 u64 page_end;
4338
4339 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4340 if (ret)
4341 goto out;
4342
4343 ret = -EINVAL;
4344 again:
4345 lock_page(page);
4346 size = i_size_read(inode);
4347 page_start = page_offset(page);
4348 page_end = page_start + PAGE_CACHE_SIZE - 1;
4349
4350 if ((page->mapping != inode->i_mapping) ||
4351 (page_start >= size)) {
4352 /* page got truncated out from underneath us */
4353 goto out_unlock;
4354 }
4355 wait_on_page_writeback(page);
4356
4357 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4358 set_page_extent_mapped(page);
4359
4360 /*
4361 * we can't set the delalloc bits if there are pending ordered
4362 * extents. Drop our locks and wait for them to finish
4363 */
4364 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4365 if (ordered) {
4366 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4367 unlock_page(page);
4368 btrfs_start_ordered_extent(inode, ordered, 1);
4369 btrfs_put_ordered_extent(ordered);
4370 goto again;
4371 }
4372
4373 btrfs_set_extent_delalloc(inode, page_start, page_end);
4374 ret = 0;
4375
4376 /* page is wholly or partially inside EOF */
4377 if (page_start + PAGE_CACHE_SIZE > size)
4378 zero_start = size & ~PAGE_CACHE_MASK;
4379 else
4380 zero_start = PAGE_CACHE_SIZE;
4381
4382 if (zero_start != PAGE_CACHE_SIZE) {
4383 kaddr = kmap(page);
4384 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4385 flush_dcache_page(page);
4386 kunmap(page);
4387 }
4388 ClearPageChecked(page);
4389 set_page_dirty(page);
4390 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4391
4392 out_unlock:
4393 unlock_page(page);
4394 out:
4395 return ret;
4396 }
4397
4398 static void btrfs_truncate(struct inode *inode)
4399 {
4400 struct btrfs_root *root = BTRFS_I(inode)->root;
4401 int ret;
4402 struct btrfs_trans_handle *trans;
4403 unsigned long nr;
4404 u64 mask = root->sectorsize - 1;
4405
4406 if (!S_ISREG(inode->i_mode))
4407 return;
4408 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4409 return;
4410
4411 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4412 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4413
4414 trans = btrfs_start_transaction(root, 1);
4415 btrfs_set_trans_block_group(trans, inode);
4416 btrfs_i_size_write(inode, inode->i_size);
4417
4418 ret = btrfs_orphan_add(trans, inode);
4419 if (ret)
4420 goto out;
4421 /* FIXME, add redo link to tree so we don't leak on crash */
4422 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4423 BTRFS_EXTENT_DATA_KEY);
4424 btrfs_update_inode(trans, root, inode);
4425
4426 ret = btrfs_orphan_del(trans, inode);
4427 BUG_ON(ret);
4428
4429 out:
4430 nr = trans->blocks_used;
4431 ret = btrfs_end_transaction_throttle(trans, root);
4432 BUG_ON(ret);
4433 btrfs_btree_balance_dirty(root, nr);
4434 }
4435
4436 /*
4437 * create a new subvolume directory/inode (helper for the ioctl).
4438 */
4439 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4440 struct btrfs_root *new_root, struct dentry *dentry,
4441 u64 new_dirid, u64 alloc_hint)
4442 {
4443 struct inode *inode;
4444 int error;
4445 u64 index = 0;
4446
4447 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4448 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4449 if (IS_ERR(inode))
4450 return PTR_ERR(inode);
4451 inode->i_op = &btrfs_dir_inode_operations;
4452 inode->i_fop = &btrfs_dir_file_operations;
4453
4454 inode->i_nlink = 1;
4455 btrfs_i_size_write(inode, 0);
4456
4457 error = btrfs_update_inode(trans, new_root, inode);
4458 if (error)
4459 return error;
4460
4461 d_instantiate(dentry, inode);
4462 return 0;
4463 }
4464
4465 /* helper function for file defrag and space balancing. This
4466 * forces readahead on a given range of bytes in an inode
4467 */
4468 unsigned long btrfs_force_ra(struct address_space *mapping,
4469 struct file_ra_state *ra, struct file *file,
4470 pgoff_t offset, pgoff_t last_index)
4471 {
4472 pgoff_t req_size = last_index - offset + 1;
4473
4474 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4475 return offset + req_size;
4476 }
4477
4478 struct inode *btrfs_alloc_inode(struct super_block *sb)
4479 {
4480 struct btrfs_inode *ei;
4481
4482 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4483 if (!ei)
4484 return NULL;
4485 ei->last_trans = 0;
4486 ei->logged_trans = 0;
4487 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4488 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4489 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4490 INIT_LIST_HEAD(&ei->i_orphan);
4491 return &ei->vfs_inode;
4492 }
4493
4494 void btrfs_destroy_inode(struct inode *inode)
4495 {
4496 struct btrfs_ordered_extent *ordered;
4497 WARN_ON(!list_empty(&inode->i_dentry));
4498 WARN_ON(inode->i_data.nrpages);
4499
4500 if (BTRFS_I(inode)->i_acl &&
4501 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4502 posix_acl_release(BTRFS_I(inode)->i_acl);
4503 if (BTRFS_I(inode)->i_default_acl &&
4504 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4505 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4506
4507 spin_lock(&BTRFS_I(inode)->root->list_lock);
4508 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4509 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4510 " list\n", inode->i_ino);
4511 dump_stack();
4512 }
4513 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4514
4515 while (1) {
4516 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4517 if (!ordered)
4518 break;
4519 else {
4520 printk(KERN_ERR "btrfs found ordered "
4521 "extent %llu %llu on inode cleanup\n",
4522 (unsigned long long)ordered->file_offset,
4523 (unsigned long long)ordered->len);
4524 btrfs_remove_ordered_extent(inode, ordered);
4525 btrfs_put_ordered_extent(ordered);
4526 btrfs_put_ordered_extent(ordered);
4527 }
4528 }
4529 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4530 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4531 }
4532
4533 static void init_once(void *foo)
4534 {
4535 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4536
4537 inode_init_once(&ei->vfs_inode);
4538 }
4539
4540 void btrfs_destroy_cachep(void)
4541 {
4542 if (btrfs_inode_cachep)
4543 kmem_cache_destroy(btrfs_inode_cachep);
4544 if (btrfs_trans_handle_cachep)
4545 kmem_cache_destroy(btrfs_trans_handle_cachep);
4546 if (btrfs_transaction_cachep)
4547 kmem_cache_destroy(btrfs_transaction_cachep);
4548 if (btrfs_bit_radix_cachep)
4549 kmem_cache_destroy(btrfs_bit_radix_cachep);
4550 if (btrfs_path_cachep)
4551 kmem_cache_destroy(btrfs_path_cachep);
4552 }
4553
4554 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4555 unsigned long extra_flags,
4556 void (*ctor)(void *))
4557 {
4558 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4559 SLAB_MEM_SPREAD | extra_flags), ctor);
4560 }
4561
4562 int btrfs_init_cachep(void)
4563 {
4564 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4565 sizeof(struct btrfs_inode),
4566 0, init_once);
4567 if (!btrfs_inode_cachep)
4568 goto fail;
4569 btrfs_trans_handle_cachep =
4570 btrfs_cache_create("btrfs_trans_handle_cache",
4571 sizeof(struct btrfs_trans_handle),
4572 0, NULL);
4573 if (!btrfs_trans_handle_cachep)
4574 goto fail;
4575 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4576 sizeof(struct btrfs_transaction),
4577 0, NULL);
4578 if (!btrfs_transaction_cachep)
4579 goto fail;
4580 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4581 sizeof(struct btrfs_path),
4582 0, NULL);
4583 if (!btrfs_path_cachep)
4584 goto fail;
4585 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4586 SLAB_DESTROY_BY_RCU, NULL);
4587 if (!btrfs_bit_radix_cachep)
4588 goto fail;
4589 return 0;
4590 fail:
4591 btrfs_destroy_cachep();
4592 return -ENOMEM;
4593 }
4594
4595 static int btrfs_getattr(struct vfsmount *mnt,
4596 struct dentry *dentry, struct kstat *stat)
4597 {
4598 struct inode *inode = dentry->d_inode;
4599 generic_fillattr(inode, stat);
4600 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4601 stat->blksize = PAGE_CACHE_SIZE;
4602 stat->blocks = (inode_get_bytes(inode) +
4603 BTRFS_I(inode)->delalloc_bytes) >> 9;
4604 return 0;
4605 }
4606
4607 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4608 struct inode *new_dir, struct dentry *new_dentry)
4609 {
4610 struct btrfs_trans_handle *trans;
4611 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4612 struct inode *new_inode = new_dentry->d_inode;
4613 struct inode *old_inode = old_dentry->d_inode;
4614 struct timespec ctime = CURRENT_TIME;
4615 u64 index = 0;
4616 int ret;
4617
4618 /* we're not allowed to rename between subvolumes */
4619 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4620 BTRFS_I(new_dir)->root->root_key.objectid)
4621 return -EXDEV;
4622
4623 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4624 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4625 return -ENOTEMPTY;
4626 }
4627
4628 /* to rename a snapshot or subvolume, we need to juggle the
4629 * backrefs. This isn't coded yet
4630 */
4631 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4632 return -EXDEV;
4633
4634 ret = btrfs_check_free_space(root, 1, 0);
4635 if (ret)
4636 goto out_unlock;
4637
4638 trans = btrfs_start_transaction(root, 1);
4639
4640 btrfs_set_trans_block_group(trans, new_dir);
4641
4642 btrfs_inc_nlink(old_dentry->d_inode);
4643 old_dir->i_ctime = old_dir->i_mtime = ctime;
4644 new_dir->i_ctime = new_dir->i_mtime = ctime;
4645 old_inode->i_ctime = ctime;
4646
4647 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4648 old_dentry->d_name.name,
4649 old_dentry->d_name.len);
4650 if (ret)
4651 goto out_fail;
4652
4653 if (new_inode) {
4654 new_inode->i_ctime = CURRENT_TIME;
4655 ret = btrfs_unlink_inode(trans, root, new_dir,
4656 new_dentry->d_inode,
4657 new_dentry->d_name.name,
4658 new_dentry->d_name.len);
4659 if (ret)
4660 goto out_fail;
4661 if (new_inode->i_nlink == 0) {
4662 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4663 if (ret)
4664 goto out_fail;
4665 }
4666
4667 }
4668 ret = btrfs_set_inode_index(new_dir, &index);
4669 if (ret)
4670 goto out_fail;
4671
4672 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4673 old_inode, new_dentry->d_name.name,
4674 new_dentry->d_name.len, 1, index);
4675 if (ret)
4676 goto out_fail;
4677
4678 out_fail:
4679 btrfs_end_transaction_throttle(trans, root);
4680 out_unlock:
4681 return ret;
4682 }
4683
4684 /*
4685 * some fairly slow code that needs optimization. This walks the list
4686 * of all the inodes with pending delalloc and forces them to disk.
4687 */
4688 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4689 {
4690 struct list_head *head = &root->fs_info->delalloc_inodes;
4691 struct btrfs_inode *binode;
4692 struct inode *inode;
4693
4694 if (root->fs_info->sb->s_flags & MS_RDONLY)
4695 return -EROFS;
4696
4697 spin_lock(&root->fs_info->delalloc_lock);
4698 while (!list_empty(head)) {
4699 binode = list_entry(head->next, struct btrfs_inode,
4700 delalloc_inodes);
4701 inode = igrab(&binode->vfs_inode);
4702 if (!inode)
4703 list_del_init(&binode->delalloc_inodes);
4704 spin_unlock(&root->fs_info->delalloc_lock);
4705 if (inode) {
4706 filemap_flush(inode->i_mapping);
4707 iput(inode);
4708 }
4709 cond_resched();
4710 spin_lock(&root->fs_info->delalloc_lock);
4711 }
4712 spin_unlock(&root->fs_info->delalloc_lock);
4713
4714 /* the filemap_flush will queue IO into the worker threads, but
4715 * we have to make sure the IO is actually started and that
4716 * ordered extents get created before we return
4717 */
4718 atomic_inc(&root->fs_info->async_submit_draining);
4719 while (atomic_read(&root->fs_info->nr_async_submits) ||
4720 atomic_read(&root->fs_info->async_delalloc_pages)) {
4721 wait_event(root->fs_info->async_submit_wait,
4722 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4723 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4724 }
4725 atomic_dec(&root->fs_info->async_submit_draining);
4726 return 0;
4727 }
4728
4729 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4730 const char *symname)
4731 {
4732 struct btrfs_trans_handle *trans;
4733 struct btrfs_root *root = BTRFS_I(dir)->root;
4734 struct btrfs_path *path;
4735 struct btrfs_key key;
4736 struct inode *inode = NULL;
4737 int err;
4738 int drop_inode = 0;
4739 u64 objectid;
4740 u64 index = 0 ;
4741 int name_len;
4742 int datasize;
4743 unsigned long ptr;
4744 struct btrfs_file_extent_item *ei;
4745 struct extent_buffer *leaf;
4746 unsigned long nr = 0;
4747
4748 name_len = strlen(symname) + 1;
4749 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4750 return -ENAMETOOLONG;
4751
4752 err = btrfs_check_free_space(root, 1, 0);
4753 if (err)
4754 goto out_fail;
4755
4756 trans = btrfs_start_transaction(root, 1);
4757 btrfs_set_trans_block_group(trans, dir);
4758
4759 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4760 if (err) {
4761 err = -ENOSPC;
4762 goto out_unlock;
4763 }
4764
4765 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4766 dentry->d_name.len,
4767 dentry->d_parent->d_inode->i_ino, objectid,
4768 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4769 &index);
4770 err = PTR_ERR(inode);
4771 if (IS_ERR(inode))
4772 goto out_unlock;
4773
4774 err = btrfs_init_inode_security(inode, dir);
4775 if (err) {
4776 drop_inode = 1;
4777 goto out_unlock;
4778 }
4779
4780 btrfs_set_trans_block_group(trans, inode);
4781 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4782 if (err)
4783 drop_inode = 1;
4784 else {
4785 inode->i_mapping->a_ops = &btrfs_aops;
4786 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4787 inode->i_fop = &btrfs_file_operations;
4788 inode->i_op = &btrfs_file_inode_operations;
4789 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4790 }
4791 dir->i_sb->s_dirt = 1;
4792 btrfs_update_inode_block_group(trans, inode);
4793 btrfs_update_inode_block_group(trans, dir);
4794 if (drop_inode)
4795 goto out_unlock;
4796
4797 path = btrfs_alloc_path();
4798 BUG_ON(!path);
4799 key.objectid = inode->i_ino;
4800 key.offset = 0;
4801 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4802 datasize = btrfs_file_extent_calc_inline_size(name_len);
4803 err = btrfs_insert_empty_item(trans, root, path, &key,
4804 datasize);
4805 if (err) {
4806 drop_inode = 1;
4807 goto out_unlock;
4808 }
4809 leaf = path->nodes[0];
4810 ei = btrfs_item_ptr(leaf, path->slots[0],
4811 struct btrfs_file_extent_item);
4812 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4813 btrfs_set_file_extent_type(leaf, ei,
4814 BTRFS_FILE_EXTENT_INLINE);
4815 btrfs_set_file_extent_encryption(leaf, ei, 0);
4816 btrfs_set_file_extent_compression(leaf, ei, 0);
4817 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4818 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4819
4820 ptr = btrfs_file_extent_inline_start(ei);
4821 write_extent_buffer(leaf, symname, ptr, name_len);
4822 btrfs_mark_buffer_dirty(leaf);
4823 btrfs_free_path(path);
4824
4825 inode->i_op = &btrfs_symlink_inode_operations;
4826 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4827 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4828 inode_set_bytes(inode, name_len);
4829 btrfs_i_size_write(inode, name_len - 1);
4830 err = btrfs_update_inode(trans, root, inode);
4831 if (err)
4832 drop_inode = 1;
4833
4834 out_unlock:
4835 nr = trans->blocks_used;
4836 btrfs_end_transaction_throttle(trans, root);
4837 out_fail:
4838 if (drop_inode) {
4839 inode_dec_link_count(inode);
4840 iput(inode);
4841 }
4842 btrfs_btree_balance_dirty(root, nr);
4843 return err;
4844 }
4845
4846 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4847 u64 alloc_hint, int mode)
4848 {
4849 struct btrfs_trans_handle *trans;
4850 struct btrfs_root *root = BTRFS_I(inode)->root;
4851 struct btrfs_key ins;
4852 u64 alloc_size;
4853 u64 cur_offset = start;
4854 u64 num_bytes = end - start;
4855 int ret = 0;
4856
4857 trans = btrfs_join_transaction(root, 1);
4858 BUG_ON(!trans);
4859 btrfs_set_trans_block_group(trans, inode);
4860
4861 while (num_bytes > 0) {
4862 alloc_size = min(num_bytes, root->fs_info->max_extent);
4863 ret = btrfs_reserve_extent(trans, root, alloc_size,
4864 root->sectorsize, 0, alloc_hint,
4865 (u64)-1, &ins, 1);
4866 if (ret) {
4867 WARN_ON(1);
4868 goto out;
4869 }
4870 ret = insert_reserved_file_extent(trans, inode,
4871 cur_offset, ins.objectid,
4872 ins.offset, ins.offset,
4873 ins.offset, 0, 0, 0,
4874 BTRFS_FILE_EXTENT_PREALLOC);
4875 BUG_ON(ret);
4876 num_bytes -= ins.offset;
4877 cur_offset += ins.offset;
4878 alloc_hint = ins.objectid + ins.offset;
4879 }
4880 out:
4881 if (cur_offset > start) {
4882 inode->i_ctime = CURRENT_TIME;
4883 btrfs_set_flag(inode, PREALLOC);
4884 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4885 cur_offset > i_size_read(inode))
4886 btrfs_i_size_write(inode, cur_offset);
4887 ret = btrfs_update_inode(trans, root, inode);
4888 BUG_ON(ret);
4889 }
4890
4891 btrfs_end_transaction(trans, root);
4892 return ret;
4893 }
4894
4895 static long btrfs_fallocate(struct inode *inode, int mode,
4896 loff_t offset, loff_t len)
4897 {
4898 u64 cur_offset;
4899 u64 last_byte;
4900 u64 alloc_start;
4901 u64 alloc_end;
4902 u64 alloc_hint = 0;
4903 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4904 struct extent_map *em;
4905 int ret;
4906
4907 alloc_start = offset & ~mask;
4908 alloc_end = (offset + len + mask) & ~mask;
4909
4910 mutex_lock(&inode->i_mutex);
4911 if (alloc_start > inode->i_size) {
4912 ret = btrfs_cont_expand(inode, alloc_start);
4913 if (ret)
4914 goto out;
4915 }
4916
4917 while (1) {
4918 struct btrfs_ordered_extent *ordered;
4919 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4920 alloc_end - 1, GFP_NOFS);
4921 ordered = btrfs_lookup_first_ordered_extent(inode,
4922 alloc_end - 1);
4923 if (ordered &&
4924 ordered->file_offset + ordered->len > alloc_start &&
4925 ordered->file_offset < alloc_end) {
4926 btrfs_put_ordered_extent(ordered);
4927 unlock_extent(&BTRFS_I(inode)->io_tree,
4928 alloc_start, alloc_end - 1, GFP_NOFS);
4929 btrfs_wait_ordered_range(inode, alloc_start,
4930 alloc_end - alloc_start);
4931 } else {
4932 if (ordered)
4933 btrfs_put_ordered_extent(ordered);
4934 break;
4935 }
4936 }
4937
4938 cur_offset = alloc_start;
4939 while (1) {
4940 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4941 alloc_end - cur_offset, 0);
4942 BUG_ON(IS_ERR(em) || !em);
4943 last_byte = min(extent_map_end(em), alloc_end);
4944 last_byte = (last_byte + mask) & ~mask;
4945 if (em->block_start == EXTENT_MAP_HOLE) {
4946 ret = prealloc_file_range(inode, cur_offset,
4947 last_byte, alloc_hint, mode);
4948 if (ret < 0) {
4949 free_extent_map(em);
4950 break;
4951 }
4952 }
4953 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4954 alloc_hint = em->block_start;
4955 free_extent_map(em);
4956
4957 cur_offset = last_byte;
4958 if (cur_offset >= alloc_end) {
4959 ret = 0;
4960 break;
4961 }
4962 }
4963 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4964 GFP_NOFS);
4965 out:
4966 mutex_unlock(&inode->i_mutex);
4967 return ret;
4968 }
4969
4970 static int btrfs_set_page_dirty(struct page *page)
4971 {
4972 return __set_page_dirty_nobuffers(page);
4973 }
4974
4975 static int btrfs_permission(struct inode *inode, int mask)
4976 {
4977 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4978 return -EACCES;
4979 return generic_permission(inode, mask, btrfs_check_acl);
4980 }
4981
4982 static struct inode_operations btrfs_dir_inode_operations = {
4983 .getattr = btrfs_getattr,
4984 .lookup = btrfs_lookup,
4985 .create = btrfs_create,
4986 .unlink = btrfs_unlink,
4987 .link = btrfs_link,
4988 .mkdir = btrfs_mkdir,
4989 .rmdir = btrfs_rmdir,
4990 .rename = btrfs_rename,
4991 .symlink = btrfs_symlink,
4992 .setattr = btrfs_setattr,
4993 .mknod = btrfs_mknod,
4994 .setxattr = btrfs_setxattr,
4995 .getxattr = btrfs_getxattr,
4996 .listxattr = btrfs_listxattr,
4997 .removexattr = btrfs_removexattr,
4998 .permission = btrfs_permission,
4999 };
5000 static struct inode_operations btrfs_dir_ro_inode_operations = {
5001 .lookup = btrfs_lookup,
5002 .permission = btrfs_permission,
5003 };
5004 static struct file_operations btrfs_dir_file_operations = {
5005 .llseek = generic_file_llseek,
5006 .read = generic_read_dir,
5007 .readdir = btrfs_real_readdir,
5008 .unlocked_ioctl = btrfs_ioctl,
5009 #ifdef CONFIG_COMPAT
5010 .compat_ioctl = btrfs_ioctl,
5011 #endif
5012 .release = btrfs_release_file,
5013 .fsync = btrfs_sync_file,
5014 };
5015
5016 static struct extent_io_ops btrfs_extent_io_ops = {
5017 .fill_delalloc = run_delalloc_range,
5018 .submit_bio_hook = btrfs_submit_bio_hook,
5019 .merge_bio_hook = btrfs_merge_bio_hook,
5020 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5021 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5022 .writepage_start_hook = btrfs_writepage_start_hook,
5023 .readpage_io_failed_hook = btrfs_io_failed_hook,
5024 .set_bit_hook = btrfs_set_bit_hook,
5025 .clear_bit_hook = btrfs_clear_bit_hook,
5026 };
5027
5028 /*
5029 * btrfs doesn't support the bmap operation because swapfiles
5030 * use bmap to make a mapping of extents in the file. They assume
5031 * these extents won't change over the life of the file and they
5032 * use the bmap result to do IO directly to the drive.
5033 *
5034 * the btrfs bmap call would return logical addresses that aren't
5035 * suitable for IO and they also will change frequently as COW
5036 * operations happen. So, swapfile + btrfs == corruption.
5037 *
5038 * For now we're avoiding this by dropping bmap.
5039 */
5040 static struct address_space_operations btrfs_aops = {
5041 .readpage = btrfs_readpage,
5042 .writepage = btrfs_writepage,
5043 .writepages = btrfs_writepages,
5044 .readpages = btrfs_readpages,
5045 .sync_page = block_sync_page,
5046 .direct_IO = btrfs_direct_IO,
5047 .invalidatepage = btrfs_invalidatepage,
5048 .releasepage = btrfs_releasepage,
5049 .set_page_dirty = btrfs_set_page_dirty,
5050 };
5051
5052 static struct address_space_operations btrfs_symlink_aops = {
5053 .readpage = btrfs_readpage,
5054 .writepage = btrfs_writepage,
5055 .invalidatepage = btrfs_invalidatepage,
5056 .releasepage = btrfs_releasepage,
5057 };
5058
5059 static struct inode_operations btrfs_file_inode_operations = {
5060 .truncate = btrfs_truncate,
5061 .getattr = btrfs_getattr,
5062 .setattr = btrfs_setattr,
5063 .setxattr = btrfs_setxattr,
5064 .getxattr = btrfs_getxattr,
5065 .listxattr = btrfs_listxattr,
5066 .removexattr = btrfs_removexattr,
5067 .permission = btrfs_permission,
5068 .fallocate = btrfs_fallocate,
5069 .fiemap = btrfs_fiemap,
5070 };
5071 static struct inode_operations btrfs_special_inode_operations = {
5072 .getattr = btrfs_getattr,
5073 .setattr = btrfs_setattr,
5074 .permission = btrfs_permission,
5075 .setxattr = btrfs_setxattr,
5076 .getxattr = btrfs_getxattr,
5077 .listxattr = btrfs_listxattr,
5078 .removexattr = btrfs_removexattr,
5079 };
5080 static struct inode_operations btrfs_symlink_inode_operations = {
5081 .readlink = generic_readlink,
5082 .follow_link = page_follow_link_light,
5083 .put_link = page_put_link,
5084 .permission = btrfs_permission,
5085 .setxattr = btrfs_setxattr,
5086 .getxattr = btrfs_getxattr,
5087 .listxattr = btrfs_listxattr,
5088 .removexattr = btrfs_removexattr,
5089 };
This page took 0.137169 seconds and 6 git commands to generate.