aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/inode.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/inode.c')
-rw-r--r--fs/btrfs/inode.c5022
1 files changed, 5022 insertions, 0 deletions
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
new file mode 100644
index 000000000000..806caacff86c
--- /dev/null
+++ b/fs/btrfs/inode.c
@@ -0,0 +1,5022 @@
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/kernel.h>
20#include <linux/bio.h>
21#include <linux/buffer_head.h>
22#include <linux/file.h>
23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
29#include <linux/smp_lock.h>
30#include <linux/backing-dev.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
36#include <linux/bit_spinlock.h>
37#include <linux/version.h>
38#include <linux/xattr.h>
39#include <linux/posix_acl.h>
40#include <linux/falloc.h>
41#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
47#include "volumes.h"
48#include "ordered-data.h"
49#include "xattr.h"
50#include "compat.h"
51#include "tree-log.h"
52#include "ref-cache.h"
53#include "compression.h"
54
55struct btrfs_iget_args {
56 u64 ino;
57 struct btrfs_root *root;
58};
59
60static struct inode_operations btrfs_dir_inode_operations;
61static struct inode_operations btrfs_symlink_inode_operations;
62static struct inode_operations btrfs_dir_ro_inode_operations;
63static struct inode_operations btrfs_special_inode_operations;
64static struct inode_operations btrfs_file_inode_operations;
65static struct address_space_operations btrfs_aops;
66static struct address_space_operations btrfs_symlink_aops;
67static struct file_operations btrfs_dir_file_operations;
68static struct extent_io_ops btrfs_extent_io_ops;
69
70static struct kmem_cache *btrfs_inode_cachep;
71struct kmem_cache *btrfs_trans_handle_cachep;
72struct kmem_cache *btrfs_transaction_cachep;
73struct kmem_cache *btrfs_bit_radix_cachep;
74struct kmem_cache *btrfs_path_cachep;
75
76#define S_SHIFT 12
77static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85};
86
87static void btrfs_truncate(struct inode *inode);
88static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
93
94/*
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
97 */
98int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
99 int for_del)
100{
101 u64 total;
102 u64 used;
103 u64 thresh;
104 unsigned long flags;
105 int ret = 0;
106
107 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
108 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
109 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
110 if (for_del)
111 thresh = total * 90;
112 else
113 thresh = total * 85;
114
115 do_div(thresh, 100);
116
117 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
118 ret = -ENOSPC;
119 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
120 return ret;
121}
122
123/*
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
128static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
132{
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
145 int use_compress = 0;
146
147 if (compressed_size && compressed_pages) {
148 use_compress = 1;
149 cur_size = compressed_size;
150 }
151
152 path = btrfs_alloc_path(); if (!path)
153 return -ENOMEM;
154
155 btrfs_set_trans_block_group(trans, inode);
156
157 key.objectid = inode->i_ino;
158 key.offset = start;
159 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160 inode_add_bytes(inode, size);
161 datasize = btrfs_file_extent_calc_inline_size(cur_size);
162
163 inode_add_bytes(inode, size);
164 ret = btrfs_insert_empty_item(trans, root, path, &key,
165 datasize);
166 BUG_ON(ret);
167 if (ret) {
168 err = ret;
169 printk("got bad ret %d\n", ret);
170 goto fail;
171 }
172 leaf = path->nodes[0];
173 ei = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_file_extent_item);
175 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
176 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
177 btrfs_set_file_extent_encryption(leaf, ei, 0);
178 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
179 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
180 ptr = btrfs_file_extent_inline_start(ei);
181
182 if (use_compress) {
183 struct page *cpage;
184 int i = 0;
185 while(compressed_size > 0) {
186 cpage = compressed_pages[i];
187 cur_size = min_t(unsigned long, compressed_size,
188 PAGE_CACHE_SIZE);
189
190 kaddr = kmap(cpage);
191 write_extent_buffer(leaf, kaddr, ptr, cur_size);
192 kunmap(cpage);
193
194 i++;
195 ptr += cur_size;
196 compressed_size -= cur_size;
197 }
198 btrfs_set_file_extent_compression(leaf, ei,
199 BTRFS_COMPRESS_ZLIB);
200 } else {
201 page = find_get_page(inode->i_mapping,
202 start >> PAGE_CACHE_SHIFT);
203 btrfs_set_file_extent_compression(leaf, ei, 0);
204 kaddr = kmap_atomic(page, KM_USER0);
205 offset = start & (PAGE_CACHE_SIZE - 1);
206 write_extent_buffer(leaf, kaddr + offset, ptr, size);
207 kunmap_atomic(kaddr, KM_USER0);
208 page_cache_release(page);
209 }
210 btrfs_mark_buffer_dirty(leaf);
211 btrfs_free_path(path);
212
213 BTRFS_I(inode)->disk_i_size = inode->i_size;
214 btrfs_update_inode(trans, root, inode);
215 return 0;
216fail:
217 btrfs_free_path(path);
218 return err;
219}
220
221
222/*
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
226 */
227static int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size,
231 struct page **compressed_pages)
232{
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = (end + root->sectorsize - 1) &
237 ~((u64)root->sectorsize - 1);
238 u64 hint_byte;
239 u64 data_len = inline_len;
240 int ret;
241
242 if (compressed_size)
243 data_len = compressed_size;
244
245 if (start > 0 ||
246 actual_end >= PAGE_CACHE_SIZE ||
247 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
248 (!compressed_size &&
249 (actual_end & (root->sectorsize - 1)) == 0) ||
250 end + 1 < isize ||
251 data_len > root->fs_info->max_inline) {
252 return 1;
253 }
254
255 ret = btrfs_drop_extents(trans, root, inode, start,
256 aligned_end, start, &hint_byte);
257 BUG_ON(ret);
258
259 if (isize > actual_end)
260 inline_len = min_t(u64, isize, actual_end);
261 ret = insert_inline_extent(trans, root, inode, start,
262 inline_len, compressed_size,
263 compressed_pages);
264 BUG_ON(ret);
265 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
266 return 0;
267}
268
269struct async_extent {
270 u64 start;
271 u64 ram_size;
272 u64 compressed_size;
273 struct page **pages;
274 unsigned long nr_pages;
275 struct list_head list;
276};
277
278struct async_cow {
279 struct inode *inode;
280 struct btrfs_root *root;
281 struct page *locked_page;
282 u64 start;
283 u64 end;
284 struct list_head extents;
285 struct btrfs_work work;
286};
287
288static noinline int add_async_extent(struct async_cow *cow,
289 u64 start, u64 ram_size,
290 u64 compressed_size,
291 struct page **pages,
292 unsigned long nr_pages)
293{
294 struct async_extent *async_extent;
295
296 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 list_add_tail(&async_extent->list, &cow->extents);
303 return 0;
304}
305
306/*
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
310 *
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
316 *
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
321 */
322static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
324 u64 start, u64 end,
325 struct async_cow *async_cow,
326 int *num_added)
327{
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
330 u64 num_bytes;
331 u64 orig_start;
332 u64 disk_num_bytes;
333 u64 blocksize = root->sectorsize;
334 u64 actual_end;
335 int ret = 0;
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
343 int i;
344 int will_compress;
345
346 orig_start = start;
347
348again:
349 will_compress = 0;
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
352
353 actual_end = min_t(u64, i_size_read(inode), end + 1);
354 total_compressed = actual_end - start;
355
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
361 *
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
365 */
366 total_compressed = min(total_compressed, max_uncompressed);
367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
368 num_bytes = max(blocksize, num_bytes);
369 disk_num_bytes = num_bytes;
370 total_in = 0;
371 ret = 0;
372
373 /*
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
377 */
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
380 WARN_ON(pages);
381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
382
383 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
384 total_compressed, pages,
385 nr_pages, &nr_pages_ret,
386 &total_in,
387 &total_compressed,
388 max_compressed);
389
390 if (!ret) {
391 unsigned long offset = total_compressed &
392 (PAGE_CACHE_SIZE - 1);
393 struct page *page = pages[nr_pages_ret - 1];
394 char *kaddr;
395
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
398 */
399 if (offset) {
400 kaddr = kmap_atomic(page, KM_USER0);
401 memset(kaddr + offset, 0,
402 PAGE_CACHE_SIZE - offset);
403 kunmap_atomic(kaddr, KM_USER0);
404 }
405 will_compress = 1;
406 }
407 }
408 if (start == 0) {
409 trans = btrfs_join_transaction(root, 1);
410 BUG_ON(!trans);
411 btrfs_set_trans_block_group(trans, inode);
412
413 /* lets try to make an inline extent */
414 if (ret || total_in < (actual_end - start)) {
415 /* we didn't compress the entire range, try
416 * to make an uncompressed inline extent.
417 */
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
420 } else {
421 /* try making a compressed inline extent */
422 ret = cow_file_range_inline(trans, root, inode,
423 start, end,
424 total_compressed, pages);
425 }
426 btrfs_end_transaction(trans, root);
427 if (ret == 0) {
428 /*
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
432 */
433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
435 start, end, NULL, 1, 0,
436 0, 1, 1, 1);
437 ret = 0;
438 goto free_pages_out;
439 }
440 }
441
442 if (will_compress) {
443 /*
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
446 * things
447 */
448 total_compressed = (total_compressed + blocksize - 1) &
449 ~(blocksize - 1);
450
451 /*
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
454 */
455 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
456 ~(PAGE_CACHE_SIZE - 1);
457 if (total_compressed >= total_in) {
458 will_compress = 0;
459 } else {
460 disk_num_bytes = total_compressed;
461 num_bytes = total_in;
462 }
463 }
464 if (!will_compress && pages) {
465 /*
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
468 */
469 for (i = 0; i < nr_pages_ret; i++) {
470 WARN_ON(pages[i]->mapping);
471 page_cache_release(pages[i]);
472 }
473 kfree(pages);
474 pages = NULL;
475 total_compressed = 0;
476 nr_pages_ret = 0;
477
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode, NOCOMPRESS);
480 }
481 if (will_compress) {
482 *num_added += 1;
483
484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
487 */
488 add_async_extent(async_cow, start, num_bytes,
489 total_compressed, pages, nr_pages_ret);
490
491 if (start + num_bytes < end) {
492 start += num_bytes;
493 pages = NULL;
494 cond_resched();
495 goto again;
496 }
497 } else {
498 /*
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
504 */
505 if (page_offset(locked_page) >= start &&
506 page_offset(locked_page) <= end) {
507 __set_page_dirty_nobuffers(locked_page);
508 /* unlocked later on in the async handlers */
509 }
510 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
511 *num_added += 1;
512 }
513
514out:
515 return 0;
516
517free_pages_out:
518 for (i = 0; i < nr_pages_ret; i++) {
519 WARN_ON(pages[i]->mapping);
520 page_cache_release(pages[i]);
521 }
522 if (pages)
523 kfree(pages);
524
525 goto out;
526}
527
528/*
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
533 */
534static noinline int submit_compressed_extents(struct inode *inode,
535 struct async_cow *async_cow)
536{
537 struct async_extent *async_extent;
538 u64 alloc_hint = 0;
539 struct btrfs_trans_handle *trans;
540 struct btrfs_key ins;
541 struct extent_map *em;
542 struct btrfs_root *root = BTRFS_I(inode)->root;
543 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
544 struct extent_io_tree *io_tree;
545 int ret;
546
547 if (list_empty(&async_cow->extents))
548 return 0;
549
550 trans = btrfs_join_transaction(root, 1);
551
552 while(!list_empty(&async_cow->extents)) {
553 async_extent = list_entry(async_cow->extents.next,
554 struct async_extent, list);
555 list_del(&async_extent->list);
556
557 io_tree = &BTRFS_I(inode)->io_tree;
558
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent->pages) {
561 int page_started = 0;
562 unsigned long nr_written = 0;
563
564 lock_extent(io_tree, async_extent->start,
565 async_extent->start + async_extent->ram_size - 1,
566 GFP_NOFS);
567
568 /* allocate blocks */
569 cow_file_range(inode, async_cow->locked_page,
570 async_extent->start,
571 async_extent->start +
572 async_extent->ram_size - 1,
573 &page_started, &nr_written, 0);
574
575 /*
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
580 */
581 if (!page_started)
582 extent_write_locked_range(io_tree,
583 inode, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1,
586 btrfs_get_extent,
587 WB_SYNC_ALL);
588 kfree(async_extent);
589 cond_resched();
590 continue;
591 }
592
593 lock_extent(io_tree, async_extent->start,
594 async_extent->start + async_extent->ram_size - 1,
595 GFP_NOFS);
596 /*
597 * here we're doing allocation and writeback of the
598 * compressed pages
599 */
600 btrfs_drop_extent_cache(inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1, 0);
603
604 ret = btrfs_reserve_extent(trans, root,
605 async_extent->compressed_size,
606 async_extent->compressed_size,
607 0, alloc_hint,
608 (u64)-1, &ins, 1);
609 BUG_ON(ret);
610 em = alloc_extent_map(GFP_NOFS);
611 em->start = async_extent->start;
612 em->len = async_extent->ram_size;
613 em->orig_start = em->start;
614
615 em->block_start = ins.objectid;
616 em->block_len = ins.offset;
617 em->bdev = root->fs_info->fs_devices->latest_bdev;
618 set_bit(EXTENT_FLAG_PINNED, &em->flags);
619 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
620
621 while(1) {
622 spin_lock(&em_tree->lock);
623 ret = add_extent_mapping(em_tree, em);
624 spin_unlock(&em_tree->lock);
625 if (ret != -EEXIST) {
626 free_extent_map(em);
627 break;
628 }
629 btrfs_drop_extent_cache(inode, async_extent->start,
630 async_extent->start +
631 async_extent->ram_size - 1, 0);
632 }
633
634 ret = btrfs_add_ordered_extent(inode, async_extent->start,
635 ins.objectid,
636 async_extent->ram_size,
637 ins.offset,
638 BTRFS_ORDERED_COMPRESSED);
639 BUG_ON(ret);
640
641 btrfs_end_transaction(trans, root);
642
643 /*
644 * clear dirty, set writeback and unlock the pages.
645 */
646 extent_clear_unlock_delalloc(inode,
647 &BTRFS_I(inode)->io_tree,
648 async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 NULL, 1, 1, 0, 1, 1, 0);
652
653 ret = btrfs_submit_compressed_write(inode,
654 async_extent->start,
655 async_extent->ram_size,
656 ins.objectid,
657 ins.offset, async_extent->pages,
658 async_extent->nr_pages);
659
660 BUG_ON(ret);
661 trans = btrfs_join_transaction(root, 1);
662 alloc_hint = ins.objectid + ins.offset;
663 kfree(async_extent);
664 cond_resched();
665 }
666
667 btrfs_end_transaction(trans, root);
668 return 0;
669}
670
671/*
672 * when extent_io.c finds a delayed allocation range in the file,
673 * the call backs end up in this code. The basic idea is to
674 * allocate extents on disk for the range, and create ordered data structs
675 * in ram to track those extents.
676 *
677 * locked_page is the page that writepage had locked already. We use
678 * it to make sure we don't do extra locks or unlocks.
679 *
680 * *page_started is set to one if we unlock locked_page and do everything
681 * required to start IO on it. It may be clean and already done with
682 * IO when we return.
683 */
684static noinline int cow_file_range(struct inode *inode,
685 struct page *locked_page,
686 u64 start, u64 end, int *page_started,
687 unsigned long *nr_written,
688 int unlock)
689{
690 struct btrfs_root *root = BTRFS_I(inode)->root;
691 struct btrfs_trans_handle *trans;
692 u64 alloc_hint = 0;
693 u64 num_bytes;
694 unsigned long ram_size;
695 u64 disk_num_bytes;
696 u64 cur_alloc_size;
697 u64 blocksize = root->sectorsize;
698 u64 actual_end;
699 struct btrfs_key ins;
700 struct extent_map *em;
701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
702 int ret = 0;
703
704 trans = btrfs_join_transaction(root, 1);
705 BUG_ON(!trans);
706 btrfs_set_trans_block_group(trans, inode);
707
708 actual_end = min_t(u64, i_size_read(inode), end + 1);
709
710 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
711 num_bytes = max(blocksize, num_bytes);
712 disk_num_bytes = num_bytes;
713 ret = 0;
714
715 if (start == 0) {
716 /* lets try to make an inline extent */
717 ret = cow_file_range_inline(trans, root, inode,
718 start, end, 0, NULL);
719 if (ret == 0) {
720 extent_clear_unlock_delalloc(inode,
721 &BTRFS_I(inode)->io_tree,
722 start, end, NULL, 1, 1,
723 1, 1, 1, 1);
724 *nr_written = *nr_written +
725 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
726 *page_started = 1;
727 ret = 0;
728 goto out;
729 }
730 }
731
732 BUG_ON(disk_num_bytes >
733 btrfs_super_total_bytes(&root->fs_info->super_copy));
734
735 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
736
737 while(disk_num_bytes > 0) {
738 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
739 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
740 root->sectorsize, 0, alloc_hint,
741 (u64)-1, &ins, 1);
742 if (ret) {
743 BUG();
744 }
745 em = alloc_extent_map(GFP_NOFS);
746 em->start = start;
747 em->orig_start = em->start;
748
749 ram_size = ins.offset;
750 em->len = ins.offset;
751
752 em->block_start = ins.objectid;
753 em->block_len = ins.offset;
754 em->bdev = root->fs_info->fs_devices->latest_bdev;
755 set_bit(EXTENT_FLAG_PINNED, &em->flags);
756
757 while(1) {
758 spin_lock(&em_tree->lock);
759 ret = add_extent_mapping(em_tree, em);
760 spin_unlock(&em_tree->lock);
761 if (ret != -EEXIST) {
762 free_extent_map(em);
763 break;
764 }
765 btrfs_drop_extent_cache(inode, start,
766 start + ram_size - 1, 0);
767 }
768
769 cur_alloc_size = ins.offset;
770 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771 ram_size, cur_alloc_size, 0);
772 BUG_ON(ret);
773
774 if (disk_num_bytes < cur_alloc_size) {
775 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
776 cur_alloc_size);
777 break;
778 }
779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
782 */
783 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784 start, start + ram_size - 1,
785 locked_page, unlock, 1,
786 1, 0, 0, 0);
787 disk_num_bytes -= cur_alloc_size;
788 num_bytes -= cur_alloc_size;
789 alloc_hint = ins.objectid + ins.offset;
790 start += cur_alloc_size;
791 }
792out:
793 ret = 0;
794 btrfs_end_transaction(trans, root);
795
796 return ret;
797}
798
799/*
800 * work queue call back to started compression on a file and pages
801 */
802static noinline void async_cow_start(struct btrfs_work *work)
803{
804 struct async_cow *async_cow;
805 int num_added = 0;
806 async_cow = container_of(work, struct async_cow, work);
807
808 compress_file_range(async_cow->inode, async_cow->locked_page,
809 async_cow->start, async_cow->end, async_cow,
810 &num_added);
811 if (num_added == 0)
812 async_cow->inode = NULL;
813}
814
815/*
816 * work queue call back to submit previously compressed pages
817 */
818static noinline void async_cow_submit(struct btrfs_work *work)
819{
820 struct async_cow *async_cow;
821 struct btrfs_root *root;
822 unsigned long nr_pages;
823
824 async_cow = container_of(work, struct async_cow, work);
825
826 root = async_cow->root;
827 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828 PAGE_CACHE_SHIFT;
829
830 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
831
832 if (atomic_read(&root->fs_info->async_delalloc_pages) <
833 5 * 1042 * 1024 &&
834 waitqueue_active(&root->fs_info->async_submit_wait))
835 wake_up(&root->fs_info->async_submit_wait);
836
837 if (async_cow->inode) {
838 submit_compressed_extents(async_cow->inode, async_cow);
839 }
840}
841
842static noinline void async_cow_free(struct btrfs_work *work)
843{
844 struct async_cow *async_cow;
845 async_cow = container_of(work, struct async_cow, work);
846 kfree(async_cow);
847}
848
849static int cow_file_range_async(struct inode *inode, struct page *locked_page,
850 u64 start, u64 end, int *page_started,
851 unsigned long *nr_written)
852{
853 struct async_cow *async_cow;
854 struct btrfs_root *root = BTRFS_I(inode)->root;
855 unsigned long nr_pages;
856 u64 cur_end;
857 int limit = 10 * 1024 * 1042;
858
859 if (!btrfs_test_opt(root, COMPRESS)) {
860 return cow_file_range(inode, locked_page, start, end,
861 page_started, nr_written, 1);
862 }
863
864 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
865 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
866 while(start < end) {
867 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
868 async_cow->inode = inode;
869 async_cow->root = root;
870 async_cow->locked_page = locked_page;
871 async_cow->start = start;
872
873 if (btrfs_test_flag(inode, NOCOMPRESS))
874 cur_end = end;
875 else
876 cur_end = min(end, start + 512 * 1024 - 1);
877
878 async_cow->end = cur_end;
879 INIT_LIST_HEAD(&async_cow->extents);
880
881 async_cow->work.func = async_cow_start;
882 async_cow->work.ordered_func = async_cow_submit;
883 async_cow->work.ordered_free = async_cow_free;
884 async_cow->work.flags = 0;
885
886 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
887 PAGE_CACHE_SHIFT;
888 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
889
890 btrfs_queue_worker(&root->fs_info->delalloc_workers,
891 &async_cow->work);
892
893 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
894 wait_event(root->fs_info->async_submit_wait,
895 (atomic_read(&root->fs_info->async_delalloc_pages) <
896 limit));
897 }
898
899 while(atomic_read(&root->fs_info->async_submit_draining) &&
900 atomic_read(&root->fs_info->async_delalloc_pages)) {
901 wait_event(root->fs_info->async_submit_wait,
902 (atomic_read(&root->fs_info->async_delalloc_pages) ==
903 0));
904 }
905
906 *nr_written += nr_pages;
907 start = cur_end + 1;
908 }
909 *page_started = 1;
910 return 0;
911}
912
913/*
914 * when nowcow writeback call back. This checks for snapshots or COW copies
915 * of the extents that exist in the file, and COWs the file as required.
916 *
917 * If no cow copies or snapshots exist, we write directly to the existing
918 * blocks on disk
919 */
920static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
921 u64 start, u64 end, int *page_started, int force,
922 unsigned long *nr_written)
923{
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 struct btrfs_trans_handle *trans;
926 struct extent_buffer *leaf;
927 struct btrfs_path *path;
928 struct btrfs_file_extent_item *fi;
929 struct btrfs_key found_key;
930 u64 cow_start;
931 u64 cur_offset;
932 u64 extent_end;
933 u64 disk_bytenr;
934 u64 num_bytes;
935 int extent_type;
936 int ret;
937 int type;
938 int nocow;
939 int check_prev = 1;
940
941 path = btrfs_alloc_path();
942 BUG_ON(!path);
943 trans = btrfs_join_transaction(root, 1);
944 BUG_ON(!trans);
945
946 cow_start = (u64)-1;
947 cur_offset = start;
948 while (1) {
949 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
950 cur_offset, 0);
951 BUG_ON(ret < 0);
952 if (ret > 0 && path->slots[0] > 0 && check_prev) {
953 leaf = path->nodes[0];
954 btrfs_item_key_to_cpu(leaf, &found_key,
955 path->slots[0] - 1);
956 if (found_key.objectid == inode->i_ino &&
957 found_key.type == BTRFS_EXTENT_DATA_KEY)
958 path->slots[0]--;
959 }
960 check_prev = 0;
961next_slot:
962 leaf = path->nodes[0];
963 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
964 ret = btrfs_next_leaf(root, path);
965 if (ret < 0)
966 BUG_ON(1);
967 if (ret > 0)
968 break;
969 leaf = path->nodes[0];
970 }
971
972 nocow = 0;
973 disk_bytenr = 0;
974 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
975
976 if (found_key.objectid > inode->i_ino ||
977 found_key.type > BTRFS_EXTENT_DATA_KEY ||
978 found_key.offset > end)
979 break;
980
981 if (found_key.offset > cur_offset) {
982 extent_end = found_key.offset;
983 goto out_check;
984 }
985
986 fi = btrfs_item_ptr(leaf, path->slots[0],
987 struct btrfs_file_extent_item);
988 extent_type = btrfs_file_extent_type(leaf, fi);
989
990 if (extent_type == BTRFS_FILE_EXTENT_REG ||
991 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
992 struct btrfs_block_group_cache *block_group;
993 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
994 extent_end = found_key.offset +
995 btrfs_file_extent_num_bytes(leaf, fi);
996 if (extent_end <= start) {
997 path->slots[0]++;
998 goto next_slot;
999 }
1000 if (btrfs_file_extent_compression(leaf, fi) ||
1001 btrfs_file_extent_encryption(leaf, fi) ||
1002 btrfs_file_extent_other_encoding(leaf, fi))
1003 goto out_check;
1004 if (disk_bytenr == 0)
1005 goto out_check;
1006 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1007 goto out_check;
1008 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1009 goto out_check;
1010 block_group = btrfs_lookup_block_group(root->fs_info,
1011 disk_bytenr);
1012 if (!block_group || block_group->ro)
1013 goto out_check;
1014 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1015 nocow = 1;
1016 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1017 extent_end = found_key.offset +
1018 btrfs_file_extent_inline_len(leaf, fi);
1019 extent_end = ALIGN(extent_end, root->sectorsize);
1020 } else {
1021 BUG_ON(1);
1022 }
1023out_check:
1024 if (extent_end <= start) {
1025 path->slots[0]++;
1026 goto next_slot;
1027 }
1028 if (!nocow) {
1029 if (cow_start == (u64)-1)
1030 cow_start = cur_offset;
1031 cur_offset = extent_end;
1032 if (cur_offset > end)
1033 break;
1034 path->slots[0]++;
1035 goto next_slot;
1036 }
1037
1038 btrfs_release_path(root, path);
1039 if (cow_start != (u64)-1) {
1040 ret = cow_file_range(inode, locked_page, cow_start,
1041 found_key.offset - 1, page_started,
1042 nr_written, 1);
1043 BUG_ON(ret);
1044 cow_start = (u64)-1;
1045 }
1046
1047 disk_bytenr += cur_offset - found_key.offset;
1048 num_bytes = min(end + 1, extent_end) - cur_offset;
1049 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1050 struct extent_map *em;
1051 struct extent_map_tree *em_tree;
1052 em_tree = &BTRFS_I(inode)->extent_tree;
1053 em = alloc_extent_map(GFP_NOFS);
1054 em->start = cur_offset;
1055 em->orig_start = em->start;
1056 em->len = num_bytes;
1057 em->block_len = num_bytes;
1058 em->block_start = disk_bytenr;
1059 em->bdev = root->fs_info->fs_devices->latest_bdev;
1060 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1061 while (1) {
1062 spin_lock(&em_tree->lock);
1063 ret = add_extent_mapping(em_tree, em);
1064 spin_unlock(&em_tree->lock);
1065 if (ret != -EEXIST) {
1066 free_extent_map(em);
1067 break;
1068 }
1069 btrfs_drop_extent_cache(inode, em->start,
1070 em->start + em->len - 1, 0);
1071 }
1072 type = BTRFS_ORDERED_PREALLOC;
1073 } else {
1074 type = BTRFS_ORDERED_NOCOW;
1075 }
1076
1077 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1078 num_bytes, num_bytes, type);
1079 BUG_ON(ret);
1080
1081 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1082 cur_offset, cur_offset + num_bytes - 1,
1083 locked_page, 1, 1, 1, 0, 0, 0);
1084 cur_offset = extent_end;
1085 if (cur_offset > end)
1086 break;
1087 }
1088 btrfs_release_path(root, path);
1089
1090 if (cur_offset <= end && cow_start == (u64)-1)
1091 cow_start = cur_offset;
1092 if (cow_start != (u64)-1) {
1093 ret = cow_file_range(inode, locked_page, cow_start, end,
1094 page_started, nr_written, 1);
1095 BUG_ON(ret);
1096 }
1097
1098 ret = btrfs_end_transaction(trans, root);
1099 BUG_ON(ret);
1100 btrfs_free_path(path);
1101 return 0;
1102}
1103
1104/*
1105 * extent_io.c call back to do delayed allocation processing
1106 */
1107static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1108 u64 start, u64 end, int *page_started,
1109 unsigned long *nr_written)
1110{
1111 struct btrfs_root *root = BTRFS_I(inode)->root;
1112 int ret;
1113
1114 if (btrfs_test_opt(root, NODATACOW) ||
1115 btrfs_test_flag(inode, NODATACOW))
1116 ret = run_delalloc_nocow(inode, locked_page, start, end,
1117 page_started, 0, nr_written);
1118 else if (btrfs_test_flag(inode, PREALLOC))
1119 ret = run_delalloc_nocow(inode, locked_page, start, end,
1120 page_started, 1, nr_written);
1121 else
1122 ret = cow_file_range_async(inode, locked_page, start, end,
1123 page_started, nr_written);
1124
1125 return ret;
1126}
1127
1128/*
1129 * extent_io.c set_bit_hook, used to track delayed allocation
1130 * bytes in this file, and to maintain the list of inodes that
1131 * have pending delalloc work to be done.
1132 */
1133int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1134 unsigned long old, unsigned long bits)
1135{
1136 unsigned long flags;
1137 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1140 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1141 root->fs_info->delalloc_bytes += end - start + 1;
1142 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1143 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1144 &root->fs_info->delalloc_inodes);
1145 }
1146 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1147 }
1148 return 0;
1149}
1150
1151/*
1152 * extent_io.c clear_bit_hook, see set_bit_hook for why
1153 */
1154int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1155 unsigned long old, unsigned long bits)
1156{
1157 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1158 struct btrfs_root *root = BTRFS_I(inode)->root;
1159 unsigned long flags;
1160
1161 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1162 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1163 printk("warning: delalloc account %Lu %Lu\n",
1164 end - start + 1, root->fs_info->delalloc_bytes);
1165 root->fs_info->delalloc_bytes = 0;
1166 BTRFS_I(inode)->delalloc_bytes = 0;
1167 } else {
1168 root->fs_info->delalloc_bytes -= end - start + 1;
1169 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1170 }
1171 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1172 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1173 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1174 }
1175 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1176 }
1177 return 0;
1178}
1179
1180/*
1181 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1182 * we don't create bios that span stripes or chunks
1183 */
1184int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1185 size_t size, struct bio *bio,
1186 unsigned long bio_flags)
1187{
1188 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1189 struct btrfs_mapping_tree *map_tree;
1190 u64 logical = (u64)bio->bi_sector << 9;
1191 u64 length = 0;
1192 u64 map_length;
1193 int ret;
1194
1195 if (bio_flags & EXTENT_BIO_COMPRESSED)
1196 return 0;
1197
1198 length = bio->bi_size;
1199 map_tree = &root->fs_info->mapping_tree;
1200 map_length = length;
1201 ret = btrfs_map_block(map_tree, READ, logical,
1202 &map_length, NULL, 0);
1203
1204 if (map_length < length + size) {
1205 return 1;
1206 }
1207 return 0;
1208}
1209
1210/*
1211 * in order to insert checksums into the metadata in large chunks,
1212 * we wait until bio submission time. All the pages in the bio are
1213 * checksummed and sums are attached onto the ordered extent record.
1214 *
1215 * At IO completion time the cums attached on the ordered extent record
1216 * are inserted into the btree
1217 */
1218int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1219 int mirror_num, unsigned long bio_flags)
1220{
1221 struct btrfs_root *root = BTRFS_I(inode)->root;
1222 int ret = 0;
1223
1224 ret = btrfs_csum_one_bio(root, inode, bio);
1225 BUG_ON(ret);
1226 return 0;
1227}
1228
1229/*
1230 * in order to insert checksums into the metadata in large chunks,
1231 * we wait until bio submission time. All the pages in the bio are
1232 * checksummed and sums are attached onto the ordered extent record.
1233 *
1234 * At IO completion time the cums attached on the ordered extent record
1235 * are inserted into the btree
1236 */
1237int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1238 int mirror_num, unsigned long bio_flags)
1239{
1240 struct btrfs_root *root = BTRFS_I(inode)->root;
1241 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1242}
1243
1244/*
1245 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1246 * or reading the csums from the tree before a read
1247 */
1248int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1249 int mirror_num, unsigned long bio_flags)
1250{
1251 struct btrfs_root *root = BTRFS_I(inode)->root;
1252 int ret = 0;
1253 int skip_sum;
1254
1255 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1256 BUG_ON(ret);
1257
1258 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1259 btrfs_test_flag(inode, NODATASUM);
1260
1261 if (!(rw & (1 << BIO_RW))) {
1262
1263 if (bio_flags & EXTENT_BIO_COMPRESSED)
1264 return btrfs_submit_compressed_read(inode, bio,
1265 mirror_num, bio_flags);
1266 else if (!skip_sum)
1267 btrfs_lookup_bio_sums(root, inode, bio);
1268 goto mapit;
1269 } else if (!skip_sum) {
1270 /* we're doing a write, do the async checksumming */
1271 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1272 inode, rw, bio, mirror_num,
1273 bio_flags, __btrfs_submit_bio_start,
1274 __btrfs_submit_bio_done);
1275 }
1276
1277mapit:
1278 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1279}
1280
1281/*
1282 * given a list of ordered sums record them in the inode. This happens
1283 * at IO completion time based on sums calculated at bio submission time.
1284 */
1285static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1286 struct inode *inode, u64 file_offset,
1287 struct list_head *list)
1288{
1289 struct list_head *cur;
1290 struct btrfs_ordered_sum *sum;
1291
1292 btrfs_set_trans_block_group(trans, inode);
1293 list_for_each(cur, list) {
1294 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1295 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1296 inode, sum);
1297 }
1298 return 0;
1299}
1300
1301int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1302{
1303 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1304 WARN_ON(1);
1305 }
1306 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1307 GFP_NOFS);
1308}
1309
1310/* see btrfs_writepage_start_hook for details on why this is required */
1311struct btrfs_writepage_fixup {
1312 struct page *page;
1313 struct btrfs_work work;
1314};
1315
1316void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1317{
1318 struct btrfs_writepage_fixup *fixup;
1319 struct btrfs_ordered_extent *ordered;
1320 struct page *page;
1321 struct inode *inode;
1322 u64 page_start;
1323 u64 page_end;
1324
1325 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1326 page = fixup->page;
1327again:
1328 lock_page(page);
1329 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1330 ClearPageChecked(page);
1331 goto out_page;
1332 }
1333
1334 inode = page->mapping->host;
1335 page_start = page_offset(page);
1336 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1337
1338 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1339
1340 /* already ordered? We're done */
1341 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1342 EXTENT_ORDERED, 0)) {
1343 goto out;
1344 }
1345
1346 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1347 if (ordered) {
1348 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1349 page_end, GFP_NOFS);
1350 unlock_page(page);
1351 btrfs_start_ordered_extent(inode, ordered, 1);
1352 goto again;
1353 }
1354
1355 btrfs_set_extent_delalloc(inode, page_start, page_end);
1356 ClearPageChecked(page);
1357out:
1358 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1359out_page:
1360 unlock_page(page);
1361 page_cache_release(page);
1362}
1363
1364/*
1365 * There are a few paths in the higher layers of the kernel that directly
1366 * set the page dirty bit without asking the filesystem if it is a
1367 * good idea. This causes problems because we want to make sure COW
1368 * properly happens and the data=ordered rules are followed.
1369 *
1370 * In our case any range that doesn't have the ORDERED bit set
1371 * hasn't been properly setup for IO. We kick off an async process
1372 * to fix it up. The async helper will wait for ordered extents, set
1373 * the delalloc bit and make it safe to write the page.
1374 */
1375int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1376{
1377 struct inode *inode = page->mapping->host;
1378 struct btrfs_writepage_fixup *fixup;
1379 struct btrfs_root *root = BTRFS_I(inode)->root;
1380 int ret;
1381
1382 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1383 EXTENT_ORDERED, 0);
1384 if (ret)
1385 return 0;
1386
1387 if (PageChecked(page))
1388 return -EAGAIN;
1389
1390 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1391 if (!fixup)
1392 return -EAGAIN;
1393
1394 SetPageChecked(page);
1395 page_cache_get(page);
1396 fixup->work.func = btrfs_writepage_fixup_worker;
1397 fixup->page = page;
1398 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1399 return -EAGAIN;
1400}
1401
1402static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1403 struct inode *inode, u64 file_pos,
1404 u64 disk_bytenr, u64 disk_num_bytes,
1405 u64 num_bytes, u64 ram_bytes,
1406 u8 compression, u8 encryption,
1407 u16 other_encoding, int extent_type)
1408{
1409 struct btrfs_root *root = BTRFS_I(inode)->root;
1410 struct btrfs_file_extent_item *fi;
1411 struct btrfs_path *path;
1412 struct extent_buffer *leaf;
1413 struct btrfs_key ins;
1414 u64 hint;
1415 int ret;
1416
1417 path = btrfs_alloc_path();
1418 BUG_ON(!path);
1419
1420 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1421 file_pos + num_bytes, file_pos, &hint);
1422 BUG_ON(ret);
1423
1424 ins.objectid = inode->i_ino;
1425 ins.offset = file_pos;
1426 ins.type = BTRFS_EXTENT_DATA_KEY;
1427 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1428 BUG_ON(ret);
1429 leaf = path->nodes[0];
1430 fi = btrfs_item_ptr(leaf, path->slots[0],
1431 struct btrfs_file_extent_item);
1432 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1433 btrfs_set_file_extent_type(leaf, fi, extent_type);
1434 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1435 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1436 btrfs_set_file_extent_offset(leaf, fi, 0);
1437 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1438 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1439 btrfs_set_file_extent_compression(leaf, fi, compression);
1440 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1441 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1442 btrfs_mark_buffer_dirty(leaf);
1443
1444 inode_add_bytes(inode, num_bytes);
1445 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1446
1447 ins.objectid = disk_bytenr;
1448 ins.offset = disk_num_bytes;
1449 ins.type = BTRFS_EXTENT_ITEM_KEY;
1450 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1451 root->root_key.objectid,
1452 trans->transid, inode->i_ino, &ins);
1453 BUG_ON(ret);
1454
1455 btrfs_free_path(path);
1456 return 0;
1457}
1458
1459/* as ordered data IO finishes, this gets called so we can finish
1460 * an ordered extent if the range of bytes in the file it covers are
1461 * fully written.
1462 */
1463static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1464{
1465 struct btrfs_root *root = BTRFS_I(inode)->root;
1466 struct btrfs_trans_handle *trans;
1467 struct btrfs_ordered_extent *ordered_extent;
1468 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1469 int compressed = 0;
1470 int ret;
1471
1472 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1473 if (!ret)
1474 return 0;
1475
1476 trans = btrfs_join_transaction(root, 1);
1477
1478 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1479 BUG_ON(!ordered_extent);
1480 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1481 goto nocow;
1482
1483 lock_extent(io_tree, ordered_extent->file_offset,
1484 ordered_extent->file_offset + ordered_extent->len - 1,
1485 GFP_NOFS);
1486
1487 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1488 compressed = 1;
1489 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1490 BUG_ON(compressed);
1491 ret = btrfs_mark_extent_written(trans, root, inode,
1492 ordered_extent->file_offset,
1493 ordered_extent->file_offset +
1494 ordered_extent->len);
1495 BUG_ON(ret);
1496 } else {
1497 ret = insert_reserved_file_extent(trans, inode,
1498 ordered_extent->file_offset,
1499 ordered_extent->start,
1500 ordered_extent->disk_len,
1501 ordered_extent->len,
1502 ordered_extent->len,
1503 compressed, 0, 0,
1504 BTRFS_FILE_EXTENT_REG);
1505 BUG_ON(ret);
1506 }
1507 unlock_extent(io_tree, ordered_extent->file_offset,
1508 ordered_extent->file_offset + ordered_extent->len - 1,
1509 GFP_NOFS);
1510nocow:
1511 add_pending_csums(trans, inode, ordered_extent->file_offset,
1512 &ordered_extent->list);
1513
1514 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1515 btrfs_ordered_update_i_size(inode, ordered_extent);
1516 btrfs_update_inode(trans, root, inode);
1517 btrfs_remove_ordered_extent(inode, ordered_extent);
1518 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1519
1520 /* once for us */
1521 btrfs_put_ordered_extent(ordered_extent);
1522 /* once for the tree */
1523 btrfs_put_ordered_extent(ordered_extent);
1524
1525 btrfs_end_transaction(trans, root);
1526 return 0;
1527}
1528
1529int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1530 struct extent_state *state, int uptodate)
1531{
1532 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1533}
1534
1535/*
1536 * When IO fails, either with EIO or csum verification fails, we
1537 * try other mirrors that might have a good copy of the data. This
1538 * io_failure_record is used to record state as we go through all the
1539 * mirrors. If another mirror has good data, the page is set up to date
1540 * and things continue. If a good mirror can't be found, the original
1541 * bio end_io callback is called to indicate things have failed.
1542 */
1543struct io_failure_record {
1544 struct page *page;
1545 u64 start;
1546 u64 len;
1547 u64 logical;
1548 int last_mirror;
1549};
1550
1551int btrfs_io_failed_hook(struct bio *failed_bio,
1552 struct page *page, u64 start, u64 end,
1553 struct extent_state *state)
1554{
1555 struct io_failure_record *failrec = NULL;
1556 u64 private;
1557 struct extent_map *em;
1558 struct inode *inode = page->mapping->host;
1559 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1560 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1561 struct bio *bio;
1562 int num_copies;
1563 int ret;
1564 int rw;
1565 u64 logical;
1566 unsigned long bio_flags = 0;
1567
1568 ret = get_state_private(failure_tree, start, &private);
1569 if (ret) {
1570 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1571 if (!failrec)
1572 return -ENOMEM;
1573 failrec->start = start;
1574 failrec->len = end - start + 1;
1575 failrec->last_mirror = 0;
1576
1577 spin_lock(&em_tree->lock);
1578 em = lookup_extent_mapping(em_tree, start, failrec->len);
1579 if (em->start > start || em->start + em->len < start) {
1580 free_extent_map(em);
1581 em = NULL;
1582 }
1583 spin_unlock(&em_tree->lock);
1584
1585 if (!em || IS_ERR(em)) {
1586 kfree(failrec);
1587 return -EIO;
1588 }
1589 logical = start - em->start;
1590 logical = em->block_start + logical;
1591 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1592 bio_flags = EXTENT_BIO_COMPRESSED;
1593 failrec->logical = logical;
1594 free_extent_map(em);
1595 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1596 EXTENT_DIRTY, GFP_NOFS);
1597 set_state_private(failure_tree, start,
1598 (u64)(unsigned long)failrec);
1599 } else {
1600 failrec = (struct io_failure_record *)(unsigned long)private;
1601 }
1602 num_copies = btrfs_num_copies(
1603 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1604 failrec->logical, failrec->len);
1605 failrec->last_mirror++;
1606 if (!state) {
1607 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1608 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1609 failrec->start,
1610 EXTENT_LOCKED);
1611 if (state && state->start != failrec->start)
1612 state = NULL;
1613 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1614 }
1615 if (!state || failrec->last_mirror > num_copies) {
1616 set_state_private(failure_tree, failrec->start, 0);
1617 clear_extent_bits(failure_tree, failrec->start,
1618 failrec->start + failrec->len - 1,
1619 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1620 kfree(failrec);
1621 return -EIO;
1622 }
1623 bio = bio_alloc(GFP_NOFS, 1);
1624 bio->bi_private = state;
1625 bio->bi_end_io = failed_bio->bi_end_io;
1626 bio->bi_sector = failrec->logical >> 9;
1627 bio->bi_bdev = failed_bio->bi_bdev;
1628 bio->bi_size = 0;
1629 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1630 if (failed_bio->bi_rw & (1 << BIO_RW))
1631 rw = WRITE;
1632 else
1633 rw = READ;
1634
1635 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1636 failrec->last_mirror,
1637 bio_flags);
1638 return 0;
1639}
1640
1641/*
1642 * each time an IO finishes, we do a fast check in the IO failure tree
1643 * to see if we need to process or clean up an io_failure_record
1644 */
1645int btrfs_clean_io_failures(struct inode *inode, u64 start)
1646{
1647 u64 private;
1648 u64 private_failure;
1649 struct io_failure_record *failure;
1650 int ret;
1651
1652 private = 0;
1653 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1654 (u64)-1, 1, EXTENT_DIRTY)) {
1655 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1656 start, &private_failure);
1657 if (ret == 0) {
1658 failure = (struct io_failure_record *)(unsigned long)
1659 private_failure;
1660 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1661 failure->start, 0);
1662 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1663 failure->start,
1664 failure->start + failure->len - 1,
1665 EXTENT_DIRTY | EXTENT_LOCKED,
1666 GFP_NOFS);
1667 kfree(failure);
1668 }
1669 }
1670 return 0;
1671}
1672
1673/*
1674 * when reads are done, we need to check csums to verify the data is correct
1675 * if there's a match, we allow the bio to finish. If not, we go through
1676 * the io_failure_record routines to find good copies
1677 */
1678int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1679 struct extent_state *state)
1680{
1681 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1682 struct inode *inode = page->mapping->host;
1683 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1684 char *kaddr;
1685 u64 private = ~(u32)0;
1686 int ret;
1687 struct btrfs_root *root = BTRFS_I(inode)->root;
1688 u32 csum = ~(u32)0;
1689 unsigned long flags;
1690
1691 if (btrfs_test_opt(root, NODATASUM) ||
1692 btrfs_test_flag(inode, NODATASUM))
1693 return 0;
1694 if (state && state->start == start) {
1695 private = state->private;
1696 ret = 0;
1697 } else {
1698 ret = get_state_private(io_tree, start, &private);
1699 }
1700 local_irq_save(flags);
1701 kaddr = kmap_atomic(page, KM_IRQ0);
1702 if (ret) {
1703 goto zeroit;
1704 }
1705 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1706 btrfs_csum_final(csum, (char *)&csum);
1707 if (csum != private) {
1708 goto zeroit;
1709 }
1710 kunmap_atomic(kaddr, KM_IRQ0);
1711 local_irq_restore(flags);
1712
1713 /* if the io failure tree for this inode is non-empty,
1714 * check to see if we've recovered from a failed IO
1715 */
1716 btrfs_clean_io_failures(inode, start);
1717 return 0;
1718
1719zeroit:
1720 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1721 page->mapping->host->i_ino, (unsigned long long)start, csum,
1722 private);
1723 memset(kaddr + offset, 1, end - start + 1);
1724 flush_dcache_page(page);
1725 kunmap_atomic(kaddr, KM_IRQ0);
1726 local_irq_restore(flags);
1727 if (private == 0)
1728 return 0;
1729 return -EIO;
1730}
1731
1732/*
1733 * This creates an orphan entry for the given inode in case something goes
1734 * wrong in the middle of an unlink/truncate.
1735 */
1736int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1737{
1738 struct btrfs_root *root = BTRFS_I(inode)->root;
1739 int ret = 0;
1740
1741 spin_lock(&root->list_lock);
1742
1743 /* already on the orphan list, we're good */
1744 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1745 spin_unlock(&root->list_lock);
1746 return 0;
1747 }
1748
1749 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1750
1751 spin_unlock(&root->list_lock);
1752
1753 /*
1754 * insert an orphan item to track this unlinked/truncated file
1755 */
1756 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1757
1758 return ret;
1759}
1760
1761/*
1762 * We have done the truncate/delete so we can go ahead and remove the orphan
1763 * item for this particular inode.
1764 */
1765int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1766{
1767 struct btrfs_root *root = BTRFS_I(inode)->root;
1768 int ret = 0;
1769
1770 spin_lock(&root->list_lock);
1771
1772 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1773 spin_unlock(&root->list_lock);
1774 return 0;
1775 }
1776
1777 list_del_init(&BTRFS_I(inode)->i_orphan);
1778 if (!trans) {
1779 spin_unlock(&root->list_lock);
1780 return 0;
1781 }
1782
1783 spin_unlock(&root->list_lock);
1784
1785 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1786
1787 return ret;
1788}
1789
1790/*
1791 * this cleans up any orphans that may be left on the list from the last use
1792 * of this root.
1793 */
1794void btrfs_orphan_cleanup(struct btrfs_root *root)
1795{
1796 struct btrfs_path *path;
1797 struct extent_buffer *leaf;
1798 struct btrfs_item *item;
1799 struct btrfs_key key, found_key;
1800 struct btrfs_trans_handle *trans;
1801 struct inode *inode;
1802 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1803
1804 path = btrfs_alloc_path();
1805 if (!path)
1806 return;
1807 path->reada = -1;
1808
1809 key.objectid = BTRFS_ORPHAN_OBJECTID;
1810 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1811 key.offset = (u64)-1;
1812
1813
1814 while (1) {
1815 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1816 if (ret < 0) {
1817 printk(KERN_ERR "Error searching slot for orphan: %d"
1818 "\n", ret);
1819 break;
1820 }
1821
1822 /*
1823 * if ret == 0 means we found what we were searching for, which
1824 * is weird, but possible, so only screw with path if we didnt
1825 * find the key and see if we have stuff that matches
1826 */
1827 if (ret > 0) {
1828 if (path->slots[0] == 0)
1829 break;
1830 path->slots[0]--;
1831 }
1832
1833 /* pull out the item */
1834 leaf = path->nodes[0];
1835 item = btrfs_item_nr(leaf, path->slots[0]);
1836 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1837
1838 /* make sure the item matches what we want */
1839 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1840 break;
1841 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1842 break;
1843
1844 /* release the path since we're done with it */
1845 btrfs_release_path(root, path);
1846
1847 /*
1848 * this is where we are basically btrfs_lookup, without the
1849 * crossing root thing. we store the inode number in the
1850 * offset of the orphan item.
1851 */
1852 inode = btrfs_iget_locked(root->fs_info->sb,
1853 found_key.offset, root);
1854 if (!inode)
1855 break;
1856
1857 if (inode->i_state & I_NEW) {
1858 BTRFS_I(inode)->root = root;
1859
1860 /* have to set the location manually */
1861 BTRFS_I(inode)->location.objectid = inode->i_ino;
1862 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1863 BTRFS_I(inode)->location.offset = 0;
1864
1865 btrfs_read_locked_inode(inode);
1866 unlock_new_inode(inode);
1867 }
1868
1869 /*
1870 * add this inode to the orphan list so btrfs_orphan_del does
1871 * the proper thing when we hit it
1872 */
1873 spin_lock(&root->list_lock);
1874 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1875 spin_unlock(&root->list_lock);
1876
1877 /*
1878 * if this is a bad inode, means we actually succeeded in
1879 * removing the inode, but not the orphan record, which means
1880 * we need to manually delete the orphan since iput will just
1881 * do a destroy_inode
1882 */
1883 if (is_bad_inode(inode)) {
1884 trans = btrfs_start_transaction(root, 1);
1885 btrfs_orphan_del(trans, inode);
1886 btrfs_end_transaction(trans, root);
1887 iput(inode);
1888 continue;
1889 }
1890
1891 /* if we have links, this was a truncate, lets do that */
1892 if (inode->i_nlink) {
1893 nr_truncate++;
1894 btrfs_truncate(inode);
1895 } else {
1896 nr_unlink++;
1897 }
1898
1899 /* this will do delete_inode and everything for us */
1900 iput(inode);
1901 }
1902
1903 if (nr_unlink)
1904 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1905 if (nr_truncate)
1906 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1907
1908 btrfs_free_path(path);
1909}
1910
1911/*
1912 * read an inode from the btree into the in-memory inode
1913 */
1914void btrfs_read_locked_inode(struct inode *inode)
1915{
1916 struct btrfs_path *path;
1917 struct extent_buffer *leaf;
1918 struct btrfs_inode_item *inode_item;
1919 struct btrfs_timespec *tspec;
1920 struct btrfs_root *root = BTRFS_I(inode)->root;
1921 struct btrfs_key location;
1922 u64 alloc_group_block;
1923 u32 rdev;
1924 int ret;
1925
1926 path = btrfs_alloc_path();
1927 BUG_ON(!path);
1928 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1929
1930 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1931 if (ret)
1932 goto make_bad;
1933
1934 leaf = path->nodes[0];
1935 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1936 struct btrfs_inode_item);
1937
1938 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1939 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1940 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1941 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1942 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1943
1944 tspec = btrfs_inode_atime(inode_item);
1945 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1946 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1947
1948 tspec = btrfs_inode_mtime(inode_item);
1949 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1950 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1951
1952 tspec = btrfs_inode_ctime(inode_item);
1953 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1954 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1955
1956 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1957 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1958 inode->i_generation = BTRFS_I(inode)->generation;
1959 inode->i_rdev = 0;
1960 rdev = btrfs_inode_rdev(leaf, inode_item);
1961
1962 BTRFS_I(inode)->index_cnt = (u64)-1;
1963
1964 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1965 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1966 alloc_group_block);
1967 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1968 if (!BTRFS_I(inode)->block_group) {
1969 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1970 NULL, 0,
1971 BTRFS_BLOCK_GROUP_METADATA, 0);
1972 }
1973 btrfs_free_path(path);
1974 inode_item = NULL;
1975
1976 switch (inode->i_mode & S_IFMT) {
1977 case S_IFREG:
1978 inode->i_mapping->a_ops = &btrfs_aops;
1979 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1980 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1981 inode->i_fop = &btrfs_file_operations;
1982 inode->i_op = &btrfs_file_inode_operations;
1983 break;
1984 case S_IFDIR:
1985 inode->i_fop = &btrfs_dir_file_operations;
1986 if (root == root->fs_info->tree_root)
1987 inode->i_op = &btrfs_dir_ro_inode_operations;
1988 else
1989 inode->i_op = &btrfs_dir_inode_operations;
1990 break;
1991 case S_IFLNK:
1992 inode->i_op = &btrfs_symlink_inode_operations;
1993 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1994 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1995 break;
1996 default:
1997 init_special_inode(inode, inode->i_mode, rdev);
1998 break;
1999 }
2000 return;
2001
2002make_bad:
2003 btrfs_free_path(path);
2004 make_bad_inode(inode);
2005}
2006
2007/*
2008 * given a leaf and an inode, copy the inode fields into the leaf
2009 */
2010static void fill_inode_item(struct btrfs_trans_handle *trans,
2011 struct extent_buffer *leaf,
2012 struct btrfs_inode_item *item,
2013 struct inode *inode)
2014{
2015 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2016 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2017 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2018 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2019 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2020
2021 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2022 inode->i_atime.tv_sec);
2023 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2024 inode->i_atime.tv_nsec);
2025
2026 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2027 inode->i_mtime.tv_sec);
2028 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2029 inode->i_mtime.tv_nsec);
2030
2031 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2032 inode->i_ctime.tv_sec);
2033 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2034 inode->i_ctime.tv_nsec);
2035
2036 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2037 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2038 btrfs_set_inode_transid(leaf, item, trans->transid);
2039 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2040 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2041 btrfs_set_inode_block_group(leaf, item,
2042 BTRFS_I(inode)->block_group->key.objectid);
2043}
2044
2045/*
2046 * copy everything in the in-memory inode into the btree.
2047 */
2048int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2049 struct btrfs_root *root,
2050 struct inode *inode)
2051{
2052 struct btrfs_inode_item *inode_item;
2053 struct btrfs_path *path;
2054 struct extent_buffer *leaf;
2055 int ret;
2056
2057 path = btrfs_alloc_path();
2058 BUG_ON(!path);
2059 ret = btrfs_lookup_inode(trans, root, path,
2060 &BTRFS_I(inode)->location, 1);
2061 if (ret) {
2062 if (ret > 0)
2063 ret = -ENOENT;
2064 goto failed;
2065 }
2066
2067 leaf = path->nodes[0];
2068 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2069 struct btrfs_inode_item);
2070
2071 fill_inode_item(trans, leaf, inode_item, inode);
2072 btrfs_mark_buffer_dirty(leaf);
2073 btrfs_set_inode_last_trans(trans, inode);
2074 ret = 0;
2075failed:
2076 btrfs_free_path(path);
2077 return ret;
2078}
2079
2080
2081/*
2082 * unlink helper that gets used here in inode.c and in the tree logging
2083 * recovery code. It remove a link in a directory with a given name, and
2084 * also drops the back refs in the inode to the directory
2085 */
2086int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct inode *dir, struct inode *inode,
2089 const char *name, int name_len)
2090{
2091 struct btrfs_path *path;
2092 int ret = 0;
2093 struct extent_buffer *leaf;
2094 struct btrfs_dir_item *di;
2095 struct btrfs_key key;
2096 u64 index;
2097
2098 path = btrfs_alloc_path();
2099 if (!path) {
2100 ret = -ENOMEM;
2101 goto err;
2102 }
2103
2104 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2105 name, name_len, -1);
2106 if (IS_ERR(di)) {
2107 ret = PTR_ERR(di);
2108 goto err;
2109 }
2110 if (!di) {
2111 ret = -ENOENT;
2112 goto err;
2113 }
2114 leaf = path->nodes[0];
2115 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2116 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2117 if (ret)
2118 goto err;
2119 btrfs_release_path(root, path);
2120
2121 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2122 inode->i_ino,
2123 dir->i_ino, &index);
2124 if (ret) {
2125 printk("failed to delete reference to %.*s, "
2126 "inode %lu parent %lu\n", name_len, name,
2127 inode->i_ino, dir->i_ino);
2128 goto err;
2129 }
2130
2131 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2132 index, name, name_len, -1);
2133 if (IS_ERR(di)) {
2134 ret = PTR_ERR(di);
2135 goto err;
2136 }
2137 if (!di) {
2138 ret = -ENOENT;
2139 goto err;
2140 }
2141 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2142 btrfs_release_path(root, path);
2143
2144 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2145 inode, dir->i_ino);
2146 BUG_ON(ret != 0 && ret != -ENOENT);
2147 if (ret != -ENOENT)
2148 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2149
2150 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2151 dir, index);
2152 BUG_ON(ret);
2153err:
2154 btrfs_free_path(path);
2155 if (ret)
2156 goto out;
2157
2158 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2159 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2160 btrfs_update_inode(trans, root, dir);
2161 btrfs_drop_nlink(inode);
2162 ret = btrfs_update_inode(trans, root, inode);
2163 dir->i_sb->s_dirt = 1;
2164out:
2165 return ret;
2166}
2167
2168static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2169{
2170 struct btrfs_root *root;
2171 struct btrfs_trans_handle *trans;
2172 struct inode *inode = dentry->d_inode;
2173 int ret;
2174 unsigned long nr = 0;
2175
2176 root = BTRFS_I(dir)->root;
2177
2178 ret = btrfs_check_free_space(root, 1, 1);
2179 if (ret)
2180 goto fail;
2181
2182 trans = btrfs_start_transaction(root, 1);
2183
2184 btrfs_set_trans_block_group(trans, dir);
2185 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2186 dentry->d_name.name, dentry->d_name.len);
2187
2188 if (inode->i_nlink == 0)
2189 ret = btrfs_orphan_add(trans, inode);
2190
2191 nr = trans->blocks_used;
2192
2193 btrfs_end_transaction_throttle(trans, root);
2194fail:
2195 btrfs_btree_balance_dirty(root, nr);
2196 return ret;
2197}
2198
2199static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2200{
2201 struct inode *inode = dentry->d_inode;
2202 int err = 0;
2203 int ret;
2204 struct btrfs_root *root = BTRFS_I(dir)->root;
2205 struct btrfs_trans_handle *trans;
2206 unsigned long nr = 0;
2207
2208 /*
2209 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2210 * the root of a subvolume or snapshot
2211 */
2212 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2213 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2214 return -ENOTEMPTY;
2215 }
2216
2217 ret = btrfs_check_free_space(root, 1, 1);
2218 if (ret)
2219 goto fail;
2220
2221 trans = btrfs_start_transaction(root, 1);
2222 btrfs_set_trans_block_group(trans, dir);
2223
2224 err = btrfs_orphan_add(trans, inode);
2225 if (err)
2226 goto fail_trans;
2227
2228 /* now the directory is empty */
2229 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2230 dentry->d_name.name, dentry->d_name.len);
2231 if (!err) {
2232 btrfs_i_size_write(inode, 0);
2233 }
2234
2235fail_trans:
2236 nr = trans->blocks_used;
2237 ret = btrfs_end_transaction_throttle(trans, root);
2238fail:
2239 btrfs_btree_balance_dirty(root, nr);
2240
2241 if (ret && !err)
2242 err = ret;
2243 return err;
2244}
2245
2246/*
2247 * when truncating bytes in a file, it is possible to avoid reading
2248 * the leaves that contain only checksum items. This can be the
2249 * majority of the IO required to delete a large file, but it must
2250 * be done carefully.
2251 *
2252 * The keys in the level just above the leaves are checked to make sure
2253 * the lowest key in a given leaf is a csum key, and starts at an offset
2254 * after the new size.
2255 *
2256 * Then the key for the next leaf is checked to make sure it also has
2257 * a checksum item for the same file. If it does, we know our target leaf
2258 * contains only checksum items, and it can be safely freed without reading
2259 * it.
2260 *
2261 * This is just an optimization targeted at large files. It may do
2262 * nothing. It will return 0 unless things went badly.
2263 */
2264static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2265 struct btrfs_root *root,
2266 struct btrfs_path *path,
2267 struct inode *inode, u64 new_size)
2268{
2269 struct btrfs_key key;
2270 int ret;
2271 int nritems;
2272 struct btrfs_key found_key;
2273 struct btrfs_key other_key;
2274 struct btrfs_leaf_ref *ref;
2275 u64 leaf_gen;
2276 u64 leaf_start;
2277
2278 path->lowest_level = 1;
2279 key.objectid = inode->i_ino;
2280 key.type = BTRFS_CSUM_ITEM_KEY;
2281 key.offset = new_size;
2282again:
2283 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2284 if (ret < 0)
2285 goto out;
2286
2287 if (path->nodes[1] == NULL) {
2288 ret = 0;
2289 goto out;
2290 }
2291 ret = 0;
2292 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2293 nritems = btrfs_header_nritems(path->nodes[1]);
2294
2295 if (!nritems)
2296 goto out;
2297
2298 if (path->slots[1] >= nritems)
2299 goto next_node;
2300
2301 /* did we find a key greater than anything we want to delete? */
2302 if (found_key.objectid > inode->i_ino ||
2303 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2304 goto out;
2305
2306 /* we check the next key in the node to make sure the leave contains
2307 * only checksum items. This comparison doesn't work if our
2308 * leaf is the last one in the node
2309 */
2310 if (path->slots[1] + 1 >= nritems) {
2311next_node:
2312 /* search forward from the last key in the node, this
2313 * will bring us into the next node in the tree
2314 */
2315 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2316
2317 /* unlikely, but we inc below, so check to be safe */
2318 if (found_key.offset == (u64)-1)
2319 goto out;
2320
2321 /* search_forward needs a path with locks held, do the
2322 * search again for the original key. It is possible
2323 * this will race with a balance and return a path that
2324 * we could modify, but this drop is just an optimization
2325 * and is allowed to miss some leaves.
2326 */
2327 btrfs_release_path(root, path);
2328 found_key.offset++;
2329
2330 /* setup a max key for search_forward */
2331 other_key.offset = (u64)-1;
2332 other_key.type = key.type;
2333 other_key.objectid = key.objectid;
2334
2335 path->keep_locks = 1;
2336 ret = btrfs_search_forward(root, &found_key, &other_key,
2337 path, 0, 0);
2338 path->keep_locks = 0;
2339 if (ret || found_key.objectid != key.objectid ||
2340 found_key.type != key.type) {
2341 ret = 0;
2342 goto out;
2343 }
2344
2345 key.offset = found_key.offset;
2346 btrfs_release_path(root, path);
2347 cond_resched();
2348 goto again;
2349 }
2350
2351 /* we know there's one more slot after us in the tree,
2352 * read that key so we can verify it is also a checksum item
2353 */
2354 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2355
2356 if (found_key.objectid < inode->i_ino)
2357 goto next_key;
2358
2359 if (found_key.type != key.type || found_key.offset < new_size)
2360 goto next_key;
2361
2362 /*
2363 * if the key for the next leaf isn't a csum key from this objectid,
2364 * we can't be sure there aren't good items inside this leaf.
2365 * Bail out
2366 */
2367 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2368 goto out;
2369
2370 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2371 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2372 /*
2373 * it is safe to delete this leaf, it contains only
2374 * csum items from this inode at an offset >= new_size
2375 */
2376 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2377 BUG_ON(ret);
2378
2379 if (root->ref_cows && leaf_gen < trans->transid) {
2380 ref = btrfs_alloc_leaf_ref(root, 0);
2381 if (ref) {
2382 ref->root_gen = root->root_key.offset;
2383 ref->bytenr = leaf_start;
2384 ref->owner = 0;
2385 ref->generation = leaf_gen;
2386 ref->nritems = 0;
2387
2388 ret = btrfs_add_leaf_ref(root, ref, 0);
2389 WARN_ON(ret);
2390 btrfs_free_leaf_ref(root, ref);
2391 } else {
2392 WARN_ON(1);
2393 }
2394 }
2395next_key:
2396 btrfs_release_path(root, path);
2397
2398 if (other_key.objectid == inode->i_ino &&
2399 other_key.type == key.type && other_key.offset > key.offset) {
2400 key.offset = other_key.offset;
2401 cond_resched();
2402 goto again;
2403 }
2404 ret = 0;
2405out:
2406 /* fixup any changes we've made to the path */
2407 path->lowest_level = 0;
2408 path->keep_locks = 0;
2409 btrfs_release_path(root, path);
2410 return ret;
2411}
2412
2413/*
2414 * this can truncate away extent items, csum items and directory items.
2415 * It starts at a high offset and removes keys until it can't find
2416 * any higher than new_size
2417 *
2418 * csum items that cross the new i_size are truncated to the new size
2419 * as well.
2420 *
2421 * min_type is the minimum key type to truncate down to. If set to 0, this
2422 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2423 */
2424noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2425 struct btrfs_root *root,
2426 struct inode *inode,
2427 u64 new_size, u32 min_type)
2428{
2429 int ret;
2430 struct btrfs_path *path;
2431 struct btrfs_key key;
2432 struct btrfs_key found_key;
2433 u32 found_type;
2434 struct extent_buffer *leaf;
2435 struct btrfs_file_extent_item *fi;
2436 u64 extent_start = 0;
2437 u64 extent_num_bytes = 0;
2438 u64 item_end = 0;
2439 u64 root_gen = 0;
2440 u64 root_owner = 0;
2441 int found_extent;
2442 int del_item;
2443 int pending_del_nr = 0;
2444 int pending_del_slot = 0;
2445 int extent_type = -1;
2446 int encoding;
2447 u64 mask = root->sectorsize - 1;
2448
2449 if (root->ref_cows)
2450 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2451 path = btrfs_alloc_path();
2452 path->reada = -1;
2453 BUG_ON(!path);
2454
2455 /* FIXME, add redo link to tree so we don't leak on crash */
2456 key.objectid = inode->i_ino;
2457 key.offset = (u64)-1;
2458 key.type = (u8)-1;
2459
2460 btrfs_init_path(path);
2461
2462 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2463 BUG_ON(ret);
2464
2465search_again:
2466 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2467 if (ret < 0) {
2468 goto error;
2469 }
2470 if (ret > 0) {
2471 /* there are no items in the tree for us to truncate, we're
2472 * done
2473 */
2474 if (path->slots[0] == 0) {
2475 ret = 0;
2476 goto error;
2477 }
2478 path->slots[0]--;
2479 }
2480
2481 while(1) {
2482 fi = NULL;
2483 leaf = path->nodes[0];
2484 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2485 found_type = btrfs_key_type(&found_key);
2486 encoding = 0;
2487
2488 if (found_key.objectid != inode->i_ino)
2489 break;
2490
2491 if (found_type < min_type)
2492 break;
2493
2494 item_end = found_key.offset;
2495 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2496 fi = btrfs_item_ptr(leaf, path->slots[0],
2497 struct btrfs_file_extent_item);
2498 extent_type = btrfs_file_extent_type(leaf, fi);
2499 encoding = btrfs_file_extent_compression(leaf, fi);
2500 encoding |= btrfs_file_extent_encryption(leaf, fi);
2501 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2502
2503 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2504 item_end +=
2505 btrfs_file_extent_num_bytes(leaf, fi);
2506 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2507 item_end += btrfs_file_extent_inline_len(leaf,
2508 fi);
2509 }
2510 item_end--;
2511 }
2512 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2513 ret = btrfs_csum_truncate(trans, root, path,
2514 new_size);
2515 BUG_ON(ret);
2516 }
2517 if (item_end < new_size) {
2518 if (found_type == BTRFS_DIR_ITEM_KEY) {
2519 found_type = BTRFS_INODE_ITEM_KEY;
2520 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2521 found_type = BTRFS_CSUM_ITEM_KEY;
2522 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2523 found_type = BTRFS_XATTR_ITEM_KEY;
2524 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2525 found_type = BTRFS_INODE_REF_KEY;
2526 } else if (found_type) {
2527 found_type--;
2528 } else {
2529 break;
2530 }
2531 btrfs_set_key_type(&key, found_type);
2532 goto next;
2533 }
2534 if (found_key.offset >= new_size)
2535 del_item = 1;
2536 else
2537 del_item = 0;
2538 found_extent = 0;
2539
2540 /* FIXME, shrink the extent if the ref count is only 1 */
2541 if (found_type != BTRFS_EXTENT_DATA_KEY)
2542 goto delete;
2543
2544 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2545 u64 num_dec;
2546 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2547 if (!del_item && !encoding) {
2548 u64 orig_num_bytes =
2549 btrfs_file_extent_num_bytes(leaf, fi);
2550 extent_num_bytes = new_size -
2551 found_key.offset + root->sectorsize - 1;
2552 extent_num_bytes = extent_num_bytes &
2553 ~((u64)root->sectorsize - 1);
2554 btrfs_set_file_extent_num_bytes(leaf, fi,
2555 extent_num_bytes);
2556 num_dec = (orig_num_bytes -
2557 extent_num_bytes);
2558 if (root->ref_cows && extent_start != 0)
2559 inode_sub_bytes(inode, num_dec);
2560 btrfs_mark_buffer_dirty(leaf);
2561 } else {
2562 extent_num_bytes =
2563 btrfs_file_extent_disk_num_bytes(leaf,
2564 fi);
2565 /* FIXME blocksize != 4096 */
2566 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2567 if (extent_start != 0) {
2568 found_extent = 1;
2569 if (root->ref_cows)
2570 inode_sub_bytes(inode, num_dec);
2571 }
2572 root_gen = btrfs_header_generation(leaf);
2573 root_owner = btrfs_header_owner(leaf);
2574 }
2575 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2576 /*
2577 * we can't truncate inline items that have had
2578 * special encodings
2579 */
2580 if (!del_item &&
2581 btrfs_file_extent_compression(leaf, fi) == 0 &&
2582 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2583 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2584 u32 size = new_size - found_key.offset;
2585
2586 if (root->ref_cows) {
2587 inode_sub_bytes(inode, item_end + 1 -
2588 new_size);
2589 }
2590 size =
2591 btrfs_file_extent_calc_inline_size(size);
2592 ret = btrfs_truncate_item(trans, root, path,
2593 size, 1);
2594 BUG_ON(ret);
2595 } else if (root->ref_cows) {
2596 inode_sub_bytes(inode, item_end + 1 -
2597 found_key.offset);
2598 }
2599 }
2600delete:
2601 if (del_item) {
2602 if (!pending_del_nr) {
2603 /* no pending yet, add ourselves */
2604 pending_del_slot = path->slots[0];
2605 pending_del_nr = 1;
2606 } else if (pending_del_nr &&
2607 path->slots[0] + 1 == pending_del_slot) {
2608 /* hop on the pending chunk */
2609 pending_del_nr++;
2610 pending_del_slot = path->slots[0];
2611 } else {
2612 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2613 }
2614 } else {
2615 break;
2616 }
2617 if (found_extent) {
2618 ret = btrfs_free_extent(trans, root, extent_start,
2619 extent_num_bytes,
2620 leaf->start, root_owner,
2621 root_gen, inode->i_ino, 0);
2622 BUG_ON(ret);
2623 }
2624next:
2625 if (path->slots[0] == 0) {
2626 if (pending_del_nr)
2627 goto del_pending;
2628 btrfs_release_path(root, path);
2629 goto search_again;
2630 }
2631
2632 path->slots[0]--;
2633 if (pending_del_nr &&
2634 path->slots[0] + 1 != pending_del_slot) {
2635 struct btrfs_key debug;
2636del_pending:
2637 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2638 pending_del_slot);
2639 ret = btrfs_del_items(trans, root, path,
2640 pending_del_slot,
2641 pending_del_nr);
2642 BUG_ON(ret);
2643 pending_del_nr = 0;
2644 btrfs_release_path(root, path);
2645 goto search_again;
2646 }
2647 }
2648 ret = 0;
2649error:
2650 if (pending_del_nr) {
2651 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2652 pending_del_nr);
2653 }
2654 btrfs_free_path(path);
2655 inode->i_sb->s_dirt = 1;
2656 return ret;
2657}
2658
2659/*
2660 * taken from block_truncate_page, but does cow as it zeros out
2661 * any bytes left in the last page in the file.
2662 */
2663static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2664{
2665 struct inode *inode = mapping->host;
2666 struct btrfs_root *root = BTRFS_I(inode)->root;
2667 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2668 struct btrfs_ordered_extent *ordered;
2669 char *kaddr;
2670 u32 blocksize = root->sectorsize;
2671 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2672 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2673 struct page *page;
2674 int ret = 0;
2675 u64 page_start;
2676 u64 page_end;
2677
2678 if ((offset & (blocksize - 1)) == 0)
2679 goto out;
2680
2681 ret = -ENOMEM;
2682again:
2683 page = grab_cache_page(mapping, index);
2684 if (!page)
2685 goto out;
2686
2687 page_start = page_offset(page);
2688 page_end = page_start + PAGE_CACHE_SIZE - 1;
2689
2690 if (!PageUptodate(page)) {
2691 ret = btrfs_readpage(NULL, page);
2692 lock_page(page);
2693 if (page->mapping != mapping) {
2694 unlock_page(page);
2695 page_cache_release(page);
2696 goto again;
2697 }
2698 if (!PageUptodate(page)) {
2699 ret = -EIO;
2700 goto out_unlock;
2701 }
2702 }
2703 wait_on_page_writeback(page);
2704
2705 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2706 set_page_extent_mapped(page);
2707
2708 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2709 if (ordered) {
2710 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2711 unlock_page(page);
2712 page_cache_release(page);
2713 btrfs_start_ordered_extent(inode, ordered, 1);
2714 btrfs_put_ordered_extent(ordered);
2715 goto again;
2716 }
2717
2718 btrfs_set_extent_delalloc(inode, page_start, page_end);
2719 ret = 0;
2720 if (offset != PAGE_CACHE_SIZE) {
2721 kaddr = kmap(page);
2722 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2723 flush_dcache_page(page);
2724 kunmap(page);
2725 }
2726 ClearPageChecked(page);
2727 set_page_dirty(page);
2728 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2729
2730out_unlock:
2731 unlock_page(page);
2732 page_cache_release(page);
2733out:
2734 return ret;
2735}
2736
2737int btrfs_cont_expand(struct inode *inode, loff_t size)
2738{
2739 struct btrfs_trans_handle *trans;
2740 struct btrfs_root *root = BTRFS_I(inode)->root;
2741 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2742 struct extent_map *em;
2743 u64 mask = root->sectorsize - 1;
2744 u64 hole_start = (inode->i_size + mask) & ~mask;
2745 u64 block_end = (size + mask) & ~mask;
2746 u64 last_byte;
2747 u64 cur_offset;
2748 u64 hole_size;
2749 int err;
2750
2751 if (size <= hole_start)
2752 return 0;
2753
2754 err = btrfs_check_free_space(root, 1, 0);
2755 if (err)
2756 return err;
2757
2758 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2759
2760 while (1) {
2761 struct btrfs_ordered_extent *ordered;
2762 btrfs_wait_ordered_range(inode, hole_start,
2763 block_end - hole_start);
2764 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2765 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2766 if (!ordered)
2767 break;
2768 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2769 btrfs_put_ordered_extent(ordered);
2770 }
2771
2772 trans = btrfs_start_transaction(root, 1);
2773 btrfs_set_trans_block_group(trans, inode);
2774
2775 cur_offset = hole_start;
2776 while (1) {
2777 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2778 block_end - cur_offset, 0);
2779 BUG_ON(IS_ERR(em) || !em);
2780 last_byte = min(extent_map_end(em), block_end);
2781 last_byte = (last_byte + mask) & ~mask;
2782 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2783 u64 hint_byte = 0;
2784 hole_size = last_byte - cur_offset;
2785 err = btrfs_drop_extents(trans, root, inode,
2786 cur_offset,
2787 cur_offset + hole_size,
2788 cur_offset, &hint_byte);
2789 if (err)
2790 break;
2791 err = btrfs_insert_file_extent(trans, root,
2792 inode->i_ino, cur_offset, 0,
2793 0, hole_size, 0, hole_size,
2794 0, 0, 0);
2795 btrfs_drop_extent_cache(inode, hole_start,
2796 last_byte - 1, 0);
2797 }
2798 free_extent_map(em);
2799 cur_offset = last_byte;
2800 if (err || cur_offset >= block_end)
2801 break;
2802 }
2803
2804 btrfs_end_transaction(trans, root);
2805 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2806 return err;
2807}
2808
2809static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2810{
2811 struct inode *inode = dentry->d_inode;
2812 int err;
2813
2814 err = inode_change_ok(inode, attr);
2815 if (err)
2816 return err;
2817
2818 if (S_ISREG(inode->i_mode) &&
2819 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2820 err = btrfs_cont_expand(inode, attr->ia_size);
2821 if (err)
2822 return err;
2823 }
2824
2825 err = inode_setattr(inode, attr);
2826
2827 if (!err && ((attr->ia_valid & ATTR_MODE)))
2828 err = btrfs_acl_chmod(inode);
2829 return err;
2830}
2831
2832void btrfs_delete_inode(struct inode *inode)
2833{
2834 struct btrfs_trans_handle *trans;
2835 struct btrfs_root *root = BTRFS_I(inode)->root;
2836 unsigned long nr;
2837 int ret;
2838
2839 truncate_inode_pages(&inode->i_data, 0);
2840 if (is_bad_inode(inode)) {
2841 btrfs_orphan_del(NULL, inode);
2842 goto no_delete;
2843 }
2844 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2845
2846 btrfs_i_size_write(inode, 0);
2847 trans = btrfs_start_transaction(root, 1);
2848
2849 btrfs_set_trans_block_group(trans, inode);
2850 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2851 if (ret) {
2852 btrfs_orphan_del(NULL, inode);
2853 goto no_delete_lock;
2854 }
2855
2856 btrfs_orphan_del(trans, inode);
2857
2858 nr = trans->blocks_used;
2859 clear_inode(inode);
2860
2861 btrfs_end_transaction(trans, root);
2862 btrfs_btree_balance_dirty(root, nr);
2863 return;
2864
2865no_delete_lock:
2866 nr = trans->blocks_used;
2867 btrfs_end_transaction(trans, root);
2868 btrfs_btree_balance_dirty(root, nr);
2869no_delete:
2870 clear_inode(inode);
2871}
2872
2873/*
2874 * this returns the key found in the dir entry in the location pointer.
2875 * If no dir entries were found, location->objectid is 0.
2876 */
2877static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2878 struct btrfs_key *location)
2879{
2880 const char *name = dentry->d_name.name;
2881 int namelen = dentry->d_name.len;
2882 struct btrfs_dir_item *di;
2883 struct btrfs_path *path;
2884 struct btrfs_root *root = BTRFS_I(dir)->root;
2885 int ret = 0;
2886
2887 path = btrfs_alloc_path();
2888 BUG_ON(!path);
2889
2890 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2891 namelen, 0);
2892 if (IS_ERR(di))
2893 ret = PTR_ERR(di);
2894 if (!di || IS_ERR(di)) {
2895 goto out_err;
2896 }
2897 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2898out:
2899 btrfs_free_path(path);
2900 return ret;
2901out_err:
2902 location->objectid = 0;
2903 goto out;
2904}
2905
2906/*
2907 * when we hit a tree root in a directory, the btrfs part of the inode
2908 * needs to be changed to reflect the root directory of the tree root. This
2909 * is kind of like crossing a mount point.
2910 */
2911static int fixup_tree_root_location(struct btrfs_root *root,
2912 struct btrfs_key *location,
2913 struct btrfs_root **sub_root,
2914 struct dentry *dentry)
2915{
2916 struct btrfs_root_item *ri;
2917
2918 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2919 return 0;
2920 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2921 return 0;
2922
2923 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2924 dentry->d_name.name,
2925 dentry->d_name.len);
2926 if (IS_ERR(*sub_root))
2927 return PTR_ERR(*sub_root);
2928
2929 ri = &(*sub_root)->root_item;
2930 location->objectid = btrfs_root_dirid(ri);
2931 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2932 location->offset = 0;
2933
2934 return 0;
2935}
2936
2937static noinline void init_btrfs_i(struct inode *inode)
2938{
2939 struct btrfs_inode *bi = BTRFS_I(inode);
2940
2941 bi->i_acl = NULL;
2942 bi->i_default_acl = NULL;
2943
2944 bi->generation = 0;
2945 bi->last_trans = 0;
2946 bi->logged_trans = 0;
2947 bi->delalloc_bytes = 0;
2948 bi->disk_i_size = 0;
2949 bi->flags = 0;
2950 bi->index_cnt = (u64)-1;
2951 bi->log_dirty_trans = 0;
2952 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2953 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2954 inode->i_mapping, GFP_NOFS);
2955 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2956 inode->i_mapping, GFP_NOFS);
2957 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2958 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2959 mutex_init(&BTRFS_I(inode)->csum_mutex);
2960 mutex_init(&BTRFS_I(inode)->extent_mutex);
2961 mutex_init(&BTRFS_I(inode)->log_mutex);
2962}
2963
2964static int btrfs_init_locked_inode(struct inode *inode, void *p)
2965{
2966 struct btrfs_iget_args *args = p;
2967 inode->i_ino = args->ino;
2968 init_btrfs_i(inode);
2969 BTRFS_I(inode)->root = args->root;
2970 return 0;
2971}
2972
2973static int btrfs_find_actor(struct inode *inode, void *opaque)
2974{
2975 struct btrfs_iget_args *args = opaque;
2976 return (args->ino == inode->i_ino &&
2977 args->root == BTRFS_I(inode)->root);
2978}
2979
2980struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2981 struct btrfs_root *root, int wait)
2982{
2983 struct inode *inode;
2984 struct btrfs_iget_args args;
2985 args.ino = objectid;
2986 args.root = root;
2987
2988 if (wait) {
2989 inode = ilookup5(s, objectid, btrfs_find_actor,
2990 (void *)&args);
2991 } else {
2992 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2993 (void *)&args);
2994 }
2995 return inode;
2996}
2997
2998struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
2999 struct btrfs_root *root)
3000{
3001 struct inode *inode;
3002 struct btrfs_iget_args args;
3003 args.ino = objectid;
3004 args.root = root;
3005
3006 inode = iget5_locked(s, objectid, btrfs_find_actor,
3007 btrfs_init_locked_inode,
3008 (void *)&args);
3009 return inode;
3010}
3011
3012/* Get an inode object given its location and corresponding root.
3013 * Returns in *is_new if the inode was read from disk
3014 */
3015struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3016 struct btrfs_root *root, int *is_new)
3017{
3018 struct inode *inode;
3019
3020 inode = btrfs_iget_locked(s, location->objectid, root);
3021 if (!inode)
3022 return ERR_PTR(-EACCES);
3023
3024 if (inode->i_state & I_NEW) {
3025 BTRFS_I(inode)->root = root;
3026 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3027 btrfs_read_locked_inode(inode);
3028 unlock_new_inode(inode);
3029 if (is_new)
3030 *is_new = 1;
3031 } else {
3032 if (is_new)
3033 *is_new = 0;
3034 }
3035
3036 return inode;
3037}
3038
3039struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3040{
3041 struct inode * inode;
3042 struct btrfs_inode *bi = BTRFS_I(dir);
3043 struct btrfs_root *root = bi->root;
3044 struct btrfs_root *sub_root = root;
3045 struct btrfs_key location;
3046 int ret, new;
3047
3048 if (dentry->d_name.len > BTRFS_NAME_LEN)
3049 return ERR_PTR(-ENAMETOOLONG);
3050
3051 ret = btrfs_inode_by_name(dir, dentry, &location);
3052
3053 if (ret < 0)
3054 return ERR_PTR(ret);
3055
3056 inode = NULL;
3057 if (location.objectid) {
3058 ret = fixup_tree_root_location(root, &location, &sub_root,
3059 dentry);
3060 if (ret < 0)
3061 return ERR_PTR(ret);
3062 if (ret > 0)
3063 return ERR_PTR(-ENOENT);
3064 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3065 if (IS_ERR(inode))
3066 return ERR_CAST(inode);
3067 }
3068 return inode;
3069}
3070
3071static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3072 struct nameidata *nd)
3073{
3074 struct inode *inode;
3075
3076 if (dentry->d_name.len > BTRFS_NAME_LEN)
3077 return ERR_PTR(-ENAMETOOLONG);
3078
3079 inode = btrfs_lookup_dentry(dir, dentry);
3080 if (IS_ERR(inode))
3081 return ERR_CAST(inode);
3082
3083 return d_splice_alias(inode, dentry);
3084}
3085
3086static unsigned char btrfs_filetype_table[] = {
3087 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3088};
3089
3090static int btrfs_real_readdir(struct file *filp, void *dirent,
3091 filldir_t filldir)
3092{
3093 struct inode *inode = filp->f_dentry->d_inode;
3094 struct btrfs_root *root = BTRFS_I(inode)->root;
3095 struct btrfs_item *item;
3096 struct btrfs_dir_item *di;
3097 struct btrfs_key key;
3098 struct btrfs_key found_key;
3099 struct btrfs_path *path;
3100 int ret;
3101 u32 nritems;
3102 struct extent_buffer *leaf;
3103 int slot;
3104 int advance;
3105 unsigned char d_type;
3106 int over = 0;
3107 u32 di_cur;
3108 u32 di_total;
3109 u32 di_len;
3110 int key_type = BTRFS_DIR_INDEX_KEY;
3111 char tmp_name[32];
3112 char *name_ptr;
3113 int name_len;
3114
3115 /* FIXME, use a real flag for deciding about the key type */
3116 if (root->fs_info->tree_root == root)
3117 key_type = BTRFS_DIR_ITEM_KEY;
3118
3119 /* special case for "." */
3120 if (filp->f_pos == 0) {
3121 over = filldir(dirent, ".", 1,
3122 1, inode->i_ino,
3123 DT_DIR);
3124 if (over)
3125 return 0;
3126 filp->f_pos = 1;
3127 }
3128 /* special case for .., just use the back ref */
3129 if (filp->f_pos == 1) {
3130 u64 pino = parent_ino(filp->f_path.dentry);
3131 over = filldir(dirent, "..", 2,
3132 2, pino, DT_DIR);
3133 if (over)
3134 return 0;
3135 filp->f_pos = 2;
3136 }
3137 path = btrfs_alloc_path();
3138 path->reada = 2;
3139
3140 btrfs_set_key_type(&key, key_type);
3141 key.offset = filp->f_pos;
3142 key.objectid = inode->i_ino;
3143
3144 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3145 if (ret < 0)
3146 goto err;
3147 advance = 0;
3148
3149 while (1) {
3150 leaf = path->nodes[0];
3151 nritems = btrfs_header_nritems(leaf);
3152 slot = path->slots[0];
3153 if (advance || slot >= nritems) {
3154 if (slot >= nritems - 1) {
3155 ret = btrfs_next_leaf(root, path);
3156 if (ret)
3157 break;
3158 leaf = path->nodes[0];
3159 nritems = btrfs_header_nritems(leaf);
3160 slot = path->slots[0];
3161 } else {
3162 slot++;
3163 path->slots[0]++;
3164 }
3165 }
3166
3167 advance = 1;
3168 item = btrfs_item_nr(leaf, slot);
3169 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3170
3171 if (found_key.objectid != key.objectid)
3172 break;
3173 if (btrfs_key_type(&found_key) != key_type)
3174 break;
3175 if (found_key.offset < filp->f_pos)
3176 continue;
3177
3178 filp->f_pos = found_key.offset;
3179
3180 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3181 di_cur = 0;
3182 di_total = btrfs_item_size(leaf, item);
3183
3184 while (di_cur < di_total) {
3185 struct btrfs_key location;
3186
3187 name_len = btrfs_dir_name_len(leaf, di);
3188 if (name_len <= sizeof(tmp_name)) {
3189 name_ptr = tmp_name;
3190 } else {
3191 name_ptr = kmalloc(name_len, GFP_NOFS);
3192 if (!name_ptr) {
3193 ret = -ENOMEM;
3194 goto err;
3195 }
3196 }
3197 read_extent_buffer(leaf, name_ptr,
3198 (unsigned long)(di + 1), name_len);
3199
3200 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3201 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3202
3203 /* is this a reference to our own snapshot? If so
3204 * skip it
3205 */
3206 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3207 location.objectid == root->root_key.objectid) {
3208 over = 0;
3209 goto skip;
3210 }
3211 over = filldir(dirent, name_ptr, name_len,
3212 found_key.offset, location.objectid,
3213 d_type);
3214
3215skip:
3216 if (name_ptr != tmp_name)
3217 kfree(name_ptr);
3218
3219 if (over)
3220 goto nopos;
3221 di_len = btrfs_dir_name_len(leaf, di) +
3222 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3223 di_cur += di_len;
3224 di = (struct btrfs_dir_item *)((char *)di + di_len);
3225 }
3226 }
3227
3228 /* Reached end of directory/root. Bump pos past the last item. */
3229 if (key_type == BTRFS_DIR_INDEX_KEY)
3230 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3231 else
3232 filp->f_pos++;
3233nopos:
3234 ret = 0;
3235err:
3236 btrfs_free_path(path);
3237 return ret;
3238}
3239
3240int btrfs_write_inode(struct inode *inode, int wait)
3241{
3242 struct btrfs_root *root = BTRFS_I(inode)->root;
3243 struct btrfs_trans_handle *trans;
3244 int ret = 0;
3245
3246 if (root->fs_info->btree_inode == inode)
3247 return 0;
3248
3249 if (wait) {
3250 trans = btrfs_join_transaction(root, 1);
3251 btrfs_set_trans_block_group(trans, inode);
3252 ret = btrfs_commit_transaction(trans, root);
3253 }
3254 return ret;
3255}
3256
3257/*
3258 * This is somewhat expensive, updating the tree every time the
3259 * inode changes. But, it is most likely to find the inode in cache.
3260 * FIXME, needs more benchmarking...there are no reasons other than performance
3261 * to keep or drop this code.
3262 */
3263void btrfs_dirty_inode(struct inode *inode)
3264{
3265 struct btrfs_root *root = BTRFS_I(inode)->root;
3266 struct btrfs_trans_handle *trans;
3267
3268 trans = btrfs_join_transaction(root, 1);
3269 btrfs_set_trans_block_group(trans, inode);
3270 btrfs_update_inode(trans, root, inode);
3271 btrfs_end_transaction(trans, root);
3272}
3273
3274/*
3275 * find the highest existing sequence number in a directory
3276 * and then set the in-memory index_cnt variable to reflect
3277 * free sequence numbers
3278 */
3279static int btrfs_set_inode_index_count(struct inode *inode)
3280{
3281 struct btrfs_root *root = BTRFS_I(inode)->root;
3282 struct btrfs_key key, found_key;
3283 struct btrfs_path *path;
3284 struct extent_buffer *leaf;
3285 int ret;
3286
3287 key.objectid = inode->i_ino;
3288 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3289 key.offset = (u64)-1;
3290
3291 path = btrfs_alloc_path();
3292 if (!path)
3293 return -ENOMEM;
3294
3295 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3296 if (ret < 0)
3297 goto out;
3298 /* FIXME: we should be able to handle this */
3299 if (ret == 0)
3300 goto out;
3301 ret = 0;
3302
3303 /*
3304 * MAGIC NUMBER EXPLANATION:
3305 * since we search a directory based on f_pos we have to start at 2
3306 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3307 * else has to start at 2
3308 */
3309 if (path->slots[0] == 0) {
3310 BTRFS_I(inode)->index_cnt = 2;
3311 goto out;
3312 }
3313
3314 path->slots[0]--;
3315
3316 leaf = path->nodes[0];
3317 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3318
3319 if (found_key.objectid != inode->i_ino ||
3320 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3321 BTRFS_I(inode)->index_cnt = 2;
3322 goto out;
3323 }
3324
3325 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3326out:
3327 btrfs_free_path(path);
3328 return ret;
3329}
3330
3331/*
3332 * helper to find a free sequence number in a given directory. This current
3333 * code is very simple, later versions will do smarter things in the btree
3334 */
3335int btrfs_set_inode_index(struct inode *dir, u64 *index)
3336{
3337 int ret = 0;
3338
3339 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3340 ret = btrfs_set_inode_index_count(dir);
3341 if (ret) {
3342 return ret;
3343 }
3344 }
3345
3346 *index = BTRFS_I(dir)->index_cnt;
3347 BTRFS_I(dir)->index_cnt++;
3348
3349 return ret;
3350}
3351
3352static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3353 struct btrfs_root *root,
3354 struct inode *dir,
3355 const char *name, int name_len,
3356 u64 ref_objectid,
3357 u64 objectid,
3358 struct btrfs_block_group_cache *group,
3359 int mode, u64 *index)
3360{
3361 struct inode *inode;
3362 struct btrfs_inode_item *inode_item;
3363 struct btrfs_block_group_cache *new_inode_group;
3364 struct btrfs_key *location;
3365 struct btrfs_path *path;
3366 struct btrfs_inode_ref *ref;
3367 struct btrfs_key key[2];
3368 u32 sizes[2];
3369 unsigned long ptr;
3370 int ret;
3371 int owner;
3372
3373 path = btrfs_alloc_path();
3374 BUG_ON(!path);
3375
3376 inode = new_inode(root->fs_info->sb);
3377 if (!inode)
3378 return ERR_PTR(-ENOMEM);
3379
3380 if (dir) {
3381 ret = btrfs_set_inode_index(dir, index);
3382 if (ret)
3383 return ERR_PTR(ret);
3384 }
3385 /*
3386 * index_cnt is ignored for everything but a dir,
3387 * btrfs_get_inode_index_count has an explanation for the magic
3388 * number
3389 */
3390 init_btrfs_i(inode);
3391 BTRFS_I(inode)->index_cnt = 2;
3392 BTRFS_I(inode)->root = root;
3393 BTRFS_I(inode)->generation = trans->transid;
3394
3395 if (mode & S_IFDIR)
3396 owner = 0;
3397 else
3398 owner = 1;
3399 new_inode_group = btrfs_find_block_group(root, group, 0,
3400 BTRFS_BLOCK_GROUP_METADATA, owner);
3401 if (!new_inode_group) {
3402 printk("find_block group failed\n");
3403 new_inode_group = group;
3404 }
3405 BTRFS_I(inode)->block_group = new_inode_group;
3406
3407 key[0].objectid = objectid;
3408 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3409 key[0].offset = 0;
3410
3411 key[1].objectid = objectid;
3412 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3413 key[1].offset = ref_objectid;
3414
3415 sizes[0] = sizeof(struct btrfs_inode_item);
3416 sizes[1] = name_len + sizeof(*ref);
3417
3418 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3419 if (ret != 0)
3420 goto fail;
3421
3422 if (objectid > root->highest_inode)
3423 root->highest_inode = objectid;
3424
3425 inode->i_uid = current->fsuid;
3426 inode->i_gid = current->fsgid;
3427 inode->i_mode = mode;
3428 inode->i_ino = objectid;
3429 inode_set_bytes(inode, 0);
3430 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3431 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3432 struct btrfs_inode_item);
3433 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3434
3435 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3436 struct btrfs_inode_ref);
3437 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3438 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3439 ptr = (unsigned long)(ref + 1);
3440 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3441
3442 btrfs_mark_buffer_dirty(path->nodes[0]);
3443 btrfs_free_path(path);
3444
3445 location = &BTRFS_I(inode)->location;
3446 location->objectid = objectid;
3447 location->offset = 0;
3448 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3449
3450 insert_inode_hash(inode);
3451 return inode;
3452fail:
3453 if (dir)
3454 BTRFS_I(dir)->index_cnt--;
3455 btrfs_free_path(path);
3456 return ERR_PTR(ret);
3457}
3458
3459static inline u8 btrfs_inode_type(struct inode *inode)
3460{
3461 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3462}
3463
3464/*
3465 * utility function to add 'inode' into 'parent_inode' with
3466 * a give name and a given sequence number.
3467 * if 'add_backref' is true, also insert a backref from the
3468 * inode to the parent directory.
3469 */
3470int btrfs_add_link(struct btrfs_trans_handle *trans,
3471 struct inode *parent_inode, struct inode *inode,
3472 const char *name, int name_len, int add_backref, u64 index)
3473{
3474 int ret;
3475 struct btrfs_key key;
3476 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3477
3478 key.objectid = inode->i_ino;
3479 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3480 key.offset = 0;
3481
3482 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3483 parent_inode->i_ino,
3484 &key, btrfs_inode_type(inode),
3485 index);
3486 if (ret == 0) {
3487 if (add_backref) {
3488 ret = btrfs_insert_inode_ref(trans, root,
3489 name, name_len,
3490 inode->i_ino,
3491 parent_inode->i_ino,
3492 index);
3493 }
3494 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3495 name_len * 2);
3496 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3497 ret = btrfs_update_inode(trans, root, parent_inode);
3498 }
3499 return ret;
3500}
3501
3502static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3503 struct dentry *dentry, struct inode *inode,
3504 int backref, u64 index)
3505{
3506 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3507 inode, dentry->d_name.name,
3508 dentry->d_name.len, backref, index);
3509 if (!err) {
3510 d_instantiate(dentry, inode);
3511 return 0;
3512 }
3513 if (err > 0)
3514 err = -EEXIST;
3515 return err;
3516}
3517
3518static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3519 int mode, dev_t rdev)
3520{
3521 struct btrfs_trans_handle *trans;
3522 struct btrfs_root *root = BTRFS_I(dir)->root;
3523 struct inode *inode = NULL;
3524 int err;
3525 int drop_inode = 0;
3526 u64 objectid;
3527 unsigned long nr = 0;
3528 u64 index = 0;
3529
3530 if (!new_valid_dev(rdev))
3531 return -EINVAL;
3532
3533 err = btrfs_check_free_space(root, 1, 0);
3534 if (err)
3535 goto fail;
3536
3537 trans = btrfs_start_transaction(root, 1);
3538 btrfs_set_trans_block_group(trans, dir);
3539
3540 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3541 if (err) {
3542 err = -ENOSPC;
3543 goto out_unlock;
3544 }
3545
3546 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3547 dentry->d_name.len,
3548 dentry->d_parent->d_inode->i_ino, objectid,
3549 BTRFS_I(dir)->block_group, mode, &index);
3550 err = PTR_ERR(inode);
3551 if (IS_ERR(inode))
3552 goto out_unlock;
3553
3554 err = btrfs_init_acl(inode, dir);
3555 if (err) {
3556 drop_inode = 1;
3557 goto out_unlock;
3558 }
3559
3560 btrfs_set_trans_block_group(trans, inode);
3561 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3562 if (err)
3563 drop_inode = 1;
3564 else {
3565 inode->i_op = &btrfs_special_inode_operations;
3566 init_special_inode(inode, inode->i_mode, rdev);
3567 btrfs_update_inode(trans, root, inode);
3568 }
3569 dir->i_sb->s_dirt = 1;
3570 btrfs_update_inode_block_group(trans, inode);
3571 btrfs_update_inode_block_group(trans, dir);
3572out_unlock:
3573 nr = trans->blocks_used;
3574 btrfs_end_transaction_throttle(trans, root);
3575fail:
3576 if (drop_inode) {
3577 inode_dec_link_count(inode);
3578 iput(inode);
3579 }
3580 btrfs_btree_balance_dirty(root, nr);
3581 return err;
3582}
3583
3584static int btrfs_create(struct inode *dir, struct dentry *dentry,
3585 int mode, struct nameidata *nd)
3586{
3587 struct btrfs_trans_handle *trans;
3588 struct btrfs_root *root = BTRFS_I(dir)->root;
3589 struct inode *inode = NULL;
3590 int err;
3591 int drop_inode = 0;
3592 unsigned long nr = 0;
3593 u64 objectid;
3594 u64 index = 0;
3595
3596 err = btrfs_check_free_space(root, 1, 0);
3597 if (err)
3598 goto fail;
3599 trans = btrfs_start_transaction(root, 1);
3600 btrfs_set_trans_block_group(trans, dir);
3601
3602 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3603 if (err) {
3604 err = -ENOSPC;
3605 goto out_unlock;
3606 }
3607
3608 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3609 dentry->d_name.len,
3610 dentry->d_parent->d_inode->i_ino,
3611 objectid, BTRFS_I(dir)->block_group, mode,
3612 &index);
3613 err = PTR_ERR(inode);
3614 if (IS_ERR(inode))
3615 goto out_unlock;
3616
3617 err = btrfs_init_acl(inode, dir);
3618 if (err) {
3619 drop_inode = 1;
3620 goto out_unlock;
3621 }
3622
3623 btrfs_set_trans_block_group(trans, inode);
3624 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3625 if (err)
3626 drop_inode = 1;
3627 else {
3628 inode->i_mapping->a_ops = &btrfs_aops;
3629 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3630 inode->i_fop = &btrfs_file_operations;
3631 inode->i_op = &btrfs_file_inode_operations;
3632 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3633 }
3634 dir->i_sb->s_dirt = 1;
3635 btrfs_update_inode_block_group(trans, inode);
3636 btrfs_update_inode_block_group(trans, dir);
3637out_unlock:
3638 nr = trans->blocks_used;
3639 btrfs_end_transaction_throttle(trans, root);
3640fail:
3641 if (drop_inode) {
3642 inode_dec_link_count(inode);
3643 iput(inode);
3644 }
3645 btrfs_btree_balance_dirty(root, nr);
3646 return err;
3647}
3648
3649static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3650 struct dentry *dentry)
3651{
3652 struct btrfs_trans_handle *trans;
3653 struct btrfs_root *root = BTRFS_I(dir)->root;
3654 struct inode *inode = old_dentry->d_inode;
3655 u64 index;
3656 unsigned long nr = 0;
3657 int err;
3658 int drop_inode = 0;
3659
3660 if (inode->i_nlink == 0)
3661 return -ENOENT;
3662
3663 btrfs_inc_nlink(inode);
3664 err = btrfs_check_free_space(root, 1, 0);
3665 if (err)
3666 goto fail;
3667 err = btrfs_set_inode_index(dir, &index);
3668 if (err)
3669 goto fail;
3670
3671 trans = btrfs_start_transaction(root, 1);
3672
3673 btrfs_set_trans_block_group(trans, dir);
3674 atomic_inc(&inode->i_count);
3675
3676 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3677
3678 if (err)
3679 drop_inode = 1;
3680
3681 dir->i_sb->s_dirt = 1;
3682 btrfs_update_inode_block_group(trans, dir);
3683 err = btrfs_update_inode(trans, root, inode);
3684
3685 if (err)
3686 drop_inode = 1;
3687
3688 nr = trans->blocks_used;
3689 btrfs_end_transaction_throttle(trans, root);
3690fail:
3691 if (drop_inode) {
3692 inode_dec_link_count(inode);
3693 iput(inode);
3694 }
3695 btrfs_btree_balance_dirty(root, nr);
3696 return err;
3697}
3698
3699static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3700{
3701 struct inode *inode = NULL;
3702 struct btrfs_trans_handle *trans;
3703 struct btrfs_root *root = BTRFS_I(dir)->root;
3704 int err = 0;
3705 int drop_on_err = 0;
3706 u64 objectid = 0;
3707 u64 index = 0;
3708 unsigned long nr = 1;
3709
3710 err = btrfs_check_free_space(root, 1, 0);
3711 if (err)
3712 goto out_unlock;
3713
3714 trans = btrfs_start_transaction(root, 1);
3715 btrfs_set_trans_block_group(trans, dir);
3716
3717 if (IS_ERR(trans)) {
3718 err = PTR_ERR(trans);
3719 goto out_unlock;
3720 }
3721
3722 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3723 if (err) {
3724 err = -ENOSPC;
3725 goto out_unlock;
3726 }
3727
3728 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3729 dentry->d_name.len,
3730 dentry->d_parent->d_inode->i_ino, objectid,
3731 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3732 &index);
3733 if (IS_ERR(inode)) {
3734 err = PTR_ERR(inode);
3735 goto out_fail;
3736 }
3737
3738 drop_on_err = 1;
3739
3740 err = btrfs_init_acl(inode, dir);
3741 if (err)
3742 goto out_fail;
3743
3744 inode->i_op = &btrfs_dir_inode_operations;
3745 inode->i_fop = &btrfs_dir_file_operations;
3746 btrfs_set_trans_block_group(trans, inode);
3747
3748 btrfs_i_size_write(inode, 0);
3749 err = btrfs_update_inode(trans, root, inode);
3750 if (err)
3751 goto out_fail;
3752
3753 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3754 inode, dentry->d_name.name,
3755 dentry->d_name.len, 0, index);
3756 if (err)
3757 goto out_fail;
3758
3759 d_instantiate(dentry, inode);
3760 drop_on_err = 0;
3761 dir->i_sb->s_dirt = 1;
3762 btrfs_update_inode_block_group(trans, inode);
3763 btrfs_update_inode_block_group(trans, dir);
3764
3765out_fail:
3766 nr = trans->blocks_used;
3767 btrfs_end_transaction_throttle(trans, root);
3768
3769out_unlock:
3770 if (drop_on_err)
3771 iput(inode);
3772 btrfs_btree_balance_dirty(root, nr);
3773 return err;
3774}
3775
3776/* helper for btfs_get_extent. Given an existing extent in the tree,
3777 * and an extent that you want to insert, deal with overlap and insert
3778 * the new extent into the tree.
3779 */
3780static int merge_extent_mapping(struct extent_map_tree *em_tree,
3781 struct extent_map *existing,
3782 struct extent_map *em,
3783 u64 map_start, u64 map_len)
3784{
3785 u64 start_diff;
3786
3787 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3788 start_diff = map_start - em->start;
3789 em->start = map_start;
3790 em->len = map_len;
3791 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3792 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3793 em->block_start += start_diff;
3794 em->block_len -= start_diff;
3795 }
3796 return add_extent_mapping(em_tree, em);
3797}
3798
3799static noinline int uncompress_inline(struct btrfs_path *path,
3800 struct inode *inode, struct page *page,
3801 size_t pg_offset, u64 extent_offset,
3802 struct btrfs_file_extent_item *item)
3803{
3804 int ret;
3805 struct extent_buffer *leaf = path->nodes[0];
3806 char *tmp;
3807 size_t max_size;
3808 unsigned long inline_size;
3809 unsigned long ptr;
3810
3811 WARN_ON(pg_offset != 0);
3812 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3813 inline_size = btrfs_file_extent_inline_item_len(leaf,
3814 btrfs_item_nr(leaf, path->slots[0]));
3815 tmp = kmalloc(inline_size, GFP_NOFS);
3816 ptr = btrfs_file_extent_inline_start(item);
3817
3818 read_extent_buffer(leaf, tmp, ptr, inline_size);
3819
3820 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3821 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3822 inline_size, max_size);
3823 if (ret) {
3824 char *kaddr = kmap_atomic(page, KM_USER0);
3825 unsigned long copy_size = min_t(u64,
3826 PAGE_CACHE_SIZE - pg_offset,
3827 max_size - extent_offset);
3828 memset(kaddr + pg_offset, 0, copy_size);
3829 kunmap_atomic(kaddr, KM_USER0);
3830 }
3831 kfree(tmp);
3832 return 0;
3833}
3834
3835/*
3836 * a bit scary, this does extent mapping from logical file offset to the disk.
3837 * the ugly parts come from merging extents from the disk with the
3838 * in-ram representation. This gets more complex because of the data=ordered code,
3839 * where the in-ram extents might be locked pending data=ordered completion.
3840 *
3841 * This also copies inline extents directly into the page.
3842 */
3843struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3844 size_t pg_offset, u64 start, u64 len,
3845 int create)
3846{
3847 int ret;
3848 int err = 0;
3849 u64 bytenr;
3850 u64 extent_start = 0;
3851 u64 extent_end = 0;
3852 u64 objectid = inode->i_ino;
3853 u32 found_type;
3854 struct btrfs_path *path = NULL;
3855 struct btrfs_root *root = BTRFS_I(inode)->root;
3856 struct btrfs_file_extent_item *item;
3857 struct extent_buffer *leaf;
3858 struct btrfs_key found_key;
3859 struct extent_map *em = NULL;
3860 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3861 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3862 struct btrfs_trans_handle *trans = NULL;
3863 int compressed;
3864
3865again:
3866 spin_lock(&em_tree->lock);
3867 em = lookup_extent_mapping(em_tree, start, len);
3868 if (em)
3869 em->bdev = root->fs_info->fs_devices->latest_bdev;
3870 spin_unlock(&em_tree->lock);
3871
3872 if (em) {
3873 if (em->start > start || em->start + em->len <= start)
3874 free_extent_map(em);
3875 else if (em->block_start == EXTENT_MAP_INLINE && page)
3876 free_extent_map(em);
3877 else
3878 goto out;
3879 }
3880 em = alloc_extent_map(GFP_NOFS);
3881 if (!em) {
3882 err = -ENOMEM;
3883 goto out;
3884 }
3885 em->bdev = root->fs_info->fs_devices->latest_bdev;
3886 em->start = EXTENT_MAP_HOLE;
3887 em->orig_start = EXTENT_MAP_HOLE;
3888 em->len = (u64)-1;
3889 em->block_len = (u64)-1;
3890
3891 if (!path) {
3892 path = btrfs_alloc_path();
3893 BUG_ON(!path);
3894 }
3895
3896 ret = btrfs_lookup_file_extent(trans, root, path,
3897 objectid, start, trans != NULL);
3898 if (ret < 0) {
3899 err = ret;
3900 goto out;
3901 }
3902
3903 if (ret != 0) {
3904 if (path->slots[0] == 0)
3905 goto not_found;
3906 path->slots[0]--;
3907 }
3908
3909 leaf = path->nodes[0];
3910 item = btrfs_item_ptr(leaf, path->slots[0],
3911 struct btrfs_file_extent_item);
3912 /* are we inside the extent that was found? */
3913 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3914 found_type = btrfs_key_type(&found_key);
3915 if (found_key.objectid != objectid ||
3916 found_type != BTRFS_EXTENT_DATA_KEY) {
3917 goto not_found;
3918 }
3919
3920 found_type = btrfs_file_extent_type(leaf, item);
3921 extent_start = found_key.offset;
3922 compressed = btrfs_file_extent_compression(leaf, item);
3923 if (found_type == BTRFS_FILE_EXTENT_REG ||
3924 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3925 extent_end = extent_start +
3926 btrfs_file_extent_num_bytes(leaf, item);
3927 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3928 size_t size;
3929 size = btrfs_file_extent_inline_len(leaf, item);
3930 extent_end = (extent_start + size + root->sectorsize - 1) &
3931 ~((u64)root->sectorsize - 1);
3932 }
3933
3934 if (start >= extent_end) {
3935 path->slots[0]++;
3936 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3937 ret = btrfs_next_leaf(root, path);
3938 if (ret < 0) {
3939 err = ret;
3940 goto out;
3941 }
3942 if (ret > 0)
3943 goto not_found;
3944 leaf = path->nodes[0];
3945 }
3946 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3947 if (found_key.objectid != objectid ||
3948 found_key.type != BTRFS_EXTENT_DATA_KEY)
3949 goto not_found;
3950 if (start + len <= found_key.offset)
3951 goto not_found;
3952 em->start = start;
3953 em->len = found_key.offset - start;
3954 goto not_found_em;
3955 }
3956
3957 if (found_type == BTRFS_FILE_EXTENT_REG ||
3958 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3959 em->start = extent_start;
3960 em->len = extent_end - extent_start;
3961 em->orig_start = extent_start -
3962 btrfs_file_extent_offset(leaf, item);
3963 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3964 if (bytenr == 0) {
3965 em->block_start = EXTENT_MAP_HOLE;
3966 goto insert;
3967 }
3968 if (compressed) {
3969 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3970 em->block_start = bytenr;
3971 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3972 item);
3973 } else {
3974 bytenr += btrfs_file_extent_offset(leaf, item);
3975 em->block_start = bytenr;
3976 em->block_len = em->len;
3977 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3978 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3979 }
3980 goto insert;
3981 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3982 unsigned long ptr;
3983 char *map;
3984 size_t size;
3985 size_t extent_offset;
3986 size_t copy_size;
3987
3988 em->block_start = EXTENT_MAP_INLINE;
3989 if (!page || create) {
3990 em->start = extent_start;
3991 em->len = extent_end - extent_start;
3992 goto out;
3993 }
3994
3995 size = btrfs_file_extent_inline_len(leaf, item);
3996 extent_offset = page_offset(page) + pg_offset - extent_start;
3997 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3998 size - extent_offset);
3999 em->start = extent_start + extent_offset;
4000 em->len = (copy_size + root->sectorsize - 1) &
4001 ~((u64)root->sectorsize - 1);
4002 em->orig_start = EXTENT_MAP_INLINE;
4003 if (compressed)
4004 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4005 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4006 if (create == 0 && !PageUptodate(page)) {
4007 if (btrfs_file_extent_compression(leaf, item) ==
4008 BTRFS_COMPRESS_ZLIB) {
4009 ret = uncompress_inline(path, inode, page,
4010 pg_offset,
4011 extent_offset, item);
4012 BUG_ON(ret);
4013 } else {
4014 map = kmap(page);
4015 read_extent_buffer(leaf, map + pg_offset, ptr,
4016 copy_size);
4017 kunmap(page);
4018 }
4019 flush_dcache_page(page);
4020 } else if (create && PageUptodate(page)) {
4021 if (!trans) {
4022 kunmap(page);
4023 free_extent_map(em);
4024 em = NULL;
4025 btrfs_release_path(root, path);
4026 trans = btrfs_join_transaction(root, 1);
4027 goto again;
4028 }
4029 map = kmap(page);
4030 write_extent_buffer(leaf, map + pg_offset, ptr,
4031 copy_size);
4032 kunmap(page);
4033 btrfs_mark_buffer_dirty(leaf);
4034 }
4035 set_extent_uptodate(io_tree, em->start,
4036 extent_map_end(em) - 1, GFP_NOFS);
4037 goto insert;
4038 } else {
4039 printk("unkknown found_type %d\n", found_type);
4040 WARN_ON(1);
4041 }
4042not_found:
4043 em->start = start;
4044 em->len = len;
4045not_found_em:
4046 em->block_start = EXTENT_MAP_HOLE;
4047 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4048insert:
4049 btrfs_release_path(root, path);
4050 if (em->start > start || extent_map_end(em) <= start) {
4051 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4052 err = -EIO;
4053 goto out;
4054 }
4055
4056 err = 0;
4057 spin_lock(&em_tree->lock);
4058 ret = add_extent_mapping(em_tree, em);
4059 /* it is possible that someone inserted the extent into the tree
4060 * while we had the lock dropped. It is also possible that
4061 * an overlapping map exists in the tree
4062 */
4063 if (ret == -EEXIST) {
4064 struct extent_map *existing;
4065
4066 ret = 0;
4067
4068 existing = lookup_extent_mapping(em_tree, start, len);
4069 if (existing && (existing->start > start ||
4070 existing->start + existing->len <= start)) {
4071 free_extent_map(existing);
4072 existing = NULL;
4073 }
4074 if (!existing) {
4075 existing = lookup_extent_mapping(em_tree, em->start,
4076 em->len);
4077 if (existing) {
4078 err = merge_extent_mapping(em_tree, existing,
4079 em, start,
4080 root->sectorsize);
4081 free_extent_map(existing);
4082 if (err) {
4083 free_extent_map(em);
4084 em = NULL;
4085 }
4086 } else {
4087 err = -EIO;
4088 printk("failing to insert %Lu %Lu\n",
4089 start, len);
4090 free_extent_map(em);
4091 em = NULL;
4092 }
4093 } else {
4094 free_extent_map(em);
4095 em = existing;
4096 err = 0;
4097 }
4098 }
4099 spin_unlock(&em_tree->lock);
4100out:
4101 if (path)
4102 btrfs_free_path(path);
4103 if (trans) {
4104 ret = btrfs_end_transaction(trans, root);
4105 if (!err) {
4106 err = ret;
4107 }
4108 }
4109 if (err) {
4110 free_extent_map(em);
4111 WARN_ON(1);
4112 return ERR_PTR(err);
4113 }
4114 return em;
4115}
4116
4117static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4118 const struct iovec *iov, loff_t offset,
4119 unsigned long nr_segs)
4120{
4121 return -EINVAL;
4122}
4123
4124static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4125{
4126 return extent_bmap(mapping, iblock, btrfs_get_extent);
4127}
4128
4129int btrfs_readpage(struct file *file, struct page *page)
4130{
4131 struct extent_io_tree *tree;
4132 tree = &BTRFS_I(page->mapping->host)->io_tree;
4133 return extent_read_full_page(tree, page, btrfs_get_extent);
4134}
4135
4136static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4137{
4138 struct extent_io_tree *tree;
4139
4140
4141 if (current->flags & PF_MEMALLOC) {
4142 redirty_page_for_writepage(wbc, page);
4143 unlock_page(page);
4144 return 0;
4145 }
4146 tree = &BTRFS_I(page->mapping->host)->io_tree;
4147 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4148}
4149
4150int btrfs_writepages(struct address_space *mapping,
4151 struct writeback_control *wbc)
4152{
4153 struct extent_io_tree *tree;
4154
4155 tree = &BTRFS_I(mapping->host)->io_tree;
4156 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4157}
4158
4159static int
4160btrfs_readpages(struct file *file, struct address_space *mapping,
4161 struct list_head *pages, unsigned nr_pages)
4162{
4163 struct extent_io_tree *tree;
4164 tree = &BTRFS_I(mapping->host)->io_tree;
4165 return extent_readpages(tree, mapping, pages, nr_pages,
4166 btrfs_get_extent);
4167}
4168static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4169{
4170 struct extent_io_tree *tree;
4171 struct extent_map_tree *map;
4172 int ret;
4173
4174 tree = &BTRFS_I(page->mapping->host)->io_tree;
4175 map = &BTRFS_I(page->mapping->host)->extent_tree;
4176 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4177 if (ret == 1) {
4178 ClearPagePrivate(page);
4179 set_page_private(page, 0);
4180 page_cache_release(page);
4181 }
4182 return ret;
4183}
4184
4185static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4186{
4187 if (PageWriteback(page) || PageDirty(page))
4188 return 0;
4189 return __btrfs_releasepage(page, gfp_flags);
4190}
4191
4192static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4193{
4194 struct extent_io_tree *tree;
4195 struct btrfs_ordered_extent *ordered;
4196 u64 page_start = page_offset(page);
4197 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4198
4199 wait_on_page_writeback(page);
4200 tree = &BTRFS_I(page->mapping->host)->io_tree;
4201 if (offset) {
4202 btrfs_releasepage(page, GFP_NOFS);
4203 return;
4204 }
4205
4206 lock_extent(tree, page_start, page_end, GFP_NOFS);
4207 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4208 page_offset(page));
4209 if (ordered) {
4210 /*
4211 * IO on this page will never be started, so we need
4212 * to account for any ordered extents now
4213 */
4214 clear_extent_bit(tree, page_start, page_end,
4215 EXTENT_DIRTY | EXTENT_DELALLOC |
4216 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4217 btrfs_finish_ordered_io(page->mapping->host,
4218 page_start, page_end);
4219 btrfs_put_ordered_extent(ordered);
4220 lock_extent(tree, page_start, page_end, GFP_NOFS);
4221 }
4222 clear_extent_bit(tree, page_start, page_end,
4223 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4224 EXTENT_ORDERED,
4225 1, 1, GFP_NOFS);
4226 __btrfs_releasepage(page, GFP_NOFS);
4227
4228 ClearPageChecked(page);
4229 if (PagePrivate(page)) {
4230 ClearPagePrivate(page);
4231 set_page_private(page, 0);
4232 page_cache_release(page);
4233 }
4234}
4235
4236/*
4237 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4238 * called from a page fault handler when a page is first dirtied. Hence we must
4239 * be careful to check for EOF conditions here. We set the page up correctly
4240 * for a written page which means we get ENOSPC checking when writing into
4241 * holes and correct delalloc and unwritten extent mapping on filesystems that
4242 * support these features.
4243 *
4244 * We are not allowed to take the i_mutex here so we have to play games to
4245 * protect against truncate races as the page could now be beyond EOF. Because
4246 * vmtruncate() writes the inode size before removing pages, once we have the
4247 * page lock we can determine safely if the page is beyond EOF. If it is not
4248 * beyond EOF, then the page is guaranteed safe against truncation until we
4249 * unlock the page.
4250 */
4251int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4252{
4253 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4254 struct btrfs_root *root = BTRFS_I(inode)->root;
4255 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4256 struct btrfs_ordered_extent *ordered;
4257 char *kaddr;
4258 unsigned long zero_start;
4259 loff_t size;
4260 int ret;
4261 u64 page_start;
4262 u64 page_end;
4263
4264 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4265 if (ret)
4266 goto out;
4267
4268 ret = -EINVAL;
4269again:
4270 lock_page(page);
4271 size = i_size_read(inode);
4272 page_start = page_offset(page);
4273 page_end = page_start + PAGE_CACHE_SIZE - 1;
4274
4275 if ((page->mapping != inode->i_mapping) ||
4276 (page_start >= size)) {
4277 /* page got truncated out from underneath us */
4278 goto out_unlock;
4279 }
4280 wait_on_page_writeback(page);
4281
4282 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4283 set_page_extent_mapped(page);
4284
4285 /*
4286 * we can't set the delalloc bits if there are pending ordered
4287 * extents. Drop our locks and wait for them to finish
4288 */
4289 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4290 if (ordered) {
4291 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4292 unlock_page(page);
4293 btrfs_start_ordered_extent(inode, ordered, 1);
4294 btrfs_put_ordered_extent(ordered);
4295 goto again;
4296 }
4297
4298 btrfs_set_extent_delalloc(inode, page_start, page_end);
4299 ret = 0;
4300
4301 /* page is wholly or partially inside EOF */
4302 if (page_start + PAGE_CACHE_SIZE > size)
4303 zero_start = size & ~PAGE_CACHE_MASK;
4304 else
4305 zero_start = PAGE_CACHE_SIZE;
4306
4307 if (zero_start != PAGE_CACHE_SIZE) {
4308 kaddr = kmap(page);
4309 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4310 flush_dcache_page(page);
4311 kunmap(page);
4312 }
4313 ClearPageChecked(page);
4314 set_page_dirty(page);
4315 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4316
4317out_unlock:
4318 unlock_page(page);
4319out:
4320 return ret;
4321}
4322
4323static void btrfs_truncate(struct inode *inode)
4324{
4325 struct btrfs_root *root = BTRFS_I(inode)->root;
4326 int ret;
4327 struct btrfs_trans_handle *trans;
4328 unsigned long nr;
4329 u64 mask = root->sectorsize - 1;
4330
4331 if (!S_ISREG(inode->i_mode))
4332 return;
4333 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4334 return;
4335
4336 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4337 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4338
4339 trans = btrfs_start_transaction(root, 1);
4340 btrfs_set_trans_block_group(trans, inode);
4341 btrfs_i_size_write(inode, inode->i_size);
4342
4343 ret = btrfs_orphan_add(trans, inode);
4344 if (ret)
4345 goto out;
4346 /* FIXME, add redo link to tree so we don't leak on crash */
4347 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4348 BTRFS_EXTENT_DATA_KEY);
4349 btrfs_update_inode(trans, root, inode);
4350
4351 ret = btrfs_orphan_del(trans, inode);
4352 BUG_ON(ret);
4353
4354out:
4355 nr = trans->blocks_used;
4356 ret = btrfs_end_transaction_throttle(trans, root);
4357 BUG_ON(ret);
4358 btrfs_btree_balance_dirty(root, nr);
4359}
4360
4361/*
4362 * Invalidate a single dcache entry at the root of the filesystem.
4363 * Needed after creation of snapshot or subvolume.
4364 */
4365void btrfs_invalidate_dcache_root(struct inode *dir, char *name,
4366 int namelen)
4367{
4368 struct dentry *alias, *entry;
4369 struct qstr qstr;
4370
4371 alias = d_find_alias(dir);
4372 if (alias) {
4373 qstr.name = name;
4374 qstr.len = namelen;
4375 /* change me if btrfs ever gets a d_hash operation */
4376 qstr.hash = full_name_hash(qstr.name, qstr.len);
4377 entry = d_lookup(alias, &qstr);
4378 dput(alias);
4379 if (entry) {
4380 d_invalidate(entry);
4381 dput(entry);
4382 }
4383 }
4384}
4385
4386/*
4387 * create a new subvolume directory/inode (helper for the ioctl).
4388 */
4389int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4390 struct btrfs_trans_handle *trans, u64 new_dirid,
4391 struct btrfs_block_group_cache *block_group)
4392{
4393 struct inode *inode;
4394 int error;
4395 u64 index = 0;
4396
4397 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4398 new_dirid, block_group, S_IFDIR | 0700, &index);
4399 if (IS_ERR(inode))
4400 return PTR_ERR(inode);
4401 inode->i_op = &btrfs_dir_inode_operations;
4402 inode->i_fop = &btrfs_dir_file_operations;
4403
4404 inode->i_nlink = 1;
4405 btrfs_i_size_write(inode, 0);
4406
4407 error = btrfs_update_inode(trans, new_root, inode);
4408 if (error)
4409 return error;
4410
4411 d_instantiate(dentry, inode);
4412 return 0;
4413}
4414
4415/* helper function for file defrag and space balancing. This
4416 * forces readahead on a given range of bytes in an inode
4417 */
4418unsigned long btrfs_force_ra(struct address_space *mapping,
4419 struct file_ra_state *ra, struct file *file,
4420 pgoff_t offset, pgoff_t last_index)
4421{
4422 pgoff_t req_size = last_index - offset + 1;
4423
4424 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4425 return offset + req_size;
4426}
4427
4428struct inode *btrfs_alloc_inode(struct super_block *sb)
4429{
4430 struct btrfs_inode *ei;
4431
4432 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4433 if (!ei)
4434 return NULL;
4435 ei->last_trans = 0;
4436 ei->logged_trans = 0;
4437 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4438 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4439 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4440 INIT_LIST_HEAD(&ei->i_orphan);
4441 return &ei->vfs_inode;
4442}
4443
4444void btrfs_destroy_inode(struct inode *inode)
4445{
4446 struct btrfs_ordered_extent *ordered;
4447 WARN_ON(!list_empty(&inode->i_dentry));
4448 WARN_ON(inode->i_data.nrpages);
4449
4450 if (BTRFS_I(inode)->i_acl &&
4451 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4452 posix_acl_release(BTRFS_I(inode)->i_acl);
4453 if (BTRFS_I(inode)->i_default_acl &&
4454 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4455 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4456
4457 spin_lock(&BTRFS_I(inode)->root->list_lock);
4458 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4459 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4460 " list\n", inode->i_ino);
4461 dump_stack();
4462 }
4463 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4464
4465 while(1) {
4466 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4467 if (!ordered)
4468 break;
4469 else {
4470 printk("found ordered extent %Lu %Lu\n",
4471 ordered->file_offset, ordered->len);
4472 btrfs_remove_ordered_extent(inode, ordered);
4473 btrfs_put_ordered_extent(ordered);
4474 btrfs_put_ordered_extent(ordered);
4475 }
4476 }
4477 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4478 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4479}
4480
4481static void init_once(void *foo)
4482{
4483 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4484
4485 inode_init_once(&ei->vfs_inode);
4486}
4487
4488void btrfs_destroy_cachep(void)
4489{
4490 if (btrfs_inode_cachep)
4491 kmem_cache_destroy(btrfs_inode_cachep);
4492 if (btrfs_trans_handle_cachep)
4493 kmem_cache_destroy(btrfs_trans_handle_cachep);
4494 if (btrfs_transaction_cachep)
4495 kmem_cache_destroy(btrfs_transaction_cachep);
4496 if (btrfs_bit_radix_cachep)
4497 kmem_cache_destroy(btrfs_bit_radix_cachep);
4498 if (btrfs_path_cachep)
4499 kmem_cache_destroy(btrfs_path_cachep);
4500}
4501
4502struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4503 unsigned long extra_flags,
4504 void (*ctor)(void *))
4505{
4506 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4507 SLAB_MEM_SPREAD | extra_flags), ctor);
4508}
4509
4510int btrfs_init_cachep(void)
4511{
4512 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4513 sizeof(struct btrfs_inode),
4514 0, init_once);
4515 if (!btrfs_inode_cachep)
4516 goto fail;
4517 btrfs_trans_handle_cachep =
4518 btrfs_cache_create("btrfs_trans_handle_cache",
4519 sizeof(struct btrfs_trans_handle),
4520 0, NULL);
4521 if (!btrfs_trans_handle_cachep)
4522 goto fail;
4523 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4524 sizeof(struct btrfs_transaction),
4525 0, NULL);
4526 if (!btrfs_transaction_cachep)
4527 goto fail;
4528 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4529 sizeof(struct btrfs_path),
4530 0, NULL);
4531 if (!btrfs_path_cachep)
4532 goto fail;
4533 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4534 SLAB_DESTROY_BY_RCU, NULL);
4535 if (!btrfs_bit_radix_cachep)
4536 goto fail;
4537 return 0;
4538fail:
4539 btrfs_destroy_cachep();
4540 return -ENOMEM;
4541}
4542
4543static int btrfs_getattr(struct vfsmount *mnt,
4544 struct dentry *dentry, struct kstat *stat)
4545{
4546 struct inode *inode = dentry->d_inode;
4547 generic_fillattr(inode, stat);
4548 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4549 stat->blksize = PAGE_CACHE_SIZE;
4550 stat->blocks = (inode_get_bytes(inode) +
4551 BTRFS_I(inode)->delalloc_bytes) >> 9;
4552 return 0;
4553}
4554
4555static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4556 struct inode * new_dir,struct dentry *new_dentry)
4557{
4558 struct btrfs_trans_handle *trans;
4559 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4560 struct inode *new_inode = new_dentry->d_inode;
4561 struct inode *old_inode = old_dentry->d_inode;
4562 struct timespec ctime = CURRENT_TIME;
4563 u64 index = 0;
4564 int ret;
4565
4566 /* we're not allowed to rename between subvolumes */
4567 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4568 BTRFS_I(new_dir)->root->root_key.objectid)
4569 return -EXDEV;
4570
4571 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4572 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4573 return -ENOTEMPTY;
4574 }
4575
4576 /* to rename a snapshot or subvolume, we need to juggle the
4577 * backrefs. This isn't coded yet
4578 */
4579 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4580 return -EXDEV;
4581
4582 ret = btrfs_check_free_space(root, 1, 0);
4583 if (ret)
4584 goto out_unlock;
4585
4586 trans = btrfs_start_transaction(root, 1);
4587
4588 btrfs_set_trans_block_group(trans, new_dir);
4589
4590 btrfs_inc_nlink(old_dentry->d_inode);
4591 old_dir->i_ctime = old_dir->i_mtime = ctime;
4592 new_dir->i_ctime = new_dir->i_mtime = ctime;
4593 old_inode->i_ctime = ctime;
4594
4595 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4596 old_dentry->d_name.name,
4597 old_dentry->d_name.len);
4598 if (ret)
4599 goto out_fail;
4600
4601 if (new_inode) {
4602 new_inode->i_ctime = CURRENT_TIME;
4603 ret = btrfs_unlink_inode(trans, root, new_dir,
4604 new_dentry->d_inode,
4605 new_dentry->d_name.name,
4606 new_dentry->d_name.len);
4607 if (ret)
4608 goto out_fail;
4609 if (new_inode->i_nlink == 0) {
4610 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4611 if (ret)
4612 goto out_fail;
4613 }
4614
4615 }
4616 ret = btrfs_set_inode_index(new_dir, &index);
4617 if (ret)
4618 goto out_fail;
4619
4620 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4621 old_inode, new_dentry->d_name.name,
4622 new_dentry->d_name.len, 1, index);
4623 if (ret)
4624 goto out_fail;
4625
4626out_fail:
4627 btrfs_end_transaction_throttle(trans, root);
4628out_unlock:
4629 return ret;
4630}
4631
4632/*
4633 * some fairly slow code that needs optimization. This walks the list
4634 * of all the inodes with pending delalloc and forces them to disk.
4635 */
4636int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4637{
4638 struct list_head *head = &root->fs_info->delalloc_inodes;
4639 struct btrfs_inode *binode;
4640 struct inode *inode;
4641 unsigned long flags;
4642
4643 if (root->fs_info->sb->s_flags & MS_RDONLY)
4644 return -EROFS;
4645
4646 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4647 while(!list_empty(head)) {
4648 binode = list_entry(head->next, struct btrfs_inode,
4649 delalloc_inodes);
4650 inode = igrab(&binode->vfs_inode);
4651 if (!inode)
4652 list_del_init(&binode->delalloc_inodes);
4653 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4654 if (inode) {
4655 filemap_flush(inode->i_mapping);
4656 iput(inode);
4657 }
4658 cond_resched();
4659 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4660 }
4661 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4662
4663 /* the filemap_flush will queue IO into the worker threads, but
4664 * we have to make sure the IO is actually started and that
4665 * ordered extents get created before we return
4666 */
4667 atomic_inc(&root->fs_info->async_submit_draining);
4668 while(atomic_read(&root->fs_info->nr_async_submits) ||
4669 atomic_read(&root->fs_info->async_delalloc_pages)) {
4670 wait_event(root->fs_info->async_submit_wait,
4671 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4672 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4673 }
4674 atomic_dec(&root->fs_info->async_submit_draining);
4675 return 0;
4676}
4677
4678static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4679 const char *symname)
4680{
4681 struct btrfs_trans_handle *trans;
4682 struct btrfs_root *root = BTRFS_I(dir)->root;
4683 struct btrfs_path *path;
4684 struct btrfs_key key;
4685 struct inode *inode = NULL;
4686 int err;
4687 int drop_inode = 0;
4688 u64 objectid;
4689 u64 index = 0 ;
4690 int name_len;
4691 int datasize;
4692 unsigned long ptr;
4693 struct btrfs_file_extent_item *ei;
4694 struct extent_buffer *leaf;
4695 unsigned long nr = 0;
4696
4697 name_len = strlen(symname) + 1;
4698 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4699 return -ENAMETOOLONG;
4700
4701 err = btrfs_check_free_space(root, 1, 0);
4702 if (err)
4703 goto out_fail;
4704
4705 trans = btrfs_start_transaction(root, 1);
4706 btrfs_set_trans_block_group(trans, dir);
4707
4708 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4709 if (err) {
4710 err = -ENOSPC;
4711 goto out_unlock;
4712 }
4713
4714 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4715 dentry->d_name.len,
4716 dentry->d_parent->d_inode->i_ino, objectid,
4717 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4718 &index);
4719 err = PTR_ERR(inode);
4720 if (IS_ERR(inode))
4721 goto out_unlock;
4722
4723 err = btrfs_init_acl(inode, dir);
4724 if (err) {
4725 drop_inode = 1;
4726 goto out_unlock;
4727 }
4728
4729 btrfs_set_trans_block_group(trans, inode);
4730 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4731 if (err)
4732 drop_inode = 1;
4733 else {
4734 inode->i_mapping->a_ops = &btrfs_aops;
4735 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4736 inode->i_fop = &btrfs_file_operations;
4737 inode->i_op = &btrfs_file_inode_operations;
4738 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4739 }
4740 dir->i_sb->s_dirt = 1;
4741 btrfs_update_inode_block_group(trans, inode);
4742 btrfs_update_inode_block_group(trans, dir);
4743 if (drop_inode)
4744 goto out_unlock;
4745
4746 path = btrfs_alloc_path();
4747 BUG_ON(!path);
4748 key.objectid = inode->i_ino;
4749 key.offset = 0;
4750 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4751 datasize = btrfs_file_extent_calc_inline_size(name_len);
4752 err = btrfs_insert_empty_item(trans, root, path, &key,
4753 datasize);
4754 if (err) {
4755 drop_inode = 1;
4756 goto out_unlock;
4757 }
4758 leaf = path->nodes[0];
4759 ei = btrfs_item_ptr(leaf, path->slots[0],
4760 struct btrfs_file_extent_item);
4761 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4762 btrfs_set_file_extent_type(leaf, ei,
4763 BTRFS_FILE_EXTENT_INLINE);
4764 btrfs_set_file_extent_encryption(leaf, ei, 0);
4765 btrfs_set_file_extent_compression(leaf, ei, 0);
4766 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4767 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4768
4769 ptr = btrfs_file_extent_inline_start(ei);
4770 write_extent_buffer(leaf, symname, ptr, name_len);
4771 btrfs_mark_buffer_dirty(leaf);
4772 btrfs_free_path(path);
4773
4774 inode->i_op = &btrfs_symlink_inode_operations;
4775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4776 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4777 inode_set_bytes(inode, name_len);
4778 btrfs_i_size_write(inode, name_len - 1);
4779 err = btrfs_update_inode(trans, root, inode);
4780 if (err)
4781 drop_inode = 1;
4782
4783out_unlock:
4784 nr = trans->blocks_used;
4785 btrfs_end_transaction_throttle(trans, root);
4786out_fail:
4787 if (drop_inode) {
4788 inode_dec_link_count(inode);
4789 iput(inode);
4790 }
4791 btrfs_btree_balance_dirty(root, nr);
4792 return err;
4793}
4794
4795static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4796 u64 alloc_hint, int mode)
4797{
4798 struct btrfs_trans_handle *trans;
4799 struct btrfs_root *root = BTRFS_I(inode)->root;
4800 struct btrfs_key ins;
4801 u64 alloc_size;
4802 u64 cur_offset = start;
4803 u64 num_bytes = end - start;
4804 int ret = 0;
4805
4806 trans = btrfs_join_transaction(root, 1);
4807 BUG_ON(!trans);
4808 btrfs_set_trans_block_group(trans, inode);
4809
4810 while (num_bytes > 0) {
4811 alloc_size = min(num_bytes, root->fs_info->max_extent);
4812 ret = btrfs_reserve_extent(trans, root, alloc_size,
4813 root->sectorsize, 0, alloc_hint,
4814 (u64)-1, &ins, 1);
4815 if (ret) {
4816 WARN_ON(1);
4817 goto out;
4818 }
4819 ret = insert_reserved_file_extent(trans, inode,
4820 cur_offset, ins.objectid,
4821 ins.offset, ins.offset,
4822 ins.offset, 0, 0, 0,
4823 BTRFS_FILE_EXTENT_PREALLOC);
4824 BUG_ON(ret);
4825 num_bytes -= ins.offset;
4826 cur_offset += ins.offset;
4827 alloc_hint = ins.objectid + ins.offset;
4828 }
4829out:
4830 if (cur_offset > start) {
4831 inode->i_ctime = CURRENT_TIME;
4832 btrfs_set_flag(inode, PREALLOC);
4833 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4834 cur_offset > i_size_read(inode))
4835 btrfs_i_size_write(inode, cur_offset);
4836 ret = btrfs_update_inode(trans, root, inode);
4837 BUG_ON(ret);
4838 }
4839
4840 btrfs_end_transaction(trans, root);
4841 return ret;
4842}
4843
4844static long btrfs_fallocate(struct inode *inode, int mode,
4845 loff_t offset, loff_t len)
4846{
4847 u64 cur_offset;
4848 u64 last_byte;
4849 u64 alloc_start;
4850 u64 alloc_end;
4851 u64 alloc_hint = 0;
4852 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4853 struct extent_map *em;
4854 int ret;
4855
4856 alloc_start = offset & ~mask;
4857 alloc_end = (offset + len + mask) & ~mask;
4858
4859 mutex_lock(&inode->i_mutex);
4860 if (alloc_start > inode->i_size) {
4861 ret = btrfs_cont_expand(inode, alloc_start);
4862 if (ret)
4863 goto out;
4864 }
4865
4866 while (1) {
4867 struct btrfs_ordered_extent *ordered;
4868 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4869 alloc_end - 1, GFP_NOFS);
4870 ordered = btrfs_lookup_first_ordered_extent(inode,
4871 alloc_end - 1);
4872 if (ordered &&
4873 ordered->file_offset + ordered->len > alloc_start &&
4874 ordered->file_offset < alloc_end) {
4875 btrfs_put_ordered_extent(ordered);
4876 unlock_extent(&BTRFS_I(inode)->io_tree,
4877 alloc_start, alloc_end - 1, GFP_NOFS);
4878 btrfs_wait_ordered_range(inode, alloc_start,
4879 alloc_end - alloc_start);
4880 } else {
4881 if (ordered)
4882 btrfs_put_ordered_extent(ordered);
4883 break;
4884 }
4885 }
4886
4887 cur_offset = alloc_start;
4888 while (1) {
4889 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4890 alloc_end - cur_offset, 0);
4891 BUG_ON(IS_ERR(em) || !em);
4892 last_byte = min(extent_map_end(em), alloc_end);
4893 last_byte = (last_byte + mask) & ~mask;
4894 if (em->block_start == EXTENT_MAP_HOLE) {
4895 ret = prealloc_file_range(inode, cur_offset,
4896 last_byte, alloc_hint, mode);
4897 if (ret < 0) {
4898 free_extent_map(em);
4899 break;
4900 }
4901 }
4902 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4903 alloc_hint = em->block_start;
4904 free_extent_map(em);
4905
4906 cur_offset = last_byte;
4907 if (cur_offset >= alloc_end) {
4908 ret = 0;
4909 break;
4910 }
4911 }
4912 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4913 GFP_NOFS);
4914out:
4915 mutex_unlock(&inode->i_mutex);
4916 return ret;
4917}
4918
4919static int btrfs_set_page_dirty(struct page *page)
4920{
4921 return __set_page_dirty_nobuffers(page);
4922}
4923
4924static int btrfs_permission(struct inode *inode, int mask)
4925{
4926 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4927 return -EACCES;
4928 return generic_permission(inode, mask, btrfs_check_acl);
4929}
4930
4931static struct inode_operations btrfs_dir_inode_operations = {
4932 .getattr = btrfs_getattr,
4933 .lookup = btrfs_lookup,
4934 .create = btrfs_create,
4935 .unlink = btrfs_unlink,
4936 .link = btrfs_link,
4937 .mkdir = btrfs_mkdir,
4938 .rmdir = btrfs_rmdir,
4939 .rename = btrfs_rename,
4940 .symlink = btrfs_symlink,
4941 .setattr = btrfs_setattr,
4942 .mknod = btrfs_mknod,
4943 .setxattr = btrfs_setxattr,
4944 .getxattr = btrfs_getxattr,
4945 .listxattr = btrfs_listxattr,
4946 .removexattr = btrfs_removexattr,
4947 .permission = btrfs_permission,
4948};
4949static struct inode_operations btrfs_dir_ro_inode_operations = {
4950 .lookup = btrfs_lookup,
4951 .permission = btrfs_permission,
4952};
4953static struct file_operations btrfs_dir_file_operations = {
4954 .llseek = generic_file_llseek,
4955 .read = generic_read_dir,
4956 .readdir = btrfs_real_readdir,
4957 .unlocked_ioctl = btrfs_ioctl,
4958#ifdef CONFIG_COMPAT
4959 .compat_ioctl = btrfs_ioctl,
4960#endif
4961 .release = btrfs_release_file,
4962 .fsync = btrfs_sync_file,
4963};
4964
4965static struct extent_io_ops btrfs_extent_io_ops = {
4966 .fill_delalloc = run_delalloc_range,
4967 .submit_bio_hook = btrfs_submit_bio_hook,
4968 .merge_bio_hook = btrfs_merge_bio_hook,
4969 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4970 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4971 .writepage_start_hook = btrfs_writepage_start_hook,
4972 .readpage_io_failed_hook = btrfs_io_failed_hook,
4973 .set_bit_hook = btrfs_set_bit_hook,
4974 .clear_bit_hook = btrfs_clear_bit_hook,
4975};
4976
4977static struct address_space_operations btrfs_aops = {
4978 .readpage = btrfs_readpage,
4979 .writepage = btrfs_writepage,
4980 .writepages = btrfs_writepages,
4981 .readpages = btrfs_readpages,
4982 .sync_page = block_sync_page,
4983 .bmap = btrfs_bmap,
4984 .direct_IO = btrfs_direct_IO,
4985 .invalidatepage = btrfs_invalidatepage,
4986 .releasepage = btrfs_releasepage,
4987 .set_page_dirty = btrfs_set_page_dirty,
4988};
4989
4990static struct address_space_operations btrfs_symlink_aops = {
4991 .readpage = btrfs_readpage,
4992 .writepage = btrfs_writepage,
4993 .invalidatepage = btrfs_invalidatepage,
4994 .releasepage = btrfs_releasepage,
4995};
4996
4997static struct inode_operations btrfs_file_inode_operations = {
4998 .truncate = btrfs_truncate,
4999 .getattr = btrfs_getattr,
5000 .setattr = btrfs_setattr,
5001 .setxattr = btrfs_setxattr,
5002 .getxattr = btrfs_getxattr,
5003 .listxattr = btrfs_listxattr,
5004 .removexattr = btrfs_removexattr,
5005 .permission = btrfs_permission,
5006 .fallocate = btrfs_fallocate,
5007};
5008static struct inode_operations btrfs_special_inode_operations = {
5009 .getattr = btrfs_getattr,
5010 .setattr = btrfs_setattr,
5011 .permission = btrfs_permission,
5012 .setxattr = btrfs_setxattr,
5013 .getxattr = btrfs_getxattr,
5014 .listxattr = btrfs_listxattr,
5015 .removexattr = btrfs_removexattr,
5016};
5017static struct inode_operations btrfs_symlink_inode_operations = {
5018 .readlink = generic_readlink,
5019 .follow_link = page_follow_link_light,
5020 .put_link = page_put_link,
5021 .permission = btrfs_permission,
5022};
generated by cgit v1.2.2 (git 2.25.0) at 2025-09-01 18:55:29 -0400