Btrfs: properly check free space for tree balancing
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM
23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
29#include <linux/smp_lock.h>
30#include <linux/backing-dev.h>
31#include <linux/mpage.h>
32#include <linux/swap.h>
33#include <linux/writeback.h>
34#include <linux/statfs.h>
35#include <linux/compat.h>
9ebefb18 36#include <linux/bit_spinlock.h>
92fee66d 37#include <linux/version.h>
5103e947 38#include <linux/xattr.h>
33268eaf 39#include <linux/posix_acl.h>
d899e052 40#include <linux/falloc.h>
4b4e25f2 41#include "compat.h"
39279cc3
CM
42#include "ctree.h"
43#include "disk-io.h"
44#include "transaction.h"
45#include "btrfs_inode.h"
46#include "ioctl.h"
47#include "print-tree.h"
0b86a832 48#include "volumes.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 51#include "tree-log.h"
5b84e8d6 52#include "ref-cache.h"
c8b97818 53#include "compression.h"
39279cc3
CM
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;
618e21d5 63static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM
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;
d1310b2e 68static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM
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
7b128766 87static void btrfs_truncate(struct inode *inode);
c8b97818 88static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM
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);
7b128766 93
d352ac68
CM
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 */
1832a6d5
CM
98int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
99 int for_del)
100{
a2135011
CM
101 u64 total;
102 u64 used;
1832a6d5
CM
103 u64 thresh;
104 int ret = 0;
105
75eff68e 106 spin_lock(&root->fs_info->delalloc_lock);
a2135011
CM
107 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
108 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
1832a6d5 109 if (for_del)
f9ef6604 110 thresh = total * 90;
1832a6d5 111 else
f9ef6604
CM
112 thresh = total * 85;
113
114 do_div(thresh, 100);
1832a6d5 115
1832a6d5
CM
116 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
117 ret = -ENOSPC;
75eff68e 118 spin_unlock(&root->fs_info->delalloc_lock);
1832a6d5
CM
119 return ret;
120}
121
c8b97818
CM
122/*
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
126 */
127static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
130 struct page **compressed_pages)
131{
132 struct btrfs_key key;
133 struct btrfs_path *path;
134 struct extent_buffer *leaf;
135 struct page *page = NULL;
136 char *kaddr;
137 unsigned long ptr;
138 struct btrfs_file_extent_item *ei;
139 int err = 0;
140 int ret;
141 size_t cur_size = size;
142 size_t datasize;
143 unsigned long offset;
144 int use_compress = 0;
145
146 if (compressed_size && compressed_pages) {
147 use_compress = 1;
148 cur_size = compressed_size;
149 }
150
151 path = btrfs_alloc_path(); if (!path)
152 return -ENOMEM;
153
154 btrfs_set_trans_block_group(trans, inode);
155
156 key.objectid = inode->i_ino;
157 key.offset = start;
158 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
159 inode_add_bytes(inode, size);
160 datasize = btrfs_file_extent_calc_inline_size(cur_size);
161
162 inode_add_bytes(inode, size);
163 ret = btrfs_insert_empty_item(trans, root, path, &key,
164 datasize);
165 BUG_ON(ret);
166 if (ret) {
167 err = ret;
168 printk("got bad ret %d\n", ret);
169 goto fail;
170 }
171 leaf = path->nodes[0];
172 ei = btrfs_item_ptr(leaf, path->slots[0],
173 struct btrfs_file_extent_item);
174 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
175 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
176 btrfs_set_file_extent_encryption(leaf, ei, 0);
177 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
178 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
179 ptr = btrfs_file_extent_inline_start(ei);
180
181 if (use_compress) {
182 struct page *cpage;
183 int i = 0;
184 while(compressed_size > 0) {
185 cpage = compressed_pages[i];
5b050f04 186 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM
187 PAGE_CACHE_SIZE);
188
189 kaddr = kmap(cpage);
190 write_extent_buffer(leaf, kaddr, ptr, cur_size);
191 kunmap(cpage);
192
193 i++;
194 ptr += cur_size;
195 compressed_size -= cur_size;
196 }
197 btrfs_set_file_extent_compression(leaf, ei,
198 BTRFS_COMPRESS_ZLIB);
199 } else {
200 page = find_get_page(inode->i_mapping,
201 start >> PAGE_CACHE_SHIFT);
202 btrfs_set_file_extent_compression(leaf, ei, 0);
203 kaddr = kmap_atomic(page, KM_USER0);
204 offset = start & (PAGE_CACHE_SIZE - 1);
205 write_extent_buffer(leaf, kaddr + offset, ptr, size);
206 kunmap_atomic(kaddr, KM_USER0);
207 page_cache_release(page);
208 }
209 btrfs_mark_buffer_dirty(leaf);
210 btrfs_free_path(path);
211
212 BTRFS_I(inode)->disk_i_size = inode->i_size;
213 btrfs_update_inode(trans, root, inode);
214 return 0;
215fail:
216 btrfs_free_path(path);
217 return err;
218}
219
220
221/*
222 * conditionally insert an inline extent into the file. This
223 * does the checks required to make sure the data is small enough
224 * to fit as an inline extent.
225 */
226static int cow_file_range_inline(struct btrfs_trans_handle *trans,
227 struct btrfs_root *root,
228 struct inode *inode, u64 start, u64 end,
229 size_t compressed_size,
230 struct page **compressed_pages)
231{
232 u64 isize = i_size_read(inode);
233 u64 actual_end = min(end + 1, isize);
234 u64 inline_len = actual_end - start;
235 u64 aligned_end = (end + root->sectorsize - 1) &
236 ~((u64)root->sectorsize - 1);
237 u64 hint_byte;
238 u64 data_len = inline_len;
239 int ret;
240
241 if (compressed_size)
242 data_len = compressed_size;
243
244 if (start > 0 ||
70b99e69 245 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM
246 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
247 (!compressed_size &&
248 (actual_end & (root->sectorsize - 1)) == 0) ||
249 end + 1 < isize ||
250 data_len > root->fs_info->max_inline) {
251 return 1;
252 }
253
c8b97818 254 ret = btrfs_drop_extents(trans, root, inode, start,
70b99e69 255 aligned_end, start, &hint_byte);
c8b97818
CM
256 BUG_ON(ret);
257
258 if (isize > actual_end)
259 inline_len = min_t(u64, isize, actual_end);
260 ret = insert_inline_extent(trans, root, inode, start,
261 inline_len, compressed_size,
262 compressed_pages);
263 BUG_ON(ret);
264 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM
265 return 0;
266}
267
771ed689
CM
268struct async_extent {
269 u64 start;
270 u64 ram_size;
271 u64 compressed_size;
272 struct page **pages;
273 unsigned long nr_pages;
274 struct list_head list;
275};
276
277struct async_cow {
278 struct inode *inode;
279 struct btrfs_root *root;
280 struct page *locked_page;
281 u64 start;
282 u64 end;
283 struct list_head extents;
284 struct btrfs_work work;
285};
286
287static noinline int add_async_extent(struct async_cow *cow,
288 u64 start, u64 ram_size,
289 u64 compressed_size,
290 struct page **pages,
291 unsigned long nr_pages)
292{
293 struct async_extent *async_extent;
294
295 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
296 async_extent->start = start;
297 async_extent->ram_size = ram_size;
298 async_extent->compressed_size = compressed_size;
299 async_extent->pages = pages;
300 async_extent->nr_pages = nr_pages;
301 list_add_tail(&async_extent->list, &cow->extents);
302 return 0;
303}
304
d352ac68 305/*
771ed689
CM
306 * we create compressed extents in two phases. The first
307 * phase compresses a range of pages that have already been
308 * locked (both pages and state bits are locked).
c8b97818 309 *
771ed689
CM
310 * This is done inside an ordered work queue, and the compression
311 * is spread across many cpus. The actual IO submission is step
312 * two, and the ordered work queue takes care of making sure that
313 * happens in the same order things were put onto the queue by
314 * writepages and friends.
c8b97818 315 *
771ed689
CM
316 * If this code finds it can't get good compression, it puts an
317 * entry onto the work queue to write the uncompressed bytes. This
318 * makes sure that both compressed inodes and uncompressed inodes
319 * are written in the same order that pdflush sent them down.
d352ac68 320 */
771ed689
CM
321static noinline int compress_file_range(struct inode *inode,
322 struct page *locked_page,
323 u64 start, u64 end,
324 struct async_cow *async_cow,
325 int *num_added)
b888db2b
CM
326{
327 struct btrfs_root *root = BTRFS_I(inode)->root;
328 struct btrfs_trans_handle *trans;
db94535d 329 u64 num_bytes;
c8b97818
CM
330 u64 orig_start;
331 u64 disk_num_bytes;
db94535d 332 u64 blocksize = root->sectorsize;
c8b97818 333 u64 actual_end;
42dc7bab 334 u64 isize = i_size_read(inode);
e6dcd2dc 335 int ret = 0;
c8b97818
CM
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;
771ed689 342 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM
343 int i;
344 int will_compress;
b888db2b 345
c8b97818
CM
346 orig_start = start;
347
42dc7bab 348 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM
349again:
350 will_compress = 0;
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 353
c8b97818
CM
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
771ed689
CM
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.
c8b97818
CM
365 */
366 total_compressed = min(total_compressed, max_uncompressed);
db94535d 367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 368 num_bytes = max(blocksize, num_bytes);
c8b97818
CM
369 disk_num_bytes = num_bytes;
370 total_in = 0;
371 ret = 0;
db94535d 372
771ed689
CM
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.
c8b97818
CM
377 */
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
380 WARN_ON(pages);
cfbc246e 381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 382
c8b97818
CM
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) {
771ed689
CM
409 trans = btrfs_join_transaction(root, 1);
410 BUG_ON(!trans);
411 btrfs_set_trans_block_group(trans, inode);
412
c8b97818 413 /* lets try to make an inline extent */
771ed689 414 if (ret || total_in < (actual_end - start)) {
c8b97818 415 /* we didn't compress the entire range, try
771ed689 416 * to make an uncompressed inline extent.
c8b97818
CM
417 */
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
420 } else {
771ed689 421 /* try making a compressed inline extent */
c8b97818
CM
422 ret = cow_file_range_inline(trans, root, inode,
423 start, end,
424 total_compressed, pages);
425 }
771ed689 426 btrfs_end_transaction(trans, root);
c8b97818 427 if (ret == 0) {
771ed689
CM
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 */
c8b97818
CM
433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
771ed689
CM
435 start, end, NULL, 1, 0,
436 0, 1, 1, 1);
c8b97818
CM
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++) {
70b99e69 470 WARN_ON(pages[i]->mapping);
c8b97818
CM
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 }
771ed689
CM
481 if (will_compress) {
482 *num_added += 1;
c8b97818 483
771ed689
CM
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);
179e29e4 490
42dc7bab 491 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM
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 }
3b951516 513
771ed689
CM
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);
c8b97818 556
771ed689
CM
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);
c8b97818 596 /*
771ed689
CM
597 * here we're doing allocation and writeback of the
598 * compressed pages
c8b97818 599 */
771ed689
CM
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;
445a6944 613 em->orig_start = em->start;
c8b97818 614
771ed689
CM
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;
42dc7bab 699 u64 isize = i_size_read(inode);
771ed689
CM
700 struct btrfs_key ins;
701 struct extent_map *em;
702 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
703 int ret = 0;
704
705 trans = btrfs_join_transaction(root, 1);
706 BUG_ON(!trans);
707 btrfs_set_trans_block_group(trans, inode);
708
42dc7bab 709 actual_end = min_t(u64, isize, end + 1);
771ed689
CM
710
711 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
712 num_bytes = max(blocksize, num_bytes);
713 disk_num_bytes = num_bytes;
714 ret = 0;
715
716 if (start == 0) {
717 /* lets try to make an inline extent */
718 ret = cow_file_range_inline(trans, root, inode,
719 start, end, 0, NULL);
720 if (ret == 0) {
721 extent_clear_unlock_delalloc(inode,
722 &BTRFS_I(inode)->io_tree,
723 start, end, NULL, 1, 1,
724 1, 1, 1, 1);
725 *nr_written = *nr_written +
726 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
727 *page_started = 1;
728 ret = 0;
729 goto out;
730 }
731 }
732
733 BUG_ON(disk_num_bytes >
734 btrfs_super_total_bytes(&root->fs_info->super_copy));
735
736 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
737
738 while(disk_num_bytes > 0) {
c8b97818 739 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 740 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 741 root->sectorsize, 0, alloc_hint,
e6dcd2dc 742 (u64)-1, &ins, 1);
c59f8951 743 if (ret) {
771ed689 744 BUG();
c59f8951 745 }
e6dcd2dc
CM
746 em = alloc_extent_map(GFP_NOFS);
747 em->start = start;
445a6944 748 em->orig_start = em->start;
c8b97818 749
771ed689
CM
750 ram_size = ins.offset;
751 em->len = ins.offset;
c8b97818 752
e6dcd2dc 753 em->block_start = ins.objectid;
c8b97818 754 em->block_len = ins.offset;
e6dcd2dc 755 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 756 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 757
e6dcd2dc
CM
758 while(1) {
759 spin_lock(&em_tree->lock);
760 ret = add_extent_mapping(em_tree, em);
761 spin_unlock(&em_tree->lock);
762 if (ret != -EEXIST) {
763 free_extent_map(em);
764 break;
765 }
766 btrfs_drop_extent_cache(inode, start,
c8b97818 767 start + ram_size - 1, 0);
e6dcd2dc
CM
768 }
769
98d20f67 770 cur_alloc_size = ins.offset;
e6dcd2dc 771 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 772 ram_size, cur_alloc_size, 0);
e6dcd2dc 773 BUG_ON(ret);
c8b97818 774
17d217fe
YZ
775 if (root->root_key.objectid ==
776 BTRFS_DATA_RELOC_TREE_OBJECTID) {
777 ret = btrfs_reloc_clone_csums(inode, start,
778 cur_alloc_size);
779 BUG_ON(ret);
780 }
781
c8b97818
CM
782 if (disk_num_bytes < cur_alloc_size) {
783 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
3b951516
CM
784 cur_alloc_size);
785 break;
786 }
c8b97818
CM
787 /* we're not doing compressed IO, don't unlock the first
788 * page (which the caller expects to stay locked), don't
789 * clear any dirty bits and don't set any writeback bits
790 */
791 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
792 start, start + ram_size - 1,
771ed689
CM
793 locked_page, unlock, 1,
794 1, 0, 0, 0);
c8b97818 795 disk_num_bytes -= cur_alloc_size;
c59f8951
CM
796 num_bytes -= cur_alloc_size;
797 alloc_hint = ins.objectid + ins.offset;
798 start += cur_alloc_size;
b888db2b 799 }
b888db2b 800out:
771ed689 801 ret = 0;
b888db2b 802 btrfs_end_transaction(trans, root);
c8b97818 803
be20aa9d 804 return ret;
771ed689 805}
c8b97818 806
771ed689
CM
807/*
808 * work queue call back to started compression on a file and pages
809 */
810static noinline void async_cow_start(struct btrfs_work *work)
811{
812 struct async_cow *async_cow;
813 int num_added = 0;
814 async_cow = container_of(work, struct async_cow, work);
815
816 compress_file_range(async_cow->inode, async_cow->locked_page,
817 async_cow->start, async_cow->end, async_cow,
818 &num_added);
819 if (num_added == 0)
820 async_cow->inode = NULL;
821}
822
823/*
824 * work queue call back to submit previously compressed pages
825 */
826static noinline void async_cow_submit(struct btrfs_work *work)
827{
828 struct async_cow *async_cow;
829 struct btrfs_root *root;
830 unsigned long nr_pages;
831
832 async_cow = container_of(work, struct async_cow, work);
833
834 root = async_cow->root;
835 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
836 PAGE_CACHE_SHIFT;
837
838 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
839
840 if (atomic_read(&root->fs_info->async_delalloc_pages) <
841 5 * 1042 * 1024 &&
842 waitqueue_active(&root->fs_info->async_submit_wait))
843 wake_up(&root->fs_info->async_submit_wait);
844
845 if (async_cow->inode) {
846 submit_compressed_extents(async_cow->inode, async_cow);
70b99e69 847 }
771ed689 848}
c8b97818 849
771ed689
CM
850static noinline void async_cow_free(struct btrfs_work *work)
851{
852 struct async_cow *async_cow;
853 async_cow = container_of(work, struct async_cow, work);
854 kfree(async_cow);
855}
856
857static int cow_file_range_async(struct inode *inode, struct page *locked_page,
858 u64 start, u64 end, int *page_started,
859 unsigned long *nr_written)
860{
861 struct async_cow *async_cow;
862 struct btrfs_root *root = BTRFS_I(inode)->root;
863 unsigned long nr_pages;
864 u64 cur_end;
865 int limit = 10 * 1024 * 1042;
866
867 if (!btrfs_test_opt(root, COMPRESS)) {
868 return cow_file_range(inode, locked_page, start, end,
869 page_started, nr_written, 1);
870 }
871
872 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
873 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
874 while(start < end) {
875 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
876 async_cow->inode = inode;
877 async_cow->root = root;
878 async_cow->locked_page = locked_page;
879 async_cow->start = start;
880
881 if (btrfs_test_flag(inode, NOCOMPRESS))
882 cur_end = end;
883 else
884 cur_end = min(end, start + 512 * 1024 - 1);
885
886 async_cow->end = cur_end;
887 INIT_LIST_HEAD(&async_cow->extents);
888
889 async_cow->work.func = async_cow_start;
890 async_cow->work.ordered_func = async_cow_submit;
891 async_cow->work.ordered_free = async_cow_free;
892 async_cow->work.flags = 0;
893
771ed689
CM
894 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
895 PAGE_CACHE_SHIFT;
896 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
897
898 btrfs_queue_worker(&root->fs_info->delalloc_workers,
899 &async_cow->work);
900
901 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
902 wait_event(root->fs_info->async_submit_wait,
903 (atomic_read(&root->fs_info->async_delalloc_pages) <
904 limit));
905 }
906
907 while(atomic_read(&root->fs_info->async_submit_draining) &&
908 atomic_read(&root->fs_info->async_delalloc_pages)) {
909 wait_event(root->fs_info->async_submit_wait,
910 (atomic_read(&root->fs_info->async_delalloc_pages) ==
911 0));
912 }
913
914 *nr_written += nr_pages;
915 start = cur_end + 1;
916 }
917 *page_started = 1;
918 return 0;
be20aa9d
CM
919}
920
17d217fe
YZ
921static int noinline csum_exist_in_range(struct btrfs_root *root,
922 u64 bytenr, u64 num_bytes)
923{
924 int ret;
925 struct btrfs_ordered_sum *sums;
926 LIST_HEAD(list);
927
928 ret = btrfs_lookup_csums_range(root, bytenr, bytenr + num_bytes - 1,
929 &list);
930 if (ret == 0 && list_empty(&list))
931 return 0;
932
933 while (!list_empty(&list)) {
934 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
935 list_del(&sums->list);
936 kfree(sums);
937 }
938 return 1;
939}
940
d352ac68
CM
941/*
942 * when nowcow writeback call back. This checks for snapshots or COW copies
943 * of the extents that exist in the file, and COWs the file as required.
944 *
945 * If no cow copies or snapshots exist, we write directly to the existing
946 * blocks on disk
947 */
c8b97818 948static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM
949 u64 start, u64 end, int *page_started, int force,
950 unsigned long *nr_written)
be20aa9d 951{
be20aa9d 952 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 953 struct btrfs_trans_handle *trans;
be20aa9d 954 struct extent_buffer *leaf;
be20aa9d 955 struct btrfs_path *path;
80ff3856 956 struct btrfs_file_extent_item *fi;
be20aa9d 957 struct btrfs_key found_key;
80ff3856
YZ
958 u64 cow_start;
959 u64 cur_offset;
960 u64 extent_end;
961 u64 disk_bytenr;
962 u64 num_bytes;
963 int extent_type;
964 int ret;
d899e052 965 int type;
80ff3856
YZ
966 int nocow;
967 int check_prev = 1;
be20aa9d
CM
968
969 path = btrfs_alloc_path();
970 BUG_ON(!path);
7ea394f1
YZ
971 trans = btrfs_join_transaction(root, 1);
972 BUG_ON(!trans);
be20aa9d 973
80ff3856
YZ
974 cow_start = (u64)-1;
975 cur_offset = start;
976 while (1) {
977 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
978 cur_offset, 0);
979 BUG_ON(ret < 0);
980 if (ret > 0 && path->slots[0] > 0 && check_prev) {
981 leaf = path->nodes[0];
982 btrfs_item_key_to_cpu(leaf, &found_key,
983 path->slots[0] - 1);
984 if (found_key.objectid == inode->i_ino &&
985 found_key.type == BTRFS_EXTENT_DATA_KEY)
986 path->slots[0]--;
987 }
988 check_prev = 0;
989next_slot:
990 leaf = path->nodes[0];
991 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
992 ret = btrfs_next_leaf(root, path);
993 if (ret < 0)
994 BUG_ON(1);
995 if (ret > 0)
996 break;
997 leaf = path->nodes[0];
998 }
be20aa9d 999
80ff3856
YZ
1000 nocow = 0;
1001 disk_bytenr = 0;
17d217fe 1002 num_bytes = 0;
80ff3856
YZ
1003 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1004
1005 if (found_key.objectid > inode->i_ino ||
1006 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1007 found_key.offset > end)
1008 break;
1009
1010 if (found_key.offset > cur_offset) {
1011 extent_end = found_key.offset;
1012 goto out_check;
1013 }
1014
1015 fi = btrfs_item_ptr(leaf, path->slots[0],
1016 struct btrfs_file_extent_item);
1017 extent_type = btrfs_file_extent_type(leaf, fi);
1018
d899e052
YZ
1019 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1020 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ
1021 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1022 extent_end = found_key.offset +
1023 btrfs_file_extent_num_bytes(leaf, fi);
1024 if (extent_end <= start) {
1025 path->slots[0]++;
1026 goto next_slot;
1027 }
17d217fe
YZ
1028 if (disk_bytenr == 0)
1029 goto out_check;
80ff3856
YZ
1030 if (btrfs_file_extent_compression(leaf, fi) ||
1031 btrfs_file_extent_encryption(leaf, fi) ||
1032 btrfs_file_extent_other_encoding(leaf, fi))
1033 goto out_check;
d899e052
YZ
1034 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1035 goto out_check;
d2fb3437 1036 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1037 goto out_check;
17d217fe
YZ
1038 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1039 disk_bytenr))
1040 goto out_check;
80ff3856 1041 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
17d217fe
YZ
1042 disk_bytenr += cur_offset - found_key.offset;
1043 num_bytes = min(end + 1, extent_end) - cur_offset;
1044 /*
1045 * force cow if csum exists in the range.
1046 * this ensure that csum for a given extent are
1047 * either valid or do not exist.
1048 */
1049 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1050 goto out_check;
80ff3856
YZ
1051 nocow = 1;
1052 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1053 extent_end = found_key.offset +
1054 btrfs_file_extent_inline_len(leaf, fi);
1055 extent_end = ALIGN(extent_end, root->sectorsize);
1056 } else {
1057 BUG_ON(1);
1058 }
1059out_check:
1060 if (extent_end <= start) {
1061 path->slots[0]++;
1062 goto next_slot;
1063 }
1064 if (!nocow) {
1065 if (cow_start == (u64)-1)
1066 cow_start = cur_offset;
1067 cur_offset = extent_end;
1068 if (cur_offset > end)
1069 break;
1070 path->slots[0]++;
1071 goto next_slot;
7ea394f1
YZ
1072 }
1073
1074 btrfs_release_path(root, path);
80ff3856
YZ
1075 if (cow_start != (u64)-1) {
1076 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM
1077 found_key.offset - 1, page_started,
1078 nr_written, 1);
80ff3856
YZ
1079 BUG_ON(ret);
1080 cow_start = (u64)-1;
7ea394f1 1081 }
80ff3856 1082
d899e052
YZ
1083 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1084 struct extent_map *em;
1085 struct extent_map_tree *em_tree;
1086 em_tree = &BTRFS_I(inode)->extent_tree;
1087 em = alloc_extent_map(GFP_NOFS);
1088 em->start = cur_offset;
445a6944 1089 em->orig_start = em->start;
d899e052
YZ
1090 em->len = num_bytes;
1091 em->block_len = num_bytes;
1092 em->block_start = disk_bytenr;
1093 em->bdev = root->fs_info->fs_devices->latest_bdev;
1094 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1095 while (1) {
1096 spin_lock(&em_tree->lock);
1097 ret = add_extent_mapping(em_tree, em);
1098 spin_unlock(&em_tree->lock);
1099 if (ret != -EEXIST) {
1100 free_extent_map(em);
1101 break;
1102 }
1103 btrfs_drop_extent_cache(inode, em->start,
1104 em->start + em->len - 1, 0);
1105 }
1106 type = BTRFS_ORDERED_PREALLOC;
1107 } else {
1108 type = BTRFS_ORDERED_NOCOW;
1109 }
80ff3856
YZ
1110
1111 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ
1112 num_bytes, num_bytes, type);
1113 BUG_ON(ret);
771ed689 1114
d899e052
YZ
1115 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1116 cur_offset, cur_offset + num_bytes - 1,
771ed689 1117 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ
1118 cur_offset = extent_end;
1119 if (cur_offset > end)
1120 break;
be20aa9d 1121 }
80ff3856
YZ
1122 btrfs_release_path(root, path);
1123
1124 if (cur_offset <= end && cow_start == (u64)-1)
1125 cow_start = cur_offset;
1126 if (cow_start != (u64)-1) {
1127 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1128 page_started, nr_written, 1);
80ff3856
YZ
1129 BUG_ON(ret);
1130 }
1131
1132 ret = btrfs_end_transaction(trans, root);
1133 BUG_ON(ret);
7ea394f1 1134 btrfs_free_path(path);
80ff3856 1135 return 0;
be20aa9d
CM
1136}
1137
d352ac68
CM
1138/*
1139 * extent_io.c call back to do delayed allocation processing
1140 */
c8b97818 1141static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM
1142 u64 start, u64 end, int *page_started,
1143 unsigned long *nr_written)
be20aa9d 1144{
be20aa9d 1145 int ret;
a2135011 1146
17d217fe 1147 if (btrfs_test_flag(inode, NODATACOW))
c8b97818 1148 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1149 page_started, 1, nr_written);
d899e052
YZ
1150 else if (btrfs_test_flag(inode, PREALLOC))
1151 ret = run_delalloc_nocow(inode, locked_page, start, end,
ce397c06 1152 page_started, 0, nr_written);
be20aa9d 1153 else
771ed689
CM
1154 ret = cow_file_range_async(inode, locked_page, start, end,
1155 page_started, nr_written);
1832a6d5 1156
b888db2b
CM
1157 return ret;
1158}
1159
d352ac68
CM
1160/*
1161 * extent_io.c set_bit_hook, used to track delayed allocation
1162 * bytes in this file, and to maintain the list of inodes that
1163 * have pending delalloc work to be done.
1164 */
b2950863 1165static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1166 unsigned long old, unsigned long bits)
291d673e 1167{
75eff68e
CM
1168 /*
1169 * set_bit and clear bit hooks normally require _irqsave/restore
1170 * but in this case, we are only testeing for the DELALLOC
1171 * bit, which is only set or cleared with irqs on
1172 */
b0c68f8b 1173 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1174 struct btrfs_root *root = BTRFS_I(inode)->root;
75eff68e 1175 spin_lock(&root->fs_info->delalloc_lock);
9069218d 1176 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1177 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM
1178 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1179 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1180 &root->fs_info->delalloc_inodes);
1181 }
75eff68e 1182 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM
1183 }
1184 return 0;
1185}
1186
d352ac68
CM
1187/*
1188 * extent_io.c clear_bit_hook, see set_bit_hook for why
1189 */
b2950863 1190static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1191 unsigned long old, unsigned long bits)
291d673e 1192{
75eff68e
CM
1193 /*
1194 * set_bit and clear bit hooks normally require _irqsave/restore
1195 * but in this case, we are only testeing for the DELALLOC
1196 * bit, which is only set or cleared with irqs on
1197 */
b0c68f8b 1198 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1199 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1200
75eff68e 1201 spin_lock(&root->fs_info->delalloc_lock);
b0c68f8b
CM
1202 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1203 printk("warning: delalloc account %Lu %Lu\n",
1204 end - start + 1, root->fs_info->delalloc_bytes);
1205 root->fs_info->delalloc_bytes = 0;
9069218d 1206 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM
1207 } else {
1208 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1209 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1210 }
ea8c2819
CM
1211 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1212 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1213 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1214 }
75eff68e 1215 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM
1216 }
1217 return 0;
1218}
1219
d352ac68
CM
1220/*
1221 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1222 * we don't create bios that span stripes or chunks
1223 */
239b14b3 1224int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM
1225 size_t size, struct bio *bio,
1226 unsigned long bio_flags)
239b14b3
CM
1227{
1228 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1229 struct btrfs_mapping_tree *map_tree;
a62b9401 1230 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM
1231 u64 length = 0;
1232 u64 map_length;
239b14b3
CM
1233 int ret;
1234
771ed689
CM
1235 if (bio_flags & EXTENT_BIO_COMPRESSED)
1236 return 0;
1237
f2d8d74d 1238 length = bio->bi_size;
239b14b3
CM
1239 map_tree = &root->fs_info->mapping_tree;
1240 map_length = length;
cea9e445 1241 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1242 &map_length, NULL, 0);
cea9e445 1243
239b14b3 1244 if (map_length < length + size) {
239b14b3
CM
1245 return 1;
1246 }
1247 return 0;
1248}
1249
d352ac68
CM
1250/*
1251 * in order to insert checksums into the metadata in large chunks,
1252 * we wait until bio submission time. All the pages in the bio are
1253 * checksummed and sums are attached onto the ordered extent record.
1254 *
1255 * At IO completion time the cums attached on the ordered extent record
1256 * are inserted into the btree
1257 */
b2950863 1258static int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
c8b97818 1259 int mirror_num, unsigned long bio_flags)
065631f6 1260{
065631f6 1261 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1262 int ret = 0;
e015640f 1263
d20f7043 1264 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1265 BUG_ON(ret);
4a69a410
CM
1266 return 0;
1267}
e015640f 1268
4a69a410
CM
1269/*
1270 * in order to insert checksums into the metadata in large chunks,
1271 * we wait until bio submission time. All the pages in the bio are
1272 * checksummed and sums are attached onto the ordered extent record.
1273 *
1274 * At IO completion time the cums attached on the ordered extent record
1275 * are inserted into the btree
1276 */
b2950863 1277static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM
1278 int mirror_num, unsigned long bio_flags)
1279{
1280 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1281 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM
1282}
1283
d352ac68
CM
1284/*
1285 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1286 * or reading the csums from the tree before a read
1287 */
b2950863 1288static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1289 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM
1290{
1291 struct btrfs_root *root = BTRFS_I(inode)->root;
1292 int ret = 0;
19b9bdb0 1293 int skip_sum;
44b8bd7e 1294
e6dcd2dc
CM
1295 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1296 BUG_ON(ret);
065631f6 1297
17d217fe 1298 skip_sum = btrfs_test_flag(inode, NODATASUM);
7ea394f1 1299
4d1b5fb4 1300 if (!(rw & (1 << BIO_RW))) {
d20f7043 1301 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM
1302 return btrfs_submit_compressed_read(inode, bio,
1303 mirror_num, bio_flags);
d20f7043
CM
1304 } else if (!skip_sum)
1305 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1306 goto mapit;
19b9bdb0 1307 } else if (!skip_sum) {
17d217fe
YZ
1308 /* csum items have already been cloned */
1309 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1310 goto mapit;
19b9bdb0
CM
1311 /* we're doing a write, do the async checksumming */
1312 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1313 inode, rw, bio, mirror_num,
4a69a410
CM
1314 bio_flags, __btrfs_submit_bio_start,
1315 __btrfs_submit_bio_done);
19b9bdb0
CM
1316 }
1317
0b86a832 1318mapit:
8b712842 1319 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1320}
6885f308 1321
d352ac68
CM
1322/*
1323 * given a list of ordered sums record them in the inode. This happens
1324 * at IO completion time based on sums calculated at bio submission time.
1325 */
ba1da2f4 1326static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM
1327 struct inode *inode, u64 file_offset,
1328 struct list_head *list)
1329{
1330 struct list_head *cur;
1331 struct btrfs_ordered_sum *sum;
1332
1333 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 1334 list_for_each(cur, list) {
e6dcd2dc 1335 sum = list_entry(cur, struct btrfs_ordered_sum, list);
d20f7043
CM
1336 btrfs_csum_file_blocks(trans,
1337 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM
1338 }
1339 return 0;
1340}
1341
ea8c2819
CM
1342int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1343{
771ed689
CM
1344 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1345 WARN_ON(1);
1346 }
ea8c2819
CM
1347 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1348 GFP_NOFS);
1349}
1350
d352ac68 1351/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM
1352struct btrfs_writepage_fixup {
1353 struct page *page;
1354 struct btrfs_work work;
1355};
1356
b2950863 1357static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM
1358{
1359 struct btrfs_writepage_fixup *fixup;
1360 struct btrfs_ordered_extent *ordered;
1361 struct page *page;
1362 struct inode *inode;
1363 u64 page_start;
1364 u64 page_end;
1365
1366 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1367 page = fixup->page;
4a096752 1368again:
247e743c
CM
1369 lock_page(page);
1370 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1371 ClearPageChecked(page);
1372 goto out_page;
1373 }
1374
1375 inode = page->mapping->host;
1376 page_start = page_offset(page);
1377 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1378
1379 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM
1380
1381 /* already ordered? We're done */
1382 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1383 EXTENT_ORDERED, 0)) {
247e743c 1384 goto out;
4a096752
CM
1385 }
1386
1387 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1388 if (ordered) {
1389 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1390 page_end, GFP_NOFS);
1391 unlock_page(page);
1392 btrfs_start_ordered_extent(inode, ordered, 1);
1393 goto again;
1394 }
247e743c 1395
ea8c2819 1396 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM
1397 ClearPageChecked(page);
1398out:
1399 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1400out_page:
1401 unlock_page(page);
1402 page_cache_release(page);
1403}
1404
1405/*
1406 * There are a few paths in the higher layers of the kernel that directly
1407 * set the page dirty bit without asking the filesystem if it is a
1408 * good idea. This causes problems because we want to make sure COW
1409 * properly happens and the data=ordered rules are followed.
1410 *
c8b97818 1411 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM
1412 * hasn't been properly setup for IO. We kick off an async process
1413 * to fix it up. The async helper will wait for ordered extents, set
1414 * the delalloc bit and make it safe to write the page.
1415 */
b2950863 1416static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM
1417{
1418 struct inode *inode = page->mapping->host;
1419 struct btrfs_writepage_fixup *fixup;
1420 struct btrfs_root *root = BTRFS_I(inode)->root;
1421 int ret;
1422
1423 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1424 EXTENT_ORDERED, 0);
1425 if (ret)
1426 return 0;
1427
1428 if (PageChecked(page))
1429 return -EAGAIN;
1430
1431 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1432 if (!fixup)
1433 return -EAGAIN;
f421950f 1434
247e743c
CM
1435 SetPageChecked(page);
1436 page_cache_get(page);
1437 fixup->work.func = btrfs_writepage_fixup_worker;
1438 fixup->page = page;
1439 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1440 return -EAGAIN;
1441}
1442
d899e052
YZ
1443static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1444 struct inode *inode, u64 file_pos,
1445 u64 disk_bytenr, u64 disk_num_bytes,
1446 u64 num_bytes, u64 ram_bytes,
1447 u8 compression, u8 encryption,
1448 u16 other_encoding, int extent_type)
1449{
1450 struct btrfs_root *root = BTRFS_I(inode)->root;
1451 struct btrfs_file_extent_item *fi;
1452 struct btrfs_path *path;
1453 struct extent_buffer *leaf;
1454 struct btrfs_key ins;
1455 u64 hint;
1456 int ret;
1457
1458 path = btrfs_alloc_path();
1459 BUG_ON(!path);
1460
1461 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1462 file_pos + num_bytes, file_pos, &hint);
1463 BUG_ON(ret);
1464
1465 ins.objectid = inode->i_ino;
1466 ins.offset = file_pos;
1467 ins.type = BTRFS_EXTENT_DATA_KEY;
1468 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1469 BUG_ON(ret);
1470 leaf = path->nodes[0];
1471 fi = btrfs_item_ptr(leaf, path->slots[0],
1472 struct btrfs_file_extent_item);
1473 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1474 btrfs_set_file_extent_type(leaf, fi, extent_type);
1475 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1476 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1477 btrfs_set_file_extent_offset(leaf, fi, 0);
1478 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1479 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1480 btrfs_set_file_extent_compression(leaf, fi, compression);
1481 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1482 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1483 btrfs_mark_buffer_dirty(leaf);
1484
1485 inode_add_bytes(inode, num_bytes);
1486 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1487
1488 ins.objectid = disk_bytenr;
1489 ins.offset = disk_num_bytes;
1490 ins.type = BTRFS_EXTENT_ITEM_KEY;
1491 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1492 root->root_key.objectid,
1493 trans->transid, inode->i_ino, &ins);
1494 BUG_ON(ret);
1495
1496 btrfs_free_path(path);
1497 return 0;
1498}
1499
d352ac68
CM
1500/* as ordered data IO finishes, this gets called so we can finish
1501 * an ordered extent if the range of bytes in the file it covers are
1502 * fully written.
1503 */
211f90e6 1504static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1505{
e6dcd2dc
CM
1506 struct btrfs_root *root = BTRFS_I(inode)->root;
1507 struct btrfs_trans_handle *trans;
1508 struct btrfs_ordered_extent *ordered_extent;
1509 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1510 int compressed = 0;
e6dcd2dc
CM
1511 int ret;
1512
1513 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1514 if (!ret)
e6dcd2dc 1515 return 0;
e6dcd2dc 1516
f9295749 1517 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM
1518
1519 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1520 BUG_ON(!ordered_extent);
7ea394f1
YZ
1521 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1522 goto nocow;
e6dcd2dc
CM
1523
1524 lock_extent(io_tree, ordered_extent->file_offset,
1525 ordered_extent->file_offset + ordered_extent->len - 1,
1526 GFP_NOFS);
1527
c8b97818 1528 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ
1529 compressed = 1;
1530 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1531 BUG_ON(compressed);
1532 ret = btrfs_mark_extent_written(trans, root, inode,
1533 ordered_extent->file_offset,
1534 ordered_extent->file_offset +
1535 ordered_extent->len);
1536 BUG_ON(ret);
1537 } else {
1538 ret = insert_reserved_file_extent(trans, inode,
1539 ordered_extent->file_offset,
1540 ordered_extent->start,
1541 ordered_extent->disk_len,
1542 ordered_extent->len,
1543 ordered_extent->len,
1544 compressed, 0, 0,
1545 BTRFS_FILE_EXTENT_REG);
1546 BUG_ON(ret);
1547 }
e6dcd2dc
CM
1548 unlock_extent(io_tree, ordered_extent->file_offset,
1549 ordered_extent->file_offset + ordered_extent->len - 1,
1550 GFP_NOFS);
7ea394f1 1551nocow:
e6dcd2dc
CM
1552 add_pending_csums(trans, inode, ordered_extent->file_offset,
1553 &ordered_extent->list);
1554
34353029 1555 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1556 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1557 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1558 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1559 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1560
e6dcd2dc
CM
1561 /* once for us */
1562 btrfs_put_ordered_extent(ordered_extent);
1563 /* once for the tree */
1564 btrfs_put_ordered_extent(ordered_extent);
1565
e6dcd2dc
CM
1566 btrfs_end_transaction(trans, root);
1567 return 0;
1568}
1569
b2950863 1570static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM
1571 struct extent_state *state, int uptodate)
1572{
1573 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1574}
1575
d352ac68
CM
1576/*
1577 * When IO fails, either with EIO or csum verification fails, we
1578 * try other mirrors that might have a good copy of the data. This
1579 * io_failure_record is used to record state as we go through all the
1580 * mirrors. If another mirror has good data, the page is set up to date
1581 * and things continue. If a good mirror can't be found, the original
1582 * bio end_io callback is called to indicate things have failed.
1583 */
7e38326f
CM
1584struct io_failure_record {
1585 struct page *page;
1586 u64 start;
1587 u64 len;
1588 u64 logical;
d20f7043 1589 unsigned long bio_flags;
7e38326f
CM
1590 int last_mirror;
1591};
1592
b2950863 1593static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM
1594 struct page *page, u64 start, u64 end,
1595 struct extent_state *state)
7e38326f
CM
1596{
1597 struct io_failure_record *failrec = NULL;
1598 u64 private;
1599 struct extent_map *em;
1600 struct inode *inode = page->mapping->host;
1601 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1602 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM
1603 struct bio *bio;
1604 int num_copies;
1605 int ret;
1259ab75 1606 int rw;
7e38326f
CM
1607 u64 logical;
1608
1609 ret = get_state_private(failure_tree, start, &private);
1610 if (ret) {
7e38326f
CM
1611 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1612 if (!failrec)
1613 return -ENOMEM;
1614 failrec->start = start;
1615 failrec->len = end - start + 1;
1616 failrec->last_mirror = 0;
d20f7043 1617 failrec->bio_flags = 0;
7e38326f 1618
3b951516
CM
1619 spin_lock(&em_tree->lock);
1620 em = lookup_extent_mapping(em_tree, start, failrec->len);
1621 if (em->start > start || em->start + em->len < start) {
1622 free_extent_map(em);
1623 em = NULL;
1624 }
1625 spin_unlock(&em_tree->lock);
7e38326f
CM
1626
1627 if (!em || IS_ERR(em)) {
1628 kfree(failrec);
1629 return -EIO;
1630 }
1631 logical = start - em->start;
1632 logical = em->block_start + logical;
d20f7043
CM
1633 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1634 logical = em->block_start;
1635 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1636 }
7e38326f
CM
1637 failrec->logical = logical;
1638 free_extent_map(em);
1639 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1640 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM
1641 set_state_private(failure_tree, start,
1642 (u64)(unsigned long)failrec);
7e38326f 1643 } else {
587f7704 1644 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM
1645 }
1646 num_copies = btrfs_num_copies(
1647 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1648 failrec->logical, failrec->len);
1649 failrec->last_mirror++;
1650 if (!state) {
1651 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1652 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1653 failrec->start,
1654 EXTENT_LOCKED);
1655 if (state && state->start != failrec->start)
1656 state = NULL;
1657 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1658 }
1659 if (!state || failrec->last_mirror > num_copies) {
1660 set_state_private(failure_tree, failrec->start, 0);
1661 clear_extent_bits(failure_tree, failrec->start,
1662 failrec->start + failrec->len - 1,
1663 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1664 kfree(failrec);
1665 return -EIO;
1666 }
1667 bio = bio_alloc(GFP_NOFS, 1);
1668 bio->bi_private = state;
1669 bio->bi_end_io = failed_bio->bi_end_io;
1670 bio->bi_sector = failrec->logical >> 9;
1671 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1672 bio->bi_size = 0;
d20f7043 1673
7e38326f 1674 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM
1675 if (failed_bio->bi_rw & (1 << BIO_RW))
1676 rw = WRITE;
1677 else
1678 rw = READ;
1679
1680 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1681 failrec->last_mirror,
d20f7043 1682 failrec->bio_flags);
1259ab75
CM
1683 return 0;
1684}
1685
d352ac68
CM
1686/*
1687 * each time an IO finishes, we do a fast check in the IO failure tree
1688 * to see if we need to process or clean up an io_failure_record
1689 */
b2950863 1690static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM
1691{
1692 u64 private;
1693 u64 private_failure;
1694 struct io_failure_record *failure;
1695 int ret;
1696
1697 private = 0;
1698 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1699 (u64)-1, 1, EXTENT_DIRTY)) {
1700 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1701 start, &private_failure);
1702 if (ret == 0) {
1703 failure = (struct io_failure_record *)(unsigned long)
1704 private_failure;
1705 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1706 failure->start, 0);
1707 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1708 failure->start,
1709 failure->start + failure->len - 1,
1710 EXTENT_DIRTY | EXTENT_LOCKED,
1711 GFP_NOFS);
1712 kfree(failure);
1713 }
1714 }
7e38326f
CM
1715 return 0;
1716}
1717
d352ac68
CM
1718/*
1719 * when reads are done, we need to check csums to verify the data is correct
1720 * if there's a match, we allow the bio to finish. If not, we go through
1721 * the io_failure_record routines to find good copies
1722 */
b2950863 1723static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1724 struct extent_state *state)
07157aac 1725{
35ebb934 1726 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1727 struct inode *inode = page->mapping->host;
d1310b2e 1728 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1729 char *kaddr;
aadfeb6e 1730 u64 private = ~(u32)0;
07157aac 1731 int ret;
ff79f819
CM
1732 struct btrfs_root *root = BTRFS_I(inode)->root;
1733 u32 csum = ~(u32)0;
bbf0d006 1734 unsigned long flags;
d1310b2e 1735
d20f7043
CM
1736 if (PageChecked(page)) {
1737 ClearPageChecked(page);
1738 goto good;
1739 }
17d217fe
YZ
1740 if (btrfs_test_flag(inode, NODATASUM))
1741 return 0;
1742
1743 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1744 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1745 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1746 GFP_NOFS);
b6cda9bc 1747 return 0;
17d217fe 1748 }
d20f7043 1749
c2e639f0 1750 if (state && state->start == start) {
70dec807
CM
1751 private = state->private;
1752 ret = 0;
1753 } else {
1754 ret = get_state_private(io_tree, start, &private);
1755 }
bbf0d006 1756 local_irq_save(flags);
07157aac
CM
1757 kaddr = kmap_atomic(page, KM_IRQ0);
1758 if (ret) {
1759 goto zeroit;
1760 }
ff79f819
CM
1761 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1762 btrfs_csum_final(csum, (char *)&csum);
1763 if (csum != private) {
07157aac
CM
1764 goto zeroit;
1765 }
1766 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1767 local_irq_restore(flags);
d20f7043 1768good:
7e38326f
CM
1769 /* if the io failure tree for this inode is non-empty,
1770 * check to see if we've recovered from a failed IO
1771 */
1259ab75 1772 btrfs_clean_io_failures(inode, start);
07157aac
CM
1773 return 0;
1774
1775zeroit:
aadfeb6e
CM
1776 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1777 page->mapping->host->i_ino, (unsigned long long)start, csum,
1778 private);
db94535d
CM
1779 memset(kaddr + offset, 1, end - start + 1);
1780 flush_dcache_page(page);
07157aac 1781 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1782 local_irq_restore(flags);
3b951516
CM
1783 if (private == 0)
1784 return 0;
7e38326f 1785 return -EIO;
07157aac 1786}
b888db2b 1787
7b128766
JB
1788/*
1789 * This creates an orphan entry for the given inode in case something goes
1790 * wrong in the middle of an unlink/truncate.
1791 */
1792int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1793{
1794 struct btrfs_root *root = BTRFS_I(inode)->root;
1795 int ret = 0;
1796
bcc63abb 1797 spin_lock(&root->list_lock);
7b128766
JB
1798
1799 /* already on the orphan list, we're good */
1800 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1801 spin_unlock(&root->list_lock);
7b128766
JB
1802 return 0;
1803 }
1804
1805 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1806
bcc63abb 1807 spin_unlock(&root->list_lock);
7b128766
JB
1808
1809 /*
1810 * insert an orphan item to track this unlinked/truncated file
1811 */
1812 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1813
1814 return ret;
1815}
1816
1817/*
1818 * We have done the truncate/delete so we can go ahead and remove the orphan
1819 * item for this particular inode.
1820 */
1821int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1822{
1823 struct btrfs_root *root = BTRFS_I(inode)->root;
1824 int ret = 0;
1825
bcc63abb 1826 spin_lock(&root->list_lock);
7b128766
JB
1827
1828 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1829 spin_unlock(&root->list_lock);
7b128766
JB
1830 return 0;
1831 }
1832
1833 list_del_init(&BTRFS_I(inode)->i_orphan);
1834 if (!trans) {
bcc63abb 1835 spin_unlock(&root->list_lock);
7b128766
JB
1836 return 0;
1837 }
1838
bcc63abb 1839 spin_unlock(&root->list_lock);
7b128766
JB
1840
1841 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1842
1843 return ret;
1844}
1845
1846/*
1847 * this cleans up any orphans that may be left on the list from the last use
1848 * of this root.
1849 */
1850void btrfs_orphan_cleanup(struct btrfs_root *root)
1851{
1852 struct btrfs_path *path;
1853 struct extent_buffer *leaf;
1854 struct btrfs_item *item;
1855 struct btrfs_key key, found_key;
1856 struct btrfs_trans_handle *trans;
1857 struct inode *inode;
1858 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1859
7b128766
JB
1860 path = btrfs_alloc_path();
1861 if (!path)
1862 return;
1863 path->reada = -1;
1864
1865 key.objectid = BTRFS_ORPHAN_OBJECTID;
1866 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1867 key.offset = (u64)-1;
1868
7b128766
JB
1869
1870 while (1) {
1871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1872 if (ret < 0) {
1873 printk(KERN_ERR "Error searching slot for orphan: %d"
1874 "\n", ret);
1875 break;
1876 }
1877
1878 /*
1879 * if ret == 0 means we found what we were searching for, which
1880 * is weird, but possible, so only screw with path if we didnt
1881 * find the key and see if we have stuff that matches
1882 */
1883 if (ret > 0) {
1884 if (path->slots[0] == 0)
1885 break;
1886 path->slots[0]--;
1887 }
1888
1889 /* pull out the item */
1890 leaf = path->nodes[0];
1891 item = btrfs_item_nr(leaf, path->slots[0]);
1892 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1893
1894 /* make sure the item matches what we want */
1895 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1896 break;
1897 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1898 break;
1899
1900 /* release the path since we're done with it */
1901 btrfs_release_path(root, path);
1902
1903 /*
1904 * this is where we are basically btrfs_lookup, without the
1905 * crossing root thing. we store the inode number in the
1906 * offset of the orphan item.
1907 */
5b21f2ed 1908 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB
1909 found_key.offset, root);
1910 if (!inode)
1911 break;
1912
1913 if (inode->i_state & I_NEW) {
1914 BTRFS_I(inode)->root = root;
1915
1916 /* have to set the location manually */
1917 BTRFS_I(inode)->location.objectid = inode->i_ino;
1918 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1919 BTRFS_I(inode)->location.offset = 0;
1920
1921 btrfs_read_locked_inode(inode);
1922 unlock_new_inode(inode);
1923 }
1924
1925 /*
1926 * add this inode to the orphan list so btrfs_orphan_del does
1927 * the proper thing when we hit it
1928 */
bcc63abb 1929 spin_lock(&root->list_lock);
7b128766 1930 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1931 spin_unlock(&root->list_lock);
7b128766
JB
1932
1933 /*
1934 * if this is a bad inode, means we actually succeeded in
1935 * removing the inode, but not the orphan record, which means
1936 * we need to manually delete the orphan since iput will just
1937 * do a destroy_inode
1938 */
1939 if (is_bad_inode(inode)) {
5b21f2ed 1940 trans = btrfs_start_transaction(root, 1);
7b128766 1941 btrfs_orphan_del(trans, inode);
5b21f2ed 1942 btrfs_end_transaction(trans, root);
7b128766
JB
1943 iput(inode);
1944 continue;
1945 }
1946
1947 /* if we have links, this was a truncate, lets do that */
1948 if (inode->i_nlink) {
1949 nr_truncate++;
1950 btrfs_truncate(inode);
1951 } else {
1952 nr_unlink++;
1953 }
1954
1955 /* this will do delete_inode and everything for us */
1956 iput(inode);
1957 }
1958
1959 if (nr_unlink)
1960 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1961 if (nr_truncate)
1962 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1963
1964 btrfs_free_path(path);
7b128766
JB
1965}
1966
d352ac68
CM
1967/*
1968 * read an inode from the btree into the in-memory inode
1969 */
39279cc3
CM
1970void btrfs_read_locked_inode(struct inode *inode)
1971{
1972 struct btrfs_path *path;
5f39d397 1973 struct extent_buffer *leaf;
39279cc3 1974 struct btrfs_inode_item *inode_item;
0b86a832 1975 struct btrfs_timespec *tspec;
39279cc3
CM
1976 struct btrfs_root *root = BTRFS_I(inode)->root;
1977 struct btrfs_key location;
1978 u64 alloc_group_block;
618e21d5 1979 u32 rdev;
39279cc3
CM
1980 int ret;
1981
1982 path = btrfs_alloc_path();
1983 BUG_ON(!path);
39279cc3 1984 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1985
39279cc3 1986 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1987 if (ret)
39279cc3 1988 goto make_bad;
39279cc3 1989
5f39d397
CM
1990 leaf = path->nodes[0];
1991 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1992 struct btrfs_inode_item);
1993
1994 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1995 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1996 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1997 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1998 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM
1999
2000 tspec = btrfs_inode_atime(inode_item);
2001 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2002 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2003
2004 tspec = btrfs_inode_mtime(inode_item);
2005 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2006 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2007
2008 tspec = btrfs_inode_ctime(inode_item);
2009 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2010 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2011
a76a3cd4 2012 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2013 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2014 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2015 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2016 inode->i_rdev = 0;
5f39d397
CM
2017 rdev = btrfs_inode_rdev(leaf, inode_item);
2018
aec7477b 2019 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2020 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2021
5f39d397 2022 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
d2fb3437
YZ
2023 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2024 alloc_group_block, 0);
39279cc3
CM
2025 btrfs_free_path(path);
2026 inode_item = NULL;
2027
39279cc3 2028 switch (inode->i_mode & S_IFMT) {
39279cc3
CM
2029 case S_IFREG:
2030 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2031 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2032 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM
2033 inode->i_fop = &btrfs_file_operations;
2034 inode->i_op = &btrfs_file_inode_operations;
2035 break;
2036 case S_IFDIR:
2037 inode->i_fop = &btrfs_dir_file_operations;
2038 if (root == root->fs_info->tree_root)
2039 inode->i_op = &btrfs_dir_ro_inode_operations;
2040 else
2041 inode->i_op = &btrfs_dir_inode_operations;
2042 break;
2043 case S_IFLNK:
2044 inode->i_op = &btrfs_symlink_inode_operations;
2045 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2046 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2047 break;
618e21d5
JB
2048 default:
2049 init_special_inode(inode, inode->i_mode, rdev);
2050 break;
39279cc3
CM
2051 }
2052 return;
2053
2054make_bad:
39279cc3 2055 btrfs_free_path(path);
39279cc3
CM
2056 make_bad_inode(inode);
2057}
2058
d352ac68
CM
2059/*
2060 * given a leaf and an inode, copy the inode fields into the leaf
2061 */
e02119d5
CM
2062static void fill_inode_item(struct btrfs_trans_handle *trans,
2063 struct extent_buffer *leaf,
5f39d397 2064 struct btrfs_inode_item *item,
39279cc3
CM
2065 struct inode *inode)
2066{
5f39d397
CM
2067 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2068 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2069 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM
2070 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2071 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2072
2073 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2074 inode->i_atime.tv_sec);
2075 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2076 inode->i_atime.tv_nsec);
2077
2078 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2079 inode->i_mtime.tv_sec);
2080 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2081 inode->i_mtime.tv_nsec);
2082
2083 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2084 inode->i_ctime.tv_sec);
2085 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2086 inode->i_ctime.tv_nsec);
2087
a76a3cd4 2088 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2089 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2090 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2091 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2092 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2093 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2094 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM
2095}
2096
d352ac68
CM
2097/*
2098 * copy everything in the in-memory inode into the btree.
2099 */
ba1da2f4 2100int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM
2101 struct btrfs_root *root,
2102 struct inode *inode)
2103{
2104 struct btrfs_inode_item *inode_item;
2105 struct btrfs_path *path;
5f39d397 2106 struct extent_buffer *leaf;
39279cc3
CM
2107 int ret;
2108
2109 path = btrfs_alloc_path();
2110 BUG_ON(!path);
39279cc3
CM
2111 ret = btrfs_lookup_inode(trans, root, path,
2112 &BTRFS_I(inode)->location, 1);
2113 if (ret) {
2114 if (ret > 0)
2115 ret = -ENOENT;
2116 goto failed;
2117 }
2118
5f39d397
CM
2119 leaf = path->nodes[0];
2120 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM
2121 struct btrfs_inode_item);
2122
e02119d5 2123 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2124 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2125 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM
2126 ret = 0;
2127failed:
39279cc3
CM
2128 btrfs_free_path(path);
2129 return ret;
2130}
2131
2132
d352ac68
CM
2133/*
2134 * unlink helper that gets used here in inode.c and in the tree logging
2135 * recovery code. It remove a link in a directory with a given name, and
2136 * also drops the back refs in the inode to the directory
2137 */
e02119d5
CM
2138int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2139 struct btrfs_root *root,
2140 struct inode *dir, struct inode *inode,
2141 const char *name, int name_len)
39279cc3
CM
2142{
2143 struct btrfs_path *path;
39279cc3 2144 int ret = 0;
5f39d397 2145 struct extent_buffer *leaf;
39279cc3 2146 struct btrfs_dir_item *di;
5f39d397 2147 struct btrfs_key key;
aec7477b 2148 u64 index;
39279cc3
CM
2149
2150 path = btrfs_alloc_path();
54aa1f4d
CM
2151 if (!path) {
2152 ret = -ENOMEM;
2153 goto err;
2154 }
2155
39279cc3
CM
2156 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2157 name, name_len, -1);
2158 if (IS_ERR(di)) {
2159 ret = PTR_ERR(di);
2160 goto err;
2161 }
2162 if (!di) {
2163 ret = -ENOENT;
2164 goto err;
2165 }
5f39d397
CM
2166 leaf = path->nodes[0];
2167 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2168 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM
2169 if (ret)
2170 goto err;
39279cc3
CM
2171 btrfs_release_path(root, path);
2172
aec7477b 2173 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM
2174 inode->i_ino,
2175 dir->i_ino, &index);
aec7477b
JB
2176 if (ret) {
2177 printk("failed to delete reference to %.*s, "
2178 "inode %lu parent %lu\n", name_len, name,
e02119d5 2179 inode->i_ino, dir->i_ino);
aec7477b
JB
2180 goto err;
2181 }
2182
39279cc3 2183 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2184 index, name, name_len, -1);
39279cc3
CM
2185 if (IS_ERR(di)) {
2186 ret = PTR_ERR(di);
2187 goto err;
2188 }
2189 if (!di) {
2190 ret = -ENOENT;
2191 goto err;
2192 }
2193 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2194 btrfs_release_path(root, path);
39279cc3 2195
e02119d5
CM
2196 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2197 inode, dir->i_ino);
49eb7e46
CM
2198 BUG_ON(ret != 0 && ret != -ENOENT);
2199 if (ret != -ENOENT)
2200 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM
2201
2202 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2203 dir, index);
2204 BUG_ON(ret);
39279cc3
CM
2205err:
2206 btrfs_free_path(path);
e02119d5
CM
2207 if (ret)
2208 goto out;
2209
2210 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2211 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2212 btrfs_update_inode(trans, root, dir);
2213 btrfs_drop_nlink(inode);
2214 ret = btrfs_update_inode(trans, root, inode);
2215 dir->i_sb->s_dirt = 1;
2216out:
39279cc3
CM
2217 return ret;
2218}
2219
2220static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2221{
2222 struct btrfs_root *root;
2223 struct btrfs_trans_handle *trans;
7b128766 2224 struct inode *inode = dentry->d_inode;
39279cc3 2225 int ret;
1832a6d5 2226 unsigned long nr = 0;
39279cc3
CM
2227
2228 root = BTRFS_I(dir)->root;
1832a6d5
CM
2229
2230 ret = btrfs_check_free_space(root, 1, 1);
2231 if (ret)
2232 goto fail;
2233
39279cc3 2234 trans = btrfs_start_transaction(root, 1);
5f39d397 2235
39279cc3 2236 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM
2237 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2238 dentry->d_name.name, dentry->d_name.len);
7b128766
JB
2239
2240 if (inode->i_nlink == 0)
2241 ret = btrfs_orphan_add(trans, inode);
2242
d3c2fdcf 2243 nr = trans->blocks_used;
5f39d397 2244
89ce8a63 2245 btrfs_end_transaction_throttle(trans, root);
1832a6d5 2246fail:
d3c2fdcf 2247 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
2248 return ret;
2249}
2250
2251static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2252{
2253 struct inode *inode = dentry->d_inode;
1832a6d5 2254 int err = 0;
39279cc3
CM
2255 int ret;
2256 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2257 struct btrfs_trans_handle *trans;
1832a6d5 2258 unsigned long nr = 0;
39279cc3 2259
3394e160
CM
2260 /*
2261 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2262 * the root of a subvolume or snapshot
2263 */
2264 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2265 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2266 return -ENOTEMPTY;
925baedd 2267 }
134d4512 2268
1832a6d5
CM
2269 ret = btrfs_check_free_space(root, 1, 1);
2270 if (ret)
2271 goto fail;
2272
39279cc3
CM
2273 trans = btrfs_start_transaction(root, 1);
2274 btrfs_set_trans_block_group(trans, dir);
39279cc3 2275
7b128766
JB
2276 err = btrfs_orphan_add(trans, inode);
2277 if (err)
2278 goto fail_trans;
2279
39279cc3 2280 /* now the directory is empty */
e02119d5
CM
2281 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2282 dentry->d_name.name, dentry->d_name.len);
39279cc3 2283 if (!err) {
dbe674a9 2284 btrfs_i_size_write(inode, 0);
39279cc3 2285 }
3954401f 2286
7b128766 2287fail_trans:
d3c2fdcf 2288 nr = trans->blocks_used;
89ce8a63 2289 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 2290fail:
d3c2fdcf 2291 btrfs_btree_balance_dirty(root, nr);
3954401f 2292
39279cc3
CM
2293 if (ret && !err)
2294 err = ret;
2295 return err;
2296}
2297
d20f7043 2298#if 0
323ac95b
CM
2299/*
2300 * when truncating bytes in a file, it is possible to avoid reading
2301 * the leaves that contain only checksum items. This can be the
2302 * majority of the IO required to delete a large file, but it must
2303 * be done carefully.
2304 *
2305 * The keys in the level just above the leaves are checked to make sure
2306 * the lowest key in a given leaf is a csum key, and starts at an offset
2307 * after the new size.
2308 *
2309 * Then the key for the next leaf is checked to make sure it also has
2310 * a checksum item for the same file. If it does, we know our target leaf
2311 * contains only checksum items, and it can be safely freed without reading
2312 * it.
2313 *
2314 * This is just an optimization targeted at large files. It may do
2315 * nothing. It will return 0 unless things went badly.
2316 */
2317static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2318 struct btrfs_root *root,
2319 struct btrfs_path *path,
2320 struct inode *inode, u64 new_size)
2321{
2322 struct btrfs_key key;
2323 int ret;
2324 int nritems;
2325 struct btrfs_key found_key;
2326 struct btrfs_key other_key;
5b84e8d6
YZ
2327 struct btrfs_leaf_ref *ref;
2328 u64 leaf_gen;
2329 u64 leaf_start;
323ac95b
CM
2330
2331 path->lowest_level = 1;
2332 key.objectid = inode->i_ino;
2333 key.type = BTRFS_CSUM_ITEM_KEY;
2334 key.offset = new_size;
2335again:
2336 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2337 if (ret < 0)
2338 goto out;
2339
2340 if (path->nodes[1] == NULL) {
2341 ret = 0;
2342 goto out;
2343 }
2344 ret = 0;
2345 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2346 nritems = btrfs_header_nritems(path->nodes[1]);
2347
2348 if (!nritems)
2349 goto out;
2350
2351 if (path->slots[1] >= nritems)
2352 goto next_node;
2353
2354 /* did we find a key greater than anything we want to delete? */
2355 if (found_key.objectid > inode->i_ino ||
2356 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2357 goto out;
2358
2359 /* we check the next key in the node to make sure the leave contains
2360 * only checksum items. This comparison doesn't work if our
2361 * leaf is the last one in the node
2362 */
2363 if (path->slots[1] + 1 >= nritems) {
2364next_node:
2365 /* search forward from the last key in the node, this
2366 * will bring us into the next node in the tree
2367 */
2368 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2369
2370 /* unlikely, but we inc below, so check to be safe */
2371 if (found_key.offset == (u64)-1)
2372 goto out;
2373
2374 /* search_forward needs a path with locks held, do the
2375 * search again for the original key. It is possible
2376 * this will race with a balance and return a path that
2377 * we could modify, but this drop is just an optimization
2378 * and is allowed to miss some leaves.
2379 */
2380 btrfs_release_path(root, path);
2381 found_key.offset++;
2382
2383 /* setup a max key for search_forward */
2384 other_key.offset = (u64)-1;
2385 other_key.type = key.type;
2386 other_key.objectid = key.objectid;
2387
2388 path->keep_locks = 1;
2389 ret = btrfs_search_forward(root, &found_key, &other_key,
2390 path, 0, 0);
2391 path->keep_locks = 0;
2392 if (ret || found_key.objectid != key.objectid ||
2393 found_key.type != key.type) {
2394 ret = 0;
2395 goto out;
2396 }
2397
2398 key.offset = found_key.offset;
2399 btrfs_release_path(root, path);
2400 cond_resched();
2401 goto again;
2402 }
2403
2404 /* we know there's one more slot after us in the tree,
2405 * read that key so we can verify it is also a checksum item
2406 */
2407 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2408
2409 if (found_key.objectid < inode->i_ino)
2410 goto next_key;
2411
2412 if (found_key.type != key.type || found_key.offset < new_size)
2413 goto next_key;
2414
2415 /*
2416 * if the key for the next leaf isn't a csum key from this objectid,
2417 * we can't be sure there aren't good items inside this leaf.
2418 * Bail out
2419 */
2420 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2421 goto out;
2422
5b84e8d6
YZ
2423 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2424 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM
2425 /*
2426 * it is safe to delete this leaf, it contains only
2427 * csum items from this inode at an offset >= new_size
2428 */
5b84e8d6 2429 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM
2430 BUG_ON(ret);
2431
5b84e8d6
YZ
2432 if (root->ref_cows && leaf_gen < trans->transid) {
2433 ref = btrfs_alloc_leaf_ref(root, 0);
2434 if (ref) {
2435 ref->root_gen = root->root_key.offset;
2436 ref->bytenr = leaf_start;
2437 ref->owner = 0;
2438 ref->generation = leaf_gen;
2439 ref->nritems = 0;
2440
2441 ret = btrfs_add_leaf_ref(root, ref, 0);
2442 WARN_ON(ret);
2443 btrfs_free_leaf_ref(root, ref);
2444 } else {
2445 WARN_ON(1);
2446 }
2447 }
323ac95b
CM
2448next_key:
2449 btrfs_release_path(root, path);
2450
2451 if (other_key.objectid == inode->i_ino &&
2452 other_key.type == key.type && other_key.offset > key.offset) {
2453 key.offset = other_key.offset;
2454 cond_resched();
2455 goto again;
2456 }
2457 ret = 0;
2458out:
2459 /* fixup any changes we've made to the path */
2460 path->lowest_level = 0;
2461 path->keep_locks = 0;
2462 btrfs_release_path(root, path);
2463 return ret;
2464}
2465
d20f7043
CM
2466#endif
2467
39279cc3
CM
2468/*
2469 * this can truncate away extent items, csum items and directory items.
2470 * It starts at a high offset and removes keys until it can't find
d352ac68 2471 * any higher than new_size
39279cc3
CM
2472 *
2473 * csum items that cross the new i_size are truncated to the new size
2474 * as well.
7b128766
JB
2475 *
2476 * min_type is the minimum key type to truncate down to. If set to 0, this
2477 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2478 */
e02119d5
CM
2479noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2480 struct btrfs_root *root,
2481 struct inode *inode,
2482 u64 new_size, u32 min_type)
39279cc3
CM
2483{
2484 int ret;
2485 struct btrfs_path *path;
2486 struct btrfs_key key;
5f39d397 2487 struct btrfs_key found_key;
39279cc3 2488 u32 found_type;
5f39d397 2489 struct extent_buffer *leaf;
39279cc3
CM
2490 struct btrfs_file_extent_item *fi;
2491 u64 extent_start = 0;
db94535d 2492 u64 extent_num_bytes = 0;
39279cc3 2493 u64 item_end = 0;
7bb86316 2494 u64 root_gen = 0;
d8d5f3e1 2495 u64 root_owner = 0;
39279cc3
CM
2496 int found_extent;
2497 int del_item;
85e21bac
CM
2498 int pending_del_nr = 0;
2499 int pending_del_slot = 0;
179e29e4 2500 int extent_type = -1;
771ed689 2501 int encoding;
3b951516 2502 u64 mask = root->sectorsize - 1;
39279cc3 2503
e02119d5 2504 if (root->ref_cows)
5b21f2ed 2505 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2506 path = btrfs_alloc_path();
3c69faec 2507 path->reada = -1;
39279cc3 2508 BUG_ON(!path);
5f39d397 2509
39279cc3
CM
2510 /* FIXME, add redo link to tree so we don't leak on crash */
2511 key.objectid = inode->i_ino;
2512 key.offset = (u64)-1;
5f39d397
CM
2513 key.type = (u8)-1;
2514
85e21bac 2515 btrfs_init_path(path);
323ac95b 2516
85e21bac
CM
2517search_again:
2518 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2519 if (ret < 0) {
2520 goto error;
2521 }
2522 if (ret > 0) {
e02119d5
CM
2523 /* there are no items in the tree for us to truncate, we're
2524 * done
2525 */
2526 if (path->slots[0] == 0) {
2527 ret = 0;
2528 goto error;
2529 }
85e21bac
CM
2530 path->slots[0]--;
2531 }
2532
39279cc3 2533 while(1) {
39279cc3 2534 fi = NULL;
5f39d397
CM
2535 leaf = path->nodes[0];
2536 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2537 found_type = btrfs_key_type(&found_key);
771ed689 2538 encoding = 0;
39279cc3 2539
5f39d397 2540 if (found_key.objectid != inode->i_ino)
39279cc3 2541 break;
5f39d397 2542
85e21bac 2543 if (found_type < min_type)
39279cc3
CM
2544 break;
2545
5f39d397 2546 item_end = found_key.offset;
39279cc3 2547 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2548 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2549 struct btrfs_file_extent_item);
179e29e4 2550 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM
2551 encoding = btrfs_file_extent_compression(leaf, fi);
2552 encoding |= btrfs_file_extent_encryption(leaf, fi);
2553 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2554
179e29e4 2555 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2556 item_end +=
db94535d 2557 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2558 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2559 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2560 fi);
39279cc3 2561 }
008630c1 2562 item_end--;
39279cc3 2563 }
e02119d5 2564 if (item_end < new_size) {
b888db2b
CM
2565 if (found_type == BTRFS_DIR_ITEM_KEY) {
2566 found_type = BTRFS_INODE_ITEM_KEY;
2567 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
d20f7043 2568 found_type = BTRFS_EXTENT_DATA_KEY;
85e21bac
CM
2569 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2570 found_type = BTRFS_XATTR_ITEM_KEY;
2571 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2572 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM
2573 } else if (found_type) {
2574 found_type--;
2575 } else {
2576 break;
39279cc3 2577 }
a61721d5 2578 btrfs_set_key_type(&key, found_type);
85e21bac 2579 goto next;
39279cc3 2580 }
e02119d5 2581 if (found_key.offset >= new_size)
39279cc3
CM
2582 del_item = 1;
2583 else
2584 del_item = 0;
2585 found_extent = 0;
2586
2587 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM
2588 if (found_type != BTRFS_EXTENT_DATA_KEY)
2589 goto delete;
2590
2591 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2592 u64 num_dec;
db94535d 2593 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2594 if (!del_item && !encoding) {
db94535d
CM
2595 u64 orig_num_bytes =
2596 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2597 extent_num_bytes = new_size -
5f39d397 2598 found_key.offset + root->sectorsize - 1;
b1632b10
Y
2599 extent_num_bytes = extent_num_bytes &
2600 ~((u64)root->sectorsize - 1);
db94535d
CM
2601 btrfs_set_file_extent_num_bytes(leaf, fi,
2602 extent_num_bytes);
2603 num_dec = (orig_num_bytes -
9069218d 2604 extent_num_bytes);
e02119d5 2605 if (root->ref_cows && extent_start != 0)
a76a3cd4 2606 inode_sub_bytes(inode, num_dec);
5f39d397 2607 btrfs_mark_buffer_dirty(leaf);
39279cc3 2608 } else {
db94535d
CM
2609 extent_num_bytes =
2610 btrfs_file_extent_disk_num_bytes(leaf,
2611 fi);
39279cc3 2612 /* FIXME blocksize != 4096 */
9069218d 2613 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM
2614 if (extent_start != 0) {
2615 found_extent = 1;
e02119d5 2616 if (root->ref_cows)
a76a3cd4 2617 inode_sub_bytes(inode, num_dec);
e02119d5 2618 }
31840ae1 2619 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2620 root_owner = btrfs_header_owner(leaf);
39279cc3 2621 }
9069218d 2622 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM
2623 /*
2624 * we can't truncate inline items that have had
2625 * special encodings
2626 */
2627 if (!del_item &&
2628 btrfs_file_extent_compression(leaf, fi) == 0 &&
2629 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2630 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM
2631 u32 size = new_size - found_key.offset;
2632
2633 if (root->ref_cows) {
a76a3cd4
YZ
2634 inode_sub_bytes(inode, item_end + 1 -
2635 new_size);
e02119d5
CM
2636 }
2637 size =
2638 btrfs_file_extent_calc_inline_size(size);
9069218d 2639 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2640 size, 1);
9069218d 2641 BUG_ON(ret);
e02119d5 2642 } else if (root->ref_cows) {
a76a3cd4
YZ
2643 inode_sub_bytes(inode, item_end + 1 -
2644 found_key.offset);
9069218d 2645 }
39279cc3 2646 }
179e29e4 2647delete:
39279cc3 2648 if (del_item) {
85e21bac
CM
2649 if (!pending_del_nr) {
2650 /* no pending yet, add ourselves */
2651 pending_del_slot = path->slots[0];
2652 pending_del_nr = 1;
2653 } else if (pending_del_nr &&
2654 path->slots[0] + 1 == pending_del_slot) {
2655 /* hop on the pending chunk */
2656 pending_del_nr++;
2657 pending_del_slot = path->slots[0];
2658 } else {
2659 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2660 }
39279cc3
CM
2661 } else {
2662 break;
2663 }
39279cc3
CM
2664 if (found_extent) {
2665 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2666 extent_num_bytes,
31840ae1 2667 leaf->start, root_owner,
3bb1a1bc 2668 root_gen, inode->i_ino, 0);
39279cc3
CM
2669 BUG_ON(ret);
2670 }
85e21bac
CM
2671next:
2672 if (path->slots[0] == 0) {
2673 if (pending_del_nr)
2674 goto del_pending;
2675 btrfs_release_path(root, path);
2676 goto search_again;
2677 }
2678
2679 path->slots[0]--;
2680 if (pending_del_nr &&
2681 path->slots[0] + 1 != pending_del_slot) {
2682 struct btrfs_key debug;
2683del_pending:
2684 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2685 pending_del_slot);
2686 ret = btrfs_del_items(trans, root, path,
2687 pending_del_slot,
2688 pending_del_nr);
2689 BUG_ON(ret);
2690 pending_del_nr = 0;
2691 btrfs_release_path(root, path);
2692 goto search_again;
2693 }
39279cc3
CM
2694 }
2695 ret = 0;
2696error:
85e21bac
CM
2697 if (pending_del_nr) {
2698 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2699 pending_del_nr);
2700 }
39279cc3
CM
2701 btrfs_free_path(path);
2702 inode->i_sb->s_dirt = 1;
2703 return ret;
2704}
2705
2706/*
2707 * taken from block_truncate_page, but does cow as it zeros out
2708 * any bytes left in the last page in the file.
2709 */
2710static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2711{
2712 struct inode *inode = mapping->host;
db94535d 2713 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM
2714 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2715 struct btrfs_ordered_extent *ordered;
2716 char *kaddr;
db94535d 2717 u32 blocksize = root->sectorsize;
39279cc3
CM
2718 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2719 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2720 struct page *page;
39279cc3 2721 int ret = 0;
a52d9a80 2722 u64 page_start;
e6dcd2dc 2723 u64 page_end;
39279cc3
CM
2724
2725 if ((offset & (blocksize - 1)) == 0)
2726 goto out;
2727
2728 ret = -ENOMEM;
211c17f5 2729again:
39279cc3
CM
2730 page = grab_cache_page(mapping, index);
2731 if (!page)
2732 goto out;
e6dcd2dc
CM
2733
2734 page_start = page_offset(page);
2735 page_end = page_start + PAGE_CACHE_SIZE - 1;
2736
39279cc3 2737 if (!PageUptodate(page)) {
9ebefb18 2738 ret = btrfs_readpage(NULL, page);
39279cc3 2739 lock_page(page);
211c17f5
CM
2740 if (page->mapping != mapping) {
2741 unlock_page(page);
2742 page_cache_release(page);
2743 goto again;
2744 }
39279cc3
CM
2745 if (!PageUptodate(page)) {
2746 ret = -EIO;
89642229 2747 goto out_unlock;
39279cc3
CM
2748 }
2749 }
211c17f5 2750 wait_on_page_writeback(page);
e6dcd2dc
CM
2751
2752 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2753 set_page_extent_mapped(page);
2754
2755 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2756 if (ordered) {
2757 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2758 unlock_page(page);
2759 page_cache_release(page);
eb84ae03 2760 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM
2761 btrfs_put_ordered_extent(ordered);
2762 goto again;
2763 }
2764
ea8c2819 2765 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM
2766 ret = 0;
2767 if (offset != PAGE_CACHE_SIZE) {
2768 kaddr = kmap(page);
2769 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2770 flush_dcache_page(page);
2771 kunmap(page);
2772 }
247e743c 2773 ClearPageChecked(page);
e6dcd2dc
CM
2774 set_page_dirty(page);
2775 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2776
89642229 2777out_unlock:
39279cc3
CM
2778 unlock_page(page);
2779 page_cache_release(page);
2780out:
2781 return ret;
2782}
2783
9036c102 2784int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2785{
9036c102
YZ
2786 struct btrfs_trans_handle *trans;
2787 struct btrfs_root *root = BTRFS_I(inode)->root;
2788 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2789 struct extent_map *em;
2790 u64 mask = root->sectorsize - 1;
2791 u64 hole_start = (inode->i_size + mask) & ~mask;
2792 u64 block_end = (size + mask) & ~mask;
2793 u64 last_byte;
2794 u64 cur_offset;
2795 u64 hole_size;
39279cc3
CM
2796 int err;
2797
9036c102
YZ
2798 if (size <= hole_start)
2799 return 0;
2800
2801 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM
2802 if (err)
2803 return err;
2804
9036c102 2805 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2806
9036c102
YZ
2807 while (1) {
2808 struct btrfs_ordered_extent *ordered;
2809 btrfs_wait_ordered_range(inode, hole_start,
2810 block_end - hole_start);
2811 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2812 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2813 if (!ordered)
2814 break;
2815 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2816 btrfs_put_ordered_extent(ordered);
2817 }
39279cc3 2818
9036c102
YZ
2819 trans = btrfs_start_transaction(root, 1);
2820 btrfs_set_trans_block_group(trans, inode);
39279cc3 2821
9036c102
YZ
2822 cur_offset = hole_start;
2823 while (1) {
2824 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2825 block_end - cur_offset, 0);
2826 BUG_ON(IS_ERR(em) || !em);
2827 last_byte = min(extent_map_end(em), block_end);
2828 last_byte = (last_byte + mask) & ~mask;
2829 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2830 u64 hint_byte = 0;
9036c102 2831 hole_size = last_byte - cur_offset;
771ed689
CM
2832 err = btrfs_drop_extents(trans, root, inode,
2833 cur_offset,
2834 cur_offset + hole_size,
2835 cur_offset, &hint_byte);
2836 if (err)
2837 break;
9036c102
YZ
2838 err = btrfs_insert_file_extent(trans, root,
2839 inode->i_ino, cur_offset, 0,
2840 0, hole_size, 0, hole_size,
2841 0, 0, 0);
2842 btrfs_drop_extent_cache(inode, hole_start,
2843 last_byte - 1, 0);
2844 }
2845 free_extent_map(em);
2846 cur_offset = last_byte;
2847 if (err || cur_offset >= block_end)
2848 break;
2849 }
1832a6d5 2850
9036c102
YZ
2851 btrfs_end_transaction(trans, root);
2852 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2853 return err;
2854}
39279cc3 2855
9036c102
YZ
2856static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2857{
2858 struct inode *inode = dentry->d_inode;
2859 int err;
39279cc3 2860
9036c102
YZ
2861 err = inode_change_ok(inode, attr);
2862 if (err)
2863 return err;
2bf5a725 2864
9036c102
YZ
2865 if (S_ISREG(inode->i_mode) &&
2866 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2867 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM
2868 if (err)
2869 return err;
39279cc3 2870 }
9036c102 2871
39279cc3 2872 err = inode_setattr(inode, attr);
33268eaf
JB
2873
2874 if (!err && ((attr->ia_valid & ATTR_MODE)))
2875 err = btrfs_acl_chmod(inode);
39279cc3
CM
2876 return err;
2877}
61295eb8 2878
39279cc3
CM
2879void btrfs_delete_inode(struct inode *inode)
2880{
2881 struct btrfs_trans_handle *trans;
2882 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2883 unsigned long nr;
39279cc3
CM
2884 int ret;
2885
2886 truncate_inode_pages(&inode->i_data, 0);
2887 if (is_bad_inode(inode)) {
7b128766 2888 btrfs_orphan_del(NULL, inode);
39279cc3
CM
2889 goto no_delete;
2890 }
4a096752 2891 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2892
dbe674a9 2893 btrfs_i_size_write(inode, 0);
39279cc3 2894 trans = btrfs_start_transaction(root, 1);
5f39d397 2895
39279cc3 2896 btrfs_set_trans_block_group(trans, inode);
e02119d5 2897 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB
2898 if (ret) {
2899 btrfs_orphan_del(NULL, inode);
54aa1f4d 2900 goto no_delete_lock;
7b128766
JB
2901 }
2902
2903 btrfs_orphan_del(trans, inode);
85e21bac 2904
d3c2fdcf 2905 nr = trans->blocks_used;
85e21bac 2906 clear_inode(inode);
5f39d397 2907
39279cc3 2908 btrfs_end_transaction(trans, root);
d3c2fdcf 2909 btrfs_btree_balance_dirty(root, nr);
39279cc3 2910 return;
54aa1f4d
CM
2911
2912no_delete_lock:
d3c2fdcf 2913 nr = trans->blocks_used;
54aa1f4d 2914 btrfs_end_transaction(trans, root);
d3c2fdcf 2915 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
2916no_delete:
2917 clear_inode(inode);
2918}
2919
2920/*
2921 * this returns the key found in the dir entry in the location pointer.
2922 * If no dir entries were found, location->objectid is 0.
2923 */
2924static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2925 struct btrfs_key *location)
2926{
2927 const char *name = dentry->d_name.name;
2928 int namelen = dentry->d_name.len;
2929 struct btrfs_dir_item *di;
2930 struct btrfs_path *path;
2931 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2932 int ret = 0;
39279cc3
CM
2933
2934 path = btrfs_alloc_path();
2935 BUG_ON(!path);
3954401f 2936
39279cc3
CM
2937 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2938 namelen, 0);
0d9f7f3e
Y
2939 if (IS_ERR(di))
2940 ret = PTR_ERR(di);
39279cc3 2941 if (!di || IS_ERR(di)) {
3954401f 2942 goto out_err;
39279cc3 2943 }
5f39d397 2944 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2945out:
39279cc3
CM
2946 btrfs_free_path(path);
2947 return ret;
3954401f
CM
2948out_err:
2949 location->objectid = 0;
2950 goto out;
39279cc3
CM
2951}
2952
2953/*
2954 * when we hit a tree root in a directory, the btrfs part of the inode
2955 * needs to be changed to reflect the root directory of the tree root. This
2956 * is kind of like crossing a mount point.
2957 */
2958static int fixup_tree_root_location(struct btrfs_root *root,
2959 struct btrfs_key *location,
58176a96
JB
2960 struct btrfs_root **sub_root,
2961 struct dentry *dentry)
39279cc3 2962{
39279cc3
CM
2963 struct btrfs_root_item *ri;
2964
2965 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2966 return 0;
2967 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2968 return 0;
2969
58176a96
JB
2970 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2971 dentry->d_name.name,
2972 dentry->d_name.len);
39279cc3
CM
2973 if (IS_ERR(*sub_root))
2974 return PTR_ERR(*sub_root);
2975
2976 ri = &(*sub_root)->root_item;
2977 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM
2978 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2979 location->offset = 0;
2980
39279cc3
CM
2981 return 0;
2982}
2983
e02119d5 2984static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2985{
e02119d5
CM
2986 struct btrfs_inode *bi = BTRFS_I(inode);
2987
2988 bi->i_acl = NULL;
2989 bi->i_default_acl = NULL;
2990
2991 bi->generation = 0;
c3027eb5 2992 bi->sequence = 0;
e02119d5
CM
2993 bi->last_trans = 0;
2994 bi->logged_trans = 0;
2995 bi->delalloc_bytes = 0;
2996 bi->disk_i_size = 0;
2997 bi->flags = 0;
2998 bi->index_cnt = (u64)-1;
49eb7e46 2999 bi->log_dirty_trans = 0;
d1310b2e
CM
3000 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3001 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 3002 inode->i_mapping, GFP_NOFS);
7e38326f
CM
3003 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3004 inode->i_mapping, GFP_NOFS);
ea8c2819 3005 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 3006 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 3007 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM
3008 mutex_init(&BTRFS_I(inode)->log_mutex);
3009}
3010
3011static int btrfs_init_locked_inode(struct inode *inode, void *p)
3012{
3013 struct btrfs_iget_args *args = p;
3014 inode->i_ino = args->ino;
3015 init_btrfs_i(inode);
3016 BTRFS_I(inode)->root = args->root;
39279cc3
CM
3017 return 0;
3018}
3019
3020static int btrfs_find_actor(struct inode *inode, void *opaque)
3021{
3022 struct btrfs_iget_args *args = opaque;
3023 return (args->ino == inode->i_ino &&
3024 args->root == BTRFS_I(inode)->root);
3025}
3026
5b21f2ed
ZY
3027struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3028 struct btrfs_root *root, int wait)
3029{
3030 struct inode *inode;
3031 struct btrfs_iget_args args;
3032 args.ino = objectid;
3033 args.root = root;
3034
3035 if (wait) {
3036 inode = ilookup5(s, objectid, btrfs_find_actor,
3037 (void *)&args);
3038 } else {
3039 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3040 (void *)&args);
3041 }
3042 return inode;
3043}
3044
39279cc3
CM
3045struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3046 struct btrfs_root *root)
3047{
3048 struct inode *inode;
3049 struct btrfs_iget_args args;
3050 args.ino = objectid;
3051 args.root = root;
3052
3053 inode = iget5_locked(s, objectid, btrfs_find_actor,
3054 btrfs_init_locked_inode,
3055 (void *)&args);
3056 return inode;
3057}
3058
1a54ef8c
BR
3059/* Get an inode object given its location and corresponding root.
3060 * Returns in *is_new if the inode was read from disk
3061 */
3062struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3063 struct btrfs_root *root, int *is_new)
3064{
3065 struct inode *inode;
3066
3067 inode = btrfs_iget_locked(s, location->objectid, root);
3068 if (!inode)
3069 return ERR_PTR(-EACCES);
3070
3071 if (inode->i_state & I_NEW) {
3072 BTRFS_I(inode)->root = root;
3073 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3074 btrfs_read_locked_inode(inode);
3075 unlock_new_inode(inode);
3076 if (is_new)
3077 *is_new = 1;
3078 } else {
3079 if (is_new)
3080 *is_new = 0;
3081 }
3082
3083 return inode;
3084}
3085
3de4586c 3086struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3
CM
3087{
3088 struct inode * inode;
3089 struct btrfs_inode *bi = BTRFS_I(dir);
3090 struct btrfs_root *root = bi->root;
3091 struct btrfs_root *sub_root = root;
3092 struct btrfs_key location;
c146afad 3093 int ret, new;
39279cc3
CM
3094
3095 if (dentry->d_name.len > BTRFS_NAME_LEN)
3096 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3097
39279cc3 3098 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3099
39279cc3
CM
3100 if (ret < 0)
3101 return ERR_PTR(ret);
5f39d397 3102
39279cc3
CM
3103 inode = NULL;
3104 if (location.objectid) {
58176a96
JB
3105 ret = fixup_tree_root_location(root, &location, &sub_root,
3106 dentry);
39279cc3
CM
3107 if (ret < 0)
3108 return ERR_PTR(ret);
3109 if (ret > 0)
3110 return ERR_PTR(-ENOENT);
1a54ef8c
BR
3111 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3112 if (IS_ERR(inode))
3113 return ERR_CAST(inode);
39279cc3 3114 }
3de4586c
CM
3115 return inode;
3116}
3117
3118static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3119 struct nameidata *nd)
3120{
3121 struct inode *inode;
3122
3123 if (dentry->d_name.len > BTRFS_NAME_LEN)
3124 return ERR_PTR(-ENAMETOOLONG);
3125
3126 inode = btrfs_lookup_dentry(dir, dentry);
3127 if (IS_ERR(inode))
3128 return ERR_CAST(inode);
7b128766 3129
39279cc3
CM
3130 return d_splice_alias(inode, dentry);
3131}
3132
39279cc3
CM
3133static unsigned char btrfs_filetype_table[] = {
3134 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3135};
3136
cbdf5a24
DW
3137static int btrfs_real_readdir(struct file *filp, void *dirent,
3138 filldir_t filldir)
39279cc3 3139{
6da6abae 3140 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM
3141 struct btrfs_root *root = BTRFS_I(inode)->root;
3142 struct btrfs_item *item;
3143 struct btrfs_dir_item *di;
3144 struct btrfs_key key;
5f39d397 3145 struct btrfs_key found_key;
39279cc3
CM
3146 struct btrfs_path *path;
3147 int ret;
3148 u32 nritems;
5f39d397 3149 struct extent_buffer *leaf;
39279cc3
CM
3150 int slot;
3151 int advance;
3152 unsigned char d_type;
3153 int over = 0;
3154 u32 di_cur;
3155 u32 di_total;
3156 u32 di_len;
3157 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM
3158 char tmp_name[32];
3159 char *name_ptr;
3160 int name_len;
39279cc3
CM
3161
3162 /* FIXME, use a real flag for deciding about the key type */
3163 if (root->fs_info->tree_root == root)
3164 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3165
3954401f
CM
3166 /* special case for "." */
3167 if (filp->f_pos == 0) {
3168 over = filldir(dirent, ".", 1,
3169 1, inode->i_ino,
3170 DT_DIR);
3171 if (over)
3172 return 0;
3173 filp->f_pos = 1;
3174 }
3954401f
CM
3175 /* special case for .., just use the back ref */
3176 if (filp->f_pos == 1) {
5ecc7e5d 3177 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3178 over = filldir(dirent, "..", 2,
5ecc7e5d 3179 2, pino, DT_DIR);
3954401f 3180 if (over)
49593bfa 3181 return 0;
3954401f
CM
3182 filp->f_pos = 2;
3183 }
49593bfa
DW
3184 path = btrfs_alloc_path();
3185 path->reada = 2;
3186
39279cc3
CM
3187 btrfs_set_key_type(&key, key_type);
3188 key.offset = filp->f_pos;
49593bfa 3189 key.objectid = inode->i_ino;
5f39d397 3190
39279cc3
CM
3191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3192 if (ret < 0)
3193 goto err;
3194 advance = 0;
49593bfa
DW
3195
3196 while (1) {
5f39d397
CM
3197 leaf = path->nodes[0];
3198 nritems = btrfs_header_nritems(leaf);
39279cc3
CM
3199 slot = path->slots[0];
3200 if (advance || slot >= nritems) {
49593bfa 3201 if (slot >= nritems - 1) {
39279cc3
CM
3202 ret = btrfs_next_leaf(root, path);
3203 if (ret)
3204 break;
5f39d397
CM
3205 leaf = path->nodes[0];
3206 nritems = btrfs_header_nritems(leaf);
39279cc3
CM
3207 slot = path->slots[0];
3208 } else {
3209 slot++;
3210 path->slots[0]++;
3211 }
3212 }
3de4586c 3213
39279cc3 3214 advance = 1;
5f39d397
CM
3215 item = btrfs_item_nr(leaf, slot);
3216 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3217
3218 if (found_key.objectid != key.objectid)
39279cc3 3219 break;
5f39d397 3220 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3221 break;
5f39d397 3222 if (found_key.offset < filp->f_pos)
39279cc3 3223 continue;
5f39d397
CM
3224
3225 filp->f_pos = found_key.offset;
49593bfa 3226
39279cc3
CM
3227 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3228 di_cur = 0;
5f39d397 3229 di_total = btrfs_item_size(leaf, item);
49593bfa
DW
3230
3231 while (di_cur < di_total) {
5f39d397
CM
3232 struct btrfs_key location;
3233
3234 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3235 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM
3236 name_ptr = tmp_name;
3237 } else {
3238 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW
3239 if (!name_ptr) {
3240 ret = -ENOMEM;
3241 goto err;
3242 }
5f39d397
CM
3243 }
3244 read_extent_buffer(leaf, name_ptr,
3245 (unsigned long)(di + 1), name_len);
3246
3247 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3248 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM
3249
3250 /* is this a reference to our own snapshot? If so
3251 * skip it
3252 */
3253 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3254 location.objectid == root->root_key.objectid) {
3255 over = 0;
3256 goto skip;
3257 }
5f39d397 3258 over = filldir(dirent, name_ptr, name_len,
49593bfa 3259 found_key.offset, location.objectid,
39279cc3 3260 d_type);
5f39d397 3261
3de4586c 3262skip:
5f39d397
CM
3263 if (name_ptr != tmp_name)
3264 kfree(name_ptr);
3265
39279cc3
CM
3266 if (over)
3267 goto nopos;
5103e947 3268 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3269 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM
3270 di_cur += di_len;
3271 di = (struct btrfs_dir_item *)((char *)di + di_len);
3272 }
3273 }
49593bfa
DW
3274
3275 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ
3276 if (key_type == BTRFS_DIR_INDEX_KEY)
3277 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3278 else
3279 filp->f_pos++;
39279cc3
CM
3280nopos:
3281 ret = 0;
3282err:
39279cc3 3283 btrfs_free_path(path);
39279cc3
CM
3284 return ret;
3285}
3286
3287int btrfs_write_inode(struct inode *inode, int wait)
3288{
3289 struct btrfs_root *root = BTRFS_I(inode)->root;
3290 struct btrfs_trans_handle *trans;
3291 int ret = 0;
3292
c146afad 3293 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM
3294 return 0;
3295
39279cc3 3296 if (wait) {
f9295749 3297 trans = btrfs_join_transaction(root, 1);
39279cc3
CM
3298 btrfs_set_trans_block_group(trans, inode);
3299 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM
3300 }
3301 return ret;
3302}
3303
3304/*
54aa1f4d 3305 * This is somewhat expensive, updating the tree every time the
39279cc3
CM
3306 * inode changes. But, it is most likely to find the inode in cache.
3307 * FIXME, needs more benchmarking...there are no reasons other than performance
3308 * to keep or drop this code.
3309 */
3310void btrfs_dirty_inode(struct inode *inode)
3311{
3312 struct btrfs_root *root = BTRFS_I(inode)->root;
3313 struct btrfs_trans_handle *trans;
3314
f9295749 3315 trans = btrfs_join_transaction(root, 1);
39279cc3
CM
3316 btrfs_set_trans_block_group(trans, inode);
3317 btrfs_update_inode(trans, root, inode);
3318 btrfs_end_transaction(trans, root);
39279cc3
CM
3319}
3320
d352ac68
CM
3321/*
3322 * find the highest existing sequence number in a directory
3323 * and then set the in-memory index_cnt variable to reflect
3324 * free sequence numbers
3325 */
aec7477b
JB
3326static int btrfs_set_inode_index_count(struct inode *inode)
3327{
3328 struct btrfs_root *root = BTRFS_I(inode)->root;
3329 struct btrfs_key key, found_key;
3330 struct btrfs_path *path;
3331 struct extent_buffer *leaf;
3332 int ret;
3333
3334 key.objectid = inode->i_ino;
3335 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3336 key.offset = (u64)-1;
3337
3338 path = btrfs_alloc_path();
3339 if (!path)
3340 return -ENOMEM;
3341
3342 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3343 if (ret < 0)
3344 goto out;
3345 /* FIXME: we should be able to handle this */
3346 if (ret == 0)
3347 goto out;
3348 ret = 0;
3349
3350 /*
3351 * MAGIC NUMBER EXPLANATION:
3352 * since we search a directory based on f_pos we have to start at 2
3353 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3354 * else has to start at 2
3355 */
3356 if (path->slots[0] == 0) {
3357 BTRFS_I(inode)->index_cnt = 2;
3358 goto out;
3359 }
3360
3361 path->slots[0]--;
3362
3363 leaf = path->nodes[0];
3364 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3365
3366 if (found_key.objectid != inode->i_ino ||
3367 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3368 BTRFS_I(inode)->index_cnt = 2;
3369 goto out;
3370 }
3371
3372 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3373out:
3374 btrfs_free_path(path);
3375 return ret;
3376}
3377
d352ac68
CM
3378/*
3379 * helper to find a free sequence number in a given directory. This current
3380 * code is very simple, later versions will do smarter things in the btree
3381 */
3de4586c 3382int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB
3383{
3384 int ret = 0;
3385
3386 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3387 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 3388 if (ret) {
aec7477b 3389 return ret;
8d5bf1cb 3390 }
aec7477b
JB
3391 }
3392
00e4e6b3 3393 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB
3394 BTRFS_I(dir)->index_cnt++;
3395
3396 return ret;
3397}
3398
39279cc3
CM
3399static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3400 struct btrfs_root *root,
aec7477b 3401 struct inode *dir,
9c58309d 3402 const char *name, int name_len,
d2fb3437
YZ
3403 u64 ref_objectid, u64 objectid,
3404 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM
3405{
3406 struct inode *inode;
5f39d397 3407 struct btrfs_inode_item *inode_item;
39279cc3 3408 struct btrfs_key *location;
5f39d397 3409 struct btrfs_path *path;
9c58309d
CM
3410 struct btrfs_inode_ref *ref;
3411 struct btrfs_key key[2];
3412 u32 sizes[2];
3413 unsigned long ptr;
39279cc3
CM
3414 int ret;
3415 int owner;
3416
5f39d397
CM
3417 path = btrfs_alloc_path();
3418 BUG_ON(!path);
3419
39279cc3
CM
3420 inode = new_inode(root->fs_info->sb);
3421 if (!inode)
3422 return ERR_PTR(-ENOMEM);
3423
aec7477b 3424 if (dir) {
3de4586c 3425 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB
3426 if (ret)
3427 return ERR_PTR(ret);
aec7477b
JB
3428 }
3429 /*
3430 * index_cnt is ignored for everything but a dir,
3431 * btrfs_get_inode_index_count has an explanation for the magic
3432 * number
3433 */
e02119d5 3434 init_btrfs_i(inode);
aec7477b 3435 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3436 BTRFS_I(inode)->root = root;
e02119d5 3437 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3438
39279cc3
CM
3439 if (mode & S_IFDIR)
3440 owner = 0;
3441 else
3442 owner = 1;
d2fb3437
YZ
3443 BTRFS_I(inode)->block_group =
3444 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ
3445 if ((mode & S_IFREG)) {
3446 if (btrfs_test_opt(root, NODATASUM))
3447 btrfs_set_flag(inode, NODATASUM);
3448 if (btrfs_test_opt(root, NODATACOW))
3449 btrfs_set_flag(inode, NODATACOW);
3450 }
9c58309d
CM
3451
3452 key[0].objectid = objectid;
3453 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3454 key[0].offset = 0;
3455
3456 key[1].objectid = objectid;
3457 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3458 key[1].offset = ref_objectid;
3459
3460 sizes[0] = sizeof(struct btrfs_inode_item);
3461 sizes[1] = name_len + sizeof(*ref);
3462
3463 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3464 if (ret != 0)
5f39d397
CM
3465 goto fail;
3466
9c58309d
CM
3467 if (objectid > root->highest_inode)
3468 root->highest_inode = objectid;
3469
79683f2d
CM
3470 inode->i_uid = current_fsuid();
3471 inode->i_gid = current_fsgid();
39279cc3
CM
3472 inode->i_mode = mode;
3473 inode->i_ino = objectid;
a76a3cd4 3474 inode_set_bytes(inode, 0);
39279cc3 3475 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM
3476 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3477 struct btrfs_inode_item);
e02119d5 3478 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM
3479
3480 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3481 struct btrfs_inode_ref);
3482 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3483 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM
3484 ptr = (unsigned long)(ref + 1);
3485 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3486
5f39d397
CM
3487 btrfs_mark_buffer_dirty(path->nodes[0]);
3488 btrfs_free_path(path);
3489
39279cc3
CM
3490 location = &BTRFS_I(inode)->location;
3491 location->objectid = objectid;
39279cc3
CM
3492 location->offset = 0;
3493 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3494
39279cc3
CM
3495 insert_inode_hash(inode);
3496 return inode;
5f39d397 3497fail:
aec7477b
JB
3498 if (dir)
3499 BTRFS_I(dir)->index_cnt--;
5f39d397
CM
3500 btrfs_free_path(path);
3501 return ERR_PTR(ret);
39279cc3
CM
3502}
3503
3504static inline u8 btrfs_inode_type(struct inode *inode)
3505{
3506 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3507}
3508
d352ac68
CM
3509/*
3510 * utility function to add 'inode' into 'parent_inode' with
3511 * a give name and a given sequence number.
3512 * if 'add_backref' is true, also insert a backref from the
3513 * inode to the parent directory.
3514 */
e02119d5
CM
3515int btrfs_add_link(struct btrfs_trans_handle *trans,
3516 struct inode *parent_inode, struct inode *inode,
3517 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM
3518{
3519 int ret;
3520 struct btrfs_key key;
e02119d5 3521 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3522
39279cc3 3523 key.objectid = inode->i_ino;
39279cc3
CM
3524 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3525 key.offset = 0;
3526
e02119d5
CM
3527 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3528 parent_inode->i_ino,
aec7477b 3529 &key, btrfs_inode_type(inode),
00e4e6b3 3530 index);
39279cc3 3531 if (ret == 0) {
9c58309d
CM
3532 if (add_backref) {
3533 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM
3534 name, name_len,
3535 inode->i_ino,
3536 parent_inode->i_ino,
3537 index);
9c58309d 3538 }
dbe674a9 3539 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3540 name_len * 2);
79c44584 3541 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3542 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM
3543 }
3544 return ret;
3545}
3546
3547static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3548 struct dentry *dentry, struct inode *inode,
00e4e6b3 3549 int backref, u64 index)
39279cc3 3550{
e02119d5
CM
3551 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3552 inode, dentry->d_name.name,
3553 dentry->d_name.len, backref, index);
39279cc3
CM
3554 if (!err) {
3555 d_instantiate(dentry, inode);
3556 return 0;
3557 }
3558 if (err > 0)
3559 err = -EEXIST;
3560 return err;
3561}
3562
618e21d5
JB
3563static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3564 int mode, dev_t rdev)
3565{
3566 struct btrfs_trans_handle *trans;
3567 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3568 struct inode *inode = NULL;
618e21d5
JB
3569 int err;
3570 int drop_inode = 0;
3571 u64 objectid;
1832a6d5 3572 unsigned long nr = 0;
00e4e6b3 3573 u64 index = 0;
618e21d5
JB
3574
3575 if (!new_valid_dev(rdev))
3576 return -EINVAL;
3577
1832a6d5
CM
3578 err = btrfs_check_free_space(root, 1, 0);
3579 if (err)
3580 goto fail;
3581
618e21d5
JB
3582 trans = btrfs_start_transaction(root, 1);
3583 btrfs_set_trans_block_group(trans, dir);
3584
3585 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3586 if (err) {
3587 err = -ENOSPC;
3588 goto out_unlock;
3589 }
3590
aec7477b 3591 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM
3592 dentry->d_name.len,
3593 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3594 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB
3595 err = PTR_ERR(inode);
3596 if (IS_ERR(inode))
3597 goto out_unlock;
3598
33268eaf
JB
3599 err = btrfs_init_acl(inode, dir);
3600 if (err) {
3601 drop_inode = 1;
3602 goto out_unlock;
3603 }
3604
618e21d5 3605 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3606 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB
3607 if (err)
3608 drop_inode = 1;
3609 else {
3610 inode->i_op = &btrfs_special_inode_operations;
3611 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3612 btrfs_update_inode(trans, root, inode);
618e21d5
JB
3613 }
3614 dir->i_sb->s_dirt = 1;
3615 btrfs_update_inode_block_group(trans, inode);
3616 btrfs_update_inode_block_group(trans, dir);
3617out_unlock:
d3c2fdcf 3618 nr = trans->blocks_used;
89ce8a63 3619 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3620fail:
618e21d5
JB
3621 if (drop_inode) {
3622 inode_dec_link_count(inode);
3623 iput(inode);
3624 }
d3c2fdcf 3625 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB
3626 return err;
3627}
3628
39279cc3
CM
3629static int btrfs_create(struct inode *dir, struct dentry *dentry,
3630 int mode, struct nameidata *nd)
3631{
3632 struct btrfs_trans_handle *trans;
3633 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3634 struct inode *inode = NULL;
39279cc3
CM
3635 int err;
3636 int drop_inode = 0;
1832a6d5 3637 unsigned long nr = 0;
39279cc3 3638 u64 objectid;
00e4e6b3 3639 u64 index = 0;
39279cc3 3640
1832a6d5
CM
3641 err = btrfs_check_free_space(root, 1, 0);
3642 if (err)
3643 goto fail;
39279cc3
CM
3644 trans = btrfs_start_transaction(root, 1);
3645 btrfs_set_trans_block_group(trans, dir);
3646
3647 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3648 if (err) {
3649 err = -ENOSPC;
3650 goto out_unlock;
3651 }
3652
aec7477b 3653 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM
3654 dentry->d_name.len,
3655 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM
3656 objectid, BTRFS_I(dir)->block_group, mode,
3657 &index);
39279cc3
CM
3658 err = PTR_ERR(inode);
3659 if (IS_ERR(inode))
3660 goto out_unlock;
3661
33268eaf
JB
3662 err = btrfs_init_acl(inode, dir);
3663 if (err) {
3664 drop_inode = 1;
3665 goto out_unlock;
3666 }
3667
39279cc3 3668 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3669 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM
3670 if (err)
3671 drop_inode = 1;
3672 else {
3673 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3674 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM
3675 inode->i_fop = &btrfs_file_operations;
3676 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3677 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM
3678 }
3679 dir->i_sb->s_dirt = 1;
3680 btrfs_update_inode_block_group(trans, inode);
3681 btrfs_update_inode_block_group(trans, dir);
3682out_unlock:
d3c2fdcf 3683 nr = trans->blocks_used;
ab78c84d 3684 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3685fail:
39279cc3
CM
3686 if (drop_inode) {
3687 inode_dec_link_count(inode);
3688 iput(inode);
3689 }
d3c2fdcf 3690 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
3691 return err;
3692}
3693
3694static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3695 struct dentry *dentry)
3696{
3697 struct btrfs_trans_handle *trans;
3698 struct btrfs_root *root = BTRFS_I(dir)->root;
3699 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3700 u64 index;
1832a6d5 3701 unsigned long nr = 0;
39279cc3
CM
3702 int err;
3703 int drop_inode = 0;
3704
3705 if (inode->i_nlink == 0)
3706 return -ENOENT;
3707
e02119d5 3708 btrfs_inc_nlink(inode);
1832a6d5
CM
3709 err = btrfs_check_free_space(root, 1, 0);
3710 if (err)
3711 goto fail;
3de4586c 3712 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB
3713 if (err)
3714 goto fail;
3715
39279cc3 3716 trans = btrfs_start_transaction(root, 1);
5f39d397 3717
39279cc3
CM
3718 btrfs_set_trans_block_group(trans, dir);
3719 atomic_inc(&inode->i_count);
aec7477b 3720
00e4e6b3 3721 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3722
39279cc3
CM
3723 if (err)
3724 drop_inode = 1;
5f39d397 3725
39279cc3
CM
3726 dir->i_sb->s_dirt = 1;
3727 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3728 err = btrfs_update_inode(trans, root, inode);
5f39d397 3729
54aa1f4d
CM
3730 if (err)
3731 drop_inode = 1;
39279cc3 3732
d3c2fdcf 3733 nr = trans->blocks_used;
ab78c84d 3734 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3735fail:
39279cc3
CM
3736 if (drop_inode) {
3737 inode_dec_link_count(inode);
3738 iput(inode);
3739 }
d3c2fdcf 3740 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
3741 return err;
3742}
3743
39279cc3
CM
3744static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3745{
b9d86667 3746 struct inode *inode = NULL;
39279cc3
CM
3747 struct btrfs_trans_handle *trans;
3748 struct btrfs_root *root = BTRFS_I(dir)->root;
3749 int err = 0;
3750 int drop_on_err = 0;
b9d86667 3751 u64 objectid = 0;
00e4e6b3 3752 u64 index = 0;
d3c2fdcf 3753 unsigned long nr = 1;
39279cc3 3754
1832a6d5
CM
3755 err = btrfs_check_free_space(root, 1, 0);
3756 if (err)
3757 goto out_unlock;
3758
39279cc3
CM
3759 trans = btrfs_start_transaction(root, 1);
3760 btrfs_set_trans_block_group(trans, dir);
5f39d397 3761
39279cc3
CM
3762 if (IS_ERR(trans)) {
3763 err = PTR_ERR(trans);
3764 goto out_unlock;
3765 }
3766
3767 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3768 if (err) {
3769 err = -ENOSPC;
3770 goto out_unlock;
3771 }
3772
aec7477b 3773 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM
3774 dentry->d_name.len,
3775 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM
3776 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3777 &index);
39279cc3
CM
3778 if (IS_ERR(inode)) {
3779 err = PTR_ERR(inode);
3780 goto out_fail;
3781 }
5f39d397 3782
39279cc3 3783 drop_on_err = 1;
33268eaf
JB
3784
3785 err = btrfs_init_acl(inode, dir);
3786 if (err)
3787 goto out_fail;
3788
39279cc3
CM
3789 inode->i_op = &btrfs_dir_inode_operations;
3790 inode->i_fop = &btrfs_dir_file_operations;
3791 btrfs_set_trans_block_group(trans, inode);
3792
dbe674a9 3793 btrfs_i_size_write(inode, 0);
39279cc3
CM
3794 err = btrfs_update_inode(trans, root, inode);
3795 if (err)
3796 goto out_fail;
5f39d397 3797
e02119d5
CM
3798 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3799 inode, dentry->d_name.name,
3800 dentry->d_name.len, 0, index);
39279cc3
CM
3801 if (err)
3802 goto out_fail;
5f39d397 3803
39279cc3
CM
3804 d_instantiate(dentry, inode);
3805 drop_on_err = 0;
3806 dir->i_sb->s_dirt = 1;
3807 btrfs_update_inode_block_group(trans, inode);
3808 btrfs_update_inode_block_group(trans, dir);
3809
3810out_fail:
d3c2fdcf 3811 nr = trans->blocks_used;
ab78c84d 3812 btrfs_end_transaction_throttle(trans, root);
5f39d397 3813
39279cc3 3814out_unlock:
39279cc3
CM
3815 if (drop_on_err)
3816 iput(inode);
d3c2fdcf 3817 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
3818 return err;
3819}
3820
d352ac68
CM
3821/* helper for btfs_get_extent. Given an existing extent in the tree,
3822 * and an extent that you want to insert, deal with overlap and insert
3823 * the new extent into the tree.
3824 */
3b951516
CM
3825static int merge_extent_mapping(struct extent_map_tree *em_tree,
3826 struct extent_map *existing,
e6dcd2dc
CM
3827 struct extent_map *em,
3828 u64 map_start, u64 map_len)
3b951516
CM
3829{
3830 u64 start_diff;
3b951516 3831
e6dcd2dc
CM
3832 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3833 start_diff = map_start - em->start;
3834 em->start = map_start;
3835 em->len = map_len;
c8b97818
CM
3836 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3837 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3838 em->block_start += start_diff;
c8b97818
CM
3839 em->block_len -= start_diff;
3840 }
e6dcd2dc 3841 return add_extent_mapping(em_tree, em);
3b951516
CM
3842}
3843
c8b97818
CM
3844static noinline int uncompress_inline(struct btrfs_path *path,
3845 struct inode *inode, struct page *page,
3846 size_t pg_offset, u64 extent_offset,
3847 struct btrfs_file_extent_item *item)
3848{
3849 int ret;
3850 struct extent_buffer *leaf = path->nodes[0];
3851 char *tmp;
3852 size_t max_size;
3853 unsigned long inline_size;
3854 unsigned long ptr;
3855
3856 WARN_ON(pg_offset != 0);
3857 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3858 inline_size = btrfs_file_extent_inline_item_len(leaf,
3859 btrfs_item_nr(leaf, path->slots[0]));
3860 tmp = kmalloc(inline_size, GFP_NOFS);
3861 ptr = btrfs_file_extent_inline_start(item);
3862
3863 read_extent_buffer(leaf, tmp, ptr, inline_size);
3864
5b050f04 3865 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM
3866 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3867 inline_size, max_size);
3868 if (ret) {
3869 char *kaddr = kmap_atomic(page, KM_USER0);
3870 unsigned long copy_size = min_t(u64,
3871 PAGE_CACHE_SIZE - pg_offset,
3872 max_size - extent_offset);
3873 memset(kaddr + pg_offset, 0, copy_size);
3874 kunmap_atomic(kaddr, KM_USER0);
3875 }
3876 kfree(tmp);
3877 return 0;
3878}
3879
d352ac68
CM
3880/*
3881 * a bit scary, this does extent mapping from logical file offset to the disk.
3882 * the ugly parts come from merging extents from the disk with the
3883 * in-ram representation. This gets more complex because of the data=ordered code,
3884 * where the in-ram extents might be locked pending data=ordered completion.
3885 *
3886 * This also copies inline extents directly into the page.
3887 */
a52d9a80 3888struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3889 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM
3890 int create)
3891{
3892 int ret;
3893 int err = 0;
db94535d 3894 u64 bytenr;
a52d9a80
CM
3895 u64 extent_start = 0;
3896 u64 extent_end = 0;
3897 u64 objectid = inode->i_ino;
3898 u32 found_type;
f421950f 3899 struct btrfs_path *path = NULL;
a52d9a80
CM
3900 struct btrfs_root *root = BTRFS_I(inode)->root;
3901 struct btrfs_file_extent_item *item;
5f39d397
CM
3902 struct extent_buffer *leaf;
3903 struct btrfs_key found_key;
a52d9a80
CM
3904 struct extent_map *em = NULL;
3905 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3906 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3907 struct btrfs_trans_handle *trans = NULL;
c8b97818 3908 int compressed;
a52d9a80 3909
a52d9a80 3910again:
d1310b2e
CM
3911 spin_lock(&em_tree->lock);
3912 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM
3913 if (em)
3914 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM
3915 spin_unlock(&em_tree->lock);
3916
a52d9a80 3917 if (em) {
e1c4b745
CM
3918 if (em->start > start || em->start + em->len <= start)
3919 free_extent_map(em);
3920 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM
3921 free_extent_map(em);
3922 else
3923 goto out;
a52d9a80 3924 }
d1310b2e 3925 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3926 if (!em) {
d1310b2e
CM
3927 err = -ENOMEM;
3928 goto out;
a52d9a80 3929 }
e6dcd2dc 3930 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3931 em->start = EXTENT_MAP_HOLE;
445a6944 3932 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3933 em->len = (u64)-1;
c8b97818 3934 em->block_len = (u64)-1;
f421950f
CM
3935
3936 if (!path) {
3937 path = btrfs_alloc_path();
3938 BUG_ON(!path);
3939 }
3940
179e29e4
CM
3941 ret = btrfs_lookup_file_extent(trans, root, path,
3942 objectid, start, trans != NULL);
a52d9a80
CM
3943 if (ret < 0) {
3944 err = ret;
3945 goto out;
3946 }
3947
3948 if (ret != 0) {
3949 if (path->slots[0] == 0)
3950 goto not_found;
3951 path->slots[0]--;
3952 }
3953
5f39d397
CM
3954 leaf = path->nodes[0];
3955 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3956 struct btrfs_file_extent_item);
a52d9a80 3957 /* are we inside the extent that was found? */
5f39d397
CM
3958 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3959 found_type = btrfs_key_type(&found_key);
3960 if (found_key.objectid != objectid ||
a52d9a80
CM
3961 found_type != BTRFS_EXTENT_DATA_KEY) {
3962 goto not_found;
3963 }
3964
5f39d397
CM
3965 found_type = btrfs_file_extent_type(leaf, item);
3966 extent_start = found_key.offset;
c8b97818 3967 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ
3968 if (found_type == BTRFS_FILE_EXTENT_REG ||
3969 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3970 extent_end = extent_start +
db94535d 3971 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ
3972 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3973 size_t size;
3974 size = btrfs_file_extent_inline_len(leaf, item);
3975 extent_end = (extent_start + size + root->sectorsize - 1) &
3976 ~((u64)root->sectorsize - 1);
3977 }
3978
3979 if (start >= extent_end) {
3980 path->slots[0]++;
3981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3982 ret = btrfs_next_leaf(root, path);
3983 if (ret < 0) {
3984 err = ret;
3985 goto out;
a52d9a80 3986 }
9036c102
YZ
3987 if (ret > 0)
3988 goto not_found;
3989 leaf = path->nodes[0];
a52d9a80 3990 }
9036c102
YZ
3991 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3992 if (found_key.objectid != objectid ||
3993 found_key.type != BTRFS_EXTENT_DATA_KEY)
3994 goto not_found;
3995 if (start + len <= found_key.offset)
3996 goto not_found;
3997 em->start = start;
3998 em->len = found_key.offset - start;
3999 goto not_found_em;
4000 }
4001
d899e052
YZ
4002 if (found_type == BTRFS_FILE_EXTENT_REG ||
4003 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ
4004 em->start = extent_start;
4005 em->len = extent_end - extent_start;
ff5b7ee3
YZ
4006 em->orig_start = extent_start -
4007 btrfs_file_extent_offset(leaf, item);
db94535d
CM
4008 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4009 if (bytenr == 0) {
5f39d397 4010 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM
4011 goto insert;
4012 }
c8b97818
CM
4013 if (compressed) {
4014 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4015 em->block_start = bytenr;
4016 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4017 item);
4018 } else {
4019 bytenr += btrfs_file_extent_offset(leaf, item);
4020 em->block_start = bytenr;
4021 em->block_len = em->len;
d899e052
YZ
4022 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4023 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4024 }
a52d9a80
CM
4025 goto insert;
4026 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4027 unsigned long ptr;
a52d9a80 4028 char *map;
3326d1b0
CM
4029 size_t size;
4030 size_t extent_offset;
4031 size_t copy_size;
a52d9a80 4032
689f9346 4033 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4034 if (!page || create) {
689f9346 4035 em->start = extent_start;
9036c102 4036 em->len = extent_end - extent_start;
689f9346
Y
4037 goto out;
4038 }
5f39d397 4039
9036c102
YZ
4040 size = btrfs_file_extent_inline_len(leaf, item);
4041 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4042 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4043 size - extent_offset);
3326d1b0 4044 em->start = extent_start + extent_offset;
70dec807
CM
4045 em->len = (copy_size + root->sectorsize - 1) &
4046 ~((u64)root->sectorsize - 1);
ff5b7ee3 4047 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM
4048 if (compressed)
4049 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4050 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4051 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM
4052 if (btrfs_file_extent_compression(leaf, item) ==
4053 BTRFS_COMPRESS_ZLIB) {
4054 ret = uncompress_inline(path, inode, page,
4055 pg_offset,
4056 extent_offset, item);
4057 BUG_ON(ret);
4058 } else {
4059 map = kmap(page);
4060 read_extent_buffer(leaf, map + pg_offset, ptr,
4061 copy_size);
4062 kunmap(page);
4063 }
179e29e4
CM
4064 flush_dcache_page(page);
4065 } else if (create && PageUptodate(page)) {
4066 if (!trans) {
4067 kunmap(page);
4068 free_extent_map(em);
4069 em = NULL;
4070 btrfs_release_path(root, path);
f9295749 4071 trans = btrfs_join_transaction(root, 1);
179e29e4
CM
4072 goto again;
4073 }
c8b97818 4074 map = kmap(page);
70dec807 4075 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4076 copy_size);
c8b97818 4077 kunmap(page);
179e29e4 4078 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4079 }
d1310b2e
CM
4080 set_extent_uptodate(io_tree, em->start,
4081 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM
4082 goto insert;
4083 } else {
4084 printk("unkknown found_type %d\n", found_type);
4085 WARN_ON(1);
4086 }
4087not_found:
4088 em->start = start;
d1310b2e 4089 em->len = len;
a52d9a80 4090not_found_em:
5f39d397 4091 em->block_start = EXTENT_MAP_HOLE;
9036c102 4092 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM
4093insert:
4094 btrfs_release_path(root, path);
d1310b2e
CM
4095 if (em->start > start || extent_map_end(em) <= start) {
4096 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM
4097 err = -EIO;
4098 goto out;
4099 }
d1310b2e
CM
4100
4101 err = 0;
4102 spin_lock(&em_tree->lock);
a52d9a80 4103 ret = add_extent_mapping(em_tree, em);
3b951516
CM
4104 /* it is possible that someone inserted the extent into the tree
4105 * while we had the lock dropped. It is also possible that
4106 * an overlapping map exists in the tree
4107 */
a52d9a80 4108 if (ret == -EEXIST) {
3b951516 4109 struct extent_map *existing;
e6dcd2dc
CM
4110
4111 ret = 0;
4112
3b951516 4113 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM
4114 if (existing && (existing->start > start ||
4115 existing->start + existing->len <= start)) {
4116 free_extent_map(existing);
4117 existing = NULL;
4118 }
3b951516
CM
4119 if (!existing) {
4120 existing = lookup_extent_mapping(em_tree, em->start,
4121 em->len);
4122 if (existing) {
4123 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM
4124 em, start,
4125 root->sectorsize);
3b951516
CM
4126 free_extent_map(existing);
4127 if (err) {
4128 free_extent_map(em);
4129 em = NULL;
4130 }
4131 } else {
4132 err = -EIO;
4133 printk("failing to insert %Lu %Lu\n",
4134 start, len);
4135 free_extent_map(em);
4136 em = NULL;
4137 }
4138 } else {
4139 free_extent_map(em);
4140 em = existing;
e6dcd2dc 4141 err = 0;
a52d9a80 4142 }
a52d9a80 4143 }
d1310b2e 4144 spin_unlock(&em_tree->lock);
a52d9a80 4145out:
f421950f
CM
4146 if (path)
4147 btrfs_free_path(path);
a52d9a80
CM
4148 if (trans) {
4149 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 4150 if (!err) {
a52d9a80 4151 err = ret;
e6dcd2dc 4152 }
a52d9a80 4153 }
a52d9a80
CM
4154 if (err) {
4155 free_extent_map(em);
4156 WARN_ON(1);
4157 return ERR_PTR(err);
4158 }
4159 return em;
4160}
4161
16432985
CM
4162static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4163 const struct iovec *iov, loff_t offset,
4164 unsigned long nr_segs)
4165{
e1c4b745 4166 return -EINVAL;
16432985
CM
4167}
4168
d396c6f5 4169static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 4170{
d396c6f5 4171 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM
4172}
4173
a52d9a80 4174int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4175{
d1310b2e
CM
4176 struct extent_io_tree *tree;
4177 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4178 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4179}
1832a6d5 4180
a52d9a80 4181static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4182{
d1310b2e 4183 struct extent_io_tree *tree;
b888db2b
CM
4184
4185
4186 if (current->flags & PF_MEMALLOC) {
4187 redirty_page_for_writepage(wbc, page);
4188 unlock_page(page);
4189 return 0;
4190 }
d1310b2e 4191 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4192 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM
4193}
4194
f421950f
CM
4195int btrfs_writepages(struct address_space *mapping,
4196 struct writeback_control *wbc)
b293f02e 4197{
d1310b2e 4198 struct extent_io_tree *tree;
771ed689 4199
d1310b2e 4200 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM
4201 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4202}
4203
3ab2fb5a
CM
4204static int
4205btrfs_readpages(struct file *file, struct address_space *mapping,
4206 struct list_head *pages, unsigned nr_pages)
4207{
d1310b2e
CM
4208 struct extent_io_tree *tree;
4209 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM
4210 return extent_readpages(tree, mapping, pages, nr_pages,
4211 btrfs_get_extent);
4212}
e6dcd2dc 4213static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4214{
d1310b2e
CM
4215 struct extent_io_tree *tree;
4216 struct extent_map_tree *map;
a52d9a80 4217 int ret;
8c2383c3 4218
d1310b2e
CM
4219 tree = &BTRFS_I(page->mapping->host)->io_tree;
4220 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4221 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM
4222 if (ret == 1) {
4223 ClearPagePrivate(page);
4224 set_page_private(page, 0);
4225 page_cache_release(page);
39279cc3 4226 }
a52d9a80 4227 return ret;
39279cc3
CM
4228}
4229
e6dcd2dc
CM
4230static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4231{
98509cfc
CM
4232 if (PageWriteback(page) || PageDirty(page))
4233 return 0;
e6dcd2dc
CM
4234 return __btrfs_releasepage(page, gfp_flags);
4235}
4236
a52d9a80 4237static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4238{
d1310b2e 4239 struct extent_io_tree *tree;
e6dcd2dc
CM
4240 struct btrfs_ordered_extent *ordered;
4241 u64 page_start = page_offset(page);
4242 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4243
e6dcd2dc 4244 wait_on_page_writeback(page);
d1310b2e 4245 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM
4246 if (offset) {
4247 btrfs_releasepage(page, GFP_NOFS);
4248 return;
4249 }
4250
4251 lock_extent(tree, page_start, page_end, GFP_NOFS);
4252 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4253 page_offset(page));
4254 if (ordered) {
eb84ae03
CM
4255 /*
4256 * IO on this page will never be started, so we need
4257 * to account for any ordered extents now
4258 */
e6dcd2dc
CM
4259 clear_extent_bit(tree, page_start, page_end,
4260 EXTENT_DIRTY | EXTENT_DELALLOC |
4261 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM
4262 btrfs_finish_ordered_io(page->mapping->host,
4263 page_start, page_end);
e6dcd2dc
CM
4264 btrfs_put_ordered_extent(ordered);
4265 lock_extent(tree, page_start, page_end, GFP_NOFS);
4266 }
4267 clear_extent_bit(tree, page_start, page_end,
4268 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4269 EXTENT_ORDERED,
4270 1, 1, GFP_NOFS);
4271 __btrfs_releasepage(page, GFP_NOFS);
4272
4a096752 4273 ClearPageChecked(page);
9ad6b7bc 4274 if (PagePrivate(page)) {
9ad6b7bc
CM
4275 ClearPagePrivate(page);
4276 set_page_private(page, 0);
4277 page_cache_release(page);
4278 }
39279cc3
CM
4279}
4280
9ebefb18
CM
4281/*
4282 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4283 * called from a page fault handler when a page is first dirtied. Hence we must
4284 * be careful to check for EOF conditions here. We set the page up correctly
4285 * for a written page which means we get ENOSPC checking when writing into
4286 * holes and correct delalloc and unwritten extent mapping on filesystems that
4287 * support these features.
4288 *
4289 * We are not allowed to take the i_mutex here so we have to play games to
4290 * protect against truncate races as the page could now be beyond EOF. Because
4291 * vmtruncate() writes the inode size before removing pages, once we have the
4292 * page lock we can determine safely if the page is beyond EOF. If it is not
4293 * beyond EOF, then the page is guaranteed safe against truncation until we
4294 * unlock the page.
4295 */
4296int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4297{
6da6abae 4298 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4299 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM
4300 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4301 struct btrfs_ordered_extent *ordered;
4302 char *kaddr;
4303 unsigned long zero_start;
9ebefb18 4304 loff_t size;
1832a6d5 4305 int ret;
a52d9a80 4306 u64 page_start;
e6dcd2dc 4307 u64 page_end;
9ebefb18 4308
1832a6d5 4309 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM
4310 if (ret)
4311 goto out;
4312
4313 ret = -EINVAL;
e6dcd2dc 4314again:
9ebefb18 4315 lock_page(page);
9ebefb18 4316 size = i_size_read(inode);
e6dcd2dc
CM
4317 page_start = page_offset(page);
4318 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4319
9ebefb18 4320 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4321 (page_start >= size)) {
9ebefb18
CM
4322 /* page got truncated out from underneath us */
4323 goto out_unlock;
4324 }
e6dcd2dc
CM
4325 wait_on_page_writeback(page);
4326
4327 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4328 set_page_extent_mapped(page);
4329
eb84ae03
CM
4330 /*
4331 * we can't set the delalloc bits if there are pending ordered
4332 * extents. Drop our locks and wait for them to finish
4333 */
e6dcd2dc
CM
4334 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4335 if (ordered) {
4336 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4337 unlock_page(page);
eb84ae03 4338 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM
4339 btrfs_put_ordered_extent(ordered);
4340 goto again;
4341 }
4342
ea8c2819 4343 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4344 ret = 0;
9ebefb18
CM
4345
4346 /* page is wholly or partially inside EOF */
a52d9a80 4347 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4348 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4349 else
e6dcd2dc 4350 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4351
e6dcd2dc
CM
4352 if (zero_start != PAGE_CACHE_SIZE) {
4353 kaddr = kmap(page);
4354 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4355 flush_dcache_page(page);
4356 kunmap(page);
4357 }
247e743c 4358 ClearPageChecked(page);
e6dcd2dc
CM
4359 set_page_dirty(page);
4360 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM
4361
4362out_unlock:
4363 unlock_page(page);
1832a6d5 4364out:
9ebefb18
CM
4365 return ret;
4366}
4367
39279cc3
CM
4368static void btrfs_truncate(struct inode *inode)
4369{
4370 struct btrfs_root *root = BTRFS_I(inode)->root;
4371 int ret;
4372 struct btrfs_trans_handle *trans;
d3c2fdcf 4373 unsigned long nr;
dbe674a9 4374 u64 mask = root->sectorsize - 1;
39279cc3
CM
4375
4376 if (!S_ISREG(inode->i_mode))
4377 return;
4378 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4379 return;
4380
4381 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4382 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4383
39279cc3
CM
4384 trans = btrfs_start_transaction(root, 1);
4385 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4386 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4387
7b128766
JB
4388 ret = btrfs_orphan_add(trans, inode);
4389 if (ret)
4390 goto out;
39279cc3 4391 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4392 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4393 BTRFS_EXTENT_DATA_KEY);
39279cc3 4394 btrfs_update_inode(trans, root, inode);
5f39d397 4395
7b128766
JB
4396 ret = btrfs_orphan_del(trans, inode);
4397 BUG_ON(ret);
4398
4399out:
4400 nr = trans->blocks_used;
89ce8a63 4401 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4402 BUG_ON(ret);
d3c2fdcf 4403 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
4404}
4405
d352ac68
CM
4406/*
4407 * create a new subvolume directory/inode (helper for the ioctl).
4408 */
d2fb3437
YZ
4409int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4410 struct btrfs_root *new_root, struct dentry *dentry,
4411 u64 new_dirid, u64 alloc_hint)
39279cc3 4412{
39279cc3 4413 struct inode *inode;
cb8e7090 4414 int error;
00e4e6b3 4415 u64 index = 0;
39279cc3 4416
aec7477b 4417 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4418 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4419 if (IS_ERR(inode))
f46b5a66 4420 return PTR_ERR(inode);
39279cc3
CM
4421 inode->i_op = &btrfs_dir_inode_operations;
4422 inode->i_fop = &btrfs_dir_file_operations;
4423
39279cc3 4424 inode->i_nlink = 1;
dbe674a9 4425 btrfs_i_size_write(inode, 0);
3b96362c 4426
cb8e7090
CH
4427 error = btrfs_update_inode(trans, new_root, inode);
4428 if (error)
4429 return error;
4430
4431 d_instantiate(dentry, inode);
4432 return 0;
39279cc3
CM
4433}
4434
d352ac68
CM
4435/* helper function for file defrag and space balancing. This
4436 * forces readahead on a given range of bytes in an inode
4437 */
edbd8d4e 4438unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM
4439 struct file_ra_state *ra, struct file *file,
4440 pgoff_t offset, pgoff_t last_index)
4441{
8e7bf94f 4442 pgoff_t req_size = last_index - offset + 1;
86479a04 4443
86479a04
CM
4444 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4445 return offset + req_size;
86479a04
CM
4446}
4447
39279cc3
CM
4448struct inode *btrfs_alloc_inode(struct super_block *sb)
4449{
4450 struct btrfs_inode *ei;
4451
4452 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4453 if (!ei)
4454 return NULL;
15ee9bc7 4455 ei->last_trans = 0;
e02119d5 4456 ei->logged_trans = 0;
e6dcd2dc 4457 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB
4458 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4459 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4460 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM
4461 return &ei->vfs_inode;
4462}
4463
4464void btrfs_destroy_inode(struct inode *inode)
4465{
e6dcd2dc 4466 struct btrfs_ordered_extent *ordered;
39279cc3
CM
4467 WARN_ON(!list_empty(&inode->i_dentry));
4468 WARN_ON(inode->i_data.nrpages);
4469
33268eaf
JB
4470 if (BTRFS_I(inode)->i_acl &&
4471 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4472 posix_acl_release(BTRFS_I(inode)->i_acl);
4473 if (BTRFS_I(inode)->i_default_acl &&
4474 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4475 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4476
bcc63abb 4477 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB
4478 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4479 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4480 " list\n", inode->i_ino);
4481 dump_stack();
4482 }
bcc63abb 4483 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4484
e6dcd2dc
CM
4485 while(1) {
4486 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4487 if (!ordered)
4488 break;
4489 else {
4490 printk("found ordered extent %Lu %Lu\n",
4491 ordered->file_offset, ordered->len);
4492 btrfs_remove_ordered_extent(inode, ordered);
4493 btrfs_put_ordered_extent(ordered);
4494 btrfs_put_ordered_extent(ordered);
4495 }
4496 }
5b21f2ed 4497 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM
4498 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4499}
4500
0ee0fda0 4501static void init_once(void *foo)
39279cc3
CM
4502{
4503 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4504
4505 inode_init_once(&ei->vfs_inode);
4506}
4507
4508void btrfs_destroy_cachep(void)
4509{
4510 if (btrfs_inode_cachep)
4511 kmem_cache_destroy(btrfs_inode_cachep);
4512 if (btrfs_trans_handle_cachep)
4513 kmem_cache_destroy(btrfs_trans_handle_cachep);
4514 if (btrfs_transaction_cachep)
4515 kmem_cache_destroy(btrfs_transaction_cachep);
4516 if (btrfs_bit_radix_cachep)
4517 kmem_cache_destroy(btrfs_bit_radix_cachep);
4518 if (btrfs_path_cachep)
4519 kmem_cache_destroy(btrfs_path_cachep);
4520}
4521
86479a04 4522struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4523 unsigned long extra_flags,
2b1f55b0 4524 void (*ctor)(void *))
92fee66d
CM
4525{
4526 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4527 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM
4528}
4529
39279cc3
CM
4530int btrfs_init_cachep(void)
4531{
86479a04 4532 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM
4533 sizeof(struct btrfs_inode),
4534 0, init_once);
39279cc3
CM
4535 if (!btrfs_inode_cachep)
4536 goto fail;
86479a04
CM
4537 btrfs_trans_handle_cachep =
4538 btrfs_cache_create("btrfs_trans_handle_cache",
4539 sizeof(struct btrfs_trans_handle),
4540 0, NULL);
39279cc3
CM
4541 if (!btrfs_trans_handle_cachep)
4542 goto fail;
86479a04 4543 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4544 sizeof(struct btrfs_transaction),
92fee66d 4545 0, NULL);
39279cc3
CM
4546 if (!btrfs_transaction_cachep)
4547 goto fail;
86479a04 4548 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4549 sizeof(struct btrfs_path),
92fee66d 4550 0, NULL);
39279cc3
CM
4551 if (!btrfs_path_cachep)
4552 goto fail;
86479a04 4553 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4554 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM
4555 if (!btrfs_bit_radix_cachep)
4556 goto fail;
4557 return 0;
4558fail:
4559 btrfs_destroy_cachep();
4560 return -ENOMEM;
4561}
4562
4563static int btrfs_getattr(struct vfsmount *mnt,
4564 struct dentry *dentry, struct kstat *stat)
4565{
4566 struct inode *inode = dentry->d_inode;
4567 generic_fillattr(inode, stat);
3394e160 4568 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4569 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ
4570 stat->blocks = (inode_get_bytes(inode) +
4571 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM
4572 return 0;
4573}
4574
4575static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4576 struct inode * new_dir,struct dentry *new_dentry)
4577{
4578 struct btrfs_trans_handle *trans;
4579 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4580 struct inode *new_inode = new_dentry->d_inode;
4581 struct inode *old_inode = old_dentry->d_inode;
4582 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4583 u64 index = 0;
39279cc3
CM
4584 int ret;
4585
3394e160
CM
4586 /* we're not allowed to rename between subvolumes */
4587 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4588 BTRFS_I(new_dir)->root->root_key.objectid)
4589 return -EXDEV;
4590
39279cc3
CM
4591 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4592 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4593 return -ENOTEMPTY;
4594 }
5f39d397 4595
0660b5af
CM
4596 /* to rename a snapshot or subvolume, we need to juggle the
4597 * backrefs. This isn't coded yet
4598 */
4599 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4600 return -EXDEV;
4601
1832a6d5
CM
4602 ret = btrfs_check_free_space(root, 1, 0);
4603 if (ret)
4604 goto out_unlock;
4605
39279cc3 4606 trans = btrfs_start_transaction(root, 1);
5f39d397 4607
39279cc3 4608 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4609
e02119d5 4610 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM
4611 old_dir->i_ctime = old_dir->i_mtime = ctime;
4612 new_dir->i_ctime = new_dir->i_mtime = ctime;
4613 old_inode->i_ctime = ctime;
5f39d397 4614
e02119d5
CM
4615 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4616 old_dentry->d_name.name,
4617 old_dentry->d_name.len);
39279cc3
CM
4618 if (ret)
4619 goto out_fail;
4620
4621 if (new_inode) {
4622 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM
4623 ret = btrfs_unlink_inode(trans, root, new_dir,
4624 new_dentry->d_inode,
4625 new_dentry->d_name.name,
4626 new_dentry->d_name.len);
39279cc3
CM
4627 if (ret)
4628 goto out_fail;
7b128766 4629 if (new_inode->i_nlink == 0) {
e02119d5 4630 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB
4631 if (ret)
4632 goto out_fail;
4633 }
e02119d5 4634
39279cc3 4635 }
3de4586c 4636 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB
4637 if (ret)
4638 goto out_fail;
4639
e02119d5
CM
4640 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4641 old_inode, new_dentry->d_name.name,
4642 new_dentry->d_name.len, 1, index);
39279cc3
CM
4643 if (ret)
4644 goto out_fail;
4645
4646out_fail:
ab78c84d 4647 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4648out_unlock:
39279cc3
CM
4649 return ret;
4650}
4651
d352ac68
CM
4652/*
4653 * some fairly slow code that needs optimization. This walks the list
4654 * of all the inodes with pending delalloc and forces them to disk.
4655 */
ea8c2819
CM
4656int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4657{
4658 struct list_head *head = &root->fs_info->delalloc_inodes;
4659 struct btrfs_inode *binode;
5b21f2ed 4660 struct inode *inode;
ea8c2819 4661
c146afad
YZ
4662 if (root->fs_info->sb->s_flags & MS_RDONLY)
4663 return -EROFS;
4664
75eff68e 4665 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819
CM
4666 while(!list_empty(head)) {
4667 binode = list_entry(head->next, struct btrfs_inode,
4668 delalloc_inodes);
5b21f2ed
ZY
4669 inode = igrab(&binode->vfs_inode);
4670 if (!inode)
4671 list_del_init(&binode->delalloc_inodes);
75eff68e 4672 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 4673 if (inode) {
8c8bee1d 4674 filemap_flush(inode->i_mapping);
5b21f2ed
ZY
4675 iput(inode);
4676 }
4677 cond_resched();
75eff68e 4678 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 4679 }
75eff68e 4680 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM
4681
4682 /* the filemap_flush will queue IO into the worker threads, but
4683 * we have to make sure the IO is actually started and that
4684 * ordered extents get created before we return
4685 */
4686 atomic_inc(&root->fs_info->async_submit_draining);
771ed689
CM
4687 while(atomic_read(&root->fs_info->nr_async_submits) ||
4688 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4689 wait_event(root->fs_info->async_submit_wait,
771ed689
CM
4690 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4691 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM
4692 }
4693 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM
4694 return 0;
4695}
4696
39279cc3
CM
4697static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4698 const char *symname)
4699{
4700 struct btrfs_trans_handle *trans;
4701 struct btrfs_root *root = BTRFS_I(dir)->root;
4702 struct btrfs_path *path;
4703 struct btrfs_key key;
1832a6d5 4704 struct inode *inode = NULL;
39279cc3
CM
4705 int err;
4706 int drop_inode = 0;
4707 u64 objectid;
00e4e6b3 4708 u64 index = 0 ;
39279cc3
CM
4709 int name_len;
4710 int datasize;
5f39d397 4711 unsigned long ptr;
39279cc3 4712 struct btrfs_file_extent_item *ei;
5f39d397 4713 struct extent_buffer *leaf;
1832a6d5 4714 unsigned long nr = 0;
39279cc3
CM
4715
4716 name_len = strlen(symname) + 1;
4717 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4718 return -ENAMETOOLONG;
1832a6d5 4719
1832a6d5
CM
4720 err = btrfs_check_free_space(root, 1, 0);
4721 if (err)
4722 goto out_fail;
4723
39279cc3
CM
4724 trans = btrfs_start_transaction(root, 1);
4725 btrfs_set_trans_block_group(trans, dir);
4726
4727 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4728 if (err) {
4729 err = -ENOSPC;
4730 goto out_unlock;
4731 }
4732
aec7477b 4733 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM
4734 dentry->d_name.len,
4735 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM
4736 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4737 &index);
39279cc3
CM
4738 err = PTR_ERR(inode);
4739 if (IS_ERR(inode))
4740 goto out_unlock;
4741
33268eaf
JB
4742 err = btrfs_init_acl(inode, dir);
4743 if (err) {
4744 drop_inode = 1;
4745 goto out_unlock;
4746 }
4747
39279cc3 4748 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4749 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM
4750 if (err)
4751 drop_inode = 1;
4752 else {
4753 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4754 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM
4755 inode->i_fop = &btrfs_file_operations;
4756 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4757 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM
4758 }
4759 dir->i_sb->s_dirt = 1;
4760 btrfs_update_inode_block_group(trans, inode);
4761 btrfs_update_inode_block_group(trans, dir);
4762 if (drop_inode)
4763 goto out_unlock;
4764
4765 path = btrfs_alloc_path();
4766 BUG_ON(!path);
4767 key.objectid = inode->i_ino;
4768 key.offset = 0;
39279cc3
CM
4769 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4770 datasize = btrfs_file_extent_calc_inline_size(name_len);
4771 err = btrfs_insert_empty_item(trans, root, path, &key,
4772 datasize);
54aa1f4d
CM
4773 if (err) {
4774 drop_inode = 1;
4775 goto out_unlock;
4776 }
5f39d397
CM
4777 leaf = path->nodes[0];
4778 ei = btrfs_item_ptr(leaf, path->slots[0],
4779 struct btrfs_file_extent_item);
4780 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4781 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4782 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM
4783 btrfs_set_file_extent_encryption(leaf, ei, 0);
4784 btrfs_set_file_extent_compression(leaf, ei, 0);
4785 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4786 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4787
39279cc3 4788 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM
4789 write_extent_buffer(leaf, symname, ptr, name_len);
4790 btrfs_mark_buffer_dirty(leaf);
39279cc3 4791 btrfs_free_path(path);
5f39d397 4792
39279cc3
CM
4793 inode->i_op = &btrfs_symlink_inode_operations;
4794 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4795 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4796 inode_set_bytes(inode, name_len);
dbe674a9 4797 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM
4798 err = btrfs_update_inode(trans, root, inode);
4799 if (err)
4800 drop_inode = 1;
39279cc3
CM
4801
4802out_unlock:
d3c2fdcf 4803 nr = trans->blocks_used;
ab78c84d 4804 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4805out_fail:
39279cc3
CM
4806 if (drop_inode) {
4807 inode_dec_link_count(inode);
4808 iput(inode);
4809 }
d3c2fdcf 4810 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM
4811 return err;
4812}
16432985 4813
d899e052
YZ
4814static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4815 u64 alloc_hint, int mode)
4816{
4817 struct btrfs_trans_handle *trans;
4818 struct btrfs_root *root = BTRFS_I(inode)->root;
4819 struct btrfs_key ins;
4820 u64 alloc_size;
4821 u64 cur_offset = start;
4822 u64 num_bytes = end - start;
4823 int ret = 0;
4824
4825 trans = btrfs_join_transaction(root, 1);
4826 BUG_ON(!trans);
4827 btrfs_set_trans_block_group(trans, inode);
4828
4829 while (num_bytes > 0) {
4830 alloc_size = min(num_bytes, root->fs_info->max_extent);
4831 ret = btrfs_reserve_extent(trans, root, alloc_size,
4832 root->sectorsize, 0, alloc_hint,
4833 (u64)-1, &ins, 1);
4834 if (ret) {
4835 WARN_ON(1);
4836 goto out;
4837 }
4838 ret = insert_reserved_file_extent(trans, inode,
4839 cur_offset, ins.objectid,
4840 ins.offset, ins.offset,
4841 ins.offset, 0, 0, 0,
4842 BTRFS_FILE_EXTENT_PREALLOC);
4843 BUG_ON(ret);
4844 num_bytes -= ins.offset;
4845 cur_offset += ins.offset;
4846 alloc_hint = ins.objectid + ins.offset;
4847 }
4848out:
4849 if (cur_offset > start) {
4850 inode->i_ctime = CURRENT_TIME;
4851 btrfs_set_flag(inode, PREALLOC);
4852 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4853 cur_offset > i_size_read(inode))
4854 btrfs_i_size_write(inode, cur_offset);
4855 ret = btrfs_update_inode(trans, root, inode);
4856 BUG_ON(ret);
4857 }
4858
4859 btrfs_end_transaction(trans, root);
4860 return ret;
4861}
4862
4863static long btrfs_fallocate(struct inode *inode, int mode,
4864 loff_t offset, loff_t len)
4865{
4866 u64 cur_offset;
4867 u64 last_byte;
4868 u64 alloc_start;
4869 u64 alloc_end;
4870 u64 alloc_hint = 0;
4871 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4872 struct extent_map *em;
4873 int ret;
4874
4875 alloc_start = offset & ~mask;
4876 alloc_end = (offset + len + mask) & ~mask;
4877
4878 mutex_lock(&inode->i_mutex);
4879 if (alloc_start > inode->i_size) {
4880 ret = btrfs_cont_expand(inode, alloc_start);
4881 if (ret)
4882 goto out;
4883 }
4884
4885 while (1) {
4886 struct btrfs_ordered_extent *ordered;
4887 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4888 alloc_end - 1, GFP_NOFS);
4889 ordered = btrfs_lookup_first_ordered_extent(inode,
4890 alloc_end - 1);
4891 if (ordered &&
4892 ordered->file_offset + ordered->len > alloc_start &&
4893 ordered->file_offset < alloc_end) {
4894 btrfs_put_ordered_extent(ordered);
4895 unlock_extent(&BTRFS_I(inode)->io_tree,
4896 alloc_start, alloc_end - 1, GFP_NOFS);
4897 btrfs_wait_ordered_range(inode, alloc_start,
4898 alloc_end - alloc_start);
4899 } else {
4900 if (ordered)
4901 btrfs_put_ordered_extent(ordered);
4902 break;
4903 }
4904 }
4905
4906 cur_offset = alloc_start;
4907 while (1) {
4908 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4909 alloc_end - cur_offset, 0);
4910 BUG_ON(IS_ERR(em) || !em);
4911 last_byte = min(extent_map_end(em), alloc_end);
4912 last_byte = (last_byte + mask) & ~mask;
4913 if (em->block_start == EXTENT_MAP_HOLE) {
4914 ret = prealloc_file_range(inode, cur_offset,
4915 last_byte, alloc_hint, mode);
4916 if (ret < 0) {
4917 free_extent_map(em);
4918 break;
4919 }
4920 }
4921 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4922 alloc_hint = em->block_start;
4923 free_extent_map(em);
4924
4925 cur_offset = last_byte;
4926 if (cur_offset >= alloc_end) {
4927 ret = 0;
4928 break;
4929 }
4930 }
4931 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4932 GFP_NOFS);
4933out:
4934 mutex_unlock(&inode->i_mutex);
4935 return ret;
4936}
4937
e6dcd2dc
CM
4938static int btrfs_set_page_dirty(struct page *page)
4939{
e6dcd2dc
CM
4940 return __set_page_dirty_nobuffers(page);
4941}
4942
0ee0fda0 4943static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y
4944{
4945 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4946 return -EACCES;
33268eaf 4947 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4948}
39279cc3
CM
4949
4950static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4951 .getattr = btrfs_getattr,
39279cc3
CM
4952 .lookup = btrfs_lookup,
4953 .create = btrfs_create,
4954 .unlink = btrfs_unlink,
4955 .link = btrfs_link,
4956 .mkdir = btrfs_mkdir,
4957 .rmdir = btrfs_rmdir,
4958 .rename = btrfs_rename,
4959 .symlink = btrfs_symlink,
4960 .setattr = btrfs_setattr,
618e21d5 4961 .mknod = btrfs_mknod,
95819c05
CH
4962 .setxattr = btrfs_setxattr,
4963 .getxattr = btrfs_getxattr,
5103e947 4964 .listxattr = btrfs_listxattr,
95819c05 4965 .removexattr = btrfs_removexattr,
fdebe2bd 4966 .permission = btrfs_permission,
39279cc3 4967};
39279cc3
CM
4968static struct inode_operations btrfs_dir_ro_inode_operations = {
4969 .lookup = btrfs_lookup,
fdebe2bd 4970 .permission = btrfs_permission,
39279cc3 4971};
39279cc3
CM
4972static struct file_operations btrfs_dir_file_operations = {
4973 .llseek = generic_file_llseek,
4974 .read = generic_read_dir,
cbdf5a24 4975 .readdir = btrfs_real_readdir,
34287aa3 4976 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4977#ifdef CONFIG_COMPAT
34287aa3 4978 .compat_ioctl = btrfs_ioctl,
39279cc3 4979#endif
6bf13c0c 4980 .release = btrfs_release_file,
e02119d5 4981 .fsync = btrfs_sync_file,
39279cc3
CM
4982};
4983
d1310b2e 4984static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4985 .fill_delalloc = run_delalloc_range,
065631f6 4986 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4987 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4988 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4989 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4990 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4991 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM
4992 .set_bit_hook = btrfs_set_bit_hook,
4993 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM
4994};
4995
39279cc3
CM
4996static struct address_space_operations btrfs_aops = {
4997 .readpage = btrfs_readpage,
4998 .writepage = btrfs_writepage,
b293f02e 4999 .writepages = btrfs_writepages,
3ab2fb5a 5000 .readpages = btrfs_readpages,
39279cc3 5001 .sync_page = block_sync_page,
39279cc3 5002 .bmap = btrfs_bmap,
16432985 5003 .direct_IO = btrfs_direct_IO,
a52d9a80
CM
5004 .invalidatepage = btrfs_invalidatepage,
5005 .releasepage = btrfs_releasepage,
e6dcd2dc 5006 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM
5007};
5008
5009static struct address_space_operations btrfs_symlink_aops = {
5010 .readpage = btrfs_readpage,
5011 .writepage = btrfs_writepage,
2bf5a725
CM
5012 .invalidatepage = btrfs_invalidatepage,
5013 .releasepage = btrfs_releasepage,
39279cc3
CM
5014};
5015
5016static struct inode_operations btrfs_file_inode_operations = {
5017 .truncate = btrfs_truncate,
5018 .getattr = btrfs_getattr,
5019 .setattr = btrfs_setattr,
95819c05
CH
5020 .setxattr = btrfs_setxattr,
5021 .getxattr = btrfs_getxattr,
5103e947 5022 .listxattr = btrfs_listxattr,
95819c05 5023 .removexattr = btrfs_removexattr,
fdebe2bd 5024 .permission = btrfs_permission,
d899e052 5025 .fallocate = btrfs_fallocate,
39279cc3 5026};
618e21d5
JB
5027static struct inode_operations btrfs_special_inode_operations = {
5028 .getattr = btrfs_getattr,
5029 .setattr = btrfs_setattr,
fdebe2bd 5030 .permission = btrfs_permission,
95819c05
CH
5031 .setxattr = btrfs_setxattr,
5032 .getxattr = btrfs_getxattr,
33268eaf 5033 .listxattr = btrfs_listxattr,
95819c05 5034 .removexattr = btrfs_removexattr,
618e21d5 5035};
39279cc3
CM
5036static struct inode_operations btrfs_symlink_inode_operations = {
5037 .readlink = generic_readlink,
5038 .follow_link = page_follow_link_light,
5039 .put_link = page_put_link,
fdebe2bd 5040 .permission = btrfs_permission,
39279cc3 5041};