?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: kill the orphan space calculation for snapshots
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / inode.c
blob5f5f8a577e697ddbffb81d4d01d7ffd2cb631aca
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include "compat.h"
42 #include "ctree.h"
43 #include "disk-io.h"
44 #include "transaction.h"
45 #include "btrfs_inode.h"
46 #include "ioctl.h"
47 #include "print-tree.h"
48 #include "volumes.h"
49 #include "ordered-data.h"
50 #include "xattr.h"
51 #include "tree-log.h"
52 #include "compression.h"
53 #include "locking.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
56
57 struct btrfs_iget_args {
58 u64 ino;
59 struct btrfs_root *root;
60 };
61
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct address_space_operations btrfs_symlink_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static struct extent_io_ops btrfs_extent_io_ops;
71
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_transaction_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
77
78 #define S_SHIFT 12
79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 };
88
89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
90 static int btrfs_truncate(struct inode *inode);
91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
92 static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
96
97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
98 struct inode *inode, struct inode *dir,
99 const struct qstr *qstr)
100 {
101 int err;
102
103 err = btrfs_init_acl(trans, inode, dir);
104 if (!err)
105 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
106 return err;
107 }
108
109 /*
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
113 */
114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
115 struct btrfs_root *root, struct inode *inode,
116 u64 start, size_t size, size_t compressed_size,
117 int compress_type,
118 struct page **compressed_pages)
119 {
120 struct btrfs_key key;
121 struct btrfs_path *path;
122 struct extent_buffer *leaf;
123 struct page *page = NULL;
124 char *kaddr;
125 unsigned long ptr;
126 struct btrfs_file_extent_item *ei;
127 int err = 0;
128 int ret;
129 size_t cur_size = size;
130 size_t datasize;
131 unsigned long offset;
132
133 if (compressed_size && compressed_pages)
134 cur_size = compressed_size;
135
136 path = btrfs_alloc_path();
137 if (!path)
138 return -ENOMEM;
139
140 path->leave_spinning = 1;
141
142 key.objectid = btrfs_ino(inode);
143 key.offset = start;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
149 datasize);
150 BUG_ON(ret);
151 if (ret) {
152 err = ret;
153 goto fail;
154 }
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
164
165 if (compress_type != BTRFS_COMPRESS_NONE) {
166 struct page *cpage;
167 int i = 0;
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
171 PAGE_CACHE_SIZE);
172
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
176
177 i++;
178 ptr += cur_size;
179 compressed_size -= cur_size;
180 }
181 btrfs_set_file_extent_compression(leaf, ei,
182 compress_type);
183 } else {
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
192 }
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
195
196 /*
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
200 *
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
204 */
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
207
208 return 0;
209 fail:
210 btrfs_free_path(path);
211 return err;
212 }
213
214
215 /*
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
219 */
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size, int compress_type,
224 struct page **compressed_pages)
225 {
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
231 u64 hint_byte;
232 u64 data_len = inline_len;
233 int ret;
234
235 if (compressed_size)
236 data_len = compressed_size;
237
238 if (start > 0 ||
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (!compressed_size &&
242 (actual_end & (root->sectorsize - 1)) == 0) ||
243 end + 1 < isize ||
244 data_len > root->fs_info->max_inline) {
245 return 1;
246 }
247
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 &hint_byte, 1);
250 BUG_ON(ret);
251
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
256 compress_type, compressed_pages);
257 BUG_ON(ret);
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
260 return 0;
261 }
262
263 struct async_extent {
264 u64 start;
265 u64 ram_size;
266 u64 compressed_size;
267 struct page **pages;
268 unsigned long nr_pages;
269 int compress_type;
270 struct list_head list;
271 };
272
273 struct async_cow {
274 struct inode *inode;
275 struct btrfs_root *root;
276 struct page *locked_page;
277 u64 start;
278 u64 end;
279 struct list_head extents;
280 struct btrfs_work work;
281 };
282
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
285 u64 compressed_size,
286 struct page **pages,
287 unsigned long nr_pages,
288 int compress_type)
289 {
290 struct async_extent *async_extent;
291
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 BUG_ON(!async_extent);
294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
299 async_extent->compress_type = compress_type;
300 list_add_tail(&async_extent->list, &cow->extents);
301 return 0;
302 }
303
304 /*
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
308 *
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
314 *
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
319 */
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
322 u64 start, u64 end,
323 struct async_cow *async_cow,
324 int *num_added)
325 {
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
328 u64 num_bytes;
329 u64 blocksize = root->sectorsize;
330 u64 actual_end;
331 u64 isize = i_size_read(inode);
332 int ret = 0;
333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
339 unsigned long max_uncompressed = 128 * 1024;
340 int i;
341 int will_compress;
342 int compress_type = root->fs_info->compress_type;
343
344 /* if this is a small write inside eof, kick off a defragbot */
345 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL, inode);
347
348 actual_end = min_t(u64, isize, end + 1);
349 again:
350 will_compress = 0;
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353
354 /*
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
360 *
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
363 */
364 if (actual_end <= start)
365 goto cleanup_and_bail_uncompressed;
366
367 total_compressed = actual_end - start;
368
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
374 *
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
378 */
379 total_compressed = min(total_compressed, max_uncompressed);
380 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
381 num_bytes = max(blocksize, num_bytes);
382 total_in = 0;
383 ret = 0;
384
385 /*
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
389 */
390 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
391 (btrfs_test_opt(root, COMPRESS) ||
392 (BTRFS_I(inode)->force_compress) ||
393 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
394 WARN_ON(pages);
395 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
396 BUG_ON(!pages);
397
398 if (BTRFS_I(inode)->force_compress)
399 compress_type = BTRFS_I(inode)->force_compress;
400
401 ret = btrfs_compress_pages(compress_type,
402 inode->i_mapping, start,
403 total_compressed, pages,
404 nr_pages, &nr_pages_ret,
405 &total_in,
406 &total_compressed,
407 max_compressed);
408
409 if (!ret) {
410 unsigned long offset = total_compressed &
411 (PAGE_CACHE_SIZE - 1);
412 struct page *page = pages[nr_pages_ret - 1];
413 char *kaddr;
414
415 /* zero the tail end of the last page, we might be
416 * sending it down to disk
417 */
418 if (offset) {
419 kaddr = kmap_atomic(page, KM_USER0);
420 memset(kaddr + offset, 0,
421 PAGE_CACHE_SIZE - offset);
422 kunmap_atomic(kaddr, KM_USER0);
423 }
424 will_compress = 1;
425 }
426 }
427 if (start == 0) {
428 trans = btrfs_join_transaction(root);
429 BUG_ON(IS_ERR(trans));
430 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
431
432 /* lets try to make an inline extent */
433 if (ret || total_in < (actual_end - start)) {
434 /* we didn't compress the entire range, try
435 * to make an uncompressed inline extent.
436 */
437 ret = cow_file_range_inline(trans, root, inode,
438 start, end, 0, 0, NULL);
439 } else {
440 /* try making a compressed inline extent */
441 ret = cow_file_range_inline(trans, root, inode,
442 start, end,
443 total_compressed,
444 compress_type, pages);
445 }
446 if (ret == 0) {
447 /*
448 * inline extent creation worked, we don't need
449 * to create any more async work items. Unlock
450 * and free up our temp pages.
451 */
452 extent_clear_unlock_delalloc(inode,
453 &BTRFS_I(inode)->io_tree,
454 start, end, NULL,
455 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
456 EXTENT_CLEAR_DELALLOC |
457 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
458
459 btrfs_end_transaction(trans, root);
460 goto free_pages_out;
461 }
462 btrfs_end_transaction(trans, root);
463 }
464
465 if (will_compress) {
466 /*
467 * we aren't doing an inline extent round the compressed size
468 * up to a block size boundary so the allocator does sane
469 * things
470 */
471 total_compressed = (total_compressed + blocksize - 1) &
472 ~(blocksize - 1);
473
474 /*
475 * one last check to make sure the compression is really a
476 * win, compare the page count read with the blocks on disk
477 */
478 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
479 ~(PAGE_CACHE_SIZE - 1);
480 if (total_compressed >= total_in) {
481 will_compress = 0;
482 } else {
483 num_bytes = total_in;
484 }
485 }
486 if (!will_compress && pages) {
487 /*
488 * the compression code ran but failed to make things smaller,
489 * free any pages it allocated and our page pointer array
490 */
491 for (i = 0; i < nr_pages_ret; i++) {
492 WARN_ON(pages[i]->mapping);
493 page_cache_release(pages[i]);
494 }
495 kfree(pages);
496 pages = NULL;
497 total_compressed = 0;
498 nr_pages_ret = 0;
499
500 /* flag the file so we don't compress in the future */
501 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
502 !(BTRFS_I(inode)->force_compress)) {
503 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
504 }
505 }
506 if (will_compress) {
507 *num_added += 1;
508
509 /* the async work queues will take care of doing actual
510 * allocation on disk for these compressed pages,
511 * and will submit them to the elevator.
512 */
513 add_async_extent(async_cow, start, num_bytes,
514 total_compressed, pages, nr_pages_ret,
515 compress_type);
516
517 if (start + num_bytes < end) {
518 start += num_bytes;
519 pages = NULL;
520 cond_resched();
521 goto again;
522 }
523 } else {
524 cleanup_and_bail_uncompressed:
525 /*
526 * No compression, but we still need to write the pages in
527 * the file we've been given so far. redirty the locked
528 * page if it corresponds to our extent and set things up
529 * for the async work queue to run cow_file_range to do
530 * the normal delalloc dance
531 */
532 if (page_offset(locked_page) >= start &&
533 page_offset(locked_page) <= end) {
534 __set_page_dirty_nobuffers(locked_page);
535 /* unlocked later on in the async handlers */
536 }
537 add_async_extent(async_cow, start, end - start + 1,
538 0, NULL, 0, BTRFS_COMPRESS_NONE);
539 *num_added += 1;
540 }
541
542 out:
543 return 0;
544
545 free_pages_out:
546 for (i = 0; i < nr_pages_ret; i++) {
547 WARN_ON(pages[i]->mapping);
548 page_cache_release(pages[i]);
549 }
550 kfree(pages);
551
552 goto out;
553 }
554
555 /*
556 * phase two of compressed writeback. This is the ordered portion
557 * of the code, which only gets called in the order the work was
558 * queued. We walk all the async extents created by compress_file_range
559 * and send them down to the disk.
560 */
561 static noinline int submit_compressed_extents(struct inode *inode,
562 struct async_cow *async_cow)
563 {
564 struct async_extent *async_extent;
565 u64 alloc_hint = 0;
566 struct btrfs_trans_handle *trans;
567 struct btrfs_key ins;
568 struct extent_map *em;
569 struct btrfs_root *root = BTRFS_I(inode)->root;
570 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
571 struct extent_io_tree *io_tree;
572 int ret = 0;
573
574 if (list_empty(&async_cow->extents))
575 return 0;
576
577
578 while (!list_empty(&async_cow->extents)) {
579 async_extent = list_entry(async_cow->extents.next,
580 struct async_extent, list);
581 list_del(&async_extent->list);
582
583 io_tree = &BTRFS_I(inode)->io_tree;
584
585 retry:
586 /* did the compression code fall back to uncompressed IO? */
587 if (!async_extent->pages) {
588 int page_started = 0;
589 unsigned long nr_written = 0;
590
591 lock_extent(io_tree, async_extent->start,
592 async_extent->start +
593 async_extent->ram_size - 1, GFP_NOFS);
594
595 /* allocate blocks */
596 ret = cow_file_range(inode, async_cow->locked_page,
597 async_extent->start,
598 async_extent->start +
599 async_extent->ram_size - 1,
600 &page_started, &nr_written, 0);
601
602 /*
603 * if page_started, cow_file_range inserted an
604 * inline extent and took care of all the unlocking
605 * and IO for us. Otherwise, we need to submit
606 * all those pages down to the drive.
607 */
608 if (!page_started && !ret)
609 extent_write_locked_range(io_tree,
610 inode, async_extent->start,
611 async_extent->start +
612 async_extent->ram_size - 1,
613 btrfs_get_extent,
614 WB_SYNC_ALL);
615 kfree(async_extent);
616 cond_resched();
617 continue;
618 }
619
620 lock_extent(io_tree, async_extent->start,
621 async_extent->start + async_extent->ram_size - 1,
622 GFP_NOFS);
623
624 trans = btrfs_join_transaction(root);
625 BUG_ON(IS_ERR(trans));
626 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
627 ret = btrfs_reserve_extent(trans, root,
628 async_extent->compressed_size,
629 async_extent->compressed_size,
630 0, alloc_hint,
631 (u64)-1, &ins, 1);
632 btrfs_end_transaction(trans, root);
633
634 if (ret) {
635 int i;
636 for (i = 0; i < async_extent->nr_pages; i++) {
637 WARN_ON(async_extent->pages[i]->mapping);
638 page_cache_release(async_extent->pages[i]);
639 }
640 kfree(async_extent->pages);
641 async_extent->nr_pages = 0;
642 async_extent->pages = NULL;
643 unlock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1, GFP_NOFS);
646 goto retry;
647 }
648
649 /*
650 * here we're doing allocation and writeback of the
651 * compressed pages
652 */
653 btrfs_drop_extent_cache(inode, async_extent->start,
654 async_extent->start +
655 async_extent->ram_size - 1, 0);
656
657 em = alloc_extent_map();
658 BUG_ON(!em);
659 em->start = async_extent->start;
660 em->len = async_extent->ram_size;
661 em->orig_start = em->start;
662
663 em->block_start = ins.objectid;
664 em->block_len = ins.offset;
665 em->bdev = root->fs_info->fs_devices->latest_bdev;
666 em->compress_type = async_extent->compress_type;
667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
668 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
669
670 while (1) {
671 write_lock(&em_tree->lock);
672 ret = add_extent_mapping(em_tree, em);
673 write_unlock(&em_tree->lock);
674 if (ret != -EEXIST) {
675 free_extent_map(em);
676 break;
677 }
678 btrfs_drop_extent_cache(inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1, 0);
681 }
682
683 ret = btrfs_add_ordered_extent_compress(inode,
684 async_extent->start,
685 ins.objectid,
686 async_extent->ram_size,
687 ins.offset,
688 BTRFS_ORDERED_COMPRESSED,
689 async_extent->compress_type);
690 BUG_ON(ret);
691
692 /*
693 * clear dirty, set writeback and unlock the pages.
694 */
695 extent_clear_unlock_delalloc(inode,
696 &BTRFS_I(inode)->io_tree,
697 async_extent->start,
698 async_extent->start +
699 async_extent->ram_size - 1,
700 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
701 EXTENT_CLEAR_UNLOCK |
702 EXTENT_CLEAR_DELALLOC |
703 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
704
705 ret = btrfs_submit_compressed_write(inode,
706 async_extent->start,
707 async_extent->ram_size,
708 ins.objectid,
709 ins.offset, async_extent->pages,
710 async_extent->nr_pages);
711
712 BUG_ON(ret);
713 alloc_hint = ins.objectid + ins.offset;
714 kfree(async_extent);
715 cond_resched();
716 }
717
718 return 0;
719 }
720
721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
722 u64 num_bytes)
723 {
724 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
725 struct extent_map *em;
726 u64 alloc_hint = 0;
727
728 read_lock(&em_tree->lock);
729 em = search_extent_mapping(em_tree, start, num_bytes);
730 if (em) {
731 /*
732 * if block start isn't an actual block number then find the
733 * first block in this inode and use that as a hint. If that
734 * block is also bogus then just don't worry about it.
735 */
736 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
737 free_extent_map(em);
738 em = search_extent_mapping(em_tree, 0, 0);
739 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
740 alloc_hint = em->block_start;
741 if (em)
742 free_extent_map(em);
743 } else {
744 alloc_hint = em->block_start;
745 free_extent_map(em);
746 }
747 }
748 read_unlock(&em_tree->lock);
749
750 return alloc_hint;
751 }
752
753 /*
754 * when extent_io.c finds a delayed allocation range in the file,
755 * the call backs end up in this code. The basic idea is to
756 * allocate extents on disk for the range, and create ordered data structs
757 * in ram to track those extents.
758 *
759 * locked_page is the page that writepage had locked already. We use
760 * it to make sure we don't do extra locks or unlocks.
761 *
762 * *page_started is set to one if we unlock locked_page and do everything
763 * required to start IO on it. It may be clean and already done with
764 * IO when we return.
765 */
766 static noinline int cow_file_range(struct inode *inode,
767 struct page *locked_page,
768 u64 start, u64 end, int *page_started,
769 unsigned long *nr_written,
770 int unlock)
771 {
772 struct btrfs_root *root = BTRFS_I(inode)->root;
773 struct btrfs_trans_handle *trans;
774 u64 alloc_hint = 0;
775 u64 num_bytes;
776 unsigned long ram_size;
777 u64 disk_num_bytes;
778 u64 cur_alloc_size;
779 u64 blocksize = root->sectorsize;
780 struct btrfs_key ins;
781 struct extent_map *em;
782 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
783 int ret = 0;
784
785 BUG_ON(btrfs_is_free_space_inode(root, inode));
786 trans = btrfs_join_transaction(root);
787 BUG_ON(IS_ERR(trans));
788 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
789
790 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
791 num_bytes = max(blocksize, num_bytes);
792 disk_num_bytes = num_bytes;
793 ret = 0;
794
795 /* if this is a small write inside eof, kick off defrag */
796 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
797 btrfs_add_inode_defrag(trans, inode);
798
799 if (start == 0) {
800 /* lets try to make an inline extent */
801 ret = cow_file_range_inline(trans, root, inode,
802 start, end, 0, 0, NULL);
803 if (ret == 0) {
804 extent_clear_unlock_delalloc(inode,
805 &BTRFS_I(inode)->io_tree,
806 start, end, NULL,
807 EXTENT_CLEAR_UNLOCK_PAGE |
808 EXTENT_CLEAR_UNLOCK |
809 EXTENT_CLEAR_DELALLOC |
810 EXTENT_CLEAR_DIRTY |
811 EXTENT_SET_WRITEBACK |
812 EXTENT_END_WRITEBACK);
813
814 *nr_written = *nr_written +
815 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
816 *page_started = 1;
817 ret = 0;
818 goto out;
819 }
820 }
821
822 BUG_ON(disk_num_bytes >
823 btrfs_super_total_bytes(&root->fs_info->super_copy));
824
825 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
826 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
827
828 while (disk_num_bytes > 0) {
829 unsigned long op;
830
831 cur_alloc_size = disk_num_bytes;
832 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
833 root->sectorsize, 0, alloc_hint,
834 (u64)-1, &ins, 1);
835 BUG_ON(ret);
836
837 em = alloc_extent_map();
838 BUG_ON(!em);
839 em->start = start;
840 em->orig_start = em->start;
841 ram_size = ins.offset;
842 em->len = ins.offset;
843
844 em->block_start = ins.objectid;
845 em->block_len = ins.offset;
846 em->bdev = root->fs_info->fs_devices->latest_bdev;
847 set_bit(EXTENT_FLAG_PINNED, &em->flags);
848
849 while (1) {
850 write_lock(&em_tree->lock);
851 ret = add_extent_mapping(em_tree, em);
852 write_unlock(&em_tree->lock);
853 if (ret != -EEXIST) {
854 free_extent_map(em);
855 break;
856 }
857 btrfs_drop_extent_cache(inode, start,
858 start + ram_size - 1, 0);
859 }
860
861 cur_alloc_size = ins.offset;
862 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
863 ram_size, cur_alloc_size, 0);
864 BUG_ON(ret);
865
866 if (root->root_key.objectid ==
867 BTRFS_DATA_RELOC_TREE_OBJECTID) {
868 ret = btrfs_reloc_clone_csums(inode, start,
869 cur_alloc_size);
870 BUG_ON(ret);
871 }
872
873 if (disk_num_bytes < cur_alloc_size)
874 break;
875
876 /* we're not doing compressed IO, don't unlock the first
877 * page (which the caller expects to stay locked), don't
878 * clear any dirty bits and don't set any writeback bits
879 *
880 * Do set the Private2 bit so we know this page was properly
881 * setup for writepage
882 */
883 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
884 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
885 EXTENT_SET_PRIVATE2;
886
887 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
888 start, start + ram_size - 1,
889 locked_page, op);
890 disk_num_bytes -= cur_alloc_size;
891 num_bytes -= cur_alloc_size;
892 alloc_hint = ins.objectid + ins.offset;
893 start += cur_alloc_size;
894 }
895 out:
896 ret = 0;
897 btrfs_end_transaction(trans, root);
898
899 return ret;
900 }
901
902 /*
903 * work queue call back to started compression on a file and pages
904 */
905 static noinline void async_cow_start(struct btrfs_work *work)
906 {
907 struct async_cow *async_cow;
908 int num_added = 0;
909 async_cow = container_of(work, struct async_cow, work);
910
911 compress_file_range(async_cow->inode, async_cow->locked_page,
912 async_cow->start, async_cow->end, async_cow,
913 &num_added);
914 if (num_added == 0)
915 async_cow->inode = NULL;
916 }
917
918 /*
919 * work queue call back to submit previously compressed pages
920 */
921 static noinline void async_cow_submit(struct btrfs_work *work)
922 {
923 struct async_cow *async_cow;
924 struct btrfs_root *root;
925 unsigned long nr_pages;
926
927 async_cow = container_of(work, struct async_cow, work);
928
929 root = async_cow->root;
930 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
931 PAGE_CACHE_SHIFT;
932
933 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
934
935 if (atomic_read(&root->fs_info->async_delalloc_pages) <
936 5 * 1042 * 1024 &&
937 waitqueue_active(&root->fs_info->async_submit_wait))
938 wake_up(&root->fs_info->async_submit_wait);
939
940 if (async_cow->inode)
941 submit_compressed_extents(async_cow->inode, async_cow);
942 }
943
944 static noinline void async_cow_free(struct btrfs_work *work)
945 {
946 struct async_cow *async_cow;
947 async_cow = container_of(work, struct async_cow, work);
948 kfree(async_cow);
949 }
950
951 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
952 u64 start, u64 end, int *page_started,
953 unsigned long *nr_written)
954 {
955 struct async_cow *async_cow;
956 struct btrfs_root *root = BTRFS_I(inode)->root;
957 unsigned long nr_pages;
958 u64 cur_end;
959 int limit = 10 * 1024 * 1042;
960
961 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
962 1, 0, NULL, GFP_NOFS);
963 while (start < end) {
964 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
965 BUG_ON(!async_cow);
966 async_cow->inode = inode;
967 async_cow->root = root;
968 async_cow->locked_page = locked_page;
969 async_cow->start = start;
970
971 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
972 cur_end = end;
973 else
974 cur_end = min(end, start + 512 * 1024 - 1);
975
976 async_cow->end = cur_end;
977 INIT_LIST_HEAD(&async_cow->extents);
978
979 async_cow->work.func = async_cow_start;
980 async_cow->work.ordered_func = async_cow_submit;
981 async_cow->work.ordered_free = async_cow_free;
982 async_cow->work.flags = 0;
983
984 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
985 PAGE_CACHE_SHIFT;
986 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
987
988 btrfs_queue_worker(&root->fs_info->delalloc_workers,
989 &async_cow->work);
990
991 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
992 wait_event(root->fs_info->async_submit_wait,
993 (atomic_read(&root->fs_info->async_delalloc_pages) <
994 limit));
995 }
996
997 while (atomic_read(&root->fs_info->async_submit_draining) &&
998 atomic_read(&root->fs_info->async_delalloc_pages)) {
999 wait_event(root->fs_info->async_submit_wait,
1000 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1001 0));
1002 }
1003
1004 *nr_written += nr_pages;
1005 start = cur_end + 1;
1006 }
1007 *page_started = 1;
1008 return 0;
1009 }
1010
1011 static noinline int csum_exist_in_range(struct btrfs_root *root,
1012 u64 bytenr, u64 num_bytes)
1013 {
1014 int ret;
1015 struct btrfs_ordered_sum *sums;
1016 LIST_HEAD(list);
1017
1018 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1019 bytenr + num_bytes - 1, &list, 0);
1020 if (ret == 0 && list_empty(&list))
1021 return 0;
1022
1023 while (!list_empty(&list)) {
1024 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025 list_del(&sums->list);
1026 kfree(sums);
1027 }
1028 return 1;
1029 }
1030
1031 /*
1032 * when nowcow writeback call back. This checks for snapshots or COW copies
1033 * of the extents that exist in the file, and COWs the file as required.
1034 *
1035 * If no cow copies or snapshots exist, we write directly to the existing
1036 * blocks on disk
1037 */
1038 static noinline int run_delalloc_nocow(struct inode *inode,
1039 struct page *locked_page,
1040 u64 start, u64 end, int *page_started, int force,
1041 unsigned long *nr_written)
1042 {
1043 struct btrfs_root *root = BTRFS_I(inode)->root;
1044 struct btrfs_trans_handle *trans;
1045 struct extent_buffer *leaf;
1046 struct btrfs_path *path;
1047 struct btrfs_file_extent_item *fi;
1048 struct btrfs_key found_key;
1049 u64 cow_start;
1050 u64 cur_offset;
1051 u64 extent_end;
1052 u64 extent_offset;
1053 u64 disk_bytenr;
1054 u64 num_bytes;
1055 int extent_type;
1056 int ret;
1057 int type;
1058 int nocow;
1059 int check_prev = 1;
1060 bool nolock;
1061 u64 ino = btrfs_ino(inode);
1062
1063 path = btrfs_alloc_path();
1064 if (!path)
1065 return -ENOMEM;
1066
1067 nolock = btrfs_is_free_space_inode(root, inode);
1068
1069 if (nolock)
1070 trans = btrfs_join_transaction_nolock(root);
1071 else
1072 trans = btrfs_join_transaction(root);
1073
1074 BUG_ON(IS_ERR(trans));
1075 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1076
1077 cow_start = (u64)-1;
1078 cur_offset = start;
1079 while (1) {
1080 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1081 cur_offset, 0);
1082 BUG_ON(ret < 0);
1083 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084 leaf = path->nodes[0];
1085 btrfs_item_key_to_cpu(leaf, &found_key,
1086 path->slots[0] - 1);
1087 if (found_key.objectid == ino &&
1088 found_key.type == BTRFS_EXTENT_DATA_KEY)
1089 path->slots[0]--;
1090 }
1091 check_prev = 0;
1092 next_slot:
1093 leaf = path->nodes[0];
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1096 if (ret < 0)
1097 BUG_ON(1);
1098 if (ret > 0)
1099 break;
1100 leaf = path->nodes[0];
1101 }
1102
1103 nocow = 0;
1104 disk_bytenr = 0;
1105 num_bytes = 0;
1106 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
1108 if (found_key.objectid > ino ||
1109 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110 found_key.offset > end)
1111 break;
1112
1113 if (found_key.offset > cur_offset) {
1114 extent_end = found_key.offset;
1115 extent_type = 0;
1116 goto out_check;
1117 }
1118
1119 fi = btrfs_item_ptr(leaf, path->slots[0],
1120 struct btrfs_file_extent_item);
1121 extent_type = btrfs_file_extent_type(leaf, fi);
1122
1123 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126 extent_offset = btrfs_file_extent_offset(leaf, fi);
1127 extent_end = found_key.offset +
1128 btrfs_file_extent_num_bytes(leaf, fi);
1129 if (extent_end <= start) {
1130 path->slots[0]++;
1131 goto next_slot;
1132 }
1133 if (disk_bytenr == 0)
1134 goto out_check;
1135 if (btrfs_file_extent_compression(leaf, fi) ||
1136 btrfs_file_extent_encryption(leaf, fi) ||
1137 btrfs_file_extent_other_encoding(leaf, fi))
1138 goto out_check;
1139 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140 goto out_check;
1141 if (btrfs_extent_readonly(root, disk_bytenr))
1142 goto out_check;
1143 if (btrfs_cross_ref_exist(trans, root, ino,
1144 found_key.offset -
1145 extent_offset, disk_bytenr))
1146 goto out_check;
1147 disk_bytenr += extent_offset;
1148 disk_bytenr += cur_offset - found_key.offset;
1149 num_bytes = min(end + 1, extent_end) - cur_offset;
1150 /*
1151 * force cow if csum exists in the range.
1152 * this ensure that csum for a given extent are
1153 * either valid or do not exist.
1154 */
1155 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156 goto out_check;
1157 nocow = 1;
1158 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159 extent_end = found_key.offset +
1160 btrfs_file_extent_inline_len(leaf, fi);
1161 extent_end = ALIGN(extent_end, root->sectorsize);
1162 } else {
1163 BUG_ON(1);
1164 }
1165 out_check:
1166 if (extent_end <= start) {
1167 path->slots[0]++;
1168 goto next_slot;
1169 }
1170 if (!nocow) {
1171 if (cow_start == (u64)-1)
1172 cow_start = cur_offset;
1173 cur_offset = extent_end;
1174 if (cur_offset > end)
1175 break;
1176 path->slots[0]++;
1177 goto next_slot;
1178 }
1179
1180 btrfs_release_path(path);
1181 if (cow_start != (u64)-1) {
1182 ret = cow_file_range(inode, locked_page, cow_start,
1183 found_key.offset - 1, page_started,
1184 nr_written, 1);
1185 BUG_ON(ret);
1186 cow_start = (u64)-1;
1187 }
1188
1189 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190 struct extent_map *em;
1191 struct extent_map_tree *em_tree;
1192 em_tree = &BTRFS_I(inode)->extent_tree;
1193 em = alloc_extent_map();
1194 BUG_ON(!em);
1195 em->start = cur_offset;
1196 em->orig_start = em->start;
1197 em->len = num_bytes;
1198 em->block_len = num_bytes;
1199 em->block_start = disk_bytenr;
1200 em->bdev = root->fs_info->fs_devices->latest_bdev;
1201 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202 while (1) {
1203 write_lock(&em_tree->lock);
1204 ret = add_extent_mapping(em_tree, em);
1205 write_unlock(&em_tree->lock);
1206 if (ret != -EEXIST) {
1207 free_extent_map(em);
1208 break;
1209 }
1210 btrfs_drop_extent_cache(inode, em->start,
1211 em->start + em->len - 1, 0);
1212 }
1213 type = BTRFS_ORDERED_PREALLOC;
1214 } else {
1215 type = BTRFS_ORDERED_NOCOW;
1216 }
1217
1218 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219 num_bytes, num_bytes, type);
1220 BUG_ON(ret);
1221
1222 if (root->root_key.objectid ==
1223 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225 num_bytes);
1226 BUG_ON(ret);
1227 }
1228
1229 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230 cur_offset, cur_offset + num_bytes - 1,
1231 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233 EXTENT_SET_PRIVATE2);
1234 cur_offset = extent_end;
1235 if (cur_offset > end)
1236 break;
1237 }
1238 btrfs_release_path(path);
1239
1240 if (cur_offset <= end && cow_start == (u64)-1)
1241 cow_start = cur_offset;
1242 if (cow_start != (u64)-1) {
1243 ret = cow_file_range(inode, locked_page, cow_start, end,
1244 page_started, nr_written, 1);
1245 BUG_ON(ret);
1246 }
1247
1248 if (nolock) {
1249 ret = btrfs_end_transaction_nolock(trans, root);
1250 BUG_ON(ret);
1251 } else {
1252 ret = btrfs_end_transaction(trans, root);
1253 BUG_ON(ret);
1254 }
1255 btrfs_free_path(path);
1256 return 0;
1257 }
1258
1259 /*
1260 * extent_io.c call back to do delayed allocation processing
1261 */
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263 u64 start, u64 end, int *page_started,
1264 unsigned long *nr_written)
1265 {
1266 int ret;
1267 struct btrfs_root *root = BTRFS_I(inode)->root;
1268
1269 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271 page_started, 1, nr_written);
1272 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274 page_started, 0, nr_written);
1275 else if (!btrfs_test_opt(root, COMPRESS) &&
1276 !(BTRFS_I(inode)->force_compress) &&
1277 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278 ret = cow_file_range(inode, locked_page, start, end,
1279 page_started, nr_written, 1);
1280 else
1281 ret = cow_file_range_async(inode, locked_page, start, end,
1282 page_started, nr_written);
1283 return ret;
1284 }
1285
1286 static void btrfs_split_extent_hook(struct inode *inode,
1287 struct extent_state *orig, u64 split)
1288 {
1289 /* not delalloc, ignore it */
1290 if (!(orig->state & EXTENT_DELALLOC))
1291 return;
1292
1293 spin_lock(&BTRFS_I(inode)->lock);
1294 BTRFS_I(inode)->outstanding_extents++;
1295 spin_unlock(&BTRFS_I(inode)->lock);
1296 }
1297
1298 /*
1299 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300 * extents so we can keep track of new extents that are just merged onto old
1301 * extents, such as when we are doing sequential writes, so we can properly
1302 * account for the metadata space we'll need.
1303 */
1304 static void btrfs_merge_extent_hook(struct inode *inode,
1305 struct extent_state *new,
1306 struct extent_state *other)
1307 {
1308 /* not delalloc, ignore it */
1309 if (!(other->state & EXTENT_DELALLOC))
1310 return;
1311
1312 spin_lock(&BTRFS_I(inode)->lock);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->lock);
1315 }
1316
1317 /*
1318 * extent_io.c set_bit_hook, used to track delayed allocation
1319 * bytes in this file, and to maintain the list of inodes that
1320 * have pending delalloc work to be done.
1321 */
1322 static void btrfs_set_bit_hook(struct inode *inode,
1323 struct extent_state *state, int *bits)
1324 {
1325
1326 /*
1327 * set_bit and clear bit hooks normally require _irqsave/restore
1328 * but in this case, we are only testing for the DELALLOC
1329 * bit, which is only set or cleared with irqs on
1330 */
1331 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1332 struct btrfs_root *root = BTRFS_I(inode)->root;
1333 u64 len = state->end + 1 - state->start;
1334 bool do_list = !btrfs_is_free_space_inode(root, inode);
1335
1336 if (*bits & EXTENT_FIRST_DELALLOC) {
1337 *bits &= ~EXTENT_FIRST_DELALLOC;
1338 } else {
1339 spin_lock(&BTRFS_I(inode)->lock);
1340 BTRFS_I(inode)->outstanding_extents++;
1341 spin_unlock(&BTRFS_I(inode)->lock);
1342 }
1343
1344 spin_lock(&root->fs_info->delalloc_lock);
1345 BTRFS_I(inode)->delalloc_bytes += len;
1346 root->fs_info->delalloc_bytes += len;
1347 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1348 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349 &root->fs_info->delalloc_inodes);
1350 }
1351 spin_unlock(&root->fs_info->delalloc_lock);
1352 }
1353 }
1354
1355 /*
1356 * extent_io.c clear_bit_hook, see set_bit_hook for why
1357 */
1358 static void btrfs_clear_bit_hook(struct inode *inode,
1359 struct extent_state *state, int *bits)
1360 {
1361 /*
1362 * set_bit and clear bit hooks normally require _irqsave/restore
1363 * but in this case, we are only testing for the DELALLOC
1364 * bit, which is only set or cleared with irqs on
1365 */
1366 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1367 struct btrfs_root *root = BTRFS_I(inode)->root;
1368 u64 len = state->end + 1 - state->start;
1369 bool do_list = !btrfs_is_free_space_inode(root, inode);
1370
1371 if (*bits & EXTENT_FIRST_DELALLOC) {
1372 *bits &= ~EXTENT_FIRST_DELALLOC;
1373 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374 spin_lock(&BTRFS_I(inode)->lock);
1375 BTRFS_I(inode)->outstanding_extents--;
1376 spin_unlock(&BTRFS_I(inode)->lock);
1377 }
1378
1379 if (*bits & EXTENT_DO_ACCOUNTING)
1380 btrfs_delalloc_release_metadata(inode, len);
1381
1382 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1383 && do_list)
1384 btrfs_free_reserved_data_space(inode, len);
1385
1386 spin_lock(&root->fs_info->delalloc_lock);
1387 root->fs_info->delalloc_bytes -= len;
1388 BTRFS_I(inode)->delalloc_bytes -= len;
1389
1390 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1391 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1393 }
1394 spin_unlock(&root->fs_info->delalloc_lock);
1395 }
1396 }
1397
1398 /*
1399 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400 * we don't create bios that span stripes or chunks
1401 */
1402 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1403 size_t size, struct bio *bio,
1404 unsigned long bio_flags)
1405 {
1406 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407 struct btrfs_mapping_tree *map_tree;
1408 u64 logical = (u64)bio->bi_sector << 9;
1409 u64 length = 0;
1410 u64 map_length;
1411 int ret;
1412
1413 if (bio_flags & EXTENT_BIO_COMPRESSED)
1414 return 0;
1415
1416 length = bio->bi_size;
1417 map_tree = &root->fs_info->mapping_tree;
1418 map_length = length;
1419 ret = btrfs_map_block(map_tree, READ, logical,
1420 &map_length, NULL, 0);
1421
1422 if (map_length < length + size)
1423 return 1;
1424 return ret;
1425 }
1426
1427 /*
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1431 *
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1434 */
1435 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436 struct bio *bio, int mirror_num,
1437 unsigned long bio_flags,
1438 u64 bio_offset)
1439 {
1440 struct btrfs_root *root = BTRFS_I(inode)->root;
1441 int ret = 0;
1442
1443 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1444 BUG_ON(ret);
1445 return 0;
1446 }
1447
1448 /*
1449 * in order to insert checksums into the metadata in large chunks,
1450 * we wait until bio submission time. All the pages in the bio are
1451 * checksummed and sums are attached onto the ordered extent record.
1452 *
1453 * At IO completion time the cums attached on the ordered extent record
1454 * are inserted into the btree
1455 */
1456 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1457 int mirror_num, unsigned long bio_flags,
1458 u64 bio_offset)
1459 {
1460 struct btrfs_root *root = BTRFS_I(inode)->root;
1461 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1462 }
1463
1464 /*
1465 * extent_io.c submission hook. This does the right thing for csum calculation
1466 * on write, or reading the csums from the tree before a read
1467 */
1468 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1469 int mirror_num, unsigned long bio_flags,
1470 u64 bio_offset)
1471 {
1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1473 int ret = 0;
1474 int skip_sum;
1475
1476 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1477
1478 if (btrfs_is_free_space_inode(root, inode))
1479 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1480 else
1481 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1482 BUG_ON(ret);
1483
1484 if (!(rw & REQ_WRITE)) {
1485 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1486 return btrfs_submit_compressed_read(inode, bio,
1487 mirror_num, bio_flags);
1488 } else if (!skip_sum) {
1489 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1490 if (ret)
1491 return ret;
1492 }
1493 goto mapit;
1494 } else if (!skip_sum) {
1495 /* csum items have already been cloned */
1496 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1497 goto mapit;
1498 /* we're doing a write, do the async checksumming */
1499 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1500 inode, rw, bio, mirror_num,
1501 bio_flags, bio_offset,
1502 __btrfs_submit_bio_start,
1503 __btrfs_submit_bio_done);
1504 }
1505
1506 mapit:
1507 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1508 }
1509
1510 /*
1511 * given a list of ordered sums record them in the inode. This happens
1512 * at IO completion time based on sums calculated at bio submission time.
1513 */
1514 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1515 struct inode *inode, u64 file_offset,
1516 struct list_head *list)
1517 {
1518 struct btrfs_ordered_sum *sum;
1519
1520 list_for_each_entry(sum, list, list) {
1521 btrfs_csum_file_blocks(trans,
1522 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1523 }
1524 return 0;
1525 }
1526
1527 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528 struct extent_state **cached_state)
1529 {
1530 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1531 WARN_ON(1);
1532 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1533 cached_state, GFP_NOFS);
1534 }
1535
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup {
1538 struct page *page;
1539 struct btrfs_work work;
1540 };
1541
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1543 {
1544 struct btrfs_writepage_fixup *fixup;
1545 struct btrfs_ordered_extent *ordered;
1546 struct extent_state *cached_state = NULL;
1547 struct page *page;
1548 struct inode *inode;
1549 u64 page_start;
1550 u64 page_end;
1551
1552 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1553 page = fixup->page;
1554 again:
1555 lock_page(page);
1556 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557 ClearPageChecked(page);
1558 goto out_page;
1559 }
1560
1561 inode = page->mapping->host;
1562 page_start = page_offset(page);
1563 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1564
1565 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566 &cached_state, GFP_NOFS);
1567
1568 /* already ordered? We're done */
1569 if (PagePrivate2(page))
1570 goto out;
1571
1572 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1573 if (ordered) {
1574 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575 page_end, &cached_state, GFP_NOFS);
1576 unlock_page(page);
1577 btrfs_start_ordered_extent(inode, ordered, 1);
1578 goto again;
1579 }
1580
1581 BUG();
1582 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1583 ClearPageChecked(page);
1584 out:
1585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586 &cached_state, GFP_NOFS);
1587 out_page:
1588 unlock_page(page);
1589 page_cache_release(page);
1590 kfree(fixup);
1591 }
1592
1593 /*
1594 * There are a few paths in the higher layers of the kernel that directly
1595 * set the page dirty bit without asking the filesystem if it is a
1596 * good idea. This causes problems because we want to make sure COW
1597 * properly happens and the data=ordered rules are followed.
1598 *
1599 * In our case any range that doesn't have the ORDERED bit set
1600 * hasn't been properly setup for IO. We kick off an async process
1601 * to fix it up. The async helper will wait for ordered extents, set
1602 * the delalloc bit and make it safe to write the page.
1603 */
1604 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1605 {
1606 struct inode *inode = page->mapping->host;
1607 struct btrfs_writepage_fixup *fixup;
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
1609
1610 /* this page is properly in the ordered list */
1611 if (TestClearPagePrivate2(page))
1612 return 0;
1613
1614 if (PageChecked(page))
1615 return -EAGAIN;
1616
1617 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1618 if (!fixup)
1619 return -EAGAIN;
1620
1621 SetPageChecked(page);
1622 page_cache_get(page);
1623 fixup->work.func = btrfs_writepage_fixup_worker;
1624 fixup->page = page;
1625 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1626 return -EAGAIN;
1627 }
1628
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630 struct inode *inode, u64 file_pos,
1631 u64 disk_bytenr, u64 disk_num_bytes,
1632 u64 num_bytes, u64 ram_bytes,
1633 u8 compression, u8 encryption,
1634 u16 other_encoding, int extent_type)
1635 {
1636 struct btrfs_root *root = BTRFS_I(inode)->root;
1637 struct btrfs_file_extent_item *fi;
1638 struct btrfs_path *path;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key ins;
1641 u64 hint;
1642 int ret;
1643
1644 path = btrfs_alloc_path();
1645 if (!path)
1646 return -ENOMEM;
1647
1648 path->leave_spinning = 1;
1649
1650 /*
1651 * we may be replacing one extent in the tree with another.
1652 * The new extent is pinned in the extent map, and we don't want
1653 * to drop it from the cache until it is completely in the btree.
1654 *
1655 * So, tell btrfs_drop_extents to leave this extent in the cache.
1656 * the caller is expected to unpin it and allow it to be merged
1657 * with the others.
1658 */
1659 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1660 &hint, 0);
1661 BUG_ON(ret);
1662
1663 ins.objectid = btrfs_ino(inode);
1664 ins.offset = file_pos;
1665 ins.type = BTRFS_EXTENT_DATA_KEY;
1666 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1667 BUG_ON(ret);
1668 leaf = path->nodes[0];
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672 btrfs_set_file_extent_type(leaf, fi, extent_type);
1673 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675 btrfs_set_file_extent_offset(leaf, fi, 0);
1676 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678 btrfs_set_file_extent_compression(leaf, fi, compression);
1679 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1681
1682 btrfs_unlock_up_safe(path, 1);
1683 btrfs_set_lock_blocking(leaf);
1684
1685 btrfs_mark_buffer_dirty(leaf);
1686
1687 inode_add_bytes(inode, num_bytes);
1688
1689 ins.objectid = disk_bytenr;
1690 ins.offset = disk_num_bytes;
1691 ins.type = BTRFS_EXTENT_ITEM_KEY;
1692 ret = btrfs_alloc_reserved_file_extent(trans, root,
1693 root->root_key.objectid,
1694 btrfs_ino(inode), file_pos, &ins);
1695 BUG_ON(ret);
1696 btrfs_free_path(path);
1697
1698 return 0;
1699 }
1700
1701 /*
1702 * helper function for btrfs_finish_ordered_io, this
1703 * just reads in some of the csum leaves to prime them into ram
1704 * before we start the transaction. It limits the amount of btree
1705 * reads required while inside the transaction.
1706 */
1707 /* as ordered data IO finishes, this gets called so we can finish
1708 * an ordered extent if the range of bytes in the file it covers are
1709 * fully written.
1710 */
1711 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1712 {
1713 struct btrfs_root *root = BTRFS_I(inode)->root;
1714 struct btrfs_trans_handle *trans = NULL;
1715 struct btrfs_ordered_extent *ordered_extent = NULL;
1716 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1717 struct extent_state *cached_state = NULL;
1718 int compress_type = 0;
1719 int ret;
1720 bool nolock;
1721
1722 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1723 end - start + 1);
1724 if (!ret)
1725 return 0;
1726 BUG_ON(!ordered_extent);
1727
1728 nolock = btrfs_is_free_space_inode(root, inode);
1729
1730 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731 BUG_ON(!list_empty(&ordered_extent->list));
1732 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1733 if (!ret) {
1734 if (nolock)
1735 trans = btrfs_join_transaction_nolock(root);
1736 else
1737 trans = btrfs_join_transaction(root);
1738 BUG_ON(IS_ERR(trans));
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740 ret = btrfs_update_inode(trans, root, inode);
1741 BUG_ON(ret);
1742 }
1743 goto out;
1744 }
1745
1746 lock_extent_bits(io_tree, ordered_extent->file_offset,
1747 ordered_extent->file_offset + ordered_extent->len - 1,
1748 0, &cached_state, GFP_NOFS);
1749
1750 if (nolock)
1751 trans = btrfs_join_transaction_nolock(root);
1752 else
1753 trans = btrfs_join_transaction(root);
1754 BUG_ON(IS_ERR(trans));
1755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756
1757 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1758 compress_type = ordered_extent->compress_type;
1759 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1760 BUG_ON(compress_type);
1761 ret = btrfs_mark_extent_written(trans, inode,
1762 ordered_extent->file_offset,
1763 ordered_extent->file_offset +
1764 ordered_extent->len);
1765 BUG_ON(ret);
1766 } else {
1767 BUG_ON(root == root->fs_info->tree_root);
1768 ret = insert_reserved_file_extent(trans, inode,
1769 ordered_extent->file_offset,
1770 ordered_extent->start,
1771 ordered_extent->disk_len,
1772 ordered_extent->len,
1773 ordered_extent->len,
1774 compress_type, 0, 0,
1775 BTRFS_FILE_EXTENT_REG);
1776 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777 ordered_extent->file_offset,
1778 ordered_extent->len);
1779 BUG_ON(ret);
1780 }
1781 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1784
1785 add_pending_csums(trans, inode, ordered_extent->file_offset,
1786 &ordered_extent->list);
1787
1788 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1789 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790 ret = btrfs_update_inode(trans, root, inode);
1791 BUG_ON(ret);
1792 }
1793 ret = 0;
1794 out:
1795 if (nolock) {
1796 if (trans)
1797 btrfs_end_transaction_nolock(trans, root);
1798 } else {
1799 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1800 if (trans)
1801 btrfs_end_transaction(trans, root);
1802 }
1803
1804 /* once for us */
1805 btrfs_put_ordered_extent(ordered_extent);
1806 /* once for the tree */
1807 btrfs_put_ordered_extent(ordered_extent);
1808
1809 return 0;
1810 }
1811
1812 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1813 struct extent_state *state, int uptodate)
1814 {
1815 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1816
1817 ClearPagePrivate2(page);
1818 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1819 }
1820
1821 /*
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1828 */
1829 struct io_failure_record {
1830 struct page *page;
1831 u64 start;
1832 u64 len;
1833 u64 logical;
1834 unsigned long bio_flags;
1835 int last_mirror;
1836 };
1837
1838 static int btrfs_io_failed_hook(struct bio *failed_bio,
1839 struct page *page, u64 start, u64 end,
1840 struct extent_state *state)
1841 {
1842 struct io_failure_record *failrec = NULL;
1843 u64 private;
1844 struct extent_map *em;
1845 struct inode *inode = page->mapping->host;
1846 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1848 struct bio *bio;
1849 int num_copies;
1850 int ret;
1851 int rw;
1852 u64 logical;
1853
1854 ret = get_state_private(failure_tree, start, &private);
1855 if (ret) {
1856 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1857 if (!failrec)
1858 return -ENOMEM;
1859 failrec->start = start;
1860 failrec->len = end - start + 1;
1861 failrec->last_mirror = 0;
1862 failrec->bio_flags = 0;
1863
1864 read_lock(&em_tree->lock);
1865 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866 if (em->start > start || em->start + em->len < start) {
1867 free_extent_map(em);
1868 em = NULL;
1869 }
1870 read_unlock(&em_tree->lock);
1871
1872 if (IS_ERR_OR_NULL(em)) {
1873 kfree(failrec);
1874 return -EIO;
1875 }
1876 logical = start - em->start;
1877 logical = em->block_start + logical;
1878 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879 logical = em->block_start;
1880 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881 extent_set_compress_type(&failrec->bio_flags,
1882 em->compress_type);
1883 }
1884 failrec->logical = logical;
1885 free_extent_map(em);
1886 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887 EXTENT_DIRTY, GFP_NOFS);
1888 set_state_private(failure_tree, start,
1889 (u64)(unsigned long)failrec);
1890 } else {
1891 failrec = (struct io_failure_record *)(unsigned long)private;
1892 }
1893 num_copies = btrfs_num_copies(
1894 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895 failrec->logical, failrec->len);
1896 failrec->last_mirror++;
1897 if (!state) {
1898 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900 failrec->start,
1901 EXTENT_LOCKED);
1902 if (state && state->start != failrec->start)
1903 state = NULL;
1904 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1905 }
1906 if (!state || failrec->last_mirror > num_copies) {
1907 set_state_private(failure_tree, failrec->start, 0);
1908 clear_extent_bits(failure_tree, failrec->start,
1909 failrec->start + failrec->len - 1,
1910 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911 kfree(failrec);
1912 return -EIO;
1913 }
1914 bio = bio_alloc(GFP_NOFS, 1);
1915 bio->bi_private = state;
1916 bio->bi_end_io = failed_bio->bi_end_io;
1917 bio->bi_sector = failrec->logical >> 9;
1918 bio->bi_bdev = failed_bio->bi_bdev;
1919 bio->bi_size = 0;
1920
1921 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922 if (failed_bio->bi_rw & REQ_WRITE)
1923 rw = WRITE;
1924 else
1925 rw = READ;
1926
1927 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928 failrec->last_mirror,
1929 failrec->bio_flags, 0);
1930 return ret;
1931 }
1932
1933 /*
1934 * each time an IO finishes, we do a fast check in the IO failure tree
1935 * to see if we need to process or clean up an io_failure_record
1936 */
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1938 {
1939 u64 private;
1940 u64 private_failure;
1941 struct io_failure_record *failure;
1942 int ret;
1943
1944 private = 0;
1945 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948 start, &private_failure);
1949 if (ret == 0) {
1950 failure = (struct io_failure_record *)(unsigned long)
1951 private_failure;
1952 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953 failure->start, 0);
1954 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955 failure->start,
1956 failure->start + failure->len - 1,
1957 EXTENT_DIRTY | EXTENT_LOCKED,
1958 GFP_NOFS);
1959 kfree(failure);
1960 }
1961 }
1962 return 0;
1963 }
1964
1965 /*
1966 * when reads are done, we need to check csums to verify the data is correct
1967 * if there's a match, we allow the bio to finish. If not, we go through
1968 * the io_failure_record routines to find good copies
1969 */
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971 struct extent_state *state)
1972 {
1973 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974 struct inode *inode = page->mapping->host;
1975 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1976 char *kaddr;
1977 u64 private = ~(u32)0;
1978 int ret;
1979 struct btrfs_root *root = BTRFS_I(inode)->root;
1980 u32 csum = ~(u32)0;
1981
1982 if (PageChecked(page)) {
1983 ClearPageChecked(page);
1984 goto good;
1985 }
1986
1987 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1988 goto good;
1989
1990 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993 GFP_NOFS);
1994 return 0;
1995 }
1996
1997 if (state && state->start == start) {
1998 private = state->private;
1999 ret = 0;
2000 } else {
2001 ret = get_state_private(io_tree, start, &private);
2002 }
2003 kaddr = kmap_atomic(page, KM_USER0);
2004 if (ret)
2005 goto zeroit;
2006
2007 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2008 btrfs_csum_final(csum, (char *)&csum);
2009 if (csum != private)
2010 goto zeroit;
2011
2012 kunmap_atomic(kaddr, KM_USER0);
2013 good:
2014 /* if the io failure tree for this inode is non-empty,
2015 * check to see if we've recovered from a failed IO
2016 */
2017 btrfs_clean_io_failures(inode, start);
2018 return 0;
2019
2020 zeroit:
2021 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2022 "private %llu\n",
2023 (unsigned long long)btrfs_ino(page->mapping->host),
2024 (unsigned long long)start, csum,
2025 (unsigned long long)private);
2026 memset(kaddr + offset, 1, end - start + 1);
2027 flush_dcache_page(page);
2028 kunmap_atomic(kaddr, KM_USER0);
2029 if (private == 0)
2030 return 0;
2031 return -EIO;
2032 }
2033
2034 struct delayed_iput {
2035 struct list_head list;
2036 struct inode *inode;
2037 };
2038
2039 void btrfs_add_delayed_iput(struct inode *inode)
2040 {
2041 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042 struct delayed_iput *delayed;
2043
2044 if (atomic_add_unless(&inode->i_count, -1, 1))
2045 return;
2046
2047 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048 delayed->inode = inode;
2049
2050 spin_lock(&fs_info->delayed_iput_lock);
2051 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052 spin_unlock(&fs_info->delayed_iput_lock);
2053 }
2054
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056 {
2057 LIST_HEAD(list);
2058 struct btrfs_fs_info *fs_info = root->fs_info;
2059 struct delayed_iput *delayed;
2060 int empty;
2061
2062 spin_lock(&fs_info->delayed_iput_lock);
2063 empty = list_empty(&fs_info->delayed_iputs);
2064 spin_unlock(&fs_info->delayed_iput_lock);
2065 if (empty)
2066 return;
2067
2068 down_read(&root->fs_info->cleanup_work_sem);
2069 spin_lock(&fs_info->delayed_iput_lock);
2070 list_splice_init(&fs_info->delayed_iputs, &list);
2071 spin_unlock(&fs_info->delayed_iput_lock);
2072
2073 while (!list_empty(&list)) {
2074 delayed = list_entry(list.next, struct delayed_iput, list);
2075 list_del(&delayed->list);
2076 iput(delayed->inode);
2077 kfree(delayed);
2078 }
2079 up_read(&root->fs_info->cleanup_work_sem);
2080 }
2081
2082 enum btrfs_orphan_cleanup_state {
2083 ORPHAN_CLEANUP_STARTED = 1,
2084 ORPHAN_CLEANUP_DONE = 2,
2085 };
2086
2087 /*
2088 * This is called in transaction commmit time. If there are no orphan
2089 * files in the subvolume, it removes orphan item and frees block_rsv
2090 * structure.
2091 */
2092 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2093 struct btrfs_root *root)
2094 {
2095 int ret;
2096
2097 if (!list_empty(&root->orphan_list) ||
2098 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2099 return;
2100
2101 if (root->orphan_item_inserted &&
2102 btrfs_root_refs(&root->root_item) > 0) {
2103 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2104 root->root_key.objectid);
2105 BUG_ON(ret);
2106 root->orphan_item_inserted = 0;
2107 }
2108
2109 if (root->orphan_block_rsv) {
2110 WARN_ON(root->orphan_block_rsv->size > 0);
2111 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2112 root->orphan_block_rsv = NULL;
2113 }
2114 }
2115
2116 /*
2117 * This creates an orphan entry for the given inode in case something goes
2118 * wrong in the middle of an unlink/truncate.
2119 *
2120 * NOTE: caller of this function should reserve 5 units of metadata for
2121 * this function.
2122 */
2123 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2124 {
2125 struct btrfs_root *root = BTRFS_I(inode)->root;
2126 struct btrfs_block_rsv *block_rsv = NULL;
2127 int reserve = 0;
2128 int insert = 0;
2129 int ret;
2130
2131 if (!root->orphan_block_rsv) {
2132 block_rsv = btrfs_alloc_block_rsv(root);
2133 if (!block_rsv)
2134 return -ENOMEM;
2135 }
2136
2137 spin_lock(&root->orphan_lock);
2138 if (!root->orphan_block_rsv) {
2139 root->orphan_block_rsv = block_rsv;
2140 } else if (block_rsv) {
2141 btrfs_free_block_rsv(root, block_rsv);
2142 block_rsv = NULL;
2143 }
2144
2145 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2146 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2147 #if 0
2148 /*
2149 * For proper ENOSPC handling, we should do orphan
2150 * cleanup when mounting. But this introduces backward
2151 * compatibility issue.
2152 */
2153 if (!xchg(&root->orphan_item_inserted, 1))
2154 insert = 2;
2155 else
2156 insert = 1;
2157 #endif
2158 insert = 1;
2159 }
2160
2161 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2162 BTRFS_I(inode)->orphan_meta_reserved = 1;
2163 reserve = 1;
2164 }
2165 spin_unlock(&root->orphan_lock);
2166
2167 if (block_rsv)
2168 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2169
2170 /* grab metadata reservation from transaction handle */
2171 if (reserve) {
2172 ret = btrfs_orphan_reserve_metadata(trans, inode);
2173 BUG_ON(ret);
2174 }
2175
2176 /* insert an orphan item to track this unlinked/truncated file */
2177 if (insert >= 1) {
2178 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2179 BUG_ON(ret);
2180 }
2181
2182 /* insert an orphan item to track subvolume contains orphan files */
2183 if (insert >= 2) {
2184 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2185 root->root_key.objectid);
2186 BUG_ON(ret);
2187 }
2188 return 0;
2189 }
2190
2191 /*
2192 * We have done the truncate/delete so we can go ahead and remove the orphan
2193 * item for this particular inode.
2194 */
2195 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2196 {
2197 struct btrfs_root *root = BTRFS_I(inode)->root;
2198 int delete_item = 0;
2199 int release_rsv = 0;
2200 int ret = 0;
2201
2202 spin_lock(&root->orphan_lock);
2203 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2204 list_del_init(&BTRFS_I(inode)->i_orphan);
2205 delete_item = 1;
2206 }
2207
2208 if (BTRFS_I(inode)->orphan_meta_reserved) {
2209 BTRFS_I(inode)->orphan_meta_reserved = 0;
2210 release_rsv = 1;
2211 }
2212 spin_unlock(&root->orphan_lock);
2213
2214 if (trans && delete_item) {
2215 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2216 BUG_ON(ret);
2217 }
2218
2219 if (release_rsv)
2220 btrfs_orphan_release_metadata(inode);
2221
2222 return 0;
2223 }
2224
2225 /*
2226 * this cleans up any orphans that may be left on the list from the last use
2227 * of this root.
2228 */
2229 int btrfs_orphan_cleanup(struct btrfs_root *root)
2230 {
2231 struct btrfs_path *path;
2232 struct extent_buffer *leaf;
2233 struct btrfs_key key, found_key;
2234 struct btrfs_trans_handle *trans;
2235 struct inode *inode;
2236 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2237
2238 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2239 return 0;
2240
2241 path = btrfs_alloc_path();
2242 if (!path) {
2243 ret = -ENOMEM;
2244 goto out;
2245 }
2246 path->reada = -1;
2247
2248 key.objectid = BTRFS_ORPHAN_OBJECTID;
2249 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2250 key.offset = (u64)-1;
2251
2252 while (1) {
2253 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2254 if (ret < 0)
2255 goto out;
2256
2257 /*
2258 * if ret == 0 means we found what we were searching for, which
2259 * is weird, but possible, so only screw with path if we didn't
2260 * find the key and see if we have stuff that matches
2261 */
2262 if (ret > 0) {
2263 ret = 0;
2264 if (path->slots[0] == 0)
2265 break;
2266 path->slots[0]--;
2267 }
2268
2269 /* pull out the item */
2270 leaf = path->nodes[0];
2271 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2272
2273 /* make sure the item matches what we want */
2274 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2275 break;
2276 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2277 break;
2278
2279 /* release the path since we're done with it */
2280 btrfs_release_path(path);
2281
2282 /*
2283 * this is where we are basically btrfs_lookup, without the
2284 * crossing root thing. we store the inode number in the
2285 * offset of the orphan item.
2286 */
2287 found_key.objectid = found_key.offset;
2288 found_key.type = BTRFS_INODE_ITEM_KEY;
2289 found_key.offset = 0;
2290 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2291 if (IS_ERR(inode)) {
2292 ret = PTR_ERR(inode);
2293 goto out;
2294 }
2295
2296 /*
2297 * add this inode to the orphan list so btrfs_orphan_del does
2298 * the proper thing when we hit it
2299 */
2300 spin_lock(&root->orphan_lock);
2301 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2302 spin_unlock(&root->orphan_lock);
2303
2304 /*
2305 * if this is a bad inode, means we actually succeeded in
2306 * removing the inode, but not the orphan record, which means
2307 * we need to manually delete the orphan since iput will just
2308 * do a destroy_inode
2309 */
2310 if (is_bad_inode(inode)) {
2311 trans = btrfs_start_transaction(root, 0);
2312 if (IS_ERR(trans)) {
2313 ret = PTR_ERR(trans);
2314 goto out;
2315 }
2316 btrfs_orphan_del(trans, inode);
2317 btrfs_end_transaction(trans, root);
2318 iput(inode);
2319 continue;
2320 }
2321
2322 /* if we have links, this was a truncate, lets do that */
2323 if (inode->i_nlink) {
2324 if (!S_ISREG(inode->i_mode)) {
2325 WARN_ON(1);
2326 iput(inode);
2327 continue;
2328 }
2329 nr_truncate++;
2330 ret = btrfs_truncate(inode);
2331 } else {
2332 nr_unlink++;
2333 }
2334
2335 /* this will do delete_inode and everything for us */
2336 iput(inode);
2337 if (ret)
2338 goto out;
2339 }
2340 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2341
2342 if (root->orphan_block_rsv)
2343 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2344 (u64)-1);
2345
2346 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2347 trans = btrfs_join_transaction(root);
2348 if (!IS_ERR(trans))
2349 btrfs_end_transaction(trans, root);
2350 }
2351
2352 if (nr_unlink)
2353 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2354 if (nr_truncate)
2355 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2356
2357 out:
2358 if (ret)
2359 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2360 btrfs_free_path(path);
2361 return ret;
2362 }
2363
2364 /*
2365 * very simple check to peek ahead in the leaf looking for xattrs. If we
2366 * don't find any xattrs, we know there can't be any acls.
2367 *
2368 * slot is the slot the inode is in, objectid is the objectid of the inode
2369 */
2370 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2371 int slot, u64 objectid)
2372 {
2373 u32 nritems = btrfs_header_nritems(leaf);
2374 struct btrfs_key found_key;
2375 int scanned = 0;
2376
2377 slot++;
2378 while (slot < nritems) {
2379 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2380
2381 /* we found a different objectid, there must not be acls */
2382 if (found_key.objectid != objectid)
2383 return 0;
2384
2385 /* we found an xattr, assume we've got an acl */
2386 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2387 return 1;
2388
2389 /*
2390 * we found a key greater than an xattr key, there can't
2391 * be any acls later on
2392 */
2393 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2394 return 0;
2395
2396 slot++;
2397 scanned++;
2398
2399 /*
2400 * it goes inode, inode backrefs, xattrs, extents,
2401 * so if there are a ton of hard links to an inode there can
2402 * be a lot of backrefs. Don't waste time searching too hard,
2403 * this is just an optimization
2404 */
2405 if (scanned >= 8)
2406 break;
2407 }
2408 /* we hit the end of the leaf before we found an xattr or
2409 * something larger than an xattr. We have to assume the inode
2410 * has acls
2411 */
2412 return 1;
2413 }
2414
2415 /*
2416 * read an inode from the btree into the in-memory inode
2417 */
2418 static void btrfs_read_locked_inode(struct inode *inode)
2419 {
2420 struct btrfs_path *path;
2421 struct extent_buffer *leaf;
2422 struct btrfs_inode_item *inode_item;
2423 struct btrfs_timespec *tspec;
2424 struct btrfs_root *root = BTRFS_I(inode)->root;
2425 struct btrfs_key location;
2426 int maybe_acls;
2427 u32 rdev;
2428 int ret;
2429 bool filled = false;
2430
2431 ret = btrfs_fill_inode(inode, &rdev);
2432 if (!ret)
2433 filled = true;
2434
2435 path = btrfs_alloc_path();
2436 if (!path)
2437 goto make_bad;
2438
2439 path->leave_spinning = 1;
2440 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2441
2442 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2443 if (ret)
2444 goto make_bad;
2445
2446 leaf = path->nodes[0];
2447
2448 if (filled)
2449 goto cache_acl;
2450
2451 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2452 struct btrfs_inode_item);
2453 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2454 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2455 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2456 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2457 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2458
2459 tspec = btrfs_inode_atime(inode_item);
2460 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2461 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2462
2463 tspec = btrfs_inode_mtime(inode_item);
2464 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2465 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2466
2467 tspec = btrfs_inode_ctime(inode_item);
2468 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2469 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2470
2471 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2472 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2473 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2474 inode->i_generation = BTRFS_I(inode)->generation;
2475 inode->i_rdev = 0;
2476 rdev = btrfs_inode_rdev(leaf, inode_item);
2477
2478 BTRFS_I(inode)->index_cnt = (u64)-1;
2479 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2480 cache_acl:
2481 /*
2482 * try to precache a NULL acl entry for files that don't have
2483 * any xattrs or acls
2484 */
2485 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2486 btrfs_ino(inode));
2487 if (!maybe_acls)
2488 cache_no_acl(inode);
2489
2490 btrfs_free_path(path);
2491
2492 switch (inode->i_mode & S_IFMT) {
2493 case S_IFREG:
2494 inode->i_mapping->a_ops = &btrfs_aops;
2495 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2496 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2497 inode->i_fop = &btrfs_file_operations;
2498 inode->i_op = &btrfs_file_inode_operations;
2499 break;
2500 case S_IFDIR:
2501 inode->i_fop = &btrfs_dir_file_operations;
2502 if (root == root->fs_info->tree_root)
2503 inode->i_op = &btrfs_dir_ro_inode_operations;
2504 else
2505 inode->i_op = &btrfs_dir_inode_operations;
2506 break;
2507 case S_IFLNK:
2508 inode->i_op = &btrfs_symlink_inode_operations;
2509 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2510 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2511 break;
2512 default:
2513 inode->i_op = &btrfs_special_inode_operations;
2514 init_special_inode(inode, inode->i_mode, rdev);
2515 break;
2516 }
2517
2518 btrfs_update_iflags(inode);
2519 return;
2520
2521 make_bad:
2522 btrfs_free_path(path);
2523 make_bad_inode(inode);
2524 }
2525
2526 /*
2527 * given a leaf and an inode, copy the inode fields into the leaf
2528 */
2529 static void fill_inode_item(struct btrfs_trans_handle *trans,
2530 struct extent_buffer *leaf,
2531 struct btrfs_inode_item *item,
2532 struct inode *inode)
2533 {
2534 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2535 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2536 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2537 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2538 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2539
2540 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2541 inode->i_atime.tv_sec);
2542 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2543 inode->i_atime.tv_nsec);
2544
2545 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2546 inode->i_mtime.tv_sec);
2547 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2548 inode->i_mtime.tv_nsec);
2549
2550 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2551 inode->i_ctime.tv_sec);
2552 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2553 inode->i_ctime.tv_nsec);
2554
2555 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2556 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2557 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2558 btrfs_set_inode_transid(leaf, item, trans->transid);
2559 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2560 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2561 btrfs_set_inode_block_group(leaf, item, 0);
2562 }
2563
2564 /*
2565 * copy everything in the in-memory inode into the btree.
2566 */
2567 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2568 struct btrfs_root *root, struct inode *inode)
2569 {
2570 struct btrfs_inode_item *inode_item;
2571 struct btrfs_path *path;
2572 struct extent_buffer *leaf;
2573 int ret;
2574
2575 /*
2576 * If the inode is a free space inode, we can deadlock during commit
2577 * if we put it into the delayed code.
2578 *
2579 * The data relocation inode should also be directly updated
2580 * without delay
2581 */
2582 if (!btrfs_is_free_space_inode(root, inode)
2583 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2584 ret = btrfs_delayed_update_inode(trans, root, inode);
2585 if (!ret)
2586 btrfs_set_inode_last_trans(trans, inode);
2587 return ret;
2588 }
2589
2590 path = btrfs_alloc_path();
2591 if (!path)
2592 return -ENOMEM;
2593
2594 path->leave_spinning = 1;
2595 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2596 1);
2597 if (ret) {
2598 if (ret > 0)
2599 ret = -ENOENT;
2600 goto failed;
2601 }
2602
2603 btrfs_unlock_up_safe(path, 1);
2604 leaf = path->nodes[0];
2605 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2606 struct btrfs_inode_item);
2607
2608 fill_inode_item(trans, leaf, inode_item, inode);
2609 btrfs_mark_buffer_dirty(leaf);
2610 btrfs_set_inode_last_trans(trans, inode);
2611 ret = 0;
2612 failed:
2613 btrfs_free_path(path);
2614 return ret;
2615 }
2616
2617 /*
2618 * unlink helper that gets used here in inode.c and in the tree logging
2619 * recovery code. It remove a link in a directory with a given name, and
2620 * also drops the back refs in the inode to the directory
2621 */
2622 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2623 struct btrfs_root *root,
2624 struct inode *dir, struct inode *inode,
2625 const char *name, int name_len)
2626 {
2627 struct btrfs_path *path;
2628 int ret = 0;
2629 struct extent_buffer *leaf;
2630 struct btrfs_dir_item *di;
2631 struct btrfs_key key;
2632 u64 index;
2633 u64 ino = btrfs_ino(inode);
2634 u64 dir_ino = btrfs_ino(dir);
2635
2636 path = btrfs_alloc_path();
2637 if (!path) {
2638 ret = -ENOMEM;
2639 goto out;
2640 }
2641
2642 path->leave_spinning = 1;
2643 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2644 name, name_len, -1);
2645 if (IS_ERR(di)) {
2646 ret = PTR_ERR(di);
2647 goto err;
2648 }
2649 if (!di) {
2650 ret = -ENOENT;
2651 goto err;
2652 }
2653 leaf = path->nodes[0];
2654 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2655 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2656 if (ret)
2657 goto err;
2658 btrfs_release_path(path);
2659
2660 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2661 dir_ino, &index);
2662 if (ret) {
2663 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2664 "inode %llu parent %llu\n", name_len, name,
2665 (unsigned long long)ino, (unsigned long long)dir_ino);
2666 goto err;
2667 }
2668
2669 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2670 if (ret)
2671 goto err;
2672
2673 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2674 inode, dir_ino);
2675 BUG_ON(ret != 0 && ret != -ENOENT);
2676
2677 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2678 dir, index);
2679 if (ret == -ENOENT)
2680 ret = 0;
2681 err:
2682 btrfs_free_path(path);
2683 if (ret)
2684 goto out;
2685
2686 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2687 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2688 btrfs_update_inode(trans, root, dir);
2689 out:
2690 return ret;
2691 }
2692
2693 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2694 struct btrfs_root *root,
2695 struct inode *dir, struct inode *inode,
2696 const char *name, int name_len)
2697 {
2698 int ret;
2699 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2700 if (!ret) {
2701 btrfs_drop_nlink(inode);
2702 ret = btrfs_update_inode(trans, root, inode);
2703 }
2704 return ret;
2705 }
2706
2707
2708 /* helper to check if there is any shared block in the path */
2709 static int check_path_shared(struct btrfs_root *root,
2710 struct btrfs_path *path)
2711 {
2712 struct extent_buffer *eb;
2713 int level;
2714 u64 refs = 1;
2715
2716 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2717 int ret;
2718
2719 if (!path->nodes[level])
2720 break;
2721 eb = path->nodes[level];
2722 if (!btrfs_block_can_be_shared(root, eb))
2723 continue;
2724 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2725 &refs, NULL);
2726 if (refs > 1)
2727 return 1;
2728 }
2729 return 0;
2730 }
2731
2732 /*
2733 * helper to start transaction for unlink and rmdir.
2734 *
2735 * unlink and rmdir are special in btrfs, they do not always free space.
2736 * so in enospc case, we should make sure they will free space before
2737 * allowing them to use the global metadata reservation.
2738 */
2739 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2740 struct dentry *dentry)
2741 {
2742 struct btrfs_trans_handle *trans;
2743 struct btrfs_root *root = BTRFS_I(dir)->root;
2744 struct btrfs_path *path;
2745 struct btrfs_inode_ref *ref;
2746 struct btrfs_dir_item *di;
2747 struct inode *inode = dentry->d_inode;
2748 u64 index;
2749 int check_link = 1;
2750 int err = -ENOSPC;
2751 int ret;
2752 u64 ino = btrfs_ino(inode);
2753 u64 dir_ino = btrfs_ino(dir);
2754
2755 trans = btrfs_start_transaction(root, 10);
2756 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2757 return trans;
2758
2759 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2760 return ERR_PTR(-ENOSPC);
2761
2762 /* check if there is someone else holds reference */
2763 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2764 return ERR_PTR(-ENOSPC);
2765
2766 if (atomic_read(&inode->i_count) > 2)
2767 return ERR_PTR(-ENOSPC);
2768
2769 if (xchg(&root->fs_info->enospc_unlink, 1))
2770 return ERR_PTR(-ENOSPC);
2771
2772 path = btrfs_alloc_path();
2773 if (!path) {
2774 root->fs_info->enospc_unlink = 0;
2775 return ERR_PTR(-ENOMEM);
2776 }
2777
2778 trans = btrfs_start_transaction(root, 0);
2779 if (IS_ERR(trans)) {
2780 btrfs_free_path(path);
2781 root->fs_info->enospc_unlink = 0;
2782 return trans;
2783 }
2784
2785 path->skip_locking = 1;
2786 path->search_commit_root = 1;
2787
2788 ret = btrfs_lookup_inode(trans, root, path,
2789 &BTRFS_I(dir)->location, 0);
2790 if (ret < 0) {
2791 err = ret;
2792 goto out;
2793 }
2794 if (ret == 0) {
2795 if (check_path_shared(root, path))
2796 goto out;
2797 } else {
2798 check_link = 0;
2799 }
2800 btrfs_release_path(path);
2801
2802 ret = btrfs_lookup_inode(trans, root, path,
2803 &BTRFS_I(inode)->location, 0);
2804 if (ret < 0) {
2805 err = ret;
2806 goto out;
2807 }
2808 if (ret == 0) {
2809 if (check_path_shared(root, path))
2810 goto out;
2811 } else {
2812 check_link = 0;
2813 }
2814 btrfs_release_path(path);
2815
2816 if (ret == 0 && S_ISREG(inode->i_mode)) {
2817 ret = btrfs_lookup_file_extent(trans, root, path,
2818 ino, (u64)-1, 0);
2819 if (ret < 0) {
2820 err = ret;
2821 goto out;
2822 }
2823 BUG_ON(ret == 0);
2824 if (check_path_shared(root, path))
2825 goto out;
2826 btrfs_release_path(path);
2827 }
2828
2829 if (!check_link) {
2830 err = 0;
2831 goto out;
2832 }
2833
2834 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2835 dentry->d_name.name, dentry->d_name.len, 0);
2836 if (IS_ERR(di)) {
2837 err = PTR_ERR(di);
2838 goto out;
2839 }
2840 if (di) {
2841 if (check_path_shared(root, path))
2842 goto out;
2843 } else {
2844 err = 0;
2845 goto out;
2846 }
2847 btrfs_release_path(path);
2848
2849 ref = btrfs_lookup_inode_ref(trans, root, path,
2850 dentry->d_name.name, dentry->d_name.len,
2851 ino, dir_ino, 0);
2852 if (IS_ERR(ref)) {
2853 err = PTR_ERR(ref);
2854 goto out;
2855 }
2856 BUG_ON(!ref);
2857 if (check_path_shared(root, path))
2858 goto out;
2859 index = btrfs_inode_ref_index(path->nodes[0], ref);
2860 btrfs_release_path(path);
2861
2862 /*
2863 * This is a commit root search, if we can lookup inode item and other
2864 * relative items in the commit root, it means the transaction of
2865 * dir/file creation has been committed, and the dir index item that we
2866 * delay to insert has also been inserted into the commit root. So
2867 * we needn't worry about the delayed insertion of the dir index item
2868 * here.
2869 */
2870 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2871 dentry->d_name.name, dentry->d_name.len, 0);
2872 if (IS_ERR(di)) {
2873 err = PTR_ERR(di);
2874 goto out;
2875 }
2876 BUG_ON(ret == -ENOENT);
2877 if (check_path_shared(root, path))
2878 goto out;
2879
2880 err = 0;
2881 out:
2882 btrfs_free_path(path);
2883 if (err) {
2884 btrfs_end_transaction(trans, root);
2885 root->fs_info->enospc_unlink = 0;
2886 return ERR_PTR(err);
2887 }
2888
2889 trans->block_rsv = &root->fs_info->global_block_rsv;
2890 return trans;
2891 }
2892
2893 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2894 struct btrfs_root *root)
2895 {
2896 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2897 BUG_ON(!root->fs_info->enospc_unlink);
2898 root->fs_info->enospc_unlink = 0;
2899 }
2900 btrfs_end_transaction_throttle(trans, root);
2901 }
2902
2903 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2904 {
2905 struct btrfs_root *root = BTRFS_I(dir)->root;
2906 struct btrfs_trans_handle *trans;
2907 struct inode *inode = dentry->d_inode;
2908 int ret;
2909 unsigned long nr = 0;
2910
2911 trans = __unlink_start_trans(dir, dentry);
2912 if (IS_ERR(trans))
2913 return PTR_ERR(trans);
2914
2915 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2916
2917 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2918 dentry->d_name.name, dentry->d_name.len);
2919 if (ret)
2920 goto out;
2921
2922 if (inode->i_nlink == 0) {
2923 ret = btrfs_orphan_add(trans, inode);
2924 if (ret)
2925 goto out;
2926 }
2927
2928 out:
2929 nr = trans->blocks_used;
2930 __unlink_end_trans(trans, root);
2931 btrfs_btree_balance_dirty(root, nr);
2932 return ret;
2933 }
2934
2935 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2936 struct btrfs_root *root,
2937 struct inode *dir, u64 objectid,
2938 const char *name, int name_len)
2939 {
2940 struct btrfs_path *path;
2941 struct extent_buffer *leaf;
2942 struct btrfs_dir_item *di;
2943 struct btrfs_key key;
2944 u64 index;
2945 int ret;
2946 u64 dir_ino = btrfs_ino(dir);
2947
2948 path = btrfs_alloc_path();
2949 if (!path)
2950 return -ENOMEM;
2951
2952 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2953 name, name_len, -1);
2954 BUG_ON(IS_ERR_OR_NULL(di));
2955
2956 leaf = path->nodes[0];
2957 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2958 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2959 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2960 BUG_ON(ret);
2961 btrfs_release_path(path);
2962
2963 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2964 objectid, root->root_key.objectid,
2965 dir_ino, &index, name, name_len);
2966 if (ret < 0) {
2967 BUG_ON(ret != -ENOENT);
2968 di = btrfs_search_dir_index_item(root, path, dir_ino,
2969 name, name_len);
2970 BUG_ON(IS_ERR_OR_NULL(di));
2971
2972 leaf = path->nodes[0];
2973 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2974 btrfs_release_path(path);
2975 index = key.offset;
2976 }
2977 btrfs_release_path(path);
2978
2979 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2980 BUG_ON(ret);
2981
2982 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2983 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2984 ret = btrfs_update_inode(trans, root, dir);
2985 BUG_ON(ret);
2986
2987 btrfs_free_path(path);
2988 return 0;
2989 }
2990
2991 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2992 {
2993 struct inode *inode = dentry->d_inode;
2994 int err = 0;
2995 struct btrfs_root *root = BTRFS_I(dir)->root;
2996 struct btrfs_trans_handle *trans;
2997 unsigned long nr = 0;
2998
2999 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3000 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3001 return -ENOTEMPTY;
3002
3003 trans = __unlink_start_trans(dir, dentry);
3004 if (IS_ERR(trans))
3005 return PTR_ERR(trans);
3006
3007 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3008 err = btrfs_unlink_subvol(trans, root, dir,
3009 BTRFS_I(inode)->location.objectid,
3010 dentry->d_name.name,
3011 dentry->d_name.len);
3012 goto out;
3013 }
3014
3015 err = btrfs_orphan_add(trans, inode);
3016 if (err)
3017 goto out;
3018
3019 /* now the directory is empty */
3020 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3021 dentry->d_name.name, dentry->d_name.len);
3022 if (!err)
3023 btrfs_i_size_write(inode, 0);
3024 out:
3025 nr = trans->blocks_used;
3026 __unlink_end_trans(trans, root);
3027 btrfs_btree_balance_dirty(root, nr);
3028
3029 return err;
3030 }
3031
3032 /*
3033 * this can truncate away extent items, csum items and directory items.
3034 * It starts at a high offset and removes keys until it can't find
3035 * any higher than new_size
3036 *
3037 * csum items that cross the new i_size are truncated to the new size
3038 * as well.
3039 *
3040 * min_type is the minimum key type to truncate down to. If set to 0, this
3041 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3042 */
3043 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3044 struct btrfs_root *root,
3045 struct inode *inode,
3046 u64 new_size, u32 min_type)
3047 {
3048 struct btrfs_path *path;
3049 struct extent_buffer *leaf;
3050 struct btrfs_file_extent_item *fi;
3051 struct btrfs_key key;
3052 struct btrfs_key found_key;
3053 u64 extent_start = 0;
3054 u64 extent_num_bytes = 0;
3055 u64 extent_offset = 0;
3056 u64 item_end = 0;
3057 u64 mask = root->sectorsize - 1;
3058 u32 found_type = (u8)-1;
3059 int found_extent;
3060 int del_item;
3061 int pending_del_nr = 0;
3062 int pending_del_slot = 0;
3063 int extent_type = -1;
3064 int encoding;
3065 int ret;
3066 int err = 0;
3067 u64 ino = btrfs_ino(inode);
3068
3069 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3070
3071 path = btrfs_alloc_path();
3072 if (!path)
3073 return -ENOMEM;
3074 path->reada = -1;
3075
3076 if (root->ref_cows || root == root->fs_info->tree_root)
3077 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3078
3079 /*
3080 * This function is also used to drop the items in the log tree before
3081 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3082 * it is used to drop the loged items. So we shouldn't kill the delayed
3083 * items.
3084 */
3085 if (min_type == 0 && root == BTRFS_I(inode)->root)
3086 btrfs_kill_delayed_inode_items(inode);
3087
3088 key.objectid = ino;
3089 key.offset = (u64)-1;
3090 key.type = (u8)-1;
3091
3092 search_again:
3093 path->leave_spinning = 1;
3094 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3095 if (ret < 0) {
3096 err = ret;
3097 goto out;
3098 }
3099
3100 if (ret > 0) {
3101 /* there are no items in the tree for us to truncate, we're
3102 * done
3103 */
3104 if (path->slots[0] == 0)
3105 goto out;
3106 path->slots[0]--;
3107 }
3108
3109 while (1) {
3110 fi = NULL;
3111 leaf = path->nodes[0];
3112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3113 found_type = btrfs_key_type(&found_key);
3114 encoding = 0;
3115
3116 if (found_key.objectid != ino)
3117 break;
3118
3119 if (found_type < min_type)
3120 break;
3121
3122 item_end = found_key.offset;
3123 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3124 fi = btrfs_item_ptr(leaf, path->slots[0],
3125 struct btrfs_file_extent_item);
3126 extent_type = btrfs_file_extent_type(leaf, fi);
3127 encoding = btrfs_file_extent_compression(leaf, fi);
3128 encoding |= btrfs_file_extent_encryption(leaf, fi);
3129 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3130
3131 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3132 item_end +=
3133 btrfs_file_extent_num_bytes(leaf, fi);
3134 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3135 item_end += btrfs_file_extent_inline_len(leaf,
3136 fi);
3137 }
3138 item_end--;
3139 }
3140 if (found_type > min_type) {
3141 del_item = 1;
3142 } else {
3143 if (item_end < new_size)
3144 break;
3145 if (found_key.offset >= new_size)
3146 del_item = 1;
3147 else
3148 del_item = 0;
3149 }
3150 found_extent = 0;
3151 /* FIXME, shrink the extent if the ref count is only 1 */
3152 if (found_type != BTRFS_EXTENT_DATA_KEY)
3153 goto delete;
3154
3155 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3156 u64 num_dec;
3157 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3158 if (!del_item && !encoding) {
3159 u64 orig_num_bytes =
3160 btrfs_file_extent_num_bytes(leaf, fi);
3161 extent_num_bytes = new_size -
3162 found_key.offset + root->sectorsize - 1;
3163 extent_num_bytes = extent_num_bytes &
3164 ~((u64)root->sectorsize - 1);
3165 btrfs_set_file_extent_num_bytes(leaf, fi,
3166 extent_num_bytes);
3167 num_dec = (orig_num_bytes -
3168 extent_num_bytes);
3169 if (root->ref_cows && extent_start != 0)
3170 inode_sub_bytes(inode, num_dec);
3171 btrfs_mark_buffer_dirty(leaf);
3172 } else {
3173 extent_num_bytes =
3174 btrfs_file_extent_disk_num_bytes(leaf,
3175 fi);
3176 extent_offset = found_key.offset -
3177 btrfs_file_extent_offset(leaf, fi);
3178
3179 /* FIXME blocksize != 4096 */
3180 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3181 if (extent_start != 0) {
3182 found_extent = 1;
3183 if (root->ref_cows)
3184 inode_sub_bytes(inode, num_dec);
3185 }
3186 }
3187 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3188 /*
3189 * we can't truncate inline items that have had
3190 * special encodings
3191 */
3192 if (!del_item &&
3193 btrfs_file_extent_compression(leaf, fi) == 0 &&
3194 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3195 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3196 u32 size = new_size - found_key.offset;
3197
3198 if (root->ref_cows) {
3199 inode_sub_bytes(inode, item_end + 1 -
3200 new_size);
3201 }
3202 size =
3203 btrfs_file_extent_calc_inline_size(size);
3204 ret = btrfs_truncate_item(trans, root, path,
3205 size, 1);
3206 } else if (root->ref_cows) {
3207 inode_sub_bytes(inode, item_end + 1 -
3208 found_key.offset);
3209 }
3210 }
3211 delete:
3212 if (del_item) {
3213 if (!pending_del_nr) {
3214 /* no pending yet, add ourselves */
3215 pending_del_slot = path->slots[0];
3216 pending_del_nr = 1;
3217 } else if (pending_del_nr &&
3218 path->slots[0] + 1 == pending_del_slot) {
3219 /* hop on the pending chunk */
3220 pending_del_nr++;
3221 pending_del_slot = path->slots[0];
3222 } else {
3223 BUG();
3224 }
3225 } else {
3226 break;
3227 }
3228 if (found_extent && (root->ref_cows ||
3229 root == root->fs_info->tree_root)) {
3230 btrfs_set_path_blocking(path);
3231 ret = btrfs_free_extent(trans, root, extent_start,
3232 extent_num_bytes, 0,
3233 btrfs_header_owner(leaf),
3234 ino, extent_offset);
3235 BUG_ON(ret);
3236 }
3237
3238 if (found_type == BTRFS_INODE_ITEM_KEY)
3239 break;
3240
3241 if (path->slots[0] == 0 ||
3242 path->slots[0] != pending_del_slot) {
3243 if (root->ref_cows &&
3244 BTRFS_I(inode)->location.objectid !=
3245 BTRFS_FREE_INO_OBJECTID) {
3246 err = -EAGAIN;
3247 goto out;
3248 }
3249 if (pending_del_nr) {
3250 ret = btrfs_del_items(trans, root, path,
3251 pending_del_slot,
3252 pending_del_nr);
3253 BUG_ON(ret);
3254 pending_del_nr = 0;
3255 }
3256 btrfs_release_path(path);
3257 goto search_again;
3258 } else {
3259 path->slots[0]--;
3260 }
3261 }
3262 out:
3263 if (pending_del_nr) {
3264 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3265 pending_del_nr);
3266 BUG_ON(ret);
3267 }
3268 btrfs_free_path(path);
3269 return err;
3270 }
3271
3272 /*
3273 * taken from block_truncate_page, but does cow as it zeros out
3274 * any bytes left in the last page in the file.
3275 */
3276 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3277 {
3278 struct inode *inode = mapping->host;
3279 struct btrfs_root *root = BTRFS_I(inode)->root;
3280 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3281 struct btrfs_ordered_extent *ordered;
3282 struct extent_state *cached_state = NULL;
3283 char *kaddr;
3284 u32 blocksize = root->sectorsize;
3285 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3286 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3287 struct page *page;
3288 int ret = 0;
3289 u64 page_start;
3290 u64 page_end;
3291
3292 if ((offset & (blocksize - 1)) == 0)
3293 goto out;
3294 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3295 if (ret)
3296 goto out;
3297
3298 ret = -ENOMEM;
3299 again:
3300 page = find_or_create_page(mapping, index, GFP_NOFS);
3301 if (!page) {
3302 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3303 goto out;
3304 }
3305
3306 page_start = page_offset(page);
3307 page_end = page_start + PAGE_CACHE_SIZE - 1;
3308
3309 if (!PageUptodate(page)) {
3310 ret = btrfs_readpage(NULL, page);
3311 lock_page(page);
3312 if (page->mapping != mapping) {
3313 unlock_page(page);
3314 page_cache_release(page);
3315 goto again;
3316 }
3317 if (!PageUptodate(page)) {
3318 ret = -EIO;
3319 goto out_unlock;
3320 }
3321 }
3322 wait_on_page_writeback(page);
3323
3324 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3325 GFP_NOFS);
3326 set_page_extent_mapped(page);
3327
3328 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3329 if (ordered) {
3330 unlock_extent_cached(io_tree, page_start, page_end,
3331 &cached_state, GFP_NOFS);
3332 unlock_page(page);
3333 page_cache_release(page);
3334 btrfs_start_ordered_extent(inode, ordered, 1);
3335 btrfs_put_ordered_extent(ordered);
3336 goto again;
3337 }
3338
3339 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3340 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3341 0, 0, &cached_state, GFP_NOFS);
3342
3343 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3344 &cached_state);
3345 if (ret) {
3346 unlock_extent_cached(io_tree, page_start, page_end,
3347 &cached_state, GFP_NOFS);
3348 goto out_unlock;
3349 }
3350
3351 ret = 0;
3352 if (offset != PAGE_CACHE_SIZE) {
3353 kaddr = kmap(page);
3354 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3355 flush_dcache_page(page);
3356 kunmap(page);
3357 }
3358 ClearPageChecked(page);
3359 set_page_dirty(page);
3360 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3361 GFP_NOFS);
3362
3363 out_unlock:
3364 if (ret)
3365 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3366 unlock_page(page);
3367 page_cache_release(page);
3368 out:
3369 return ret;
3370 }
3371
3372 /*
3373 * This function puts in dummy file extents for the area we're creating a hole
3374 * for. So if we are truncating this file to a larger size we need to insert
3375 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3376 * the range between oldsize and size
3377 */
3378 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3379 {
3380 struct btrfs_trans_handle *trans;
3381 struct btrfs_root *root = BTRFS_I(inode)->root;
3382 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3383 struct extent_map *em = NULL;
3384 struct extent_state *cached_state = NULL;
3385 u64 mask = root->sectorsize - 1;
3386 u64 hole_start = (oldsize + mask) & ~mask;
3387 u64 block_end = (size + mask) & ~mask;
3388 u64 last_byte;
3389 u64 cur_offset;
3390 u64 hole_size;
3391 int err = 0;
3392
3393 if (size <= hole_start)
3394 return 0;
3395
3396 while (1) {
3397 struct btrfs_ordered_extent *ordered;
3398 btrfs_wait_ordered_range(inode, hole_start,
3399 block_end - hole_start);
3400 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3401 &cached_state, GFP_NOFS);
3402 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3403 if (!ordered)
3404 break;
3405 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3406 &cached_state, GFP_NOFS);
3407 btrfs_put_ordered_extent(ordered);
3408 }
3409
3410 cur_offset = hole_start;
3411 while (1) {
3412 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3413 block_end - cur_offset, 0);
3414 BUG_ON(IS_ERR_OR_NULL(em));
3415 last_byte = min(extent_map_end(em), block_end);
3416 last_byte = (last_byte + mask) & ~mask;
3417 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3418 u64 hint_byte = 0;
3419 hole_size = last_byte - cur_offset;
3420
3421 trans = btrfs_start_transaction(root, 2);
3422 if (IS_ERR(trans)) {
3423 err = PTR_ERR(trans);
3424 break;
3425 }
3426
3427 err = btrfs_drop_extents(trans, inode, cur_offset,
3428 cur_offset + hole_size,
3429 &hint_byte, 1);
3430 if (err) {
3431 btrfs_end_transaction(trans, root);
3432 break;
3433 }
3434
3435 err = btrfs_insert_file_extent(trans, root,
3436 btrfs_ino(inode), cur_offset, 0,
3437 0, hole_size, 0, hole_size,
3438 0, 0, 0);
3439 if (err) {
3440 btrfs_end_transaction(trans, root);
3441 break;
3442 }
3443
3444 btrfs_drop_extent_cache(inode, hole_start,
3445 last_byte - 1, 0);
3446
3447 btrfs_end_transaction(trans, root);
3448 }
3449 free_extent_map(em);
3450 em = NULL;
3451 cur_offset = last_byte;
3452 if (cur_offset >= block_end)
3453 break;
3454 }
3455
3456 free_extent_map(em);
3457 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3458 GFP_NOFS);
3459 return err;
3460 }
3461
3462 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3463 {
3464 loff_t oldsize = i_size_read(inode);
3465 int ret;
3466
3467 if (newsize == oldsize)
3468 return 0;
3469
3470 if (newsize > oldsize) {
3471 i_size_write(inode, newsize);
3472 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3473 truncate_pagecache(inode, oldsize, newsize);
3474 ret = btrfs_cont_expand(inode, oldsize, newsize);
3475 if (ret) {
3476 btrfs_setsize(inode, oldsize);
3477 return ret;
3478 }
3479
3480 mark_inode_dirty(inode);
3481 } else {
3482
3483 /*
3484 * We're truncating a file that used to have good data down to
3485 * zero. Make sure it gets into the ordered flush list so that
3486 * any new writes get down to disk quickly.
3487 */
3488 if (newsize == 0)
3489 BTRFS_I(inode)->ordered_data_close = 1;
3490
3491 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3492 truncate_setsize(inode, newsize);
3493 ret = btrfs_truncate(inode);
3494 }
3495
3496 return ret;
3497 }
3498
3499 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3500 {
3501 struct inode *inode = dentry->d_inode;
3502 struct btrfs_root *root = BTRFS_I(inode)->root;
3503 int err;
3504
3505 if (btrfs_root_readonly(root))
3506 return -EROFS;
3507
3508 err = inode_change_ok(inode, attr);
3509 if (err)
3510 return err;
3511
3512 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3513 err = btrfs_setsize(inode, attr->ia_size);
3514 if (err)
3515 return err;
3516 }
3517
3518 if (attr->ia_valid) {
3519 setattr_copy(inode, attr);
3520 mark_inode_dirty(inode);
3521
3522 if (attr->ia_valid & ATTR_MODE)
3523 err = btrfs_acl_chmod(inode);
3524 }
3525
3526 return err;
3527 }
3528
3529 void btrfs_evict_inode(struct inode *inode)
3530 {
3531 struct btrfs_trans_handle *trans;
3532 struct btrfs_root *root = BTRFS_I(inode)->root;
3533 unsigned long nr;
3534 int ret;
3535
3536 trace_btrfs_inode_evict(inode);
3537
3538 truncate_inode_pages(&inode->i_data, 0);
3539 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3540 btrfs_is_free_space_inode(root, inode)))
3541 goto no_delete;
3542
3543 if (is_bad_inode(inode)) {
3544 btrfs_orphan_del(NULL, inode);
3545 goto no_delete;
3546 }
3547 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3548 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3549
3550 if (root->fs_info->log_root_recovering) {
3551 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3552 goto no_delete;
3553 }
3554
3555 if (inode->i_nlink > 0) {
3556 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3557 goto no_delete;
3558 }
3559
3560 btrfs_i_size_write(inode, 0);
3561
3562 while (1) {
3563 trans = btrfs_join_transaction(root);
3564 BUG_ON(IS_ERR(trans));
3565 trans->block_rsv = root->orphan_block_rsv;
3566
3567 ret = btrfs_block_rsv_check(trans, root,
3568 root->orphan_block_rsv, 0, 5);
3569 if (ret) {
3570 BUG_ON(ret != -EAGAIN);
3571 ret = btrfs_commit_transaction(trans, root);
3572 BUG_ON(ret);
3573 continue;
3574 }
3575
3576 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3577 if (ret != -EAGAIN)
3578 break;
3579
3580 nr = trans->blocks_used;
3581 btrfs_end_transaction(trans, root);
3582 trans = NULL;
3583 btrfs_btree_balance_dirty(root, nr);
3584
3585 }
3586
3587 if (ret == 0) {
3588 ret = btrfs_orphan_del(trans, inode);
3589 BUG_ON(ret);
3590 }
3591
3592 if (!(root == root->fs_info->tree_root ||
3593 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3594 btrfs_return_ino(root, btrfs_ino(inode));
3595
3596 nr = trans->blocks_used;
3597 btrfs_end_transaction(trans, root);
3598 btrfs_btree_balance_dirty(root, nr);
3599 no_delete:
3600 end_writeback(inode);
3601 return;
3602 }
3603
3604 /*
3605 * this returns the key found in the dir entry in the location pointer.
3606 * If no dir entries were found, location->objectid is 0.
3607 */
3608 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3609 struct btrfs_key *location)
3610 {
3611 const char *name = dentry->d_name.name;
3612 int namelen = dentry->d_name.len;
3613 struct btrfs_dir_item *di;
3614 struct btrfs_path *path;
3615 struct btrfs_root *root = BTRFS_I(dir)->root;
3616 int ret = 0;
3617
3618 path = btrfs_alloc_path();
3619 if (!path)
3620 return -ENOMEM;
3621
3622 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3623 namelen, 0);
3624 if (IS_ERR(di))
3625 ret = PTR_ERR(di);
3626
3627 if (IS_ERR_OR_NULL(di))
3628 goto out_err;
3629
3630 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3631 out:
3632 btrfs_free_path(path);
3633 return ret;
3634 out_err:
3635 location->objectid = 0;
3636 goto out;
3637 }
3638
3639 /*
3640 * when we hit a tree root in a directory, the btrfs part of the inode
3641 * needs to be changed to reflect the root directory of the tree root. This
3642 * is kind of like crossing a mount point.
3643 */
3644 static int fixup_tree_root_location(struct btrfs_root *root,
3645 struct inode *dir,
3646 struct dentry *dentry,
3647 struct btrfs_key *location,
3648 struct btrfs_root **sub_root)
3649 {
3650 struct btrfs_path *path;
3651 struct btrfs_root *new_root;
3652 struct btrfs_root_ref *ref;
3653 struct extent_buffer *leaf;
3654 int ret;
3655 int err = 0;
3656
3657 path = btrfs_alloc_path();
3658 if (!path) {
3659 err = -ENOMEM;
3660 goto out;
3661 }
3662
3663 err = -ENOENT;
3664 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3665 BTRFS_I(dir)->root->root_key.objectid,
3666 location->objectid);
3667 if (ret) {
3668 if (ret < 0)
3669 err = ret;
3670 goto out;
3671 }
3672
3673 leaf = path->nodes[0];
3674 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3675 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3676 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3677 goto out;
3678
3679 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3680 (unsigned long)(ref + 1),
3681 dentry->d_name.len);
3682 if (ret)
3683 goto out;
3684
3685 btrfs_release_path(path);
3686
3687 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3688 if (IS_ERR(new_root)) {
3689 err = PTR_ERR(new_root);
3690 goto out;
3691 }
3692
3693 if (btrfs_root_refs(&new_root->root_item) == 0) {
3694 err = -ENOENT;
3695 goto out;
3696 }
3697
3698 *sub_root = new_root;
3699 location->objectid = btrfs_root_dirid(&new_root->root_item);
3700 location->type = BTRFS_INODE_ITEM_KEY;
3701 location->offset = 0;
3702 err = 0;
3703 out:
3704 btrfs_free_path(path);
3705 return err;
3706 }
3707
3708 static void inode_tree_add(struct inode *inode)
3709 {
3710 struct btrfs_root *root = BTRFS_I(inode)->root;
3711 struct btrfs_inode *entry;
3712 struct rb_node **p;
3713 struct rb_node *parent;
3714 u64 ino = btrfs_ino(inode);
3715 again:
3716 p = &root->inode_tree.rb_node;
3717 parent = NULL;
3718
3719 if (inode_unhashed(inode))
3720 return;
3721
3722 spin_lock(&root->inode_lock);
3723 while (*p) {
3724 parent = *p;
3725 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3726
3727 if (ino < btrfs_ino(&entry->vfs_inode))
3728 p = &parent->rb_left;
3729 else if (ino > btrfs_ino(&entry->vfs_inode))
3730 p = &parent->rb_right;
3731 else {
3732 WARN_ON(!(entry->vfs_inode.i_state &
3733 (I_WILL_FREE | I_FREEING)));
3734 rb_erase(parent, &root->inode_tree);
3735 RB_CLEAR_NODE(parent);
3736 spin_unlock(&root->inode_lock);
3737 goto again;
3738 }
3739 }
3740 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3741 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3742 spin_unlock(&root->inode_lock);
3743 }
3744
3745 static void inode_tree_del(struct inode *inode)
3746 {
3747 struct btrfs_root *root = BTRFS_I(inode)->root;
3748 int empty = 0;
3749
3750 spin_lock(&root->inode_lock);
3751 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3752 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3753 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3754 empty = RB_EMPTY_ROOT(&root->inode_tree);
3755 }
3756 spin_unlock(&root->inode_lock);
3757
3758 /*
3759 * Free space cache has inodes in the tree root, but the tree root has a
3760 * root_refs of 0, so this could end up dropping the tree root as a
3761 * snapshot, so we need the extra !root->fs_info->tree_root check to
3762 * make sure we don't drop it.
3763 */
3764 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3765 root != root->fs_info->tree_root) {
3766 synchronize_srcu(&root->fs_info->subvol_srcu);
3767 spin_lock(&root->inode_lock);
3768 empty = RB_EMPTY_ROOT(&root->inode_tree);
3769 spin_unlock(&root->inode_lock);
3770 if (empty)
3771 btrfs_add_dead_root(root);
3772 }
3773 }
3774
3775 int btrfs_invalidate_inodes(struct btrfs_root *root)
3776 {
3777 struct rb_node *node;
3778 struct rb_node *prev;
3779 struct btrfs_inode *entry;
3780 struct inode *inode;
3781 u64 objectid = 0;
3782
3783 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3784
3785 spin_lock(&root->inode_lock);
3786 again:
3787 node = root->inode_tree.rb_node;
3788 prev = NULL;
3789 while (node) {
3790 prev = node;
3791 entry = rb_entry(node, struct btrfs_inode, rb_node);
3792
3793 if (objectid < btrfs_ino(&entry->vfs_inode))
3794 node = node->rb_left;
3795 else if (objectid > btrfs_ino(&entry->vfs_inode))
3796 node = node->rb_right;
3797 else
3798 break;
3799 }
3800 if (!node) {
3801 while (prev) {
3802 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3803 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3804 node = prev;
3805 break;
3806 }
3807 prev = rb_next(prev);
3808 }
3809 }
3810 while (node) {
3811 entry = rb_entry(node, struct btrfs_inode, rb_node);
3812 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3813 inode = igrab(&entry->vfs_inode);
3814 if (inode) {
3815 spin_unlock(&root->inode_lock);
3816 if (atomic_read(&inode->i_count) > 1)
3817 d_prune_aliases(inode);
3818 /*
3819 * btrfs_drop_inode will have it removed from
3820 * the inode cache when its usage count
3821 * hits zero.
3822 */
3823 iput(inode);
3824 cond_resched();
3825 spin_lock(&root->inode_lock);
3826 goto again;
3827 }
3828
3829 if (cond_resched_lock(&root->inode_lock))
3830 goto again;
3831
3832 node = rb_next(node);
3833 }
3834 spin_unlock(&root->inode_lock);
3835 return 0;
3836 }
3837
3838 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3839 {
3840 struct btrfs_iget_args *args = p;
3841 inode->i_ino = args->ino;
3842 BTRFS_I(inode)->root = args->root;
3843 btrfs_set_inode_space_info(args->root, inode);
3844 return 0;
3845 }
3846
3847 static int btrfs_find_actor(struct inode *inode, void *opaque)
3848 {
3849 struct btrfs_iget_args *args = opaque;
3850 return args->ino == btrfs_ino(inode) &&
3851 args->root == BTRFS_I(inode)->root;
3852 }
3853
3854 static struct inode *btrfs_iget_locked(struct super_block *s,
3855 u64 objectid,
3856 struct btrfs_root *root)
3857 {
3858 struct inode *inode;
3859 struct btrfs_iget_args args;
3860 args.ino = objectid;
3861 args.root = root;
3862
3863 inode = iget5_locked(s, objectid, btrfs_find_actor,
3864 btrfs_init_locked_inode,
3865 (void *)&args);
3866 return inode;
3867 }
3868
3869 /* Get an inode object given its location and corresponding root.
3870 * Returns in *is_new if the inode was read from disk
3871 */
3872 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3873 struct btrfs_root *root, int *new)
3874 {
3875 struct inode *inode;
3876
3877 inode = btrfs_iget_locked(s, location->objectid, root);
3878 if (!inode)
3879 return ERR_PTR(-ENOMEM);
3880
3881 if (inode->i_state & I_NEW) {
3882 BTRFS_I(inode)->root = root;
3883 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3884 btrfs_read_locked_inode(inode);
3885 if (!is_bad_inode(inode)) {
3886 inode_tree_add(inode);
3887 unlock_new_inode(inode);
3888 if (new)
3889 *new = 1;
3890 } else {
3891 unlock_new_inode(inode);
3892 iput(inode);
3893 inode = ERR_PTR(-ESTALE);
3894 }
3895 }
3896
3897 return inode;
3898 }
3899
3900 static struct inode *new_simple_dir(struct super_block *s,
3901 struct btrfs_key *key,
3902 struct btrfs_root *root)
3903 {
3904 struct inode *inode = new_inode(s);
3905
3906 if (!inode)
3907 return ERR_PTR(-ENOMEM);
3908
3909 BTRFS_I(inode)->root = root;
3910 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3911 BTRFS_I(inode)->dummy_inode = 1;
3912
3913 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3914 inode->i_op = &simple_dir_inode_operations;
3915 inode->i_fop = &simple_dir_operations;
3916 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3917 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3918
3919 return inode;
3920 }
3921
3922 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3923 {
3924 struct inode *inode;
3925 struct btrfs_root *root = BTRFS_I(dir)->root;
3926 struct btrfs_root *sub_root = root;
3927 struct btrfs_key location;
3928 int index;
3929 int ret = 0;
3930
3931 if (dentry->d_name.len > BTRFS_NAME_LEN)
3932 return ERR_PTR(-ENAMETOOLONG);
3933
3934 if (unlikely(d_need_lookup(dentry))) {
3935 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3936 kfree(dentry->d_fsdata);
3937 dentry->d_fsdata = NULL;
3938 /* This thing is hashed, drop it for now */
3939 d_drop(dentry);
3940 } else {
3941 ret = btrfs_inode_by_name(dir, dentry, &location);
3942 }
3943
3944 if (ret < 0)
3945 return ERR_PTR(ret);
3946
3947 if (location.objectid == 0)
3948 return NULL;
3949
3950 if (location.type == BTRFS_INODE_ITEM_KEY) {
3951 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3952 return inode;
3953 }
3954
3955 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3956
3957 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3958 ret = fixup_tree_root_location(root, dir, dentry,
3959 &location, &sub_root);
3960 if (ret < 0) {
3961 if (ret != -ENOENT)
3962 inode = ERR_PTR(ret);
3963 else
3964 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3965 } else {
3966 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3967 }
3968 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3969
3970 if (!IS_ERR(inode) && root != sub_root) {
3971 down_read(&root->fs_info->cleanup_work_sem);
3972 if (!(inode->i_sb->s_flags & MS_RDONLY))
3973 ret = btrfs_orphan_cleanup(sub_root);
3974 up_read(&root->fs_info->cleanup_work_sem);
3975 if (ret)
3976 inode = ERR_PTR(ret);
3977 }
3978
3979 return inode;
3980 }
3981
3982 static int btrfs_dentry_delete(const struct dentry *dentry)
3983 {
3984 struct btrfs_root *root;
3985
3986 if (!dentry->d_inode && !IS_ROOT(dentry))
3987 dentry = dentry->d_parent;
3988
3989 if (dentry->d_inode) {
3990 root = BTRFS_I(dentry->d_inode)->root;
3991 if (btrfs_root_refs(&root->root_item) == 0)
3992 return 1;
3993 }
3994 return 0;
3995 }
3996
3997 static void btrfs_dentry_release(struct dentry *dentry)
3998 {
3999 if (dentry->d_fsdata)
4000 kfree(dentry->d_fsdata);
4001 }
4002
4003 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4004 struct nameidata *nd)
4005 {
4006 struct dentry *ret;
4007
4008 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4009 if (unlikely(d_need_lookup(dentry))) {
4010 spin_lock(&dentry->d_lock);
4011 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4012 spin_unlock(&dentry->d_lock);
4013 }
4014 return ret;
4015 }
4016
4017 unsigned char btrfs_filetype_table[] = {
4018 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4019 };
4020
4021 static int btrfs_real_readdir(struct file *filp, void *dirent,
4022 filldir_t filldir)
4023 {
4024 struct inode *inode = filp->f_dentry->d_inode;
4025 struct btrfs_root *root = BTRFS_I(inode)->root;
4026 struct btrfs_item *item;
4027 struct btrfs_dir_item *di;
4028 struct btrfs_key key;
4029 struct btrfs_key found_key;
4030 struct btrfs_path *path;
4031 struct list_head ins_list;
4032 struct list_head del_list;
4033 struct qstr q;
4034 int ret;
4035 struct extent_buffer *leaf;
4036 int slot;
4037 unsigned char d_type;
4038 int over = 0;
4039 u32 di_cur;
4040 u32 di_total;
4041 u32 di_len;
4042 int key_type = BTRFS_DIR_INDEX_KEY;
4043 char tmp_name[32];
4044 char *name_ptr;
4045 int name_len;
4046 int is_curr = 0; /* filp->f_pos points to the current index? */
4047
4048 /* FIXME, use a real flag for deciding about the key type */
4049 if (root->fs_info->tree_root == root)
4050 key_type = BTRFS_DIR_ITEM_KEY;
4051
4052 /* special case for "." */
4053 if (filp->f_pos == 0) {
4054 over = filldir(dirent, ".", 1,
4055 filp->f_pos, btrfs_ino(inode), DT_DIR);
4056 if (over)
4057 return 0;
4058 filp->f_pos = 1;
4059 }
4060 /* special case for .., just use the back ref */
4061 if (filp->f_pos == 1) {
4062 u64 pino = parent_ino(filp->f_path.dentry);
4063 over = filldir(dirent, "..", 2,
4064 filp->f_pos, pino, DT_DIR);
4065 if (over)
4066 return 0;
4067 filp->f_pos = 2;
4068 }
4069 path = btrfs_alloc_path();
4070 if (!path)
4071 return -ENOMEM;
4072
4073 path->reada = 1;
4074
4075 if (key_type == BTRFS_DIR_INDEX_KEY) {
4076 INIT_LIST_HEAD(&ins_list);
4077 INIT_LIST_HEAD(&del_list);
4078 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4079 }
4080
4081 btrfs_set_key_type(&key, key_type);
4082 key.offset = filp->f_pos;
4083 key.objectid = btrfs_ino(inode);
4084
4085 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4086 if (ret < 0)
4087 goto err;
4088
4089 while (1) {
4090 leaf = path->nodes[0];
4091 slot = path->slots[0];
4092 if (slot >= btrfs_header_nritems(leaf)) {
4093 ret = btrfs_next_leaf(root, path);
4094 if (ret < 0)
4095 goto err;
4096 else if (ret > 0)
4097 break;
4098 continue;
4099 }
4100
4101 item = btrfs_item_nr(leaf, slot);
4102 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4103
4104 if (found_key.objectid != key.objectid)
4105 break;
4106 if (btrfs_key_type(&found_key) != key_type)
4107 break;
4108 if (found_key.offset < filp->f_pos)
4109 goto next;
4110 if (key_type == BTRFS_DIR_INDEX_KEY &&
4111 btrfs_should_delete_dir_index(&del_list,
4112 found_key.offset))
4113 goto next;
4114
4115 filp->f_pos = found_key.offset;
4116 is_curr = 1;
4117
4118 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4119 di_cur = 0;
4120 di_total = btrfs_item_size(leaf, item);
4121
4122 while (di_cur < di_total) {
4123 struct btrfs_key location;
4124 struct dentry *tmp;
4125
4126 if (verify_dir_item(root, leaf, di))
4127 break;
4128
4129 name_len = btrfs_dir_name_len(leaf, di);
4130 if (name_len <= sizeof(tmp_name)) {
4131 name_ptr = tmp_name;
4132 } else {
4133 name_ptr = kmalloc(name_len, GFP_NOFS);
4134 if (!name_ptr) {
4135 ret = -ENOMEM;
4136 goto err;
4137 }
4138 }
4139 read_extent_buffer(leaf, name_ptr,
4140 (unsigned long)(di + 1), name_len);
4141
4142 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4143 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4144
4145 q.name = name_ptr;
4146 q.len = name_len;
4147 q.hash = full_name_hash(q.name, q.len);
4148 tmp = d_lookup(filp->f_dentry, &q);
4149 if (!tmp) {
4150 struct btrfs_key *newkey;
4151
4152 newkey = kzalloc(sizeof(struct btrfs_key),
4153 GFP_NOFS);
4154 if (!newkey)
4155 goto no_dentry;
4156 tmp = d_alloc(filp->f_dentry, &q);
4157 if (!tmp) {
4158 kfree(newkey);
4159 dput(tmp);
4160 goto no_dentry;
4161 }
4162 memcpy(newkey, &location,
4163 sizeof(struct btrfs_key));
4164 tmp->d_fsdata = newkey;
4165 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4166 d_rehash(tmp);
4167 dput(tmp);
4168 } else {
4169 dput(tmp);
4170 }
4171 no_dentry:
4172 /* is this a reference to our own snapshot? If so
4173 * skip it
4174 */
4175 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4176 location.objectid == root->root_key.objectid) {
4177 over = 0;
4178 goto skip;
4179 }
4180 over = filldir(dirent, name_ptr, name_len,
4181 found_key.offset, location.objectid,
4182 d_type);
4183
4184 skip:
4185 if (name_ptr != tmp_name)
4186 kfree(name_ptr);
4187
4188 if (over)
4189 goto nopos;
4190 di_len = btrfs_dir_name_len(leaf, di) +
4191 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4192 di_cur += di_len;
4193 di = (struct btrfs_dir_item *)((char *)di + di_len);
4194 }
4195 next:
4196 path->slots[0]++;
4197 }
4198
4199 if (key_type == BTRFS_DIR_INDEX_KEY) {
4200 if (is_curr)
4201 filp->f_pos++;
4202 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4203 &ins_list);
4204 if (ret)
4205 goto nopos;
4206 }
4207
4208 /* Reached end of directory/root. Bump pos past the last item. */
4209 if (key_type == BTRFS_DIR_INDEX_KEY)
4210 /*
4211 * 32-bit glibc will use getdents64, but then strtol -
4212 * so the last number we can serve is this.
4213 */
4214 filp->f_pos = 0x7fffffff;
4215 else
4216 filp->f_pos++;
4217 nopos:
4218 ret = 0;
4219 err:
4220 if (key_type == BTRFS_DIR_INDEX_KEY)
4221 btrfs_put_delayed_items(&ins_list, &del_list);
4222 btrfs_free_path(path);
4223 return ret;
4224 }
4225
4226 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4227 {
4228 struct btrfs_root *root = BTRFS_I(inode)->root;
4229 struct btrfs_trans_handle *trans;
4230 int ret = 0;
4231 bool nolock = false;
4232
4233 if (BTRFS_I(inode)->dummy_inode)
4234 return 0;
4235
4236 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4237 nolock = true;
4238
4239 if (wbc->sync_mode == WB_SYNC_ALL) {
4240 if (nolock)
4241 trans = btrfs_join_transaction_nolock(root);
4242 else
4243 trans = btrfs_join_transaction(root);
4244 if (IS_ERR(trans))
4245 return PTR_ERR(trans);
4246 if (nolock)
4247 ret = btrfs_end_transaction_nolock(trans, root);
4248 else
4249 ret = btrfs_commit_transaction(trans, root);
4250 }
4251 return ret;
4252 }
4253
4254 /*
4255 * This is somewhat expensive, updating the tree every time the
4256 * inode changes. But, it is most likely to find the inode in cache.
4257 * FIXME, needs more benchmarking...there are no reasons other than performance
4258 * to keep or drop this code.
4259 */
4260 void btrfs_dirty_inode(struct inode *inode, int flags)
4261 {
4262 struct btrfs_root *root = BTRFS_I(inode)->root;
4263 struct btrfs_trans_handle *trans;
4264 int ret;
4265
4266 if (BTRFS_I(inode)->dummy_inode)
4267 return;
4268
4269 trans = btrfs_join_transaction(root);
4270 BUG_ON(IS_ERR(trans));
4271
4272 ret = btrfs_update_inode(trans, root, inode);
4273 if (ret && ret == -ENOSPC) {
4274 /* whoops, lets try again with the full transaction */
4275 btrfs_end_transaction(trans, root);
4276 trans = btrfs_start_transaction(root, 1);
4277 if (IS_ERR(trans)) {
4278 printk_ratelimited(KERN_ERR "btrfs: fail to "
4279 "dirty inode %llu error %ld\n",
4280 (unsigned long long)btrfs_ino(inode),
4281 PTR_ERR(trans));
4282 return;
4283 }
4284
4285 ret = btrfs_update_inode(trans, root, inode);
4286 if (ret) {
4287 printk_ratelimited(KERN_ERR "btrfs: fail to "
4288 "dirty inode %llu error %d\n",
4289 (unsigned long long)btrfs_ino(inode),
4290 ret);
4291 }
4292 }
4293 btrfs_end_transaction(trans, root);
4294 if (BTRFS_I(inode)->delayed_node)
4295 btrfs_balance_delayed_items(root);
4296 }
4297
4298 /*
4299 * find the highest existing sequence number in a directory
4300 * and then set the in-memory index_cnt variable to reflect
4301 * free sequence numbers
4302 */
4303 static int btrfs_set_inode_index_count(struct inode *inode)
4304 {
4305 struct btrfs_root *root = BTRFS_I(inode)->root;
4306 struct btrfs_key key, found_key;
4307 struct btrfs_path *path;
4308 struct extent_buffer *leaf;
4309 int ret;
4310
4311 key.objectid = btrfs_ino(inode);
4312 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4313 key.offset = (u64)-1;
4314
4315 path = btrfs_alloc_path();
4316 if (!path)
4317 return -ENOMEM;
4318
4319 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4320 if (ret < 0)
4321 goto out;
4322 /* FIXME: we should be able to handle this */
4323 if (ret == 0)
4324 goto out;
4325 ret = 0;
4326
4327 /*
4328 * MAGIC NUMBER EXPLANATION:
4329 * since we search a directory based on f_pos we have to start at 2
4330 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4331 * else has to start at 2
4332 */
4333 if (path->slots[0] == 0) {
4334 BTRFS_I(inode)->index_cnt = 2;
4335 goto out;
4336 }
4337
4338 path->slots[0]--;
4339
4340 leaf = path->nodes[0];
4341 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4342
4343 if (found_key.objectid != btrfs_ino(inode) ||
4344 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4345 BTRFS_I(inode)->index_cnt = 2;
4346 goto out;
4347 }
4348
4349 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4350 out:
4351 btrfs_free_path(path);
4352 return ret;
4353 }
4354
4355 /*
4356 * helper to find a free sequence number in a given directory. This current
4357 * code is very simple, later versions will do smarter things in the btree
4358 */
4359 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4360 {
4361 int ret = 0;
4362
4363 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4364 ret = btrfs_inode_delayed_dir_index_count(dir);
4365 if (ret) {
4366 ret = btrfs_set_inode_index_count(dir);
4367 if (ret)
4368 return ret;
4369 }
4370 }
4371
4372 *index = BTRFS_I(dir)->index_cnt;
4373 BTRFS_I(dir)->index_cnt++;
4374
4375 return ret;
4376 }
4377
4378 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4379 struct btrfs_root *root,
4380 struct inode *dir,
4381 const char *name, int name_len,
4382 u64 ref_objectid, u64 objectid, int mode,
4383 u64 *index)
4384 {
4385 struct inode *inode;
4386 struct btrfs_inode_item *inode_item;
4387 struct btrfs_key *location;
4388 struct btrfs_path *path;
4389 struct btrfs_inode_ref *ref;
4390 struct btrfs_key key[2];
4391 u32 sizes[2];
4392 unsigned long ptr;
4393 int ret;
4394 int owner;
4395
4396 path = btrfs_alloc_path();
4397 if (!path)
4398 return ERR_PTR(-ENOMEM);
4399
4400 inode = new_inode(root->fs_info->sb);
4401 if (!inode) {
4402 btrfs_free_path(path);
4403 return ERR_PTR(-ENOMEM);
4404 }
4405
4406 /*
4407 * we have to initialize this early, so we can reclaim the inode
4408 * number if we fail afterwards in this function.
4409 */
4410 inode->i_ino = objectid;
4411
4412 if (dir) {
4413 trace_btrfs_inode_request(dir);
4414
4415 ret = btrfs_set_inode_index(dir, index);
4416 if (ret) {
4417 btrfs_free_path(path);
4418 iput(inode);
4419 return ERR_PTR(ret);
4420 }
4421 }
4422 /*
4423 * index_cnt is ignored for everything but a dir,
4424 * btrfs_get_inode_index_count has an explanation for the magic
4425 * number
4426 */
4427 BTRFS_I(inode)->index_cnt = 2;
4428 BTRFS_I(inode)->root = root;
4429 BTRFS_I(inode)->generation = trans->transid;
4430 inode->i_generation = BTRFS_I(inode)->generation;
4431 btrfs_set_inode_space_info(root, inode);
4432
4433 if (S_ISDIR(mode))
4434 owner = 0;
4435 else
4436 owner = 1;
4437
4438 key[0].objectid = objectid;
4439 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4440 key[0].offset = 0;
4441
4442 key[1].objectid = objectid;
4443 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4444 key[1].offset = ref_objectid;
4445
4446 sizes[0] = sizeof(struct btrfs_inode_item);
4447 sizes[1] = name_len + sizeof(*ref);
4448
4449 path->leave_spinning = 1;
4450 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4451 if (ret != 0)
4452 goto fail;
4453
4454 inode_init_owner(inode, dir, mode);
4455 inode_set_bytes(inode, 0);
4456 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4457 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4458 struct btrfs_inode_item);
4459 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4460
4461 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4462 struct btrfs_inode_ref);
4463 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4464 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4465 ptr = (unsigned long)(ref + 1);
4466 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4467
4468 btrfs_mark_buffer_dirty(path->nodes[0]);
4469 btrfs_free_path(path);
4470
4471 location = &BTRFS_I(inode)->location;
4472 location->objectid = objectid;
4473 location->offset = 0;
4474 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4475
4476 btrfs_inherit_iflags(inode, dir);
4477
4478 if (S_ISREG(mode)) {
4479 if (btrfs_test_opt(root, NODATASUM))
4480 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4481 if (btrfs_test_opt(root, NODATACOW) ||
4482 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4483 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4484 }
4485
4486 insert_inode_hash(inode);
4487 inode_tree_add(inode);
4488
4489 trace_btrfs_inode_new(inode);
4490 btrfs_set_inode_last_trans(trans, inode);
4491
4492 return inode;
4493 fail:
4494 if (dir)
4495 BTRFS_I(dir)->index_cnt--;
4496 btrfs_free_path(path);
4497 iput(inode);
4498 return ERR_PTR(ret);
4499 }
4500
4501 static inline u8 btrfs_inode_type(struct inode *inode)
4502 {
4503 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4504 }
4505
4506 /*
4507 * utility function to add 'inode' into 'parent_inode' with
4508 * a give name and a given sequence number.
4509 * if 'add_backref' is true, also insert a backref from the
4510 * inode to the parent directory.
4511 */
4512 int btrfs_add_link(struct btrfs_trans_handle *trans,
4513 struct inode *parent_inode, struct inode *inode,
4514 const char *name, int name_len, int add_backref, u64 index)
4515 {
4516 int ret = 0;
4517 struct btrfs_key key;
4518 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4519 u64 ino = btrfs_ino(inode);
4520 u64 parent_ino = btrfs_ino(parent_inode);
4521
4522 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4523 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4524 } else {
4525 key.objectid = ino;
4526 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4527 key.offset = 0;
4528 }
4529
4530 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4531 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4532 key.objectid, root->root_key.objectid,
4533 parent_ino, index, name, name_len);
4534 } else if (add_backref) {
4535 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4536 parent_ino, index);
4537 }
4538
4539 if (ret == 0) {
4540 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4541 parent_inode, &key,
4542 btrfs_inode_type(inode), index);
4543 BUG_ON(ret);
4544
4545 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4546 name_len * 2);
4547 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4548 ret = btrfs_update_inode(trans, root, parent_inode);
4549 }
4550 return ret;
4551 }
4552
4553 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4554 struct inode *dir, struct dentry *dentry,
4555 struct inode *inode, int backref, u64 index)
4556 {
4557 int err = btrfs_add_link(trans, dir, inode,
4558 dentry->d_name.name, dentry->d_name.len,
4559 backref, index);
4560 if (!err) {
4561 d_instantiate(dentry, inode);
4562 return 0;
4563 }
4564 if (err > 0)
4565 err = -EEXIST;
4566 return err;
4567 }
4568
4569 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4570 int mode, dev_t rdev)
4571 {
4572 struct btrfs_trans_handle *trans;
4573 struct btrfs_root *root = BTRFS_I(dir)->root;
4574 struct inode *inode = NULL;
4575 int err;
4576 int drop_inode = 0;
4577 u64 objectid;
4578 unsigned long nr = 0;
4579 u64 index = 0;
4580
4581 if (!new_valid_dev(rdev))
4582 return -EINVAL;
4583
4584 /*
4585 * 2 for inode item and ref
4586 * 2 for dir items
4587 * 1 for xattr if selinux is on
4588 */
4589 trans = btrfs_start_transaction(root, 5);
4590 if (IS_ERR(trans))
4591 return PTR_ERR(trans);
4592
4593 err = btrfs_find_free_ino(root, &objectid);
4594 if (err)
4595 goto out_unlock;
4596
4597 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4598 dentry->d_name.len, btrfs_ino(dir), objectid,
4599 mode, &index);
4600 if (IS_ERR(inode)) {
4601 err = PTR_ERR(inode);
4602 goto out_unlock;
4603 }
4604
4605 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4606 if (err) {
4607 drop_inode = 1;
4608 goto out_unlock;
4609 }
4610
4611 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4612 if (err)
4613 drop_inode = 1;
4614 else {
4615 inode->i_op = &btrfs_special_inode_operations;
4616 init_special_inode(inode, inode->i_mode, rdev);
4617 btrfs_update_inode(trans, root, inode);
4618 }
4619 out_unlock:
4620 nr = trans->blocks_used;
4621 btrfs_end_transaction_throttle(trans, root);
4622 btrfs_btree_balance_dirty(root, nr);
4623 if (drop_inode) {
4624 inode_dec_link_count(inode);
4625 iput(inode);
4626 }
4627 return err;
4628 }
4629
4630 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4631 int mode, struct nameidata *nd)
4632 {
4633 struct btrfs_trans_handle *trans;
4634 struct btrfs_root *root = BTRFS_I(dir)->root;
4635 struct inode *inode = NULL;
4636 int drop_inode = 0;
4637 int err;
4638 unsigned long nr = 0;
4639 u64 objectid;
4640 u64 index = 0;
4641
4642 /*
4643 * 2 for inode item and ref
4644 * 2 for dir items
4645 * 1 for xattr if selinux is on
4646 */
4647 trans = btrfs_start_transaction(root, 5);
4648 if (IS_ERR(trans))
4649 return PTR_ERR(trans);
4650
4651 err = btrfs_find_free_ino(root, &objectid);
4652 if (err)
4653 goto out_unlock;
4654
4655 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4656 dentry->d_name.len, btrfs_ino(dir), objectid,
4657 mode, &index);
4658 if (IS_ERR(inode)) {
4659 err = PTR_ERR(inode);
4660 goto out_unlock;
4661 }
4662
4663 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4664 if (err) {
4665 drop_inode = 1;
4666 goto out_unlock;
4667 }
4668
4669 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4670 if (err)
4671 drop_inode = 1;
4672 else {
4673 inode->i_mapping->a_ops = &btrfs_aops;
4674 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4675 inode->i_fop = &btrfs_file_operations;
4676 inode->i_op = &btrfs_file_inode_operations;
4677 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4678 }
4679 out_unlock:
4680 nr = trans->blocks_used;
4681 btrfs_end_transaction_throttle(trans, root);
4682 if (drop_inode) {
4683 inode_dec_link_count(inode);
4684 iput(inode);
4685 }
4686 btrfs_btree_balance_dirty(root, nr);
4687 return err;
4688 }
4689
4690 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4691 struct dentry *dentry)
4692 {
4693 struct btrfs_trans_handle *trans;
4694 struct btrfs_root *root = BTRFS_I(dir)->root;
4695 struct inode *inode = old_dentry->d_inode;
4696 u64 index;
4697 unsigned long nr = 0;
4698 int err;
4699 int drop_inode = 0;
4700
4701 /* do not allow sys_link's with other subvols of the same device */
4702 if (root->objectid != BTRFS_I(inode)->root->objectid)
4703 return -EXDEV;
4704
4705 if (inode->i_nlink == ~0U)
4706 return -EMLINK;
4707
4708 err = btrfs_set_inode_index(dir, &index);
4709 if (err)
4710 goto fail;
4711
4712 /*
4713 * 2 items for inode and inode ref
4714 * 2 items for dir items
4715 * 1 item for parent inode
4716 */
4717 trans = btrfs_start_transaction(root, 5);
4718 if (IS_ERR(trans)) {
4719 err = PTR_ERR(trans);
4720 goto fail;
4721 }
4722
4723 btrfs_inc_nlink(inode);
4724 inode->i_ctime = CURRENT_TIME;
4725 ihold(inode);
4726
4727 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4728
4729 if (err) {
4730 drop_inode = 1;
4731 } else {
4732 struct dentry *parent = dentry->d_parent;
4733 err = btrfs_update_inode(trans, root, inode);
4734 BUG_ON(err);
4735 btrfs_log_new_name(trans, inode, NULL, parent);
4736 }
4737
4738 nr = trans->blocks_used;
4739 btrfs_end_transaction_throttle(trans, root);
4740 fail:
4741 if (drop_inode) {
4742 inode_dec_link_count(inode);
4743 iput(inode);
4744 }
4745 btrfs_btree_balance_dirty(root, nr);
4746 return err;
4747 }
4748
4749 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4750 {
4751 struct inode *inode = NULL;
4752 struct btrfs_trans_handle *trans;
4753 struct btrfs_root *root = BTRFS_I(dir)->root;
4754 int err = 0;
4755 int drop_on_err = 0;
4756 u64 objectid = 0;
4757 u64 index = 0;
4758 unsigned long nr = 1;
4759
4760 /*
4761 * 2 items for inode and ref
4762 * 2 items for dir items
4763 * 1 for xattr if selinux is on
4764 */
4765 trans = btrfs_start_transaction(root, 5);
4766 if (IS_ERR(trans))
4767 return PTR_ERR(trans);
4768
4769 err = btrfs_find_free_ino(root, &objectid);
4770 if (err)
4771 goto out_fail;
4772
4773 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4774 dentry->d_name.len, btrfs_ino(dir), objectid,
4775 S_IFDIR | mode, &index);
4776 if (IS_ERR(inode)) {
4777 err = PTR_ERR(inode);
4778 goto out_fail;
4779 }
4780
4781 drop_on_err = 1;
4782
4783 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4784 if (err)
4785 goto out_fail;
4786
4787 inode->i_op = &btrfs_dir_inode_operations;
4788 inode->i_fop = &btrfs_dir_file_operations;
4789
4790 btrfs_i_size_write(inode, 0);
4791 err = btrfs_update_inode(trans, root, inode);
4792 if (err)
4793 goto out_fail;
4794
4795 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4796 dentry->d_name.len, 0, index);
4797 if (err)
4798 goto out_fail;
4799
4800 d_instantiate(dentry, inode);
4801 drop_on_err = 0;
4802
4803 out_fail:
4804 nr = trans->blocks_used;
4805 btrfs_end_transaction_throttle(trans, root);
4806 if (drop_on_err)
4807 iput(inode);
4808 btrfs_btree_balance_dirty(root, nr);
4809 return err;
4810 }
4811
4812 /* helper for btfs_get_extent. Given an existing extent in the tree,
4813 * and an extent that you want to insert, deal with overlap and insert
4814 * the new extent into the tree.
4815 */
4816 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4817 struct extent_map *existing,
4818 struct extent_map *em,
4819 u64 map_start, u64 map_len)
4820 {
4821 u64 start_diff;
4822
4823 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4824 start_diff = map_start - em->start;
4825 em->start = map_start;
4826 em->len = map_len;
4827 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4828 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4829 em->block_start += start_diff;
4830 em->block_len -= start_diff;
4831 }
4832 return add_extent_mapping(em_tree, em);
4833 }
4834
4835 static noinline int uncompress_inline(struct btrfs_path *path,
4836 struct inode *inode, struct page *page,
4837 size_t pg_offset, u64 extent_offset,
4838 struct btrfs_file_extent_item *item)
4839 {
4840 int ret;
4841 struct extent_buffer *leaf = path->nodes[0];
4842 char *tmp;
4843 size_t max_size;
4844 unsigned long inline_size;
4845 unsigned long ptr;
4846 int compress_type;
4847
4848 WARN_ON(pg_offset != 0);
4849 compress_type = btrfs_file_extent_compression(leaf, item);
4850 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4851 inline_size = btrfs_file_extent_inline_item_len(leaf,
4852 btrfs_item_nr(leaf, path->slots[0]));
4853 tmp = kmalloc(inline_size, GFP_NOFS);
4854 if (!tmp)
4855 return -ENOMEM;
4856 ptr = btrfs_file_extent_inline_start(item);
4857
4858 read_extent_buffer(leaf, tmp, ptr, inline_size);
4859
4860 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4861 ret = btrfs_decompress(compress_type, tmp, page,
4862 extent_offset, inline_size, max_size);
4863 if (ret) {
4864 char *kaddr = kmap_atomic(page, KM_USER0);
4865 unsigned long copy_size = min_t(u64,
4866 PAGE_CACHE_SIZE - pg_offset,
4867 max_size - extent_offset);
4868 memset(kaddr + pg_offset, 0, copy_size);
4869 kunmap_atomic(kaddr, KM_USER0);
4870 }
4871 kfree(tmp);
4872 return 0;
4873 }
4874
4875 /*
4876 * a bit scary, this does extent mapping from logical file offset to the disk.
4877 * the ugly parts come from merging extents from the disk with the in-ram
4878 * representation. This gets more complex because of the data=ordered code,
4879 * where the in-ram extents might be locked pending data=ordered completion.
4880 *
4881 * This also copies inline extents directly into the page.
4882 */
4883
4884 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4885 size_t pg_offset, u64 start, u64 len,
4886 int create)
4887 {
4888 int ret;
4889 int err = 0;
4890 u64 bytenr;
4891 u64 extent_start = 0;
4892 u64 extent_end = 0;
4893 u64 objectid = btrfs_ino(inode);
4894 u32 found_type;
4895 struct btrfs_path *path = NULL;
4896 struct btrfs_root *root = BTRFS_I(inode)->root;
4897 struct btrfs_file_extent_item *item;
4898 struct extent_buffer *leaf;
4899 struct btrfs_key found_key;
4900 struct extent_map *em = NULL;
4901 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4902 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4903 struct btrfs_trans_handle *trans = NULL;
4904 int compress_type;
4905
4906 again:
4907 read_lock(&em_tree->lock);
4908 em = lookup_extent_mapping(em_tree, start, len);
4909 if (em)
4910 em->bdev = root->fs_info->fs_devices->latest_bdev;
4911 read_unlock(&em_tree->lock);
4912
4913 if (em) {
4914 if (em->start > start || em->start + em->len <= start)
4915 free_extent_map(em);
4916 else if (em->block_start == EXTENT_MAP_INLINE && page)
4917 free_extent_map(em);
4918 else
4919 goto out;
4920 }
4921 em = alloc_extent_map();
4922 if (!em) {
4923 err = -ENOMEM;
4924 goto out;
4925 }
4926 em->bdev = root->fs_info->fs_devices->latest_bdev;
4927 em->start = EXTENT_MAP_HOLE;
4928 em->orig_start = EXTENT_MAP_HOLE;
4929 em->len = (u64)-1;
4930 em->block_len = (u64)-1;
4931
4932 if (!path) {
4933 path = btrfs_alloc_path();
4934 if (!path) {
4935 err = -ENOMEM;
4936 goto out;
4937 }
4938 /*
4939 * Chances are we'll be called again, so go ahead and do
4940 * readahead
4941 */
4942 path->reada = 1;
4943 }
4944
4945 ret = btrfs_lookup_file_extent(trans, root, path,
4946 objectid, start, trans != NULL);
4947 if (ret < 0) {
4948 err = ret;
4949 goto out;
4950 }
4951
4952 if (ret != 0) {
4953 if (path->slots[0] == 0)
4954 goto not_found;
4955 path->slots[0]--;
4956 }
4957
4958 leaf = path->nodes[0];
4959 item = btrfs_item_ptr(leaf, path->slots[0],
4960 struct btrfs_file_extent_item);
4961 /* are we inside the extent that was found? */
4962 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4963 found_type = btrfs_key_type(&found_key);
4964 if (found_key.objectid != objectid ||
4965 found_type != BTRFS_EXTENT_DATA_KEY) {
4966 goto not_found;
4967 }
4968
4969 found_type = btrfs_file_extent_type(leaf, item);
4970 extent_start = found_key.offset;
4971 compress_type = btrfs_file_extent_compression(leaf, item);
4972 if (found_type == BTRFS_FILE_EXTENT_REG ||
4973 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4974 extent_end = extent_start +
4975 btrfs_file_extent_num_bytes(leaf, item);
4976 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4977 size_t size;
4978 size = btrfs_file_extent_inline_len(leaf, item);
4979 extent_end = (extent_start + size + root->sectorsize - 1) &
4980 ~((u64)root->sectorsize - 1);
4981 }
4982
4983 if (start >= extent_end) {
4984 path->slots[0]++;
4985 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4986 ret = btrfs_next_leaf(root, path);
4987 if (ret < 0) {
4988 err = ret;
4989 goto out;
4990 }
4991 if (ret > 0)
4992 goto not_found;
4993 leaf = path->nodes[0];
4994 }
4995 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4996 if (found_key.objectid != objectid ||
4997 found_key.type != BTRFS_EXTENT_DATA_KEY)
4998 goto not_found;
4999 if (start + len <= found_key.offset)
5000 goto not_found;
5001 em->start = start;
5002 em->len = found_key.offset - start;
5003 goto not_found_em;
5004 }
5005
5006 if (found_type == BTRFS_FILE_EXTENT_REG ||
5007 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5008 em->start = extent_start;
5009 em->len = extent_end - extent_start;
5010 em->orig_start = extent_start -
5011 btrfs_file_extent_offset(leaf, item);
5012 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5013 if (bytenr == 0) {
5014 em->block_start = EXTENT_MAP_HOLE;
5015 goto insert;
5016 }
5017 if (compress_type != BTRFS_COMPRESS_NONE) {
5018 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5019 em->compress_type = compress_type;
5020 em->block_start = bytenr;
5021 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5022 item);
5023 } else {
5024 bytenr += btrfs_file_extent_offset(leaf, item);
5025 em->block_start = bytenr;
5026 em->block_len = em->len;
5027 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5028 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5029 }
5030 goto insert;
5031 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5032 unsigned long ptr;
5033 char *map;
5034 size_t size;
5035 size_t extent_offset;
5036 size_t copy_size;
5037
5038 em->block_start = EXTENT_MAP_INLINE;
5039 if (!page || create) {
5040 em->start = extent_start;
5041 em->len = extent_end - extent_start;
5042 goto out;
5043 }
5044
5045 size = btrfs_file_extent_inline_len(leaf, item);
5046 extent_offset = page_offset(page) + pg_offset - extent_start;
5047 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5048 size - extent_offset);
5049 em->start = extent_start + extent_offset;
5050 em->len = (copy_size + root->sectorsize - 1) &
5051 ~((u64)root->sectorsize - 1);
5052 em->orig_start = EXTENT_MAP_INLINE;
5053 if (compress_type) {
5054 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5055 em->compress_type = compress_type;
5056 }
5057 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5058 if (create == 0 && !PageUptodate(page)) {
5059 if (btrfs_file_extent_compression(leaf, item) !=
5060 BTRFS_COMPRESS_NONE) {
5061 ret = uncompress_inline(path, inode, page,
5062 pg_offset,
5063 extent_offset, item);
5064 BUG_ON(ret);
5065 } else {
5066 map = kmap(page);
5067 read_extent_buffer(leaf, map + pg_offset, ptr,
5068 copy_size);
5069 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5070 memset(map + pg_offset + copy_size, 0,
5071 PAGE_CACHE_SIZE - pg_offset -
5072 copy_size);
5073 }
5074 kunmap(page);
5075 }
5076 flush_dcache_page(page);
5077 } else if (create && PageUptodate(page)) {
5078 WARN_ON(1);
5079 if (!trans) {
5080 kunmap(page);
5081 free_extent_map(em);
5082 em = NULL;
5083
5084 btrfs_release_path(path);
5085 trans = btrfs_join_transaction(root);
5086
5087 if (IS_ERR(trans))
5088 return ERR_CAST(trans);
5089 goto again;
5090 }
5091 map = kmap(page);
5092 write_extent_buffer(leaf, map + pg_offset, ptr,
5093 copy_size);
5094 kunmap(page);
5095 btrfs_mark_buffer_dirty(leaf);
5096 }
5097 set_extent_uptodate(io_tree, em->start,
5098 extent_map_end(em) - 1, NULL, GFP_NOFS);
5099 goto insert;
5100 } else {
5101 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5102 WARN_ON(1);
5103 }
5104 not_found:
5105 em->start = start;
5106 em->len = len;
5107 not_found_em:
5108 em->block_start = EXTENT_MAP_HOLE;
5109 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5110 insert:
5111 btrfs_release_path(path);
5112 if (em->start > start || extent_map_end(em) <= start) {
5113 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5114 "[%llu %llu]\n", (unsigned long long)em->start,
5115 (unsigned long long)em->len,
5116 (unsigned long long)start,
5117 (unsigned long long)len);
5118 err = -EIO;
5119 goto out;
5120 }
5121
5122 err = 0;
5123 write_lock(&em_tree->lock);
5124 ret = add_extent_mapping(em_tree, em);
5125 /* it is possible that someone inserted the extent into the tree
5126 * while we had the lock dropped. It is also possible that
5127 * an overlapping map exists in the tree
5128 */
5129 if (ret == -EEXIST) {
5130 struct extent_map *existing;
5131
5132 ret = 0;
5133
5134 existing = lookup_extent_mapping(em_tree, start, len);
5135 if (existing && (existing->start > start ||
5136 existing->start + existing->len <= start)) {
5137 free_extent_map(existing);
5138 existing = NULL;
5139 }
5140 if (!existing) {
5141 existing = lookup_extent_mapping(em_tree, em->start,
5142 em->len);
5143 if (existing) {
5144 err = merge_extent_mapping(em_tree, existing,
5145 em, start,
5146 root->sectorsize);
5147 free_extent_map(existing);
5148 if (err) {
5149 free_extent_map(em);
5150 em = NULL;
5151 }
5152 } else {
5153 err = -EIO;
5154 free_extent_map(em);
5155 em = NULL;
5156 }
5157 } else {
5158 free_extent_map(em);
5159 em = existing;
5160 err = 0;
5161 }
5162 }
5163 write_unlock(&em_tree->lock);
5164 out:
5165
5166 trace_btrfs_get_extent(root, em);
5167
5168 if (path)
5169 btrfs_free_path(path);
5170 if (trans) {
5171 ret = btrfs_end_transaction(trans, root);
5172 if (!err)
5173 err = ret;
5174 }
5175 if (err) {
5176 free_extent_map(em);
5177 return ERR_PTR(err);
5178 }
5179 return em;
5180 }
5181
5182 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5183 size_t pg_offset, u64 start, u64 len,
5184 int create)
5185 {
5186 struct extent_map *em;
5187 struct extent_map *hole_em = NULL;
5188 u64 range_start = start;
5189 u64 end;
5190 u64 found;
5191 u64 found_end;
5192 int err = 0;
5193
5194 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5195 if (IS_ERR(em))
5196 return em;
5197 if (em) {
5198 /*
5199 * if our em maps to a hole, there might
5200 * actually be delalloc bytes behind it
5201 */
5202 if (em->block_start != EXTENT_MAP_HOLE)
5203 return em;
5204 else
5205 hole_em = em;
5206 }
5207
5208 /* check to see if we've wrapped (len == -1 or similar) */
5209 end = start + len;
5210 if (end < start)
5211 end = (u64)-1;
5212 else
5213 end -= 1;
5214
5215 em = NULL;
5216
5217 /* ok, we didn't find anything, lets look for delalloc */
5218 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5219 end, len, EXTENT_DELALLOC, 1);
5220 found_end = range_start + found;
5221 if (found_end < range_start)
5222 found_end = (u64)-1;
5223
5224 /*
5225 * we didn't find anything useful, return
5226 * the original results from get_extent()
5227 */
5228 if (range_start > end || found_end <= start) {
5229 em = hole_em;
5230 hole_em = NULL;
5231 goto out;
5232 }
5233
5234 /* adjust the range_start to make sure it doesn't
5235 * go backwards from the start they passed in
5236 */
5237 range_start = max(start,range_start);
5238 found = found_end - range_start;
5239
5240 if (found > 0) {
5241 u64 hole_start = start;
5242 u64 hole_len = len;
5243
5244 em = alloc_extent_map();
5245 if (!em) {
5246 err = -ENOMEM;
5247 goto out;
5248 }
5249 /*
5250 * when btrfs_get_extent can't find anything it
5251 * returns one huge hole
5252 *
5253 * make sure what it found really fits our range, and
5254 * adjust to make sure it is based on the start from
5255 * the caller
5256 */
5257 if (hole_em) {
5258 u64 calc_end = extent_map_end(hole_em);
5259
5260 if (calc_end <= start || (hole_em->start > end)) {
5261 free_extent_map(hole_em);
5262 hole_em = NULL;
5263 } else {
5264 hole_start = max(hole_em->start, start);
5265 hole_len = calc_end - hole_start;
5266 }
5267 }
5268 em->bdev = NULL;
5269 if (hole_em && range_start > hole_start) {
5270 /* our hole starts before our delalloc, so we
5271 * have to return just the parts of the hole
5272 * that go until the delalloc starts
5273 */
5274 em->len = min(hole_len,
5275 range_start - hole_start);
5276 em->start = hole_start;
5277 em->orig_start = hole_start;
5278 /*
5279 * don't adjust block start at all,
5280 * it is fixed at EXTENT_MAP_HOLE
5281 */
5282 em->block_start = hole_em->block_start;
5283 em->block_len = hole_len;
5284 } else {
5285 em->start = range_start;
5286 em->len = found;
5287 em->orig_start = range_start;
5288 em->block_start = EXTENT_MAP_DELALLOC;
5289 em->block_len = found;
5290 }
5291 } else if (hole_em) {
5292 return hole_em;
5293 }
5294 out:
5295
5296 free_extent_map(hole_em);
5297 if (err) {
5298 free_extent_map(em);
5299 return ERR_PTR(err);
5300 }
5301 return em;
5302 }
5303
5304 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5305 struct extent_map *em,
5306 u64 start, u64 len)
5307 {
5308 struct btrfs_root *root = BTRFS_I(inode)->root;
5309 struct btrfs_trans_handle *trans;
5310 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5311 struct btrfs_key ins;
5312 u64 alloc_hint;
5313 int ret;
5314 bool insert = false;
5315
5316 /*
5317 * Ok if the extent map we looked up is a hole and is for the exact
5318 * range we want, there is no reason to allocate a new one, however if
5319 * it is not right then we need to free this one and drop the cache for
5320 * our range.
5321 */
5322 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5323 em->len != len) {
5324 free_extent_map(em);
5325 em = NULL;
5326 insert = true;
5327 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5328 }
5329
5330 trans = btrfs_join_transaction(root);
5331 if (IS_ERR(trans))
5332 return ERR_CAST(trans);
5333
5334 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5335 btrfs_add_inode_defrag(trans, inode);
5336
5337 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5338
5339 alloc_hint = get_extent_allocation_hint(inode, start, len);
5340 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5341 alloc_hint, (u64)-1, &ins, 1);
5342 if (ret) {
5343 em = ERR_PTR(ret);
5344 goto out;
5345 }
5346
5347 if (!em) {
5348 em = alloc_extent_map();
5349 if (!em) {
5350 em = ERR_PTR(-ENOMEM);
5351 goto out;
5352 }
5353 }
5354
5355 em->start = start;
5356 em->orig_start = em->start;
5357 em->len = ins.offset;
5358
5359 em->block_start = ins.objectid;
5360 em->block_len = ins.offset;
5361 em->bdev = root->fs_info->fs_devices->latest_bdev;
5362
5363 /*
5364 * We need to do this because if we're using the original em we searched
5365 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5366 */
5367 em->flags = 0;
5368 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5369
5370 while (insert) {
5371 write_lock(&em_tree->lock);
5372 ret = add_extent_mapping(em_tree, em);
5373 write_unlock(&em_tree->lock);
5374 if (ret != -EEXIST)
5375 break;
5376 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5377 }
5378
5379 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5380 ins.offset, ins.offset, 0);
5381 if (ret) {
5382 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5383 em = ERR_PTR(ret);
5384 }
5385 out:
5386 btrfs_end_transaction(trans, root);
5387 return em;
5388 }
5389
5390 /*
5391 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5392 * block must be cow'd
5393 */
5394 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5395 struct inode *inode, u64 offset, u64 len)
5396 {
5397 struct btrfs_path *path;
5398 int ret;
5399 struct extent_buffer *leaf;
5400 struct btrfs_root *root = BTRFS_I(inode)->root;
5401 struct btrfs_file_extent_item *fi;
5402 struct btrfs_key key;
5403 u64 disk_bytenr;
5404 u64 backref_offset;
5405 u64 extent_end;
5406 u64 num_bytes;
5407 int slot;
5408 int found_type;
5409
5410 path = btrfs_alloc_path();
5411 if (!path)
5412 return -ENOMEM;
5413
5414 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5415 offset, 0);
5416 if (ret < 0)
5417 goto out;
5418
5419 slot = path->slots[0];
5420 if (ret == 1) {
5421 if (slot == 0) {
5422 /* can't find the item, must cow */
5423 ret = 0;
5424 goto out;
5425 }
5426 slot--;
5427 }
5428 ret = 0;
5429 leaf = path->nodes[0];
5430 btrfs_item_key_to_cpu(leaf, &key, slot);
5431 if (key.objectid != btrfs_ino(inode) ||
5432 key.type != BTRFS_EXTENT_DATA_KEY) {
5433 /* not our file or wrong item type, must cow */
5434 goto out;
5435 }
5436
5437 if (key.offset > offset) {
5438 /* Wrong offset, must cow */
5439 goto out;
5440 }
5441
5442 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5443 found_type = btrfs_file_extent_type(leaf, fi);
5444 if (found_type != BTRFS_FILE_EXTENT_REG &&
5445 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5446 /* not a regular extent, must cow */
5447 goto out;
5448 }
5449 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5450 backref_offset = btrfs_file_extent_offset(leaf, fi);
5451
5452 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5453 if (extent_end < offset + len) {
5454 /* extent doesn't include our full range, must cow */
5455 goto out;
5456 }
5457
5458 if (btrfs_extent_readonly(root, disk_bytenr))
5459 goto out;
5460
5461 /*
5462 * look for other files referencing this extent, if we
5463 * find any we must cow
5464 */
5465 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5466 key.offset - backref_offset, disk_bytenr))
5467 goto out;
5468
5469 /*
5470 * adjust disk_bytenr and num_bytes to cover just the bytes
5471 * in this extent we are about to write. If there
5472 * are any csums in that range we have to cow in order
5473 * to keep the csums correct
5474 */
5475 disk_bytenr += backref_offset;
5476 disk_bytenr += offset - key.offset;
5477 num_bytes = min(offset + len, extent_end) - offset;
5478 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5479 goto out;
5480 /*
5481 * all of the above have passed, it is safe to overwrite this extent
5482 * without cow
5483 */
5484 ret = 1;
5485 out:
5486 btrfs_free_path(path);
5487 return ret;
5488 }
5489
5490 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5491 struct buffer_head *bh_result, int create)
5492 {
5493 struct extent_map *em;
5494 struct btrfs_root *root = BTRFS_I(inode)->root;
5495 u64 start = iblock << inode->i_blkbits;
5496 u64 len = bh_result->b_size;
5497 struct btrfs_trans_handle *trans;
5498
5499 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5500 if (IS_ERR(em))
5501 return PTR_ERR(em);
5502
5503 /*
5504 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5505 * io. INLINE is special, and we could probably kludge it in here, but
5506 * it's still buffered so for safety lets just fall back to the generic
5507 * buffered path.
5508 *
5509 * For COMPRESSED we _have_ to read the entire extent in so we can
5510 * decompress it, so there will be buffering required no matter what we
5511 * do, so go ahead and fallback to buffered.
5512 *
5513 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5514 * to buffered IO. Don't blame me, this is the price we pay for using
5515 * the generic code.
5516 */
5517 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5518 em->block_start == EXTENT_MAP_INLINE) {
5519 free_extent_map(em);
5520 return -ENOTBLK;
5521 }
5522
5523 /* Just a good old fashioned hole, return */
5524 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5525 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5526 free_extent_map(em);
5527 /* DIO will do one hole at a time, so just unlock a sector */
5528 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5529 start + root->sectorsize - 1, GFP_NOFS);
5530 return 0;
5531 }
5532
5533 /*
5534 * We don't allocate a new extent in the following cases
5535 *
5536 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5537 * existing extent.
5538 * 2) The extent is marked as PREALLOC. We're good to go here and can
5539 * just use the extent.
5540 *
5541 */
5542 if (!create) {
5543 len = em->len - (start - em->start);
5544 goto map;
5545 }
5546
5547 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5548 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5549 em->block_start != EXTENT_MAP_HOLE)) {
5550 int type;
5551 int ret;
5552 u64 block_start;
5553
5554 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5555 type = BTRFS_ORDERED_PREALLOC;
5556 else
5557 type = BTRFS_ORDERED_NOCOW;
5558 len = min(len, em->len - (start - em->start));
5559 block_start = em->block_start + (start - em->start);
5560
5561 /*
5562 * we're not going to log anything, but we do need
5563 * to make sure the current transaction stays open
5564 * while we look for nocow cross refs
5565 */
5566 trans = btrfs_join_transaction(root);
5567 if (IS_ERR(trans))
5568 goto must_cow;
5569
5570 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5571 ret = btrfs_add_ordered_extent_dio(inode, start,
5572 block_start, len, len, type);
5573 btrfs_end_transaction(trans, root);
5574 if (ret) {
5575 free_extent_map(em);
5576 return ret;
5577 }
5578 goto unlock;
5579 }
5580 btrfs_end_transaction(trans, root);
5581 }
5582 must_cow:
5583 /*
5584 * this will cow the extent, reset the len in case we changed
5585 * it above
5586 */
5587 len = bh_result->b_size;
5588 em = btrfs_new_extent_direct(inode, em, start, len);
5589 if (IS_ERR(em))
5590 return PTR_ERR(em);
5591 len = min(len, em->len - (start - em->start));
5592 unlock:
5593 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5594 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5595 0, NULL, GFP_NOFS);
5596 map:
5597 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5598 inode->i_blkbits;
5599 bh_result->b_size = len;
5600 bh_result->b_bdev = em->bdev;
5601 set_buffer_mapped(bh_result);
5602 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5603 set_buffer_new(bh_result);
5604
5605 free_extent_map(em);
5606
5607 return 0;
5608 }
5609
5610 struct btrfs_dio_private {
5611 struct inode *inode;
5612 u64 logical_offset;
5613 u64 disk_bytenr;
5614 u64 bytes;
5615 u32 *csums;
5616 void *private;
5617
5618 /* number of bios pending for this dio */
5619 atomic_t pending_bios;
5620
5621 /* IO errors */
5622 int errors;
5623
5624 struct bio *orig_bio;
5625 };
5626
5627 static void btrfs_endio_direct_read(struct bio *bio, int err)
5628 {
5629 struct btrfs_dio_private *dip = bio->bi_private;
5630 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5631 struct bio_vec *bvec = bio->bi_io_vec;
5632 struct inode *inode = dip->inode;
5633 struct btrfs_root *root = BTRFS_I(inode)->root;
5634 u64 start;
5635 u32 *private = dip->csums;
5636
5637 start = dip->logical_offset;
5638 do {
5639 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5640 struct page *page = bvec->bv_page;
5641 char *kaddr;
5642 u32 csum = ~(u32)0;
5643 unsigned long flags;
5644
5645 local_irq_save(flags);
5646 kaddr = kmap_atomic(page, KM_IRQ0);
5647 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5648 csum, bvec->bv_len);
5649 btrfs_csum_final(csum, (char *)&csum);
5650 kunmap_atomic(kaddr, KM_IRQ0);
5651 local_irq_restore(flags);
5652
5653 flush_dcache_page(bvec->bv_page);
5654 if (csum != *private) {
5655 printk(KERN_ERR "btrfs csum failed ino %llu off"
5656 " %llu csum %u private %u\n",
5657 (unsigned long long)btrfs_ino(inode),
5658 (unsigned long long)start,
5659 csum, *private);
5660 err = -EIO;
5661 }
5662 }
5663
5664 start += bvec->bv_len;
5665 private++;
5666 bvec++;
5667 } while (bvec <= bvec_end);
5668
5669 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5670 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5671 bio->bi_private = dip->private;
5672
5673 kfree(dip->csums);
5674 kfree(dip);
5675
5676 /* If we had a csum failure make sure to clear the uptodate flag */
5677 if (err)
5678 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5679 dio_end_io(bio, err);
5680 }
5681
5682 static void btrfs_endio_direct_write(struct bio *bio, int err)
5683 {
5684 struct btrfs_dio_private *dip = bio->bi_private;
5685 struct inode *inode = dip->inode;
5686 struct btrfs_root *root = BTRFS_I(inode)->root;
5687 struct btrfs_trans_handle *trans;
5688 struct btrfs_ordered_extent *ordered = NULL;
5689 struct extent_state *cached_state = NULL;
5690 u64 ordered_offset = dip->logical_offset;
5691 u64 ordered_bytes = dip->bytes;
5692 int ret;
5693
5694 if (err)
5695 goto out_done;
5696 again:
5697 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5698 &ordered_offset,
5699 ordered_bytes);
5700 if (!ret)
5701 goto out_test;
5702
5703 BUG_ON(!ordered);
5704
5705 trans = btrfs_join_transaction(root);
5706 if (IS_ERR(trans)) {
5707 err = -ENOMEM;
5708 goto out;
5709 }
5710 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5711
5712 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5713 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5714 if (!ret)
5715 ret = btrfs_update_inode(trans, root, inode);
5716 err = ret;
5717 goto out;
5718 }
5719
5720 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5721 ordered->file_offset + ordered->len - 1, 0,
5722 &cached_state, GFP_NOFS);
5723
5724 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5725 ret = btrfs_mark_extent_written(trans, inode,
5726 ordered->file_offset,
5727 ordered->file_offset +
5728 ordered->len);
5729 if (ret) {
5730 err = ret;
5731 goto out_unlock;
5732 }
5733 } else {
5734 ret = insert_reserved_file_extent(trans, inode,
5735 ordered->file_offset,
5736 ordered->start,
5737 ordered->disk_len,
5738 ordered->len,
5739 ordered->len,
5740 0, 0, 0,
5741 BTRFS_FILE_EXTENT_REG);
5742 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5743 ordered->file_offset, ordered->len);
5744 if (ret) {
5745 err = ret;
5746 WARN_ON(1);
5747 goto out_unlock;
5748 }
5749 }
5750
5751 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5752 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5753 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5754 btrfs_update_inode(trans, root, inode);
5755 ret = 0;
5756 out_unlock:
5757 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5758 ordered->file_offset + ordered->len - 1,
5759 &cached_state, GFP_NOFS);
5760 out:
5761 btrfs_delalloc_release_metadata(inode, ordered->len);
5762 btrfs_end_transaction(trans, root);
5763 ordered_offset = ordered->file_offset + ordered->len;
5764 btrfs_put_ordered_extent(ordered);
5765 btrfs_put_ordered_extent(ordered);
5766
5767 out_test:
5768 /*
5769 * our bio might span multiple ordered extents. If we haven't
5770 * completed the accounting for the whole dio, go back and try again
5771 */
5772 if (ordered_offset < dip->logical_offset + dip->bytes) {
5773 ordered_bytes = dip->logical_offset + dip->bytes -
5774 ordered_offset;
5775 goto again;
5776 }
5777 out_done:
5778 bio->bi_private = dip->private;
5779
5780 kfree(dip->csums);
5781 kfree(dip);
5782
5783 /* If we had an error make sure to clear the uptodate flag */
5784 if (err)
5785 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5786 dio_end_io(bio, err);
5787 }
5788
5789 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5790 struct bio *bio, int mirror_num,
5791 unsigned long bio_flags, u64 offset)
5792 {
5793 int ret;
5794 struct btrfs_root *root = BTRFS_I(inode)->root;
5795 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5796 BUG_ON(ret);
5797 return 0;
5798 }
5799
5800 static void btrfs_end_dio_bio(struct bio *bio, int err)
5801 {
5802 struct btrfs_dio_private *dip = bio->bi_private;
5803
5804 if (err) {
5805 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5806 "sector %#Lx len %u err no %d\n",
5807 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5808 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5809 dip->errors = 1;
5810
5811 /*
5812 * before atomic variable goto zero, we must make sure
5813 * dip->errors is perceived to be set.
5814 */
5815 smp_mb__before_atomic_dec();
5816 }
5817
5818 /* if there are more bios still pending for this dio, just exit */
5819 if (!atomic_dec_and_test(&dip->pending_bios))
5820 goto out;
5821
5822 if (dip->errors)
5823 bio_io_error(dip->orig_bio);
5824 else {
5825 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5826 bio_endio(dip->orig_bio, 0);
5827 }
5828 out:
5829 bio_put(bio);
5830 }
5831
5832 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5833 u64 first_sector, gfp_t gfp_flags)
5834 {
5835 int nr_vecs = bio_get_nr_vecs(bdev);
5836 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5837 }
5838
5839 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5840 int rw, u64 file_offset, int skip_sum,
5841 u32 *csums, int async_submit)
5842 {
5843 int write = rw & REQ_WRITE;
5844 struct btrfs_root *root = BTRFS_I(inode)->root;
5845 int ret;
5846
5847 bio_get(bio);
5848 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5849 if (ret)
5850 goto err;
5851
5852 if (skip_sum)
5853 goto map;
5854
5855 if (write && async_submit) {
5856 ret = btrfs_wq_submit_bio(root->fs_info,
5857 inode, rw, bio, 0, 0,
5858 file_offset,
5859 __btrfs_submit_bio_start_direct_io,
5860 __btrfs_submit_bio_done);
5861 goto err;
5862 } else if (write) {
5863 /*
5864 * If we aren't doing async submit, calculate the csum of the
5865 * bio now.
5866 */
5867 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5868 if (ret)
5869 goto err;
5870 } else if (!skip_sum) {
5871 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5872 file_offset, csums);
5873 if (ret)
5874 goto err;
5875 }
5876
5877 map:
5878 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5879 err:
5880 bio_put(bio);
5881 return ret;
5882 }
5883
5884 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5885 int skip_sum)
5886 {
5887 struct inode *inode = dip->inode;
5888 struct btrfs_root *root = BTRFS_I(inode)->root;
5889 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5890 struct bio *bio;
5891 struct bio *orig_bio = dip->orig_bio;
5892 struct bio_vec *bvec = orig_bio->bi_io_vec;
5893 u64 start_sector = orig_bio->bi_sector;
5894 u64 file_offset = dip->logical_offset;
5895 u64 submit_len = 0;
5896 u64 map_length;
5897 int nr_pages = 0;
5898 u32 *csums = dip->csums;
5899 int ret = 0;
5900 int async_submit = 0;
5901 int write = rw & REQ_WRITE;
5902
5903 map_length = orig_bio->bi_size;
5904 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5905 &map_length, NULL, 0);
5906 if (ret) {
5907 bio_put(orig_bio);
5908 return -EIO;
5909 }
5910
5911 if (map_length >= orig_bio->bi_size) {
5912 bio = orig_bio;
5913 goto submit;
5914 }
5915
5916 async_submit = 1;
5917 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5918 if (!bio)
5919 return -ENOMEM;
5920 bio->bi_private = dip;
5921 bio->bi_end_io = btrfs_end_dio_bio;
5922 atomic_inc(&dip->pending_bios);
5923
5924 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5925 if (unlikely(map_length < submit_len + bvec->bv_len ||
5926 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5927 bvec->bv_offset) < bvec->bv_len)) {
5928 /*
5929 * inc the count before we submit the bio so
5930 * we know the end IO handler won't happen before
5931 * we inc the count. Otherwise, the dip might get freed
5932 * before we're done setting it up
5933 */
5934 atomic_inc(&dip->pending_bios);
5935 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5936 file_offset, skip_sum,
5937 csums, async_submit);
5938 if (ret) {
5939 bio_put(bio);
5940 atomic_dec(&dip->pending_bios);
5941 goto out_err;
5942 }
5943
5944 /* Write's use the ordered csums */
5945 if (!write && !skip_sum)
5946 csums = csums + nr_pages;
5947 start_sector += submit_len >> 9;
5948 file_offset += submit_len;
5949
5950 submit_len = 0;
5951 nr_pages = 0;
5952
5953 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5954 start_sector, GFP_NOFS);
5955 if (!bio)
5956 goto out_err;
5957 bio->bi_private = dip;
5958 bio->bi_end_io = btrfs_end_dio_bio;
5959
5960 map_length = orig_bio->bi_size;
5961 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5962 &map_length, NULL, 0);
5963 if (ret) {
5964 bio_put(bio);
5965 goto out_err;
5966 }
5967 } else {
5968 submit_len += bvec->bv_len;
5969 nr_pages ++;
5970 bvec++;
5971 }
5972 }
5973
5974 submit:
5975 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5976 csums, async_submit);
5977 if (!ret)
5978 return 0;
5979
5980 bio_put(bio);
5981 out_err:
5982 dip->errors = 1;
5983 /*
5984 * before atomic variable goto zero, we must
5985 * make sure dip->errors is perceived to be set.
5986 */
5987 smp_mb__before_atomic_dec();
5988 if (atomic_dec_and_test(&dip->pending_bios))
5989 bio_io_error(dip->orig_bio);
5990
5991 /* bio_end_io() will handle error, so we needn't return it */
5992 return 0;
5993 }
5994
5995 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5996 loff_t file_offset)
5997 {
5998 struct btrfs_root *root = BTRFS_I(inode)->root;
5999 struct btrfs_dio_private *dip;
6000 struct bio_vec *bvec = bio->bi_io_vec;
6001 int skip_sum;
6002 int write = rw & REQ_WRITE;
6003 int ret = 0;
6004
6005 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6006
6007 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6008 if (!dip) {
6009 ret = -ENOMEM;
6010 goto free_ordered;
6011 }
6012 dip->csums = NULL;
6013
6014 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6015 if (!write && !skip_sum) {
6016 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6017 if (!dip->csums) {
6018 kfree(dip);
6019 ret = -ENOMEM;
6020 goto free_ordered;
6021 }
6022 }
6023
6024 dip->private = bio->bi_private;
6025 dip->inode = inode;
6026 dip->logical_offset = file_offset;
6027
6028 dip->bytes = 0;
6029 do {
6030 dip->bytes += bvec->bv_len;
6031 bvec++;
6032 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6033
6034 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6035 bio->bi_private = dip;
6036 dip->errors = 0;
6037 dip->orig_bio = bio;
6038 atomic_set(&dip->pending_bios, 0);
6039
6040 if (write)
6041 bio->bi_end_io = btrfs_endio_direct_write;
6042 else
6043 bio->bi_end_io = btrfs_endio_direct_read;
6044
6045 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6046 if (!ret)
6047 return;
6048 free_ordered:
6049 /*
6050 * If this is a write, we need to clean up the reserved space and kill
6051 * the ordered extent.
6052 */
6053 if (write) {
6054 struct btrfs_ordered_extent *ordered;
6055 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6056 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6057 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6058 btrfs_free_reserved_extent(root, ordered->start,
6059 ordered->disk_len);
6060 btrfs_put_ordered_extent(ordered);
6061 btrfs_put_ordered_extent(ordered);
6062 }
6063 bio_endio(bio, ret);
6064 }
6065
6066 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6067 const struct iovec *iov, loff_t offset,
6068 unsigned long nr_segs)
6069 {
6070 int seg;
6071 int i;
6072 size_t size;
6073 unsigned long addr;
6074 unsigned blocksize_mask = root->sectorsize - 1;
6075 ssize_t retval = -EINVAL;
6076 loff_t end = offset;
6077
6078 if (offset & blocksize_mask)
6079 goto out;
6080
6081 /* Check the memory alignment. Blocks cannot straddle pages */
6082 for (seg = 0; seg < nr_segs; seg++) {
6083 addr = (unsigned long)iov[seg].iov_base;
6084 size = iov[seg].iov_len;
6085 end += size;
6086 if ((addr & blocksize_mask) || (size & blocksize_mask))
6087 goto out;
6088
6089 /* If this is a write we don't need to check anymore */
6090 if (rw & WRITE)
6091 continue;
6092
6093 /*
6094 * Check to make sure we don't have duplicate iov_base's in this
6095 * iovec, if so return EINVAL, otherwise we'll get csum errors
6096 * when reading back.
6097 */
6098 for (i = seg + 1; i < nr_segs; i++) {
6099 if (iov[seg].iov_base == iov[i].iov_base)
6100 goto out;
6101 }
6102 }
6103 retval = 0;
6104 out:
6105 return retval;
6106 }
6107 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6108 const struct iovec *iov, loff_t offset,
6109 unsigned long nr_segs)
6110 {
6111 struct file *file = iocb->ki_filp;
6112 struct inode *inode = file->f_mapping->host;
6113 struct btrfs_ordered_extent *ordered;
6114 struct extent_state *cached_state = NULL;
6115 u64 lockstart, lockend;
6116 ssize_t ret;
6117 int writing = rw & WRITE;
6118 int write_bits = 0;
6119 size_t count = iov_length(iov, nr_segs);
6120
6121 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6122 offset, nr_segs)) {
6123 return 0;
6124 }
6125
6126 lockstart = offset;
6127 lockend = offset + count - 1;
6128
6129 if (writing) {
6130 ret = btrfs_delalloc_reserve_space(inode, count);
6131 if (ret)
6132 goto out;
6133 }
6134
6135 while (1) {
6136 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6137 0, &cached_state, GFP_NOFS);
6138 /*
6139 * We're concerned with the entire range that we're going to be
6140 * doing DIO to, so we need to make sure theres no ordered
6141 * extents in this range.
6142 */
6143 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6144 lockend - lockstart + 1);
6145 if (!ordered)
6146 break;
6147 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6148 &cached_state, GFP_NOFS);
6149 btrfs_start_ordered_extent(inode, ordered, 1);
6150 btrfs_put_ordered_extent(ordered);
6151 cond_resched();
6152 }
6153
6154 /*
6155 * we don't use btrfs_set_extent_delalloc because we don't want
6156 * the dirty or uptodate bits
6157 */
6158 if (writing) {
6159 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6160 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6161 EXTENT_DELALLOC, 0, NULL, &cached_state,
6162 GFP_NOFS);
6163 if (ret) {
6164 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6165 lockend, EXTENT_LOCKED | write_bits,
6166 1, 0, &cached_state, GFP_NOFS);
6167 goto out;
6168 }
6169 }
6170
6171 free_extent_state(cached_state);
6172 cached_state = NULL;
6173
6174 ret = __blockdev_direct_IO(rw, iocb, inode,
6175 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6176 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6177 btrfs_submit_direct, 0);
6178
6179 if (ret < 0 && ret != -EIOCBQUEUED) {
6180 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6181 offset + iov_length(iov, nr_segs) - 1,
6182 EXTENT_LOCKED | write_bits, 1, 0,
6183 &cached_state, GFP_NOFS);
6184 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6185 /*
6186 * We're falling back to buffered, unlock the section we didn't
6187 * do IO on.
6188 */
6189 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6190 offset + iov_length(iov, nr_segs) - 1,
6191 EXTENT_LOCKED | write_bits, 1, 0,
6192 &cached_state, GFP_NOFS);
6193 }
6194 out:
6195 free_extent_state(cached_state);
6196 return ret;
6197 }
6198
6199 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6200 __u64 start, __u64 len)
6201 {
6202 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6203 }
6204
6205 int btrfs_readpage(struct file *file, struct page *page)
6206 {
6207 struct extent_io_tree *tree;
6208 tree = &BTRFS_I(page->mapping->host)->io_tree;
6209 return extent_read_full_page(tree, page, btrfs_get_extent);
6210 }
6211
6212 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6213 {
6214 struct extent_io_tree *tree;
6215
6216
6217 if (current->flags & PF_MEMALLOC) {
6218 redirty_page_for_writepage(wbc, page);
6219 unlock_page(page);
6220 return 0;
6221 }
6222 tree = &BTRFS_I(page->mapping->host)->io_tree;
6223 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6224 }
6225
6226 int btrfs_writepages(struct address_space *mapping,
6227 struct writeback_control *wbc)
6228 {
6229 struct extent_io_tree *tree;
6230
6231 tree = &BTRFS_I(mapping->host)->io_tree;
6232 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6233 }
6234
6235 static int
6236 btrfs_readpages(struct file *file, struct address_space *mapping,
6237 struct list_head *pages, unsigned nr_pages)
6238 {
6239 struct extent_io_tree *tree;
6240 tree = &BTRFS_I(mapping->host)->io_tree;
6241 return extent_readpages(tree, mapping, pages, nr_pages,
6242 btrfs_get_extent);
6243 }
6244 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6245 {
6246 struct extent_io_tree *tree;
6247 struct extent_map_tree *map;
6248 int ret;
6249
6250 tree = &BTRFS_I(page->mapping->host)->io_tree;
6251 map = &BTRFS_I(page->mapping->host)->extent_tree;
6252 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6253 if (ret == 1) {
6254 ClearPagePrivate(page);
6255 set_page_private(page, 0);
6256 page_cache_release(page);
6257 }
6258 return ret;
6259 }
6260
6261 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6262 {
6263 if (PageWriteback(page) || PageDirty(page))
6264 return 0;
6265 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6266 }
6267
6268 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6269 {
6270 struct extent_io_tree *tree;
6271 struct btrfs_ordered_extent *ordered;
6272 struct extent_state *cached_state = NULL;
6273 u64 page_start = page_offset(page);
6274 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6275
6276
6277 /*
6278 * we have the page locked, so new writeback can't start,
6279 * and the dirty bit won't be cleared while we are here.
6280 *
6281 * Wait for IO on this page so that we can safely clear
6282 * the PagePrivate2 bit and do ordered accounting
6283 */
6284 wait_on_page_writeback(page);
6285
6286 tree = &BTRFS_I(page->mapping->host)->io_tree;
6287 if (offset) {
6288 btrfs_releasepage(page, GFP_NOFS);
6289 return;
6290 }
6291 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6292 GFP_NOFS);
6293 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6294 page_offset(page));
6295 if (ordered) {
6296 /*
6297 * IO on this page will never be started, so we need
6298 * to account for any ordered extents now
6299 */
6300 clear_extent_bit(tree, page_start, page_end,
6301 EXTENT_DIRTY | EXTENT_DELALLOC |
6302 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6303 &cached_state, GFP_NOFS);
6304 /*
6305 * whoever cleared the private bit is responsible
6306 * for the finish_ordered_io
6307 */
6308 if (TestClearPagePrivate2(page)) {
6309 btrfs_finish_ordered_io(page->mapping->host,
6310 page_start, page_end);
6311 }
6312 btrfs_put_ordered_extent(ordered);
6313 cached_state = NULL;
6314 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6315 GFP_NOFS);
6316 }
6317 clear_extent_bit(tree, page_start, page_end,
6318 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6319 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6320 __btrfs_releasepage(page, GFP_NOFS);
6321
6322 ClearPageChecked(page);
6323 if (PagePrivate(page)) {
6324 ClearPagePrivate(page);
6325 set_page_private(page, 0);
6326 page_cache_release(page);
6327 }
6328 }
6329
6330 /*
6331 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6332 * called from a page fault handler when a page is first dirtied. Hence we must
6333 * be careful to check for EOF conditions here. We set the page up correctly
6334 * for a written page which means we get ENOSPC checking when writing into
6335 * holes and correct delalloc and unwritten extent mapping on filesystems that
6336 * support these features.
6337 *
6338 * We are not allowed to take the i_mutex here so we have to play games to
6339 * protect against truncate races as the page could now be beyond EOF. Because
6340 * vmtruncate() writes the inode size before removing pages, once we have the
6341 * page lock we can determine safely if the page is beyond EOF. If it is not
6342 * beyond EOF, then the page is guaranteed safe against truncation until we
6343 * unlock the page.
6344 */
6345 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6346 {
6347 struct page *page = vmf->page;
6348 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6349 struct btrfs_root *root = BTRFS_I(inode)->root;
6350 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6351 struct btrfs_ordered_extent *ordered;
6352 struct extent_state *cached_state = NULL;
6353 char *kaddr;
6354 unsigned long zero_start;
6355 loff_t size;
6356 int ret;
6357 u64 page_start;
6358 u64 page_end;
6359
6360 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6361 if (ret) {
6362 if (ret == -ENOMEM)
6363 ret = VM_FAULT_OOM;
6364 else /* -ENOSPC, -EIO, etc */
6365 ret = VM_FAULT_SIGBUS;
6366 goto out;
6367 }
6368
6369 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6370 again:
6371 lock_page(page);
6372 size = i_size_read(inode);
6373 page_start = page_offset(page);
6374 page_end = page_start + PAGE_CACHE_SIZE - 1;
6375
6376 if ((page->mapping != inode->i_mapping) ||
6377 (page_start >= size)) {
6378 /* page got truncated out from underneath us */
6379 goto out_unlock;
6380 }
6381 wait_on_page_writeback(page);
6382
6383 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6384 GFP_NOFS);
6385 set_page_extent_mapped(page);
6386
6387 /*
6388 * we can't set the delalloc bits if there are pending ordered
6389 * extents. Drop our locks and wait for them to finish
6390 */
6391 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6392 if (ordered) {
6393 unlock_extent_cached(io_tree, page_start, page_end,
6394 &cached_state, GFP_NOFS);
6395 unlock_page(page);
6396 btrfs_start_ordered_extent(inode, ordered, 1);
6397 btrfs_put_ordered_extent(ordered);
6398 goto again;
6399 }
6400
6401 /*
6402 * XXX - page_mkwrite gets called every time the page is dirtied, even
6403 * if it was already dirty, so for space accounting reasons we need to
6404 * clear any delalloc bits for the range we are fixing to save. There
6405 * is probably a better way to do this, but for now keep consistent with
6406 * prepare_pages in the normal write path.
6407 */
6408 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6409 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6410 0, 0, &cached_state, GFP_NOFS);
6411
6412 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6413 &cached_state);
6414 if (ret) {
6415 unlock_extent_cached(io_tree, page_start, page_end,
6416 &cached_state, GFP_NOFS);
6417 ret = VM_FAULT_SIGBUS;
6418 goto out_unlock;
6419 }
6420 ret = 0;
6421
6422 /* page is wholly or partially inside EOF */
6423 if (page_start + PAGE_CACHE_SIZE > size)
6424 zero_start = size & ~PAGE_CACHE_MASK;
6425 else
6426 zero_start = PAGE_CACHE_SIZE;
6427
6428 if (zero_start != PAGE_CACHE_SIZE) {
6429 kaddr = kmap(page);
6430 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6431 flush_dcache_page(page);
6432 kunmap(page);
6433 }
6434 ClearPageChecked(page);
6435 set_page_dirty(page);
6436 SetPageUptodate(page);
6437
6438 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6439 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6440
6441 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6442
6443 out_unlock:
6444 if (!ret)
6445 return VM_FAULT_LOCKED;
6446 unlock_page(page);
6447 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6448 out:
6449 return ret;
6450 }
6451
6452 static int btrfs_truncate(struct inode *inode)
6453 {
6454 struct btrfs_root *root = BTRFS_I(inode)->root;
6455 struct btrfs_block_rsv *rsv;
6456 int ret;
6457 int err = 0;
6458 struct btrfs_trans_handle *trans;
6459 unsigned long nr;
6460 u64 mask = root->sectorsize - 1;
6461
6462 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6463 if (ret)
6464 return ret;
6465
6466 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6467 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6468
6469 /*
6470 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6471 * 3 things going on here
6472 *
6473 * 1) We need to reserve space for our orphan item and the space to
6474 * delete our orphan item. Lord knows we don't want to have a dangling
6475 * orphan item because we didn't reserve space to remove it.
6476 *
6477 * 2) We need to reserve space to update our inode.
6478 *
6479 * 3) We need to have something to cache all the space that is going to
6480 * be free'd up by the truncate operation, but also have some slack
6481 * space reserved in case it uses space during the truncate (thank you
6482 * very much snapshotting).
6483 *
6484 * And we need these to all be seperate. The fact is we can use alot of
6485 * space doing the truncate, and we have no earthly idea how much space
6486 * we will use, so we need the truncate reservation to be seperate so it
6487 * doesn't end up using space reserved for updating the inode or
6488 * removing the orphan item. We also need to be able to stop the
6489 * transaction and start a new one, which means we need to be able to
6490 * update the inode several times, and we have no idea of knowing how
6491 * many times that will be, so we can't just reserve 1 item for the
6492 * entirety of the opration, so that has to be done seperately as well.
6493 * Then there is the orphan item, which does indeed need to be held on
6494 * to for the whole operation, and we need nobody to touch this reserved
6495 * space except the orphan code.
6496 *
6497 * So that leaves us with
6498 *
6499 * 1) root->orphan_block_rsv - for the orphan deletion.
6500 * 2) rsv - for the truncate reservation, which we will steal from the
6501 * transaction reservation.
6502 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6503 * updating the inode.
6504 */
6505 rsv = btrfs_alloc_block_rsv(root);
6506 if (!rsv)
6507 return -ENOMEM;
6508 btrfs_add_durable_block_rsv(root->fs_info, rsv);
6509
6510 trans = btrfs_start_transaction(root, 4);
6511 if (IS_ERR(trans)) {
6512 err = PTR_ERR(trans);
6513 goto out;
6514 }
6515
6516 /*
6517 * Reserve space for the truncate process. Truncate should be adding
6518 * space, but if there are snapshots it may end up using space.
6519 */
6520 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6521 BUG_ON(ret);
6522
6523 ret = btrfs_orphan_add(trans, inode);
6524 if (ret) {
6525 btrfs_end_transaction(trans, root);
6526 goto out;
6527 }
6528
6529 nr = trans->blocks_used;
6530 btrfs_end_transaction(trans, root);
6531 btrfs_btree_balance_dirty(root, nr);
6532
6533 /*
6534 * Ok so we've already migrated our bytes over for the truncate, so here
6535 * just reserve the one slot we need for updating the inode.
6536 */
6537 trans = btrfs_start_transaction(root, 1);
6538 if (IS_ERR(trans)) {
6539 err = PTR_ERR(trans);
6540 goto out;
6541 }
6542 trans->block_rsv = rsv;
6543
6544 /*
6545 * setattr is responsible for setting the ordered_data_close flag,
6546 * but that is only tested during the last file release. That
6547 * could happen well after the next commit, leaving a great big
6548 * window where new writes may get lost if someone chooses to write
6549 * to this file after truncating to zero
6550 *
6551 * The inode doesn't have any dirty data here, and so if we commit
6552 * this is a noop. If someone immediately starts writing to the inode
6553 * it is very likely we'll catch some of their writes in this
6554 * transaction, and the commit will find this file on the ordered
6555 * data list with good things to send down.
6556 *
6557 * This is a best effort solution, there is still a window where
6558 * using truncate to replace the contents of the file will
6559 * end up with a zero length file after a crash.
6560 */
6561 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6562 btrfs_add_ordered_operation(trans, root, inode);
6563
6564 while (1) {
6565 if (!trans) {
6566 trans = btrfs_start_transaction(root, 3);
6567 if (IS_ERR(trans)) {
6568 err = PTR_ERR(trans);
6569 goto out;
6570 }
6571
6572 ret = btrfs_truncate_reserve_metadata(trans, root,
6573 rsv);
6574 BUG_ON(ret);
6575
6576 trans->block_rsv = rsv;
6577 }
6578
6579 ret = btrfs_truncate_inode_items(trans, root, inode,
6580 inode->i_size,
6581 BTRFS_EXTENT_DATA_KEY);
6582 if (ret != -EAGAIN) {
6583 err = ret;
6584 break;
6585 }
6586
6587 trans->block_rsv = &root->fs_info->trans_block_rsv;
6588 ret = btrfs_update_inode(trans, root, inode);
6589 if (ret) {
6590 err = ret;
6591 break;
6592 }
6593
6594 nr = trans->blocks_used;
6595 btrfs_end_transaction(trans, root);
6596 trans = NULL;
6597 btrfs_btree_balance_dirty(root, nr);
6598 }
6599
6600 if (ret == 0 && inode->i_nlink > 0) {
6601 trans->block_rsv = root->orphan_block_rsv;
6602 ret = btrfs_orphan_del(trans, inode);
6603 if (ret)
6604 err = ret;
6605 } else if (ret && inode->i_nlink > 0) {
6606 /*
6607 * Failed to do the truncate, remove us from the in memory
6608 * orphan list.
6609 */
6610 ret = btrfs_orphan_del(NULL, inode);
6611 }
6612
6613 trans->block_rsv = &root->fs_info->trans_block_rsv;
6614 ret = btrfs_update_inode(trans, root, inode);
6615 if (ret && !err)
6616 err = ret;
6617
6618 nr = trans->blocks_used;
6619 ret = btrfs_end_transaction_throttle(trans, root);
6620 btrfs_btree_balance_dirty(root, nr);
6621
6622 out:
6623 btrfs_free_block_rsv(root, rsv);
6624
6625 if (ret && !err)
6626 err = ret;
6627
6628 return err;
6629 }
6630
6631 /*
6632 * create a new subvolume directory/inode (helper for the ioctl).
6633 */
6634 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6635 struct btrfs_root *new_root, u64 new_dirid)
6636 {
6637 struct inode *inode;
6638 int err;
6639 u64 index = 0;
6640
6641 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6642 new_dirid, S_IFDIR | 0700, &index);
6643 if (IS_ERR(inode))
6644 return PTR_ERR(inode);
6645 inode->i_op = &btrfs_dir_inode_operations;
6646 inode->i_fop = &btrfs_dir_file_operations;
6647
6648 inode->i_nlink = 1;
6649 btrfs_i_size_write(inode, 0);
6650
6651 err = btrfs_update_inode(trans, new_root, inode);
6652 BUG_ON(err);
6653
6654 iput(inode);
6655 return 0;
6656 }
6657
6658 struct inode *btrfs_alloc_inode(struct super_block *sb)
6659 {
6660 struct btrfs_inode *ei;
6661 struct inode *inode;
6662
6663 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6664 if (!ei)
6665 return NULL;
6666
6667 ei->root = NULL;
6668 ei->space_info = NULL;
6669 ei->generation = 0;
6670 ei->sequence = 0;
6671 ei->last_trans = 0;
6672 ei->last_sub_trans = 0;
6673 ei->logged_trans = 0;
6674 ei->delalloc_bytes = 0;
6675 ei->disk_i_size = 0;
6676 ei->flags = 0;
6677 ei->csum_bytes = 0;
6678 ei->index_cnt = (u64)-1;
6679 ei->last_unlink_trans = 0;
6680
6681 spin_lock_init(&ei->lock);
6682 ei->outstanding_extents = 0;
6683 ei->reserved_extents = 0;
6684
6685 ei->ordered_data_close = 0;
6686 ei->orphan_meta_reserved = 0;
6687 ei->dummy_inode = 0;
6688 ei->in_defrag = 0;
6689 ei->force_compress = BTRFS_COMPRESS_NONE;
6690
6691 ei->delayed_node = NULL;
6692
6693 inode = &ei->vfs_inode;
6694 extent_map_tree_init(&ei->extent_tree);
6695 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6696 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6697 mutex_init(&ei->log_mutex);
6698 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6699 INIT_LIST_HEAD(&ei->i_orphan);
6700 INIT_LIST_HEAD(&ei->delalloc_inodes);
6701 INIT_LIST_HEAD(&ei->ordered_operations);
6702 RB_CLEAR_NODE(&ei->rb_node);
6703
6704 return inode;
6705 }
6706
6707 static void btrfs_i_callback(struct rcu_head *head)
6708 {
6709 struct inode *inode = container_of(head, struct inode, i_rcu);
6710 INIT_LIST_HEAD(&inode->i_dentry);
6711 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6712 }
6713
6714 void btrfs_destroy_inode(struct inode *inode)
6715 {
6716 struct btrfs_ordered_extent *ordered;
6717 struct btrfs_root *root = BTRFS_I(inode)->root;
6718
6719 WARN_ON(!list_empty(&inode->i_dentry));
6720 WARN_ON(inode->i_data.nrpages);
6721 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6722 WARN_ON(BTRFS_I(inode)->reserved_extents);
6723 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6724 WARN_ON(BTRFS_I(inode)->csum_bytes);
6725
6726 /*
6727 * This can happen where we create an inode, but somebody else also
6728 * created the same inode and we need to destroy the one we already
6729 * created.
6730 */
6731 if (!root)
6732 goto free;
6733
6734 /*
6735 * Make sure we're properly removed from the ordered operation
6736 * lists.
6737 */
6738 smp_mb();
6739 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6740 spin_lock(&root->fs_info->ordered_extent_lock);
6741 list_del_init(&BTRFS_I(inode)->ordered_operations);
6742 spin_unlock(&root->fs_info->ordered_extent_lock);
6743 }
6744
6745 spin_lock(&root->orphan_lock);
6746 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6747 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6748 (unsigned long long)btrfs_ino(inode));
6749 list_del_init(&BTRFS_I(inode)->i_orphan);
6750 }
6751 spin_unlock(&root->orphan_lock);
6752
6753 while (1) {
6754 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6755 if (!ordered)
6756 break;
6757 else {
6758 printk(KERN_ERR "btrfs found ordered "
6759 "extent %llu %llu on inode cleanup\n",
6760 (unsigned long long)ordered->file_offset,
6761 (unsigned long long)ordered->len);
6762 btrfs_remove_ordered_extent(inode, ordered);
6763 btrfs_put_ordered_extent(ordered);
6764 btrfs_put_ordered_extent(ordered);
6765 }
6766 }
6767 inode_tree_del(inode);
6768 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6769 free:
6770 btrfs_remove_delayed_node(inode);
6771 call_rcu(&inode->i_rcu, btrfs_i_callback);
6772 }
6773
6774 int btrfs_drop_inode(struct inode *inode)
6775 {
6776 struct btrfs_root *root = BTRFS_I(inode)->root;
6777
6778 if (btrfs_root_refs(&root->root_item) == 0 &&
6779 !btrfs_is_free_space_inode(root, inode))
6780 return 1;
6781 else
6782 return generic_drop_inode(inode);
6783 }
6784
6785 static void init_once(void *foo)
6786 {
6787 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6788
6789 inode_init_once(&ei->vfs_inode);
6790 }
6791
6792 void btrfs_destroy_cachep(void)
6793 {
6794 if (btrfs_inode_cachep)
6795 kmem_cache_destroy(btrfs_inode_cachep);
6796 if (btrfs_trans_handle_cachep)
6797 kmem_cache_destroy(btrfs_trans_handle_cachep);
6798 if (btrfs_transaction_cachep)
6799 kmem_cache_destroy(btrfs_transaction_cachep);
6800 if (btrfs_path_cachep)
6801 kmem_cache_destroy(btrfs_path_cachep);
6802 if (btrfs_free_space_cachep)
6803 kmem_cache_destroy(btrfs_free_space_cachep);
6804 }
6805
6806 int btrfs_init_cachep(void)
6807 {
6808 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6809 sizeof(struct btrfs_inode), 0,
6810 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6811 if (!btrfs_inode_cachep)
6812 goto fail;
6813
6814 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6815 sizeof(struct btrfs_trans_handle), 0,
6816 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6817 if (!btrfs_trans_handle_cachep)
6818 goto fail;
6819
6820 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6821 sizeof(struct btrfs_transaction), 0,
6822 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6823 if (!btrfs_transaction_cachep)
6824 goto fail;
6825
6826 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6827 sizeof(struct btrfs_path), 0,
6828 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6829 if (!btrfs_path_cachep)
6830 goto fail;
6831
6832 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6833 sizeof(struct btrfs_free_space), 0,
6834 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6835 if (!btrfs_free_space_cachep)
6836 goto fail;
6837
6838 return 0;
6839 fail:
6840 btrfs_destroy_cachep();
6841 return -ENOMEM;
6842 }
6843
6844 static int btrfs_getattr(struct vfsmount *mnt,
6845 struct dentry *dentry, struct kstat *stat)
6846 {
6847 struct inode *inode = dentry->d_inode;
6848 generic_fillattr(inode, stat);
6849 stat->dev = BTRFS_I(inode)->root->anon_dev;
6850 stat->blksize = PAGE_CACHE_SIZE;
6851 stat->blocks = (inode_get_bytes(inode) +
6852 BTRFS_I(inode)->delalloc_bytes) >> 9;
6853 return 0;
6854 }
6855
6856 /*
6857 * If a file is moved, it will inherit the cow and compression flags of the new
6858 * directory.
6859 */
6860 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6861 {
6862 struct btrfs_inode *b_dir = BTRFS_I(dir);
6863 struct btrfs_inode *b_inode = BTRFS_I(inode);
6864
6865 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6866 b_inode->flags |= BTRFS_INODE_NODATACOW;
6867 else
6868 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6869
6870 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6871 b_inode->flags |= BTRFS_INODE_COMPRESS;
6872 else
6873 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6874 }
6875
6876 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6877 struct inode *new_dir, struct dentry *new_dentry)
6878 {
6879 struct btrfs_trans_handle *trans;
6880 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6881 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6882 struct inode *new_inode = new_dentry->d_inode;
6883 struct inode *old_inode = old_dentry->d_inode;
6884 struct timespec ctime = CURRENT_TIME;
6885 u64 index = 0;
6886 u64 root_objectid;
6887 int ret;
6888 u64 old_ino = btrfs_ino(old_inode);
6889
6890 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6891 return -EPERM;
6892
6893 /* we only allow rename subvolume link between subvolumes */
6894 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6895 return -EXDEV;
6896
6897 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6898 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6899 return -ENOTEMPTY;
6900
6901 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6902 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6903 return -ENOTEMPTY;
6904 /*
6905 * we're using rename to replace one file with another.
6906 * and the replacement file is large. Start IO on it now so
6907 * we don't add too much work to the end of the transaction
6908 */
6909 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6910 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6911 filemap_flush(old_inode->i_mapping);
6912
6913 /* close the racy window with snapshot create/destroy ioctl */
6914 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6915 down_read(&root->fs_info->subvol_sem);
6916 /*
6917 * We want to reserve the absolute worst case amount of items. So if
6918 * both inodes are subvols and we need to unlink them then that would
6919 * require 4 item modifications, but if they are both normal inodes it
6920 * would require 5 item modifications, so we'll assume their normal
6921 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6922 * should cover the worst case number of items we'll modify.
6923 */
6924 trans = btrfs_start_transaction(root, 20);
6925 if (IS_ERR(trans)) {
6926 ret = PTR_ERR(trans);
6927 goto out_notrans;
6928 }
6929
6930 if (dest != root)
6931 btrfs_record_root_in_trans(trans, dest);
6932
6933 ret = btrfs_set_inode_index(new_dir, &index);
6934 if (ret)
6935 goto out_fail;
6936
6937 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6938 /* force full log commit if subvolume involved. */
6939 root->fs_info->last_trans_log_full_commit = trans->transid;
6940 } else {
6941 ret = btrfs_insert_inode_ref(trans, dest,
6942 new_dentry->d_name.name,
6943 new_dentry->d_name.len,
6944 old_ino,
6945 btrfs_ino(new_dir), index);
6946 if (ret)
6947 goto out_fail;
6948 /*
6949 * this is an ugly little race, but the rename is required
6950 * to make sure that if we crash, the inode is either at the
6951 * old name or the new one. pinning the log transaction lets
6952 * us make sure we don't allow a log commit to come in after
6953 * we unlink the name but before we add the new name back in.
6954 */
6955 btrfs_pin_log_trans(root);
6956 }
6957 /*
6958 * make sure the inode gets flushed if it is replacing
6959 * something.
6960 */
6961 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6962 btrfs_add_ordered_operation(trans, root, old_inode);
6963
6964 old_dir->i_ctime = old_dir->i_mtime = ctime;
6965 new_dir->i_ctime = new_dir->i_mtime = ctime;
6966 old_inode->i_ctime = ctime;
6967
6968 if (old_dentry->d_parent != new_dentry->d_parent)
6969 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6970
6971 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6972 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6973 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6974 old_dentry->d_name.name,
6975 old_dentry->d_name.len);
6976 } else {
6977 ret = __btrfs_unlink_inode(trans, root, old_dir,
6978 old_dentry->d_inode,
6979 old_dentry->d_name.name,
6980 old_dentry->d_name.len);
6981 if (!ret)
6982 ret = btrfs_update_inode(trans, root, old_inode);
6983 }
6984 BUG_ON(ret);
6985
6986 if (new_inode) {
6987 new_inode->i_ctime = CURRENT_TIME;
6988 if (unlikely(btrfs_ino(new_inode) ==
6989 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6990 root_objectid = BTRFS_I(new_inode)->location.objectid;
6991 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6992 root_objectid,
6993 new_dentry->d_name.name,
6994 new_dentry->d_name.len);
6995 BUG_ON(new_inode->i_nlink == 0);
6996 } else {
6997 ret = btrfs_unlink_inode(trans, dest, new_dir,
6998 new_dentry->d_inode,
6999 new_dentry->d_name.name,
7000 new_dentry->d_name.len);
7001 }
7002 BUG_ON(ret);
7003 if (new_inode->i_nlink == 0) {
7004 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7005 BUG_ON(ret);
7006 }
7007 }
7008
7009 fixup_inode_flags(new_dir, old_inode);
7010
7011 ret = btrfs_add_link(trans, new_dir, old_inode,
7012 new_dentry->d_name.name,
7013 new_dentry->d_name.len, 0, index);
7014 BUG_ON(ret);
7015
7016 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7017 struct dentry *parent = new_dentry->d_parent;
7018 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7019 btrfs_end_log_trans(root);
7020 }
7021 out_fail:
7022 btrfs_end_transaction_throttle(trans, root);
7023 out_notrans:
7024 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7025 up_read(&root->fs_info->subvol_sem);
7026
7027 return ret;
7028 }
7029
7030 /*
7031 * some fairly slow code that needs optimization. This walks the list
7032 * of all the inodes with pending delalloc and forces them to disk.
7033 */
7034 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7035 {
7036 struct list_head *head = &root->fs_info->delalloc_inodes;
7037 struct btrfs_inode *binode;
7038 struct inode *inode;
7039
7040 if (root->fs_info->sb->s_flags & MS_RDONLY)
7041 return -EROFS;
7042
7043 spin_lock(&root->fs_info->delalloc_lock);
7044 while (!list_empty(head)) {
7045 binode = list_entry(head->next, struct btrfs_inode,
7046 delalloc_inodes);
7047 inode = igrab(&binode->vfs_inode);
7048 if (!inode)
7049 list_del_init(&binode->delalloc_inodes);
7050 spin_unlock(&root->fs_info->delalloc_lock);
7051 if (inode) {
7052 filemap_flush(inode->i_mapping);
7053 if (delay_iput)
7054 btrfs_add_delayed_iput(inode);
7055 else
7056 iput(inode);
7057 }
7058 cond_resched();
7059 spin_lock(&root->fs_info->delalloc_lock);
7060 }
7061 spin_unlock(&root->fs_info->delalloc_lock);
7062
7063 /* the filemap_flush will queue IO into the worker threads, but
7064 * we have to make sure the IO is actually started and that
7065 * ordered extents get created before we return
7066 */
7067 atomic_inc(&root->fs_info->async_submit_draining);
7068 while (atomic_read(&root->fs_info->nr_async_submits) ||
7069 atomic_read(&root->fs_info->async_delalloc_pages)) {
7070 wait_event(root->fs_info->async_submit_wait,
7071 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7072 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7073 }
7074 atomic_dec(&root->fs_info->async_submit_draining);
7075 return 0;
7076 }
7077
7078 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7079 const char *symname)
7080 {
7081 struct btrfs_trans_handle *trans;
7082 struct btrfs_root *root = BTRFS_I(dir)->root;
7083 struct btrfs_path *path;
7084 struct btrfs_key key;
7085 struct inode *inode = NULL;
7086 int err;
7087 int drop_inode = 0;
7088 u64 objectid;
7089 u64 index = 0 ;
7090 int name_len;
7091 int datasize;
7092 unsigned long ptr;
7093 struct btrfs_file_extent_item *ei;
7094 struct extent_buffer *leaf;
7095 unsigned long nr = 0;
7096
7097 name_len = strlen(symname) + 1;
7098 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7099 return -ENAMETOOLONG;
7100
7101 /*
7102 * 2 items for inode item and ref
7103 * 2 items for dir items
7104 * 1 item for xattr if selinux is on
7105 */
7106 trans = btrfs_start_transaction(root, 5);
7107 if (IS_ERR(trans))
7108 return PTR_ERR(trans);
7109
7110 err = btrfs_find_free_ino(root, &objectid);
7111 if (err)
7112 goto out_unlock;
7113
7114 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7115 dentry->d_name.len, btrfs_ino(dir), objectid,
7116 S_IFLNK|S_IRWXUGO, &index);
7117 if (IS_ERR(inode)) {
7118 err = PTR_ERR(inode);
7119 goto out_unlock;
7120 }
7121
7122 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7123 if (err) {
7124 drop_inode = 1;
7125 goto out_unlock;
7126 }
7127
7128 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7129 if (err)
7130 drop_inode = 1;
7131 else {
7132 inode->i_mapping->a_ops = &btrfs_aops;
7133 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7134 inode->i_fop = &btrfs_file_operations;
7135 inode->i_op = &btrfs_file_inode_operations;
7136 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7137 }
7138 if (drop_inode)
7139 goto out_unlock;
7140
7141 path = btrfs_alloc_path();
7142 if (!path) {
7143 err = -ENOMEM;
7144 drop_inode = 1;
7145 goto out_unlock;
7146 }
7147 key.objectid = btrfs_ino(inode);
7148 key.offset = 0;
7149 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7150 datasize = btrfs_file_extent_calc_inline_size(name_len);
7151 err = btrfs_insert_empty_item(trans, root, path, &key,
7152 datasize);
7153 if (err) {
7154 drop_inode = 1;
7155 btrfs_free_path(path);
7156 goto out_unlock;
7157 }
7158 leaf = path->nodes[0];
7159 ei = btrfs_item_ptr(leaf, path->slots[0],
7160 struct btrfs_file_extent_item);
7161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7162 btrfs_set_file_extent_type(leaf, ei,
7163 BTRFS_FILE_EXTENT_INLINE);
7164 btrfs_set_file_extent_encryption(leaf, ei, 0);
7165 btrfs_set_file_extent_compression(leaf, ei, 0);
7166 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7167 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7168
7169 ptr = btrfs_file_extent_inline_start(ei);
7170 write_extent_buffer(leaf, symname, ptr, name_len);
7171 btrfs_mark_buffer_dirty(leaf);
7172 btrfs_free_path(path);
7173
7174 inode->i_op = &btrfs_symlink_inode_operations;
7175 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7176 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7177 inode_set_bytes(inode, name_len);
7178 btrfs_i_size_write(inode, name_len - 1);
7179 err = btrfs_update_inode(trans, root, inode);
7180 if (err)
7181 drop_inode = 1;
7182
7183 out_unlock:
7184 nr = trans->blocks_used;
7185 btrfs_end_transaction_throttle(trans, root);
7186 if (drop_inode) {
7187 inode_dec_link_count(inode);
7188 iput(inode);
7189 }
7190 btrfs_btree_balance_dirty(root, nr);
7191 return err;
7192 }
7193
7194 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7195 u64 start, u64 num_bytes, u64 min_size,
7196 loff_t actual_len, u64 *alloc_hint,
7197 struct btrfs_trans_handle *trans)
7198 {
7199 struct btrfs_root *root = BTRFS_I(inode)->root;
7200 struct btrfs_key ins;
7201 u64 cur_offset = start;
7202 u64 i_size;
7203 int ret = 0;
7204 bool own_trans = true;
7205
7206 if (trans)
7207 own_trans = false;
7208 while (num_bytes > 0) {
7209 if (own_trans) {
7210 trans = btrfs_start_transaction(root, 3);
7211 if (IS_ERR(trans)) {
7212 ret = PTR_ERR(trans);
7213 break;
7214 }
7215 }
7216
7217 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7218 0, *alloc_hint, (u64)-1, &ins, 1);
7219 if (ret) {
7220 if (own_trans)
7221 btrfs_end_transaction(trans, root);
7222 break;
7223 }
7224
7225 ret = insert_reserved_file_extent(trans, inode,
7226 cur_offset, ins.objectid,
7227 ins.offset, ins.offset,
7228 ins.offset, 0, 0, 0,
7229 BTRFS_FILE_EXTENT_PREALLOC);
7230 BUG_ON(ret);
7231 btrfs_drop_extent_cache(inode, cur_offset,
7232 cur_offset + ins.offset -1, 0);
7233
7234 num_bytes -= ins.offset;
7235 cur_offset += ins.offset;
7236 *alloc_hint = ins.objectid + ins.offset;
7237
7238 inode->i_ctime = CURRENT_TIME;
7239 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7240 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7241 (actual_len > inode->i_size) &&
7242 (cur_offset > inode->i_size)) {
7243 if (cur_offset > actual_len)
7244 i_size = actual_len;
7245 else
7246 i_size = cur_offset;
7247 i_size_write(inode, i_size);
7248 btrfs_ordered_update_i_size(inode, i_size, NULL);
7249 }
7250
7251 ret = btrfs_update_inode(trans, root, inode);
7252 BUG_ON(ret);
7253
7254 if (own_trans)
7255 btrfs_end_transaction(trans, root);
7256 }
7257 return ret;
7258 }
7259
7260 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7261 u64 start, u64 num_bytes, u64 min_size,
7262 loff_t actual_len, u64 *alloc_hint)
7263 {
7264 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7265 min_size, actual_len, alloc_hint,
7266 NULL);
7267 }
7268
7269 int btrfs_prealloc_file_range_trans(struct inode *inode,
7270 struct btrfs_trans_handle *trans, int mode,
7271 u64 start, u64 num_bytes, u64 min_size,
7272 loff_t actual_len, u64 *alloc_hint)
7273 {
7274 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7275 min_size, actual_len, alloc_hint, trans);
7276 }
7277
7278 static int btrfs_set_page_dirty(struct page *page)
7279 {
7280 return __set_page_dirty_nobuffers(page);
7281 }
7282
7283 static int btrfs_permission(struct inode *inode, int mask)
7284 {
7285 struct btrfs_root *root = BTRFS_I(inode)->root;
7286 umode_t mode = inode->i_mode;
7287
7288 if (mask & MAY_WRITE &&
7289 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7290 if (btrfs_root_readonly(root))
7291 return -EROFS;
7292 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7293 return -EACCES;
7294 }
7295 return generic_permission(inode, mask);
7296 }
7297
7298 static const struct inode_operations btrfs_dir_inode_operations = {
7299 .getattr = btrfs_getattr,
7300 .lookup = btrfs_lookup,
7301 .create = btrfs_create,
7302 .unlink = btrfs_unlink,
7303 .link = btrfs_link,
7304 .mkdir = btrfs_mkdir,
7305 .rmdir = btrfs_rmdir,
7306 .rename = btrfs_rename,
7307 .symlink = btrfs_symlink,
7308 .setattr = btrfs_setattr,
7309 .mknod = btrfs_mknod,
7310 .setxattr = btrfs_setxattr,
7311 .getxattr = btrfs_getxattr,
7312 .listxattr = btrfs_listxattr,
7313 .removexattr = btrfs_removexattr,
7314 .permission = btrfs_permission,
7315 .get_acl = btrfs_get_acl,
7316 };
7317 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7318 .lookup = btrfs_lookup,
7319 .permission = btrfs_permission,
7320 .get_acl = btrfs_get_acl,
7321 };
7322
7323 static const struct file_operations btrfs_dir_file_operations = {
7324 .llseek = generic_file_llseek,
7325 .read = generic_read_dir,
7326 .readdir = btrfs_real_readdir,
7327 .unlocked_ioctl = btrfs_ioctl,
7328 #ifdef CONFIG_COMPAT
7329 .compat_ioctl = btrfs_ioctl,
7330 #endif
7331 .release = btrfs_release_file,
7332 .fsync = btrfs_sync_file,
7333 };
7334
7335 static struct extent_io_ops btrfs_extent_io_ops = {
7336 .fill_delalloc = run_delalloc_range,
7337 .submit_bio_hook = btrfs_submit_bio_hook,
7338 .merge_bio_hook = btrfs_merge_bio_hook,
7339 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7340 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7341 .writepage_start_hook = btrfs_writepage_start_hook,
7342 .readpage_io_failed_hook = btrfs_io_failed_hook,
7343 .set_bit_hook = btrfs_set_bit_hook,
7344 .clear_bit_hook = btrfs_clear_bit_hook,
7345 .merge_extent_hook = btrfs_merge_extent_hook,
7346 .split_extent_hook = btrfs_split_extent_hook,
7347 };
7348
7349 /*
7350 * btrfs doesn't support the bmap operation because swapfiles
7351 * use bmap to make a mapping of extents in the file. They assume
7352 * these extents won't change over the life of the file and they
7353 * use the bmap result to do IO directly to the drive.
7354 *
7355 * the btrfs bmap call would return logical addresses that aren't
7356 * suitable for IO and they also will change frequently as COW
7357 * operations happen. So, swapfile + btrfs == corruption.
7358 *
7359 * For now we're avoiding this by dropping bmap.
7360 */
7361 static const struct address_space_operations btrfs_aops = {
7362 .readpage = btrfs_readpage,
7363 .writepage = btrfs_writepage,
7364 .writepages = btrfs_writepages,
7365 .readpages = btrfs_readpages,
7366 .direct_IO = btrfs_direct_IO,
7367 .invalidatepage = btrfs_invalidatepage,
7368 .releasepage = btrfs_releasepage,
7369 .set_page_dirty = btrfs_set_page_dirty,
7370 .error_remove_page = generic_error_remove_page,
7371 };
7372
7373 static const struct address_space_operations btrfs_symlink_aops = {
7374 .readpage = btrfs_readpage,
7375 .writepage = btrfs_writepage,
7376 .invalidatepage = btrfs_invalidatepage,
7377 .releasepage = btrfs_releasepage,
7378 };
7379
7380 static const struct inode_operations btrfs_file_inode_operations = {
7381 .getattr = btrfs_getattr,
7382 .setattr = btrfs_setattr,
7383 .setxattr = btrfs_setxattr,
7384 .getxattr = btrfs_getxattr,
7385 .listxattr = btrfs_listxattr,
7386 .removexattr = btrfs_removexattr,
7387 .permission = btrfs_permission,
7388 .fiemap = btrfs_fiemap,
7389 .get_acl = btrfs_get_acl,
7390 };
7391 static const struct inode_operations btrfs_special_inode_operations = {
7392 .getattr = btrfs_getattr,
7393 .setattr = btrfs_setattr,
7394 .permission = btrfs_permission,
7395 .setxattr = btrfs_setxattr,
7396 .getxattr = btrfs_getxattr,
7397 .listxattr = btrfs_listxattr,
7398 .removexattr = btrfs_removexattr,
7399 .get_acl = btrfs_get_acl,
7400 };
7401 static const struct inode_operations btrfs_symlink_inode_operations = {
7402 .readlink = generic_readlink,
7403 .follow_link = page_follow_link_light,
7404 .put_link = page_put_link,
7405 .getattr = btrfs_getattr,
7406 .permission = btrfs_permission,
7407 .setxattr = btrfs_setxattr,
7408 .getxattr = btrfs_getxattr,
7409 .listxattr = btrfs_listxattr,
7410 .removexattr = btrfs_removexattr,
7411 .get_acl = btrfs_get_acl,
7412 };
7413
7414 const struct dentry_operations btrfs_dentry_operations = {
7415 .d_delete = btrfs_dentry_delete,
7416 .d_release = btrfs_dentry_release,
7417 };
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree