projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Add fallocate support v2
[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>
39279cc3
CM		 41#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
0b86a832 47#include "volumes.h"
e6dcd2dc 48#include "ordered-data.h"
95819c05 49#include "xattr.h"
e02119d5
CM		 50#include "compat.h"
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);
7b128766 89
d352ac68
CM		 90/*
91 * a very lame attempt at stopping writes when the FS is 85% full. There
92 * are countless ways this is incorrect, but it is better than nothing.
93 */
1832a6d5
CM		 94int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
95 int for_del)
96{
a2135011
CM		 97 u64 total;
98 u64 used;
1832a6d5 99 u64 thresh;
bcbfce8a 100 unsigned long flags;
1832a6d5
CM		 101 int ret = 0;
102
a2135011
CM		 103 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
104 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
105 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
1832a6d5 106 if (for_del)
f9ef6604 107 thresh = total * 90;
1832a6d5 108 else
f9ef6604
CM		 109 thresh = total * 85;
110
111 do_div(thresh, 100);
1832a6d5 112
1832a6d5
CM		 113 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
114 ret = -ENOSPC;
bcbfce8a 115 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1832a6d5
CM		 116 return ret;
117}
118
c8b97818
CM		 119/*
120 * this does all the hard work for inserting an inline extent into
121 * the btree. The caller should have done a btrfs_drop_extents so that
122 * no overlapping inline items exist in the btree
123 */
124static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
125 struct btrfs_root *root, struct inode *inode,
126 u64 start, size_t size, size_t compressed_size,
127 struct page **compressed_pages)
128{
129 struct btrfs_key key;
130 struct btrfs_path *path;
131 struct extent_buffer *leaf;
132 struct page *page = NULL;
133 char *kaddr;
134 unsigned long ptr;
135 struct btrfs_file_extent_item *ei;
136 int err = 0;
137 int ret;
138 size_t cur_size = size;
139 size_t datasize;
140 unsigned long offset;
141 int use_compress = 0;
142
143 if (compressed_size && compressed_pages) {
144 use_compress = 1;
145 cur_size = compressed_size;
146 }
147
148 path = btrfs_alloc_path(); if (!path)
149 return -ENOMEM;
150
151 btrfs_set_trans_block_group(trans, inode);
152
153 key.objectid = inode->i_ino;
154 key.offset = start;
155 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
156 inode_add_bytes(inode, size);
157 datasize = btrfs_file_extent_calc_inline_size(cur_size);
158
159 inode_add_bytes(inode, size);
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 datasize);
162 BUG_ON(ret);
163 if (ret) {
164 err = ret;
165 printk("got bad ret %d\n", ret);
166 goto fail;
167 }
168 leaf = path->nodes[0];
169 ei = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_file_extent_item);
171 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
172 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
173 btrfs_set_file_extent_encryption(leaf, ei, 0);
174 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
175 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
176 ptr = btrfs_file_extent_inline_start(ei);
177
178 if (use_compress) {
179 struct page *cpage;
180 int i = 0;
181 while(compressed_size > 0) {
182 cpage = compressed_pages[i];
183 cur_size = min(compressed_size,
184 PAGE_CACHE_SIZE);
185
186 kaddr = kmap(cpage);
187 write_extent_buffer(leaf, kaddr, ptr, cur_size);
188 kunmap(cpage);
189
190 i++;
191 ptr += cur_size;
192 compressed_size -= cur_size;
193 }
194 btrfs_set_file_extent_compression(leaf, ei,
195 BTRFS_COMPRESS_ZLIB);
196 } else {
197 page = find_get_page(inode->i_mapping,
198 start >> PAGE_CACHE_SHIFT);
199 btrfs_set_file_extent_compression(leaf, ei, 0);
200 kaddr = kmap_atomic(page, KM_USER0);
201 offset = start & (PAGE_CACHE_SIZE - 1);
202 write_extent_buffer(leaf, kaddr + offset, ptr, size);
203 kunmap_atomic(kaddr, KM_USER0);
204 page_cache_release(page);
205 }
206 btrfs_mark_buffer_dirty(leaf);
207 btrfs_free_path(path);
208
209 BTRFS_I(inode)->disk_i_size = inode->i_size;
210 btrfs_update_inode(trans, root, inode);
211 return 0;
212fail:
213 btrfs_free_path(path);
214 return err;
215}
216
217
218/*
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
222 */
223static int cow_file_range_inline(struct btrfs_trans_handle *trans,
224 struct btrfs_root *root,
225 struct inode *inode, u64 start, u64 end,
226 size_t compressed_size,
227 struct page **compressed_pages)
228{
229 u64 isize = i_size_read(inode);
230 u64 actual_end = min(end + 1, isize);
231 u64 inline_len = actual_end - start;
232 u64 aligned_end = (end + root->sectorsize - 1) &
233 ~((u64)root->sectorsize - 1);
234 u64 hint_byte;
235 u64 data_len = inline_len;
236 int ret;
237
238 if (compressed_size)
239 data_len = compressed_size;
240
241 if (start > 0 ||
242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243 (!compressed_size &&
244 (actual_end & (root->sectorsize - 1)) == 0) ||
245 end + 1 < isize ||
246 data_len > root->fs_info->max_inline) {
247 return 1;
248 }
249
c8b97818
CM		 250 ret = btrfs_drop_extents(trans, root, inode, start,
251 aligned_end, aligned_end, &hint_byte);
252 BUG_ON(ret);
253
254 if (isize > actual_end)
255 inline_len = min_t(u64, isize, actual_end);
256 ret = insert_inline_extent(trans, root, inode, start,
257 inline_len, compressed_size,
258 compressed_pages);
259 BUG_ON(ret);
260 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 261 return 0;
262}
263
d352ac68
CM		 264/*
265 * when extent_io.c finds a delayed allocation range in the file,
266 * the call backs end up in this code. The basic idea is to
267 * allocate extents on disk for the range, and create ordered data structs
268 * in ram to track those extents.
c8b97818
CM		 269 *
270 * locked_page is the page that writepage had locked already. We use
271 * it to make sure we don't do extra locks or unlocks.
272 *
273 * *page_started is set to one if we unlock locked_page and do everything
274 * required to start IO on it. It may be clean and already done with
275 * IO when we return.
d352ac68 276 */
c8b97818
CM		 277static int cow_file_range(struct inode *inode, struct page *locked_page,
278 u64 start, u64 end, int *page_started)
b888db2b
CM		 279{
280 struct btrfs_root *root = BTRFS_I(inode)->root;
281 struct btrfs_trans_handle *trans;
b888db2b 282 u64 alloc_hint = 0;
db94535d 283 u64 num_bytes;
c8b97818
CM		 284 unsigned long ram_size;
285 u64 orig_start;
286 u64 disk_num_bytes;
c59f8951 287 u64 cur_alloc_size;
db94535d 288 u64 blocksize = root->sectorsize;
c8b97818 289 u64 actual_end;
be20aa9d 290 struct btrfs_key ins;
e6dcd2dc
CM		 291 struct extent_map *em;
292 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
293 int ret = 0;
c8b97818
CM		 294 struct page **pages = NULL;
295 unsigned long nr_pages;
296 unsigned long nr_pages_ret = 0;
297 unsigned long total_compressed = 0;
298 unsigned long total_in = 0;
299 unsigned long max_compressed = 128 * 1024;
300 unsigned long max_uncompressed = 256 * 1024;
301 int i;
80ff3856 302 int ordered_type;
c8b97818 303 int will_compress;
b888db2b 304
f9295749 305 trans = btrfs_join_transaction(root, 1);
b888db2b 306 BUG_ON(!trans);
be20aa9d 307 btrfs_set_trans_block_group(trans, inode);
c8b97818
CM		 308 orig_start = start;
309
310 /*
311 * compression made this loop a bit ugly, but the basic idea is to
312 * compress some pages but keep the total size of the compressed
313 * extent relatively small. If compression is off, this goto target
314 * is never used.
315 */
316again:
317 will_compress = 0;
318 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
319 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 320
c8b97818
CM		 321 actual_end = min_t(u64, i_size_read(inode), end + 1);
322 total_compressed = actual_end - start;
323
324 /* we want to make sure that amount of ram required to uncompress
325 * an extent is reasonable, so we limit the total size in ram
326 * of a compressed extent to 256k
327 */
328 total_compressed = min(total_compressed, max_uncompressed);
db94535d 329 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 330 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 331 disk_num_bytes = num_bytes;
332 total_in = 0;
333 ret = 0;
db94535d 334
c8b97818
CM		 335 /* we do compression for mount -o compress and when the
336 * inode has not been flagged as nocompress
337 */
338 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
339 btrfs_test_opt(root, COMPRESS)) {
340 WARN_ON(pages);
cfbc246e 341 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818
CM		 342
343 /* we want to make sure the amount of IO required to satisfy
344 * a random read is reasonably small, so we limit the size
345 * of a compressed extent to 128k
346 */
347 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
348 total_compressed, pages,
349 nr_pages, &nr_pages_ret,
350 &total_in,
351 &total_compressed,
352 max_compressed);
353
354 if (!ret) {
355 unsigned long offset = total_compressed &
356 (PAGE_CACHE_SIZE - 1);
357 struct page *page = pages[nr_pages_ret - 1];
358 char *kaddr;
359
360 /* zero the tail end of the last page, we might be
361 * sending it down to disk
362 */
363 if (offset) {
364 kaddr = kmap_atomic(page, KM_USER0);
365 memset(kaddr + offset, 0,
366 PAGE_CACHE_SIZE - offset);
367 kunmap_atomic(kaddr, KM_USER0);
368 }
369 will_compress = 1;
370 }
371 }
372 if (start == 0) {
373 /* lets try to make an inline extent */
374 if (ret || total_in < (end - start + 1)) {
375 /* we didn't compress the entire range, try
376 * to make an uncompressed inline extent. This
377 * is almost sure to fail, but maybe inline sizes
378 * will get bigger later
379 */
380 ret = cow_file_range_inline(trans, root, inode,
381 start, end, 0, NULL);
382 } else {
383 ret = cow_file_range_inline(trans, root, inode,
384 start, end,
385 total_compressed, pages);
386 }
387 if (ret == 0) {
388 extent_clear_unlock_delalloc(inode,
389 &BTRFS_I(inode)->io_tree,
390 start, end, NULL,
391 1, 1, 1);
392 *page_started = 1;
393 ret = 0;
394 goto free_pages_out;
395 }
396 }
397
398 if (will_compress) {
399 /*
400 * we aren't doing an inline extent round the compressed size
401 * up to a block size boundary so the allocator does sane
402 * things
403 */
404 total_compressed = (total_compressed + blocksize - 1) &
405 ~(blocksize - 1);
406
407 /*
408 * one last check to make sure the compression is really a
409 * win, compare the page count read with the blocks on disk
410 */
411 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
412 ~(PAGE_CACHE_SIZE - 1);
413 if (total_compressed >= total_in) {
414 will_compress = 0;
415 } else {
416 disk_num_bytes = total_compressed;
417 num_bytes = total_in;
418 }
419 }
420 if (!will_compress && pages) {
421 /*
422 * the compression code ran but failed to make things smaller,
423 * free any pages it allocated and our page pointer array
424 */
425 for (i = 0; i < nr_pages_ret; i++) {
426 page_cache_release(pages[i]);
427 }
428 kfree(pages);
429 pages = NULL;
430 total_compressed = 0;
431 nr_pages_ret = 0;
432
433 /* flag the file so we don't compress in the future */
434 btrfs_set_flag(inode, NOCOMPRESS);
435 }
436
437 BUG_ON(disk_num_bytes >
438 btrfs_super_total_bytes(&root->fs_info->super_copy));
179e29e4 439
5b21f2ed 440 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
3b951516 441
c8b97818
CM		 442 while(disk_num_bytes > 0) {
443 unsigned long min_bytes;
444
445 /*
446 * the max size of a compressed extent is pretty small,
447 * make the code a little less complex by forcing
448 * the allocator to find a whole compressed extent at once
449 */
450 if (will_compress)
451 min_bytes = disk_num_bytes;
452 else
453 min_bytes = root->sectorsize;
454
455 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 456 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
c8b97818 457 min_bytes, 0, alloc_hint,
e6dcd2dc 458 (u64)-1, &ins, 1);
c59f8951
CM		 459 if (ret) {
460 WARN_ON(1);
c8b97818 461 goto free_pages_out_fail;
c59f8951 462 }
e6dcd2dc
CM		 463 em = alloc_extent_map(GFP_NOFS);
464 em->start = start;
c8b97818
CM		 465
466 if (will_compress) {
467 ram_size = num_bytes;
468 em->len = num_bytes;
469 } else {
470 /* ramsize == disk size */
471 ram_size = ins.offset;
472 em->len = ins.offset;
473 }
474
e6dcd2dc 475 em->block_start = ins.objectid;
c8b97818 476 em->block_len = ins.offset;
e6dcd2dc 477 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 478 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818
CM		 479
480 if (will_compress)
481 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
482
e6dcd2dc
CM		 483 while(1) {
484 spin_lock(&em_tree->lock);
485 ret = add_extent_mapping(em_tree, em);
486 spin_unlock(&em_tree->lock);
487 if (ret != -EEXIST) {
488 free_extent_map(em);
489 break;
490 }
491 btrfs_drop_extent_cache(inode, start,
c8b97818 492 start + ram_size - 1, 0);
e6dcd2dc
CM		 493 }
494
98d20f67 495 cur_alloc_size = ins.offset;
80ff3856 496 ordered_type = will_compress ? BTRFS_ORDERED_COMPRESSED : 0;
e6dcd2dc 497 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
80ff3856
YZ		 498 ram_size, cur_alloc_size,
499 ordered_type);
e6dcd2dc 500 BUG_ON(ret);
c8b97818
CM		 501
502 if (disk_num_bytes < cur_alloc_size) {
503 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
3b951516
CM		 504 cur_alloc_size);
505 break;
506 }
c8b97818
CM		 507
508 if (will_compress) {
509 /*
510 * we're doing compression, we and we need to
511 * submit the compressed extents down to the device.
512 *
513 * We lock down all the file pages, clearing their
514 * dirty bits and setting them writeback. Everyone
515 * that wants to modify the page will wait on the
516 * ordered extent above.
517 *
518 * The writeback bits on the file pages are
519 * cleared when the compressed pages are on disk
520 */
521 btrfs_end_transaction(trans, root);
522
523 if (start <= page_offset(locked_page) &&
524 page_offset(locked_page) < start + ram_size) {
525 *page_started = 1;
526 }
527
528 extent_clear_unlock_delalloc(inode,
529 &BTRFS_I(inode)->io_tree,
530 start,
531 start + ram_size - 1,
532 NULL, 1, 1, 0);
533
534 ret = btrfs_submit_compressed_write(inode, start,
535 ram_size, ins.objectid,
536 cur_alloc_size, pages,
537 nr_pages_ret);
538
539 BUG_ON(ret);
540 trans = btrfs_join_transaction(root, 1);
541 if (start + ram_size < end) {
542 start += ram_size;
543 alloc_hint = ins.objectid + ins.offset;
544 /* pages will be freed at end_bio time */
545 pages = NULL;
546 goto again;
547 } else {
548 /* we've written everything, time to go */
549 break;
550 }
551 }
552 /* we're not doing compressed IO, don't unlock the first
553 * page (which the caller expects to stay locked), don't
554 * clear any dirty bits and don't set any writeback bits
555 */
556 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
557 start, start + ram_size - 1,
558 locked_page, 0, 0, 0);
559 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 560 num_bytes -= cur_alloc_size;
561 alloc_hint = ins.objectid + ins.offset;
562 start += cur_alloc_size;
b888db2b 563 }
c8b97818
CM		 564
565 ret = 0;
b888db2b
CM		 566out:
567 btrfs_end_transaction(trans, root);
c8b97818 568
be20aa9d 569 return ret;
c8b97818
CM		 570
571free_pages_out_fail:
572 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
573 start, end, locked_page, 0, 0, 0);
574free_pages_out:
575 for (i = 0; i < nr_pages_ret; i++)
576 page_cache_release(pages[i]);
577 if (pages)
578 kfree(pages);
579
580 goto out;
be20aa9d
CM		 581}
582
d352ac68
CM		 583/*
584 * when nowcow writeback call back. This checks for snapshots or COW copies
585 * of the extents that exist in the file, and COWs the file as required.
586 *
587 * If no cow copies or snapshots exist, we write directly to the existing
588 * blocks on disk
589 */
c8b97818 590static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
d899e052 591 u64 start, u64 end, int *page_started, int force)
be20aa9d 592{
be20aa9d 593 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 594 struct btrfs_trans_handle *trans;
be20aa9d 595 struct extent_buffer *leaf;
be20aa9d 596 struct btrfs_path *path;
80ff3856 597 struct btrfs_file_extent_item *fi;
be20aa9d 598 struct btrfs_key found_key;
80ff3856
YZ		 599 u64 cow_start;
600 u64 cur_offset;
601 u64 extent_end;
602 u64 disk_bytenr;
603 u64 num_bytes;
604 int extent_type;
605 int ret;
d899e052 606 int type;
80ff3856
YZ		 607 int nocow;
608 int check_prev = 1;
be20aa9d
CM		 609
610 path = btrfs_alloc_path();
611 BUG_ON(!path);
7ea394f1
YZ		 612 trans = btrfs_join_transaction(root, 1);
613 BUG_ON(!trans);
be20aa9d 614
80ff3856
YZ		 615 cow_start = (u64)-1;
616 cur_offset = start;
617 while (1) {
618 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
619 cur_offset, 0);
620 BUG_ON(ret < 0);
621 if (ret > 0 && path->slots[0] > 0 && check_prev) {
622 leaf = path->nodes[0];
623 btrfs_item_key_to_cpu(leaf, &found_key,
624 path->slots[0] - 1);
625 if (found_key.objectid == inode->i_ino &&
626 found_key.type == BTRFS_EXTENT_DATA_KEY)
627 path->slots[0]--;
628 }
629 check_prev = 0;
630next_slot:
631 leaf = path->nodes[0];
632 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
633 ret = btrfs_next_leaf(root, path);
634 if (ret < 0)
635 BUG_ON(1);
636 if (ret > 0)
637 break;
638 leaf = path->nodes[0];
639 }
be20aa9d 640
80ff3856
YZ		 641 nocow = 0;
642 disk_bytenr = 0;
643 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
644
645 if (found_key.objectid > inode->i_ino ||
646 found_key.type > BTRFS_EXTENT_DATA_KEY ||
647 found_key.offset > end)
648 break;
649
650 if (found_key.offset > cur_offset) {
651 extent_end = found_key.offset;
652 goto out_check;
653 }
654
655 fi = btrfs_item_ptr(leaf, path->slots[0],
656 struct btrfs_file_extent_item);
657 extent_type = btrfs_file_extent_type(leaf, fi);
658
d899e052
YZ		 659 if (extent_type == BTRFS_FILE_EXTENT_REG ||
660 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 661 struct btrfs_block_group_cache *block_group;
662 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
663 extent_end = found_key.offset +
664 btrfs_file_extent_num_bytes(leaf, fi);
665 if (extent_end <= start) {
666 path->slots[0]++;
667 goto next_slot;
668 }
669 if (btrfs_file_extent_compression(leaf, fi) ||
670 btrfs_file_extent_encryption(leaf, fi) ||
671 btrfs_file_extent_other_encoding(leaf, fi))
672 goto out_check;
673 if (disk_bytenr == 0)
674 goto out_check;
d899e052
YZ		 675 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
676 goto out_check;
80ff3856
YZ		 677 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
678 goto out_check;
679 block_group = btrfs_lookup_block_group(root->fs_info,
680 disk_bytenr);
681 if (!block_group || block_group->ro)
682 goto out_check;
683 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
684 nocow = 1;
685 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
686 extent_end = found_key.offset +
687 btrfs_file_extent_inline_len(leaf, fi);
688 extent_end = ALIGN(extent_end, root->sectorsize);
689 } else {
690 BUG_ON(1);
691 }
692out_check:
693 if (extent_end <= start) {
694 path->slots[0]++;
695 goto next_slot;
696 }
697 if (!nocow) {
698 if (cow_start == (u64)-1)
699 cow_start = cur_offset;
700 cur_offset = extent_end;
701 if (cur_offset > end)
702 break;
703 path->slots[0]++;
704 goto next_slot;
7ea394f1
YZ		 705 }
706
707 btrfs_release_path(root, path);
80ff3856
YZ		 708 if (cow_start != (u64)-1) {
709 ret = cow_file_range(inode, locked_page, cow_start,
710 found_key.offset - 1, page_started);
711 BUG_ON(ret);
712 cow_start = (u64)-1;
7ea394f1 713 }
80ff3856
YZ		 714
715 disk_bytenr += cur_offset - found_key.offset;
716 num_bytes = min(end + 1, extent_end) - cur_offset;
d899e052
YZ		 717 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
718 struct extent_map *em;
719 struct extent_map_tree *em_tree;
720 em_tree = &BTRFS_I(inode)->extent_tree;
721 em = alloc_extent_map(GFP_NOFS);
722 em->start = cur_offset;
723 em->len = num_bytes;
724 em->block_len = num_bytes;
725 em->block_start = disk_bytenr;
726 em->bdev = root->fs_info->fs_devices->latest_bdev;
727 set_bit(EXTENT_FLAG_PINNED, &em->flags);
728 while (1) {
729 spin_lock(&em_tree->lock);
730 ret = add_extent_mapping(em_tree, em);
731 spin_unlock(&em_tree->lock);
732 if (ret != -EEXIST) {
733 free_extent_map(em);
734 break;
735 }
736 btrfs_drop_extent_cache(inode, em->start,
737 em->start + em->len - 1, 0);
738 }
739 type = BTRFS_ORDERED_PREALLOC;
740 } else {
741 type = BTRFS_ORDERED_NOCOW;
742 }
80ff3856
YZ		 743
744 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 745 num_bytes, num_bytes, type);
746 BUG_ON(ret);
747 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
748 cur_offset, cur_offset + num_bytes - 1,
749 locked_page, 0, 0, 0);
80ff3856
YZ		 750 cur_offset = extent_end;
751 if (cur_offset > end)
752 break;
be20aa9d 753 }
80ff3856
YZ		 754 btrfs_release_path(root, path);
755
756 if (cur_offset <= end && cow_start == (u64)-1)
757 cow_start = cur_offset;
758 if (cow_start != (u64)-1) {
759 ret = cow_file_range(inode, locked_page, cow_start, end,
760 page_started);
761 BUG_ON(ret);
762 }
763
764 ret = btrfs_end_transaction(trans, root);
765 BUG_ON(ret);
7ea394f1 766 btrfs_free_path(path);
80ff3856 767 return 0;
be20aa9d
CM		 768}
769
d352ac68
CM		 770/*
771 * extent_io.c call back to do delayed allocation processing
772 */
c8b97818
CM		 773static int run_delalloc_range(struct inode *inode, struct page *locked_page,
774 u64 start, u64 end, int *page_started)
be20aa9d
CM		 775{
776 struct btrfs_root *root = BTRFS_I(inode)->root;
777 int ret;
a2135011 778
b98b6767
Y		 779 if (btrfs_test_opt(root, NODATACOW) ||
780 btrfs_test_flag(inode, NODATACOW))
c8b97818 781 ret = run_delalloc_nocow(inode, locked_page, start, end,
d899e052
YZ		 782 page_started, 0);
783 else if (btrfs_test_flag(inode, PREALLOC))
784 ret = run_delalloc_nocow(inode, locked_page, start, end,
785 page_started, 1);
be20aa9d 786 else
c8b97818
CM		 787 ret = cow_file_range(inode, locked_page, start, end,
788 page_started);
1832a6d5 789
b888db2b
CM		 790 return ret;
791}
792
d352ac68
CM		 793/*
794 * extent_io.c set_bit_hook, used to track delayed allocation
795 * bytes in this file, and to maintain the list of inodes that
796 * have pending delalloc work to be done.
797 */
291d673e 798int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 799 unsigned long old, unsigned long bits)
291d673e 800{
bcbfce8a 801 unsigned long flags;
b0c68f8b 802 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 803 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 804 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
9069218d 805 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 806 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 807 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
808 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
809 &root->fs_info->delalloc_inodes);
810 }
bcbfce8a 811 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 812 }
813 return 0;
814}
815
d352ac68
CM		 816/*
817 * extent_io.c clear_bit_hook, see set_bit_hook for why
818 */
291d673e 819int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 820 unsigned long old, unsigned long bits)
291d673e 821{
b0c68f8b 822 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 823 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a
CM		 824 unsigned long flags;
825
826 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
b0c68f8b
CM		 827 if (end - start + 1 > root->fs_info->delalloc_bytes) {
828 printk("warning: delalloc account %Lu %Lu\n",
829 end - start + 1, root->fs_info->delalloc_bytes);
830 root->fs_info->delalloc_bytes = 0;
9069218d 831 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 832 } else {
833 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 834 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 835 }
ea8c2819
CM		 836 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
837 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
838 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
839 }
bcbfce8a 840 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 841 }
842 return 0;
843}
844
d352ac68
CM		 845/*
846 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
847 * we don't create bios that span stripes or chunks
848 */
239b14b3 849int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 850 size_t size, struct bio *bio,
851 unsigned long bio_flags)
239b14b3
CM		 852{
853 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
854 struct btrfs_mapping_tree *map_tree;
a62b9401 855 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 856 u64 length = 0;
857 u64 map_length;
239b14b3
CM		 858 int ret;
859
f2d8d74d 860 length = bio->bi_size;
239b14b3
CM		 861 map_tree = &root->fs_info->mapping_tree;
862 map_length = length;
cea9e445 863 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 864 &map_length, NULL, 0);
cea9e445 865
239b14b3 866 if (map_length < length + size) {
239b14b3
CM		 867 return 1;
868 }
869 return 0;
870}
871
d352ac68
CM		 872/*
873 * in order to insert checksums into the metadata in large chunks,
874 * we wait until bio submission time. All the pages in the bio are
875 * checksummed and sums are attached onto the ordered extent record.
876 *
877 * At IO completion time the cums attached on the ordered extent record
878 * are inserted into the btree
879 */
44b8bd7e 880int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 881 int mirror_num, unsigned long bio_flags)
065631f6 882{
065631f6 883 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 884 int ret = 0;
e015640f 885
3edf7d33 886 ret = btrfs_csum_one_bio(root, inode, bio);
44b8bd7e 887 BUG_ON(ret);
e015640f 888
8b712842 889 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 890}
891
d352ac68
CM		 892/*
893 * extent_io.c submission hook. This does the right thing for csum calculation on write,
894 * or reading the csums from the tree before a read
895 */
44b8bd7e 896int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 897 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 898{
899 struct btrfs_root *root = BTRFS_I(inode)->root;
900 int ret = 0;
19b9bdb0 901 int skip_sum;
44b8bd7e 902
e6dcd2dc
CM		 903 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
904 BUG_ON(ret);
065631f6 905
19b9bdb0
CM		 906 skip_sum = btrfs_test_opt(root, NODATASUM) ||
907 btrfs_test_flag(inode, NODATASUM);
7ea394f1 908
4d1b5fb4 909 if (!(rw & (1 << BIO_RW))) {
19b9bdb0
CM		 910 if (!skip_sum)
911 btrfs_lookup_bio_sums(root, inode, bio);
c8b97818 912
19b9bdb0 913 if (bio_flags & EXTENT_BIO_COMPRESSED)
c8b97818
CM		 914 return btrfs_submit_compressed_read(inode, bio,
915 mirror_num, bio_flags);
4d1b5fb4 916 goto mapit;
19b9bdb0
CM		 917 } else if (!skip_sum) {
918 /* we're doing a write, do the async checksumming */
919 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 920 inode, rw, bio, mirror_num,
c8b97818 921 bio_flags, __btrfs_submit_bio_hook);
19b9bdb0
CM		 922 }
923
0b86a832 924mapit:
8b712842 925 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 926}
6885f308 927
d352ac68
CM		 928/*
929 * given a list of ordered sums record them in the inode. This happens
930 * at IO completion time based on sums calculated at bio submission time.
931 */
ba1da2f4 932static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 933 struct inode *inode, u64 file_offset,
934 struct list_head *list)
935{
936 struct list_head *cur;
937 struct btrfs_ordered_sum *sum;
938
939 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 940 list_for_each(cur, list) {
e6dcd2dc 941 sum = list_entry(cur, struct btrfs_ordered_sum, list);
e6dcd2dc
CM		 942 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
943 inode, sum);
e6dcd2dc
CM		 944 }
945 return 0;
946}
947
ea8c2819
CM		 948int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
949{
950 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
951 GFP_NOFS);
952}
953
d352ac68 954/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 955struct btrfs_writepage_fixup {
956 struct page *page;
957 struct btrfs_work work;
958};
959
247e743c
CM		 960void btrfs_writepage_fixup_worker(struct btrfs_work *work)
961{
962 struct btrfs_writepage_fixup *fixup;
963 struct btrfs_ordered_extent *ordered;
964 struct page *page;
965 struct inode *inode;
966 u64 page_start;
967 u64 page_end;
968
969 fixup = container_of(work, struct btrfs_writepage_fixup, work);
970 page = fixup->page;
4a096752 971again:
247e743c
CM		 972 lock_page(page);
973 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
974 ClearPageChecked(page);
975 goto out_page;
976 }
977
978 inode = page->mapping->host;
979 page_start = page_offset(page);
980 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
981
982 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 983
984 /* already ordered? We're done */
985 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
986 EXTENT_ORDERED, 0)) {
247e743c 987 goto out;
4a096752
CM		 988 }
989
990 ordered = btrfs_lookup_ordered_extent(inode, page_start);
991 if (ordered) {
992 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
993 page_end, GFP_NOFS);
994 unlock_page(page);
995 btrfs_start_ordered_extent(inode, ordered, 1);
996 goto again;
997 }
247e743c 998
ea8c2819 999 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1000 ClearPageChecked(page);
1001out:
1002 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1003out_page:
1004 unlock_page(page);
1005 page_cache_release(page);
1006}
1007
1008/*
1009 * There are a few paths in the higher layers of the kernel that directly
1010 * set the page dirty bit without asking the filesystem if it is a
1011 * good idea. This causes problems because we want to make sure COW
1012 * properly happens and the data=ordered rules are followed.
1013 *
c8b97818 1014 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1015 * hasn't been properly setup for IO. We kick off an async process
1016 * to fix it up. The async helper will wait for ordered extents, set
1017 * the delalloc bit and make it safe to write the page.
1018 */
1019int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1020{
1021 struct inode *inode = page->mapping->host;
1022 struct btrfs_writepage_fixup *fixup;
1023 struct btrfs_root *root = BTRFS_I(inode)->root;
1024 int ret;
1025
1026 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1027 EXTENT_ORDERED, 0);
1028 if (ret)
1029 return 0;
1030
1031 if (PageChecked(page))
1032 return -EAGAIN;
1033
1034 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1035 if (!fixup)
1036 return -EAGAIN;
f421950f 1037
247e743c
CM		 1038 SetPageChecked(page);
1039 page_cache_get(page);
1040 fixup->work.func = btrfs_writepage_fixup_worker;
1041 fixup->page = page;
1042 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1043 return -EAGAIN;
1044}
1045
d899e052
YZ		 1046static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1047 struct inode *inode, u64 file_pos,
1048 u64 disk_bytenr, u64 disk_num_bytes,
1049 u64 num_bytes, u64 ram_bytes,
1050 u8 compression, u8 encryption,
1051 u16 other_encoding, int extent_type)
1052{
1053 struct btrfs_root *root = BTRFS_I(inode)->root;
1054 struct btrfs_file_extent_item *fi;
1055 struct btrfs_path *path;
1056 struct extent_buffer *leaf;
1057 struct btrfs_key ins;
1058 u64 hint;
1059 int ret;
1060
1061 path = btrfs_alloc_path();
1062 BUG_ON(!path);
1063
1064 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1065 file_pos + num_bytes, file_pos, &hint);
1066 BUG_ON(ret);
1067
1068 ins.objectid = inode->i_ino;
1069 ins.offset = file_pos;
1070 ins.type = BTRFS_EXTENT_DATA_KEY;
1071 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1072 BUG_ON(ret);
1073 leaf = path->nodes[0];
1074 fi = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_file_extent_item);
1076 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1077 btrfs_set_file_extent_type(leaf, fi, extent_type);
1078 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1079 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1080 btrfs_set_file_extent_offset(leaf, fi, 0);
1081 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1082 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1083 btrfs_set_file_extent_compression(leaf, fi, compression);
1084 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1085 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1086 btrfs_mark_buffer_dirty(leaf);
1087
1088 inode_add_bytes(inode, num_bytes);
1089 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1090
1091 ins.objectid = disk_bytenr;
1092 ins.offset = disk_num_bytes;
1093 ins.type = BTRFS_EXTENT_ITEM_KEY;
1094 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1095 root->root_key.objectid,
1096 trans->transid, inode->i_ino, &ins);
1097 BUG_ON(ret);
1098
1099 btrfs_free_path(path);
1100 return 0;
1101}
1102
d352ac68
CM		 1103/* as ordered data IO finishes, this gets called so we can finish
1104 * an ordered extent if the range of bytes in the file it covers are
1105 * fully written.
1106 */
211f90e6 1107static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1108{
e6dcd2dc
CM		 1109 struct btrfs_root *root = BTRFS_I(inode)->root;
1110 struct btrfs_trans_handle *trans;
1111 struct btrfs_ordered_extent *ordered_extent;
1112 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1113 int compressed = 0;
e6dcd2dc
CM		 1114 int ret;
1115
1116 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1117 if (!ret)
e6dcd2dc 1118 return 0;
e6dcd2dc 1119
f9295749 1120 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1121
1122 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1123 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1124 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1125 goto nocow;
e6dcd2dc
CM		 1126
1127 lock_extent(io_tree, ordered_extent->file_offset,
1128 ordered_extent->file_offset + ordered_extent->len - 1,
1129 GFP_NOFS);
1130
c8b97818 1131 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1132 compressed = 1;
1133 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1134 BUG_ON(compressed);
1135 ret = btrfs_mark_extent_written(trans, root, inode,
1136 ordered_extent->file_offset,
1137 ordered_extent->file_offset +
1138 ordered_extent->len);
1139 BUG_ON(ret);
1140 } else {
1141 ret = insert_reserved_file_extent(trans, inode,
1142 ordered_extent->file_offset,
1143 ordered_extent->start,
1144 ordered_extent->disk_len,
1145 ordered_extent->len,
1146 ordered_extent->len,
1147 compressed, 0, 0,
1148 BTRFS_FILE_EXTENT_REG);
1149 BUG_ON(ret);
1150 }
e6dcd2dc
CM		 1151 unlock_extent(io_tree, ordered_extent->file_offset,
1152 ordered_extent->file_offset + ordered_extent->len - 1,
1153 GFP_NOFS);
7ea394f1 1154nocow:
e6dcd2dc
CM		 1155 add_pending_csums(trans, inode, ordered_extent->file_offset,
1156 &ordered_extent->list);
1157
34353029 1158 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1159 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1160 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1161 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1162 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1163
e6dcd2dc
CM		 1164 /* once for us */
1165 btrfs_put_ordered_extent(ordered_extent);
1166 /* once for the tree */
1167 btrfs_put_ordered_extent(ordered_extent);
1168
e6dcd2dc
CM		 1169 btrfs_end_transaction(trans, root);
1170 return 0;
1171}
1172
211f90e6
CM		 1173int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1174 struct extent_state *state, int uptodate)
1175{
1176 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1177}
1178
d352ac68
CM		 1179/*
1180 * When IO fails, either with EIO or csum verification fails, we
1181 * try other mirrors that might have a good copy of the data. This
1182 * io_failure_record is used to record state as we go through all the
1183 * mirrors. If another mirror has good data, the page is set up to date
1184 * and things continue. If a good mirror can't be found, the original
1185 * bio end_io callback is called to indicate things have failed.
1186 */
7e38326f
CM		 1187struct io_failure_record {
1188 struct page *page;
1189 u64 start;
1190 u64 len;
1191 u64 logical;
1192 int last_mirror;
1193};
1194
1259ab75
CM		 1195int btrfs_io_failed_hook(struct bio *failed_bio,
1196 struct page *page, u64 start, u64 end,
1197 struct extent_state *state)
7e38326f
CM		 1198{
1199 struct io_failure_record *failrec = NULL;
1200 u64 private;
1201 struct extent_map *em;
1202 struct inode *inode = page->mapping->host;
1203 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1205 struct bio *bio;
1206 int num_copies;
1207 int ret;
1259ab75 1208 int rw;
7e38326f 1209 u64 logical;
c8b97818 1210 unsigned long bio_flags = 0;
7e38326f
CM		 1211
1212 ret = get_state_private(failure_tree, start, &private);
1213 if (ret) {
7e38326f
CM		 1214 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1215 if (!failrec)
1216 return -ENOMEM;
1217 failrec->start = start;
1218 failrec->len = end - start + 1;
1219 failrec->last_mirror = 0;
1220
3b951516
CM		 1221 spin_lock(&em_tree->lock);
1222 em = lookup_extent_mapping(em_tree, start, failrec->len);
1223 if (em->start > start || em->start + em->len < start) {
1224 free_extent_map(em);
1225 em = NULL;
1226 }
1227 spin_unlock(&em_tree->lock);
7e38326f
CM		 1228
1229 if (!em || IS_ERR(em)) {
1230 kfree(failrec);
1231 return -EIO;
1232 }
1233 logical = start - em->start;
1234 logical = em->block_start + logical;
c8b97818
CM		 1235 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1236 bio_flags = EXTENT_BIO_COMPRESSED;
7e38326f
CM		 1237 failrec->logical = logical;
1238 free_extent_map(em);
1239 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1240 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1241 set_state_private(failure_tree, start,
1242 (u64)(unsigned long)failrec);
7e38326f 1243 } else {
587f7704 1244 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1245 }
1246 num_copies = btrfs_num_copies(
1247 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1248 failrec->logical, failrec->len);
1249 failrec->last_mirror++;
1250 if (!state) {
1251 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1252 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1253 failrec->start,
1254 EXTENT_LOCKED);
1255 if (state && state->start != failrec->start)
1256 state = NULL;
1257 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1258 }
1259 if (!state || failrec->last_mirror > num_copies) {
1260 set_state_private(failure_tree, failrec->start, 0);
1261 clear_extent_bits(failure_tree, failrec->start,
1262 failrec->start + failrec->len - 1,
1263 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1264 kfree(failrec);
1265 return -EIO;
1266 }
1267 bio = bio_alloc(GFP_NOFS, 1);
1268 bio->bi_private = state;
1269 bio->bi_end_io = failed_bio->bi_end_io;
1270 bio->bi_sector = failrec->logical >> 9;
1271 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1272 bio->bi_size = 0;
7e38326f 1273 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1274 if (failed_bio->bi_rw & (1 << BIO_RW))
1275 rw = WRITE;
1276 else
1277 rw = READ;
1278
1279 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818
CM		 1280 failrec->last_mirror,
1281 bio_flags);
1259ab75
CM		 1282 return 0;
1283}
1284
d352ac68
CM		 1285/*
1286 * each time an IO finishes, we do a fast check in the IO failure tree
1287 * to see if we need to process or clean up an io_failure_record
1288 */
1259ab75
CM		 1289int btrfs_clean_io_failures(struct inode *inode, u64 start)
1290{
1291 u64 private;
1292 u64 private_failure;
1293 struct io_failure_record *failure;
1294 int ret;
1295
1296 private = 0;
1297 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1298 (u64)-1, 1, EXTENT_DIRTY)) {
1299 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1300 start, &private_failure);
1301 if (ret == 0) {
1302 failure = (struct io_failure_record *)(unsigned long)
1303 private_failure;
1304 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1305 failure->start, 0);
1306 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1307 failure->start,
1308 failure->start + failure->len - 1,
1309 EXTENT_DIRTY | EXTENT_LOCKED,
1310 GFP_NOFS);
1311 kfree(failure);
1312 }
1313 }
7e38326f
CM		 1314 return 0;
1315}
1316
d352ac68
CM		 1317/*
1318 * when reads are done, we need to check csums to verify the data is correct
1319 * if there's a match, we allow the bio to finish. If not, we go through
1320 * the io_failure_record routines to find good copies
1321 */
70dec807
CM		 1322int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1323 struct extent_state *state)
07157aac 1324{
35ebb934 1325 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1326 struct inode *inode = page->mapping->host;
d1310b2e 1327 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1328 char *kaddr;
aadfeb6e 1329 u64 private = ~(u32)0;
07157aac 1330 int ret;
ff79f819
CM		 1331 struct btrfs_root *root = BTRFS_I(inode)->root;
1332 u32 csum = ~(u32)0;
bbf0d006 1333 unsigned long flags;
d1310b2e 1334
b98b6767
Y		 1335 if (btrfs_test_opt(root, NODATASUM) ||
1336 btrfs_test_flag(inode, NODATASUM))
b6cda9bc 1337 return 0;
c2e639f0 1338 if (state && state->start == start) {
70dec807
CM		 1339 private = state->private;
1340 ret = 0;
1341 } else {
1342 ret = get_state_private(io_tree, start, &private);
1343 }
bbf0d006 1344 local_irq_save(flags);
07157aac
CM		 1345 kaddr = kmap_atomic(page, KM_IRQ0);
1346 if (ret) {
1347 goto zeroit;
1348 }
ff79f819
CM		 1349 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1350 btrfs_csum_final(csum, (char *)&csum);
1351 if (csum != private) {
07157aac
CM		 1352 goto zeroit;
1353 }
1354 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1355 local_irq_restore(flags);
7e38326f
CM		 1356
1357 /* if the io failure tree for this inode is non-empty,
1358 * check to see if we've recovered from a failed IO
1359 */
1259ab75 1360 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1361 return 0;
1362
1363zeroit:
aadfeb6e
CM		 1364 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1365 page->mapping->host->i_ino, (unsigned long long)start, csum,
1366 private);
db94535d
CM		 1367 memset(kaddr + offset, 1, end - start + 1);
1368 flush_dcache_page(page);
07157aac 1369 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1370 local_irq_restore(flags);
3b951516
CM		 1371 if (private == 0)
1372 return 0;
7e38326f 1373 return -EIO;
07157aac 1374}
b888db2b 1375
7b128766
JB		 1376/*
1377 * This creates an orphan entry for the given inode in case something goes
1378 * wrong in the middle of an unlink/truncate.
1379 */
1380int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1381{
1382 struct btrfs_root *root = BTRFS_I(inode)->root;
1383 int ret = 0;
1384
bcc63abb 1385 spin_lock(&root->list_lock);
7b128766
JB		 1386
1387 /* already on the orphan list, we're good */
1388 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1389 spin_unlock(&root->list_lock);
7b128766
JB		 1390 return 0;
1391 }
1392
1393 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1394
bcc63abb 1395 spin_unlock(&root->list_lock);
7b128766
JB		 1396
1397 /*
1398 * insert an orphan item to track this unlinked/truncated file
1399 */
1400 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1401
1402 return ret;
1403}
1404
1405/*
1406 * We have done the truncate/delete so we can go ahead and remove the orphan
1407 * item for this particular inode.
1408 */
1409int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1410{
1411 struct btrfs_root *root = BTRFS_I(inode)->root;
1412 int ret = 0;
1413
bcc63abb 1414 spin_lock(&root->list_lock);
7b128766
JB		 1415
1416 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1417 spin_unlock(&root->list_lock);
7b128766
JB		 1418 return 0;
1419 }
1420
1421 list_del_init(&BTRFS_I(inode)->i_orphan);
1422 if (!trans) {
bcc63abb 1423 spin_unlock(&root->list_lock);
7b128766
JB		 1424 return 0;
1425 }
1426
bcc63abb 1427 spin_unlock(&root->list_lock);
7b128766
JB		 1428
1429 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1430
1431 return ret;
1432}
1433
1434/*
1435 * this cleans up any orphans that may be left on the list from the last use
1436 * of this root.
1437 */
1438void btrfs_orphan_cleanup(struct btrfs_root *root)
1439{
1440 struct btrfs_path *path;
1441 struct extent_buffer *leaf;
1442 struct btrfs_item *item;
1443 struct btrfs_key key, found_key;
1444 struct btrfs_trans_handle *trans;
1445 struct inode *inode;
1446 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1447
1448 /* don't do orphan cleanup if the fs is readonly. */
5b21f2ed 1449 if (root->fs_info->sb->s_flags & MS_RDONLY)
7b128766
JB		 1450 return;
1451
1452 path = btrfs_alloc_path();
1453 if (!path)
1454 return;
1455 path->reada = -1;
1456
1457 key.objectid = BTRFS_ORPHAN_OBJECTID;
1458 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1459 key.offset = (u64)-1;
1460
7b128766
JB		 1461
1462 while (1) {
1463 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1464 if (ret < 0) {
1465 printk(KERN_ERR "Error searching slot for orphan: %d"
1466 "\n", ret);
1467 break;
1468 }
1469
1470 /*
1471 * if ret == 0 means we found what we were searching for, which
1472 * is weird, but possible, so only screw with path if we didnt
1473 * find the key and see if we have stuff that matches
1474 */
1475 if (ret > 0) {
1476 if (path->slots[0] == 0)
1477 break;
1478 path->slots[0]--;
1479 }
1480
1481 /* pull out the item */
1482 leaf = path->nodes[0];
1483 item = btrfs_item_nr(leaf, path->slots[0]);
1484 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1485
1486 /* make sure the item matches what we want */
1487 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1488 break;
1489 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1490 break;
1491
1492 /* release the path since we're done with it */
1493 btrfs_release_path(root, path);
1494
1495 /*
1496 * this is where we are basically btrfs_lookup, without the
1497 * crossing root thing. we store the inode number in the
1498 * offset of the orphan item.
1499 */
5b21f2ed 1500 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1501 found_key.offset, root);
1502 if (!inode)
1503 break;
1504
1505 if (inode->i_state & I_NEW) {
1506 BTRFS_I(inode)->root = root;
1507
1508 /* have to set the location manually */
1509 BTRFS_I(inode)->location.objectid = inode->i_ino;
1510 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1511 BTRFS_I(inode)->location.offset = 0;
1512
1513 btrfs_read_locked_inode(inode);
1514 unlock_new_inode(inode);
1515 }
1516
1517 /*
1518 * add this inode to the orphan list so btrfs_orphan_del does
1519 * the proper thing when we hit it
1520 */
bcc63abb 1521 spin_lock(&root->list_lock);
7b128766 1522 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1523 spin_unlock(&root->list_lock);
7b128766
JB		 1524
1525 /*
1526 * if this is a bad inode, means we actually succeeded in
1527 * removing the inode, but not the orphan record, which means
1528 * we need to manually delete the orphan since iput will just
1529 * do a destroy_inode
1530 */
1531 if (is_bad_inode(inode)) {
5b21f2ed 1532 trans = btrfs_start_transaction(root, 1);
7b128766 1533 btrfs_orphan_del(trans, inode);
5b21f2ed 1534 btrfs_end_transaction(trans, root);
7b128766
JB		 1535 iput(inode);
1536 continue;
1537 }
1538
1539 /* if we have links, this was a truncate, lets do that */
1540 if (inode->i_nlink) {
1541 nr_truncate++;
1542 btrfs_truncate(inode);
1543 } else {
1544 nr_unlink++;
1545 }
1546
1547 /* this will do delete_inode and everything for us */
1548 iput(inode);
1549 }
1550
1551 if (nr_unlink)
1552 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1553 if (nr_truncate)
1554 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1555
1556 btrfs_free_path(path);
7b128766
JB		 1557}
1558
d352ac68
CM		 1559/*
1560 * read an inode from the btree into the in-memory inode
1561 */
39279cc3
CM		 1562void btrfs_read_locked_inode(struct inode *inode)
1563{
1564 struct btrfs_path *path;
5f39d397 1565 struct extent_buffer *leaf;
39279cc3 1566 struct btrfs_inode_item *inode_item;
0b86a832 1567 struct btrfs_timespec *tspec;
39279cc3
CM		 1568 struct btrfs_root *root = BTRFS_I(inode)->root;
1569 struct btrfs_key location;
1570 u64 alloc_group_block;
618e21d5 1571 u32 rdev;
39279cc3
CM		 1572 int ret;
1573
1574 path = btrfs_alloc_path();
1575 BUG_ON(!path);
39279cc3 1576 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1577
39279cc3 1578 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1579 if (ret)
39279cc3 1580 goto make_bad;
39279cc3 1581
5f39d397
CM		 1582 leaf = path->nodes[0];
1583 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1584 struct btrfs_inode_item);
1585
1586 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1587 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1588 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1589 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1590 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1591
1592 tspec = btrfs_inode_atime(inode_item);
1593 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1594 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1595
1596 tspec = btrfs_inode_mtime(inode_item);
1597 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1598 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1599
1600 tspec = btrfs_inode_ctime(inode_item);
1601 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1602 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1603
a76a3cd4 1604 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5
CM		 1605 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1606 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 1607 inode->i_rdev = 0;
5f39d397
CM		 1608 rdev = btrfs_inode_rdev(leaf, inode_item);
1609
aec7477b
JB		 1610 BTRFS_I(inode)->index_cnt = (u64)-1;
1611
5f39d397 1612 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
39279cc3
CM		 1613 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1614 alloc_group_block);
b98b6767 1615 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
e52ec0eb
CM		 1616 if (!BTRFS_I(inode)->block_group) {
1617 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
0b86a832
CM		 1618 NULL, 0,
1619 BTRFS_BLOCK_GROUP_METADATA, 0);
e52ec0eb 1620 }
39279cc3
CM		 1621 btrfs_free_path(path);
1622 inode_item = NULL;
1623
39279cc3 1624 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 1625 case S_IFREG:
1626 inode->i_mapping->a_ops = &btrfs_aops;
04160088 1627 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 1628 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 1629 inode->i_fop = &btrfs_file_operations;
1630 inode->i_op = &btrfs_file_inode_operations;
1631 break;
1632 case S_IFDIR:
1633 inode->i_fop = &btrfs_dir_file_operations;
1634 if (root == root->fs_info->tree_root)
1635 inode->i_op = &btrfs_dir_ro_inode_operations;
1636 else
1637 inode->i_op = &btrfs_dir_inode_operations;
1638 break;
1639 case S_IFLNK:
1640 inode->i_op = &btrfs_symlink_inode_operations;
1641 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 1642 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 1643 break;
618e21d5
JB		 1644 default:
1645 init_special_inode(inode, inode->i_mode, rdev);
1646 break;
39279cc3
CM		 1647 }
1648 return;
1649
1650make_bad:
39279cc3 1651 btrfs_free_path(path);
39279cc3
CM		 1652 make_bad_inode(inode);
1653}
1654
d352ac68
CM		 1655/*
1656 * given a leaf and an inode, copy the inode fields into the leaf
1657 */
e02119d5
CM		 1658static void fill_inode_item(struct btrfs_trans_handle *trans,
1659 struct extent_buffer *leaf,
5f39d397 1660 struct btrfs_inode_item *item,
39279cc3
CM		 1661 struct inode *inode)
1662{
5f39d397
CM		 1663 btrfs_set_inode_uid(leaf, item, inode->i_uid);
1664 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 1665 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 1666 btrfs_set_inode_mode(leaf, item, inode->i_mode);
1667 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1668
1669 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1670 inode->i_atime.tv_sec);
1671 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1672 inode->i_atime.tv_nsec);
1673
1674 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1675 inode->i_mtime.tv_sec);
1676 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1677 inode->i_mtime.tv_nsec);
1678
1679 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1680 inode->i_ctime.tv_sec);
1681 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1682 inode->i_ctime.tv_nsec);
1683
a76a3cd4 1684 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5
CM		 1685 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1686 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 1687 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 1688 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
5f39d397 1689 btrfs_set_inode_block_group(leaf, item,
39279cc3
CM		 1690 BTRFS_I(inode)->block_group->key.objectid);
1691}
1692
d352ac68
CM		 1693/*
1694 * copy everything in the in-memory inode into the btree.
1695 */
ba1da2f4 1696int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM		 1697 struct btrfs_root *root,
1698 struct inode *inode)
1699{
1700 struct btrfs_inode_item *inode_item;
1701 struct btrfs_path *path;
5f39d397 1702 struct extent_buffer *leaf;
39279cc3
CM		 1703 int ret;
1704
1705 path = btrfs_alloc_path();
1706 BUG_ON(!path);
39279cc3
CM		 1707 ret = btrfs_lookup_inode(trans, root, path,
1708 &BTRFS_I(inode)->location, 1);
1709 if (ret) {
1710 if (ret > 0)
1711 ret = -ENOENT;
1712 goto failed;
1713 }
1714
5f39d397
CM		 1715 leaf = path->nodes[0];
1716 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 1717 struct btrfs_inode_item);
1718
e02119d5 1719 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 1720 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 1721 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 1722 ret = 0;
1723failed:
39279cc3
CM		 1724 btrfs_free_path(path);
1725 return ret;
1726}
1727
1728
d352ac68
CM		 1729/*
1730 * unlink helper that gets used here in inode.c and in the tree logging
1731 * recovery code. It remove a link in a directory with a given name, and
1732 * also drops the back refs in the inode to the directory
1733 */
e02119d5
CM		 1734int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1735 struct btrfs_root *root,
1736 struct inode *dir, struct inode *inode,
1737 const char *name, int name_len)
39279cc3
CM		 1738{
1739 struct btrfs_path *path;
39279cc3 1740 int ret = 0;
5f39d397 1741 struct extent_buffer *leaf;
39279cc3 1742 struct btrfs_dir_item *di;
5f39d397 1743 struct btrfs_key key;
aec7477b 1744 u64 index;
39279cc3
CM		 1745
1746 path = btrfs_alloc_path();
54aa1f4d
CM		 1747 if (!path) {
1748 ret = -ENOMEM;
1749 goto err;
1750 }
1751
39279cc3
CM		 1752 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1753 name, name_len, -1);
1754 if (IS_ERR(di)) {
1755 ret = PTR_ERR(di);
1756 goto err;
1757 }
1758 if (!di) {
1759 ret = -ENOENT;
1760 goto err;
1761 }
5f39d397
CM		 1762 leaf = path->nodes[0];
1763 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 1764 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 1765 if (ret)
1766 goto err;
39279cc3
CM		 1767 btrfs_release_path(root, path);
1768
aec7477b 1769 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 1770 inode->i_ino,
1771 dir->i_ino, &index);
aec7477b
JB		 1772 if (ret) {
1773 printk("failed to delete reference to %.*s, "
1774 "inode %lu parent %lu\n", name_len, name,
e02119d5 1775 inode->i_ino, dir->i_ino);
aec7477b
JB		 1776 goto err;
1777 }
1778
39279cc3 1779 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 1780 index, name, name_len, -1);
39279cc3
CM		 1781 if (IS_ERR(di)) {
1782 ret = PTR_ERR(di);
1783 goto err;
1784 }
1785 if (!di) {
1786 ret = -ENOENT;
1787 goto err;
1788 }
1789 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 1790 btrfs_release_path(root, path);
39279cc3 1791
e02119d5
CM		 1792 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1793 inode, dir->i_ino);
49eb7e46
CM		 1794 BUG_ON(ret != 0 && ret != -ENOENT);
1795 if (ret != -ENOENT)
1796 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 1797
1798 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1799 dir, index);
1800 BUG_ON(ret);
39279cc3
CM		 1801err:
1802 btrfs_free_path(path);
e02119d5
CM		 1803 if (ret)
1804 goto out;
1805
1806 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1807 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1808 btrfs_update_inode(trans, root, dir);
1809 btrfs_drop_nlink(inode);
1810 ret = btrfs_update_inode(trans, root, inode);
1811 dir->i_sb->s_dirt = 1;
1812out:
39279cc3
CM		 1813 return ret;
1814}
1815
1816static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1817{
1818 struct btrfs_root *root;
1819 struct btrfs_trans_handle *trans;
7b128766 1820 struct inode *inode = dentry->d_inode;
39279cc3 1821 int ret;
1832a6d5 1822 unsigned long nr = 0;
39279cc3
CM		 1823
1824 root = BTRFS_I(dir)->root;
1832a6d5
CM		 1825
1826 ret = btrfs_check_free_space(root, 1, 1);
1827 if (ret)
1828 goto fail;
1829
39279cc3 1830 trans = btrfs_start_transaction(root, 1);
5f39d397 1831
39279cc3 1832 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 1833 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1834 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 1835
1836 if (inode->i_nlink == 0)
1837 ret = btrfs_orphan_add(trans, inode);
1838
d3c2fdcf 1839 nr = trans->blocks_used;
5f39d397 1840
89ce8a63 1841 btrfs_end_transaction_throttle(trans, root);
1832a6d5 1842fail:
d3c2fdcf 1843 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 1844 return ret;
1845}
1846
1847static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1848{
1849 struct inode *inode = dentry->d_inode;
1832a6d5 1850 int err = 0;
39279cc3
CM		 1851 int ret;
1852 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 1853 struct btrfs_trans_handle *trans;
1832a6d5 1854 unsigned long nr = 0;
39279cc3 1855
925baedd 1856 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
134d4512 1857 return -ENOTEMPTY;
925baedd 1858 }
134d4512 1859
1832a6d5
CM		 1860 ret = btrfs_check_free_space(root, 1, 1);
1861 if (ret)
1862 goto fail;
1863
39279cc3
CM		 1864 trans = btrfs_start_transaction(root, 1);
1865 btrfs_set_trans_block_group(trans, dir);
39279cc3 1866
7b128766
JB		 1867 err = btrfs_orphan_add(trans, inode);
1868 if (err)
1869 goto fail_trans;
1870
39279cc3 1871 /* now the directory is empty */
e02119d5
CM		 1872 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1873 dentry->d_name.name, dentry->d_name.len);
39279cc3 1874 if (!err) {
dbe674a9 1875 btrfs_i_size_write(inode, 0);
39279cc3 1876 }
3954401f 1877
7b128766 1878fail_trans:
d3c2fdcf 1879 nr = trans->blocks_used;
89ce8a63 1880 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 1881fail:
d3c2fdcf 1882 btrfs_btree_balance_dirty(root, nr);
3954401f 1883
39279cc3
CM		 1884 if (ret && !err)
1885 err = ret;
1886 return err;
1887}
1888
323ac95b
CM		 1889/*
1890 * when truncating bytes in a file, it is possible to avoid reading
1891 * the leaves that contain only checksum items. This can be the
1892 * majority of the IO required to delete a large file, but it must
1893 * be done carefully.
1894 *
1895 * The keys in the level just above the leaves are checked to make sure
1896 * the lowest key in a given leaf is a csum key, and starts at an offset
1897 * after the new size.
1898 *
1899 * Then the key for the next leaf is checked to make sure it also has
1900 * a checksum item for the same file. If it does, we know our target leaf
1901 * contains only checksum items, and it can be safely freed without reading
1902 * it.
1903 *
1904 * This is just an optimization targeted at large files. It may do
1905 * nothing. It will return 0 unless things went badly.
1906 */
1907static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root,
1909 struct btrfs_path *path,
1910 struct inode *inode, u64 new_size)
1911{
1912 struct btrfs_key key;
1913 int ret;
1914 int nritems;
1915 struct btrfs_key found_key;
1916 struct btrfs_key other_key;
5b84e8d6
YZ		 1917 struct btrfs_leaf_ref *ref;
1918 u64 leaf_gen;
1919 u64 leaf_start;
323ac95b
CM		 1920
1921 path->lowest_level = 1;
1922 key.objectid = inode->i_ino;
1923 key.type = BTRFS_CSUM_ITEM_KEY;
1924 key.offset = new_size;
1925again:
1926 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1927 if (ret < 0)
1928 goto out;
1929
1930 if (path->nodes[1] == NULL) {
1931 ret = 0;
1932 goto out;
1933 }
1934 ret = 0;
1935 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
1936 nritems = btrfs_header_nritems(path->nodes[1]);
1937
1938 if (!nritems)
1939 goto out;
1940
1941 if (path->slots[1] >= nritems)
1942 goto next_node;
1943
1944 /* did we find a key greater than anything we want to delete? */
1945 if (found_key.objectid > inode->i_ino ||
1946 (found_key.objectid == inode->i_ino && found_key.type > key.type))
1947 goto out;
1948
1949 /* we check the next key in the node to make sure the leave contains
1950 * only checksum items. This comparison doesn't work if our
1951 * leaf is the last one in the node
1952 */
1953 if (path->slots[1] + 1 >= nritems) {
1954next_node:
1955 /* search forward from the last key in the node, this
1956 * will bring us into the next node in the tree
1957 */
1958 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
1959
1960 /* unlikely, but we inc below, so check to be safe */
1961 if (found_key.offset == (u64)-1)
1962 goto out;
1963
1964 /* search_forward needs a path with locks held, do the
1965 * search again for the original key. It is possible
1966 * this will race with a balance and return a path that
1967 * we could modify, but this drop is just an optimization
1968 * and is allowed to miss some leaves.
1969 */
1970 btrfs_release_path(root, path);
1971 found_key.offset++;
1972
1973 /* setup a max key for search_forward */
1974 other_key.offset = (u64)-1;
1975 other_key.type = key.type;
1976 other_key.objectid = key.objectid;
1977
1978 path->keep_locks = 1;
1979 ret = btrfs_search_forward(root, &found_key, &other_key,
1980 path, 0, 0);
1981 path->keep_locks = 0;
1982 if (ret || found_key.objectid != key.objectid ||
1983 found_key.type != key.type) {
1984 ret = 0;
1985 goto out;
1986 }
1987
1988 key.offset = found_key.offset;
1989 btrfs_release_path(root, path);
1990 cond_resched();
1991 goto again;
1992 }
1993
1994 /* we know there's one more slot after us in the tree,
1995 * read that key so we can verify it is also a checksum item
1996 */
1997 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
1998
1999 if (found_key.objectid < inode->i_ino)
2000 goto next_key;
2001
2002 if (found_key.type != key.type || found_key.offset < new_size)
2003 goto next_key;
2004
2005 /*
2006 * if the key for the next leaf isn't a csum key from this objectid,
2007 * we can't be sure there aren't good items inside this leaf.
2008 * Bail out
2009 */
2010 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2011 goto out;
2012
5b84e8d6
YZ		 2013 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2014 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2015 /*
2016 * it is safe to delete this leaf, it contains only
2017 * csum items from this inode at an offset >= new_size
2018 */
5b84e8d6 2019 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2020 BUG_ON(ret);
2021
5b84e8d6
YZ		 2022 if (root->ref_cows && leaf_gen < trans->transid) {
2023 ref = btrfs_alloc_leaf_ref(root, 0);
2024 if (ref) {
2025 ref->root_gen = root->root_key.offset;
2026 ref->bytenr = leaf_start;
2027 ref->owner = 0;
2028 ref->generation = leaf_gen;
2029 ref->nritems = 0;
2030
2031 ret = btrfs_add_leaf_ref(root, ref, 0);
2032 WARN_ON(ret);
2033 btrfs_free_leaf_ref(root, ref);
2034 } else {
2035 WARN_ON(1);
2036 }
2037 }
323ac95b
CM		 2038next_key:
2039 btrfs_release_path(root, path);
2040
2041 if (other_key.objectid == inode->i_ino &&
2042 other_key.type == key.type && other_key.offset > key.offset) {
2043 key.offset = other_key.offset;
2044 cond_resched();
2045 goto again;
2046 }
2047 ret = 0;
2048out:
2049 /* fixup any changes we've made to the path */
2050 path->lowest_level = 0;
2051 path->keep_locks = 0;
2052 btrfs_release_path(root, path);
2053 return ret;
2054}
2055
39279cc3
CM		 2056/*
2057 * this can truncate away extent items, csum items and directory items.
2058 * It starts at a high offset and removes keys until it can't find
d352ac68 2059 * any higher than new_size
39279cc3
CM		 2060 *
2061 * csum items that cross the new i_size are truncated to the new size
2062 * as well.
7b128766
JB		 2063 *
2064 * min_type is the minimum key type to truncate down to. If set to 0, this
2065 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2066 */
e02119d5
CM		 2067noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2068 struct btrfs_root *root,
2069 struct inode *inode,
2070 u64 new_size, u32 min_type)
39279cc3
CM		 2071{
2072 int ret;
2073 struct btrfs_path *path;
2074 struct btrfs_key key;
5f39d397 2075 struct btrfs_key found_key;
39279cc3 2076 u32 found_type;
5f39d397 2077 struct extent_buffer *leaf;
39279cc3
CM		 2078 struct btrfs_file_extent_item *fi;
2079 u64 extent_start = 0;
db94535d 2080 u64 extent_num_bytes = 0;
39279cc3 2081 u64 item_end = 0;
7bb86316 2082 u64 root_gen = 0;
d8d5f3e1 2083 u64 root_owner = 0;
39279cc3
CM		 2084 int found_extent;
2085 int del_item;
85e21bac
CM		 2086 int pending_del_nr = 0;
2087 int pending_del_slot = 0;
179e29e4 2088 int extent_type = -1;
3b951516 2089 u64 mask = root->sectorsize - 1;
39279cc3 2090
e02119d5 2091 if (root->ref_cows)
5b21f2ed 2092 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2093 path = btrfs_alloc_path();
3c69faec 2094 path->reada = -1;
39279cc3 2095 BUG_ON(!path);
5f39d397 2096
39279cc3
CM		 2097 /* FIXME, add redo link to tree so we don't leak on crash */
2098 key.objectid = inode->i_ino;
2099 key.offset = (u64)-1;
5f39d397
CM		 2100 key.type = (u8)-1;
2101
85e21bac 2102 btrfs_init_path(path);
323ac95b
CM		 2103
2104 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2105 BUG_ON(ret);
2106
85e21bac
CM		 2107search_again:
2108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2109 if (ret < 0) {
2110 goto error;
2111 }
2112 if (ret > 0) {
e02119d5
CM		 2113 /* there are no items in the tree for us to truncate, we're
2114 * done
2115 */
2116 if (path->slots[0] == 0) {
2117 ret = 0;
2118 goto error;
2119 }
85e21bac
CM		 2120 path->slots[0]--;
2121 }
2122
39279cc3 2123 while(1) {
39279cc3 2124 fi = NULL;
5f39d397
CM		 2125 leaf = path->nodes[0];
2126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2127 found_type = btrfs_key_type(&found_key);
39279cc3 2128
5f39d397 2129 if (found_key.objectid != inode->i_ino)
39279cc3 2130 break;
5f39d397 2131
85e21bac 2132 if (found_type < min_type)
39279cc3
CM		 2133 break;
2134
5f39d397 2135 item_end = found_key.offset;
39279cc3 2136 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2137 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2138 struct btrfs_file_extent_item);
179e29e4
CM		 2139 extent_type = btrfs_file_extent_type(leaf, fi);
2140 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2141 item_end +=
db94535d 2142 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2143 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2144 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2145 fi);
39279cc3 2146 }
008630c1 2147 item_end--;
39279cc3
CM		 2148 }
2149 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2150 ret = btrfs_csum_truncate(trans, root, path,
e02119d5 2151 new_size);
39279cc3
CM		 2152 BUG_ON(ret);
2153 }
e02119d5 2154 if (item_end < new_size) {
b888db2b
CM		 2155 if (found_type == BTRFS_DIR_ITEM_KEY) {
2156 found_type = BTRFS_INODE_ITEM_KEY;
2157 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2158 found_type = BTRFS_CSUM_ITEM_KEY;
85e21bac
CM		 2159 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2160 found_type = BTRFS_XATTR_ITEM_KEY;
2161 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2162 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM		 2163 } else if (found_type) {
2164 found_type--;
2165 } else {
2166 break;
39279cc3 2167 }
a61721d5 2168 btrfs_set_key_type(&key, found_type);
85e21bac 2169 goto next;
39279cc3 2170 }
e02119d5 2171 if (found_key.offset >= new_size)
39279cc3
CM		 2172 del_item = 1;
2173 else
2174 del_item = 0;
2175 found_extent = 0;
2176
2177 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2178 if (found_type != BTRFS_EXTENT_DATA_KEY)
2179 goto delete;
2180
2181 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2182 u64 num_dec;
db94535d 2183 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
39279cc3 2184 if (!del_item) {
db94535d
CM		 2185 u64 orig_num_bytes =
2186 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2187 extent_num_bytes = new_size -
5f39d397 2188 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2189 extent_num_bytes = extent_num_bytes &
2190 ~((u64)root->sectorsize - 1);
db94535d
CM		 2191 btrfs_set_file_extent_num_bytes(leaf, fi,
2192 extent_num_bytes);
2193 num_dec = (orig_num_bytes -
9069218d 2194 extent_num_bytes);
e02119d5 2195 if (root->ref_cows && extent_start != 0)
a76a3cd4 2196 inode_sub_bytes(inode, num_dec);
5f39d397 2197 btrfs_mark_buffer_dirty(leaf);
39279cc3 2198 } else {
db94535d
CM		 2199 extent_num_bytes =
2200 btrfs_file_extent_disk_num_bytes(leaf,
2201 fi);
39279cc3 2202 /* FIXME blocksize != 4096 */
9069218d 2203 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2204 if (extent_start != 0) {
2205 found_extent = 1;
e02119d5 2206 if (root->ref_cows)
a76a3cd4 2207 inode_sub_bytes(inode, num_dec);
e02119d5 2208 }
31840ae1 2209 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2210 root_owner = btrfs_header_owner(leaf);
39279cc3 2211 }
9069218d 2212 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2213 /*
2214 * we can't truncate inline items that have had
2215 * special encodings
2216 */
2217 if (!del_item &&
2218 btrfs_file_extent_compression(leaf, fi) == 0 &&
2219 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2220 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2221 u32 size = new_size - found_key.offset;
2222
2223 if (root->ref_cows) {
a76a3cd4
YZ		 2224 inode_sub_bytes(inode, item_end + 1 -
2225 new_size);
e02119d5
CM		 2226 }
2227 size =
2228 btrfs_file_extent_calc_inline_size(size);
9069218d 2229 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2230 size, 1);
9069218d 2231 BUG_ON(ret);
e02119d5 2232 } else if (root->ref_cows) {
a76a3cd4
YZ		 2233 inode_sub_bytes(inode, item_end + 1 -
2234 found_key.offset);
9069218d 2235 }
39279cc3 2236 }
179e29e4 2237delete:
39279cc3 2238 if (del_item) {
85e21bac
CM		 2239 if (!pending_del_nr) {
2240 /* no pending yet, add ourselves */
2241 pending_del_slot = path->slots[0];
2242 pending_del_nr = 1;
2243 } else if (pending_del_nr &&
2244 path->slots[0] + 1 == pending_del_slot) {
2245 /* hop on the pending chunk */
2246 pending_del_nr++;
2247 pending_del_slot = path->slots[0];
2248 } else {
2249 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2250 }
39279cc3
CM		 2251 } else {
2252 break;
2253 }
39279cc3
CM		 2254 if (found_extent) {
2255 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2256 extent_num_bytes,
31840ae1 2257 leaf->start, root_owner,
3bb1a1bc 2258 root_gen, inode->i_ino, 0);
39279cc3
CM		 2259 BUG_ON(ret);
2260 }
85e21bac
CM		 2261next:
2262 if (path->slots[0] == 0) {
2263 if (pending_del_nr)
2264 goto del_pending;
2265 btrfs_release_path(root, path);
2266 goto search_again;
2267 }
2268
2269 path->slots[0]--;
2270 if (pending_del_nr &&
2271 path->slots[0] + 1 != pending_del_slot) {
2272 struct btrfs_key debug;
2273del_pending:
2274 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2275 pending_del_slot);
2276 ret = btrfs_del_items(trans, root, path,
2277 pending_del_slot,
2278 pending_del_nr);
2279 BUG_ON(ret);
2280 pending_del_nr = 0;
2281 btrfs_release_path(root, path);
2282 goto search_again;
2283 }
39279cc3
CM		 2284 }
2285 ret = 0;
2286error:
85e21bac
CM		 2287 if (pending_del_nr) {
2288 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2289 pending_del_nr);
2290 }
39279cc3
CM		 2291 btrfs_free_path(path);
2292 inode->i_sb->s_dirt = 1;
2293 return ret;
2294}
2295
2296/*
2297 * taken from block_truncate_page, but does cow as it zeros out
2298 * any bytes left in the last page in the file.
2299 */
2300static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2301{
2302 struct inode *inode = mapping->host;
db94535d 2303 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2304 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2305 struct btrfs_ordered_extent *ordered;
2306 char *kaddr;
db94535d 2307 u32 blocksize = root->sectorsize;
39279cc3
CM		 2308 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2309 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2310 struct page *page;
39279cc3 2311 int ret = 0;
a52d9a80 2312 u64 page_start;
e6dcd2dc 2313 u64 page_end;
39279cc3
CM		 2314
2315 if ((offset & (blocksize - 1)) == 0)
2316 goto out;
2317
2318 ret = -ENOMEM;
211c17f5 2319again:
39279cc3
CM		 2320 page = grab_cache_page(mapping, index);
2321 if (!page)
2322 goto out;
e6dcd2dc
CM		 2323
2324 page_start = page_offset(page);
2325 page_end = page_start + PAGE_CACHE_SIZE - 1;
2326
39279cc3 2327 if (!PageUptodate(page)) {
9ebefb18 2328 ret = btrfs_readpage(NULL, page);
39279cc3 2329 lock_page(page);
211c17f5
CM		 2330 if (page->mapping != mapping) {
2331 unlock_page(page);
2332 page_cache_release(page);
2333 goto again;
2334 }
39279cc3
CM		 2335 if (!PageUptodate(page)) {
2336 ret = -EIO;
89642229 2337 goto out_unlock;
39279cc3
CM		 2338 }
2339 }
211c17f5 2340 wait_on_page_writeback(page);
e6dcd2dc
CM		 2341
2342 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2343 set_page_extent_mapped(page);
2344
2345 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2346 if (ordered) {
2347 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2348 unlock_page(page);
2349 page_cache_release(page);
eb84ae03 2350 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2351 btrfs_put_ordered_extent(ordered);
2352 goto again;
2353 }
2354
ea8c2819 2355 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2356 ret = 0;
2357 if (offset != PAGE_CACHE_SIZE) {
2358 kaddr = kmap(page);
2359 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2360 flush_dcache_page(page);
2361 kunmap(page);
2362 }
247e743c 2363 ClearPageChecked(page);
e6dcd2dc
CM		 2364 set_page_dirty(page);
2365 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2366
89642229 2367out_unlock:
39279cc3
CM		 2368 unlock_page(page);
2369 page_cache_release(page);
2370out:
2371 return ret;
2372}
2373
9036c102 2374int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2375{
9036c102
YZ		 2376 struct btrfs_trans_handle *trans;
2377 struct btrfs_root *root = BTRFS_I(inode)->root;
2378 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2379 struct extent_map *em;
2380 u64 mask = root->sectorsize - 1;
2381 u64 hole_start = (inode->i_size + mask) & ~mask;
2382 u64 block_end = (size + mask) & ~mask;
2383 u64 last_byte;
2384 u64 cur_offset;
2385 u64 hole_size;
39279cc3
CM		 2386 int err;
2387
9036c102
YZ		 2388 if (size <= hole_start)
2389 return 0;
2390
2391 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2392 if (err)
2393 return err;
2394
9036c102 2395 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2396
9036c102
YZ		 2397 while (1) {
2398 struct btrfs_ordered_extent *ordered;
2399 btrfs_wait_ordered_range(inode, hole_start,
2400 block_end - hole_start);
2401 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2402 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2403 if (!ordered)
2404 break;
2405 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2406 btrfs_put_ordered_extent(ordered);
2407 }
39279cc3 2408
9036c102
YZ		 2409 trans = btrfs_start_transaction(root, 1);
2410 btrfs_set_trans_block_group(trans, inode);
39279cc3 2411
9036c102
YZ		 2412 cur_offset = hole_start;
2413 while (1) {
2414 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2415 block_end - cur_offset, 0);
2416 BUG_ON(IS_ERR(em) || !em);
2417 last_byte = min(extent_map_end(em), block_end);
2418 last_byte = (last_byte + mask) & ~mask;
2419 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2420 hole_size = last_byte - cur_offset;
2421 err = btrfs_insert_file_extent(trans, root,
2422 inode->i_ino, cur_offset, 0,
2423 0, hole_size, 0, hole_size,
2424 0, 0, 0);
2425 btrfs_drop_extent_cache(inode, hole_start,
2426 last_byte - 1, 0);
2427 }
2428 free_extent_map(em);
2429 cur_offset = last_byte;
2430 if (err || cur_offset >= block_end)
2431 break;
2432 }
1832a6d5 2433
9036c102
YZ		 2434 btrfs_end_transaction(trans, root);
2435 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2436 return err;
2437}
39279cc3 2438
9036c102
YZ		 2439static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2440{
2441 struct inode *inode = dentry->d_inode;
2442 int err;
39279cc3 2443
9036c102
YZ		 2444 err = inode_change_ok(inode, attr);
2445 if (err)
2446 return err;
2bf5a725 2447
9036c102
YZ		 2448 if (S_ISREG(inode->i_mode) &&
2449 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2450 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2451 if (err)
2452 return err;
39279cc3 2453 }
9036c102 2454
39279cc3 2455 err = inode_setattr(inode, attr);
33268eaf
JB		 2456
2457 if (!err && ((attr->ia_valid & ATTR_MODE)))
2458 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2459 return err;
2460}
61295eb8 2461
39279cc3
CM		 2462void btrfs_delete_inode(struct inode *inode)
2463{
2464 struct btrfs_trans_handle *trans;
2465 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2466 unsigned long nr;
39279cc3
CM		 2467 int ret;
2468
2469 truncate_inode_pages(&inode->i_data, 0);
2470 if (is_bad_inode(inode)) {
7b128766 2471 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2472 goto no_delete;
2473 }
4a096752 2474 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2475
dbe674a9 2476 btrfs_i_size_write(inode, 0);
39279cc3 2477 trans = btrfs_start_transaction(root, 1);
5f39d397 2478
39279cc3 2479 btrfs_set_trans_block_group(trans, inode);
e02119d5 2480 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2481 if (ret) {
2482 btrfs_orphan_del(NULL, inode);
54aa1f4d 2483 goto no_delete_lock;
7b128766
JB		 2484 }
2485
2486 btrfs_orphan_del(trans, inode);
85e21bac 2487
d3c2fdcf 2488 nr = trans->blocks_used;
85e21bac 2489 clear_inode(inode);
5f39d397 2490
39279cc3 2491 btrfs_end_transaction(trans, root);
d3c2fdcf 2492 btrfs_btree_balance_dirty(root, nr);
39279cc3 2493 return;
54aa1f4d
CM		 2494
2495no_delete_lock:
d3c2fdcf 2496 nr = trans->blocks_used;
54aa1f4d 2497 btrfs_end_transaction(trans, root);
d3c2fdcf 2498 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2499no_delete:
2500 clear_inode(inode);
2501}
2502
2503/*
2504 * this returns the key found in the dir entry in the location pointer.
2505 * If no dir entries were found, location->objectid is 0.
2506 */
2507static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2508 struct btrfs_key *location)
2509{
2510 const char *name = dentry->d_name.name;
2511 int namelen = dentry->d_name.len;
2512 struct btrfs_dir_item *di;
2513 struct btrfs_path *path;
2514 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2515 int ret = 0;
39279cc3
CM		 2516
2517 path = btrfs_alloc_path();
2518 BUG_ON(!path);
3954401f 2519
39279cc3
CM		 2520 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2521 namelen, 0);
0d9f7f3e
Y		 2522 if (IS_ERR(di))
2523 ret = PTR_ERR(di);
39279cc3 2524 if (!di || IS_ERR(di)) {
3954401f 2525 goto out_err;
39279cc3 2526 }
5f39d397 2527 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2528out:
39279cc3
CM		 2529 btrfs_free_path(path);
2530 return ret;
3954401f
CM		 2531out_err:
2532 location->objectid = 0;
2533 goto out;
39279cc3
CM		 2534}
2535
2536/*
2537 * when we hit a tree root in a directory, the btrfs part of the inode
2538 * needs to be changed to reflect the root directory of the tree root. This
2539 * is kind of like crossing a mount point.
2540 */
2541static int fixup_tree_root_location(struct btrfs_root *root,
2542 struct btrfs_key *location,
58176a96
JB		 2543 struct btrfs_root **sub_root,
2544 struct dentry *dentry)
39279cc3 2545{
39279cc3
CM		 2546 struct btrfs_root_item *ri;
2547
2548 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2549 return 0;
2550 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2551 return 0;
2552
58176a96
JB		 2553 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2554 dentry->d_name.name,
2555 dentry->d_name.len);
39279cc3
CM		 2556 if (IS_ERR(*sub_root))
2557 return PTR_ERR(*sub_root);
2558
2559 ri = &(*sub_root)->root_item;
2560 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2561 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2562 location->offset = 0;
2563
39279cc3
CM		 2564 return 0;
2565}
2566
e02119d5 2567static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2568{
e02119d5
CM		 2569 struct btrfs_inode *bi = BTRFS_I(inode);
2570
2571 bi->i_acl = NULL;
2572 bi->i_default_acl = NULL;
2573
2574 bi->generation = 0;
2575 bi->last_trans = 0;
2576 bi->logged_trans = 0;
2577 bi->delalloc_bytes = 0;
2578 bi->disk_i_size = 0;
2579 bi->flags = 0;
2580 bi->index_cnt = (u64)-1;
49eb7e46 2581 bi->log_dirty_trans = 0;
d1310b2e
CM		 2582 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2583 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2584 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2585 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2586 inode->i_mapping, GFP_NOFS);
ea8c2819 2587 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2588 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1b1e2135 2589 mutex_init(&BTRFS_I(inode)->csum_mutex);
ee6e6504 2590 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2591 mutex_init(&BTRFS_I(inode)->log_mutex);
2592}
2593
2594static int btrfs_init_locked_inode(struct inode *inode, void *p)
2595{
2596 struct btrfs_iget_args *args = p;
2597 inode->i_ino = args->ino;
2598 init_btrfs_i(inode);
2599 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 2600 return 0;
2601}
2602
2603static int btrfs_find_actor(struct inode *inode, void *opaque)
2604{
2605 struct btrfs_iget_args *args = opaque;
2606 return (args->ino == inode->i_ino &&
2607 args->root == BTRFS_I(inode)->root);
2608}
2609
5b21f2ed
ZY		 2610struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2611 struct btrfs_root *root, int wait)
2612{
2613 struct inode *inode;
2614 struct btrfs_iget_args args;
2615 args.ino = objectid;
2616 args.root = root;
2617
2618 if (wait) {
2619 inode = ilookup5(s, objectid, btrfs_find_actor,
2620 (void *)&args);
2621 } else {
2622 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2623 (void *)&args);
2624 }
2625 return inode;
2626}
2627
39279cc3
CM		 2628struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
2629 struct btrfs_root *root)
2630{
2631 struct inode *inode;
2632 struct btrfs_iget_args args;
2633 args.ino = objectid;
2634 args.root = root;
2635
2636 inode = iget5_locked(s, objectid, btrfs_find_actor,
2637 btrfs_init_locked_inode,
2638 (void *)&args);
2639 return inode;
2640}
2641
1a54ef8c
BR		 2642/* Get an inode object given its location and corresponding root.
2643 * Returns in *is_new if the inode was read from disk
2644 */
2645struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
2646 struct btrfs_root *root, int *is_new)
2647{
2648 struct inode *inode;
2649
2650 inode = btrfs_iget_locked(s, location->objectid, root);
2651 if (!inode)
2652 return ERR_PTR(-EACCES);
2653
2654 if (inode->i_state & I_NEW) {
2655 BTRFS_I(inode)->root = root;
2656 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
2657 btrfs_read_locked_inode(inode);
2658 unlock_new_inode(inode);
2659 if (is_new)
2660 *is_new = 1;
2661 } else {
2662 if (is_new)
2663 *is_new = 0;
2664 }
2665
2666 return inode;
2667}
2668
39279cc3
CM		 2669static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
2670 struct nameidata *nd)
2671{
2672 struct inode * inode;
2673 struct btrfs_inode *bi = BTRFS_I(dir);
2674 struct btrfs_root *root = bi->root;
2675 struct btrfs_root *sub_root = root;
2676 struct btrfs_key location;
1a54ef8c 2677 int ret, new, do_orphan = 0;
39279cc3
CM		 2678
2679 if (dentry->d_name.len > BTRFS_NAME_LEN)
2680 return ERR_PTR(-ENAMETOOLONG);
5f39d397 2681
39279cc3 2682 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 2683
39279cc3
CM		 2684 if (ret < 0)
2685 return ERR_PTR(ret);
5f39d397 2686
39279cc3
CM		 2687 inode = NULL;
2688 if (location.objectid) {
58176a96
JB		 2689 ret = fixup_tree_root_location(root, &location, &sub_root,
2690 dentry);
39279cc3
CM		 2691 if (ret < 0)
2692 return ERR_PTR(ret);
2693 if (ret > 0)
2694 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 2695 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
2696 if (IS_ERR(inode))
2697 return ERR_CAST(inode);
2698
2699 /* the inode and parent dir are two different roots */
2700 if (new && root != sub_root) {
2701 igrab(inode);
2702 sub_root->inode = inode;
2703 do_orphan = 1;
39279cc3
CM		 2704 }
2705 }
7b128766
JB		 2706
2707 if (unlikely(do_orphan))
2708 btrfs_orphan_cleanup(sub_root);
2709
39279cc3
CM		 2710 return d_splice_alias(inode, dentry);
2711}
2712
39279cc3
CM		 2713static unsigned char btrfs_filetype_table[] = {
2714 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
2715};
2716
cbdf5a24
DW		 2717static int btrfs_real_readdir(struct file *filp, void *dirent,
2718 filldir_t filldir)
39279cc3 2719{
6da6abae 2720 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 2721 struct btrfs_root *root = BTRFS_I(inode)->root;
2722 struct btrfs_item *item;
2723 struct btrfs_dir_item *di;
2724 struct btrfs_key key;
5f39d397 2725 struct btrfs_key found_key;
39279cc3
CM		 2726 struct btrfs_path *path;
2727 int ret;
2728 u32 nritems;
5f39d397 2729 struct extent_buffer *leaf;
39279cc3
CM		 2730 int slot;
2731 int advance;
2732 unsigned char d_type;
2733 int over = 0;
2734 u32 di_cur;
2735 u32 di_total;
2736 u32 di_len;
2737 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 2738 char tmp_name[32];
2739 char *name_ptr;
2740 int name_len;
39279cc3
CM		 2741
2742 /* FIXME, use a real flag for deciding about the key type */
2743 if (root->fs_info->tree_root == root)
2744 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 2745
3954401f
CM		 2746 /* special case for "." */
2747 if (filp->f_pos == 0) {
2748 over = filldir(dirent, ".", 1,
2749 1, inode->i_ino,
2750 DT_DIR);
2751 if (over)
2752 return 0;
2753 filp->f_pos = 1;
2754 }
3954401f
CM		 2755 /* special case for .., just use the back ref */
2756 if (filp->f_pos == 1) {
5ecc7e5d 2757 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 2758 over = filldir(dirent, "..", 2,
5ecc7e5d 2759 2, pino, DT_DIR);
3954401f 2760 if (over)
49593bfa 2761 return 0;
3954401f
CM		 2762 filp->f_pos = 2;
2763 }
2764
49593bfa
DW		 2765 path = btrfs_alloc_path();
2766 path->reada = 2;
2767
39279cc3
CM		 2768 btrfs_set_key_type(&key, key_type);
2769 key.offset = filp->f_pos;
49593bfa 2770 key.objectid = inode->i_ino;
5f39d397 2771
39279cc3
CM		 2772 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2773 if (ret < 0)
2774 goto err;
2775 advance = 0;
49593bfa
DW		 2776
2777 while (1) {
5f39d397
CM		 2778 leaf = path->nodes[0];
2779 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 2780 slot = path->slots[0];
2781 if (advance || slot >= nritems) {
49593bfa 2782 if (slot >= nritems - 1) {
39279cc3
CM		 2783 ret = btrfs_next_leaf(root, path);
2784 if (ret)
2785 break;
5f39d397
CM		 2786 leaf = path->nodes[0];
2787 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 2788 slot = path->slots[0];
2789 } else {
2790 slot++;
2791 path->slots[0]++;
2792 }
2793 }
2794 advance = 1;
5f39d397
CM		 2795 item = btrfs_item_nr(leaf, slot);
2796 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2797
2798 if (found_key.objectid != key.objectid)
39279cc3 2799 break;
5f39d397 2800 if (btrfs_key_type(&found_key) != key_type)
39279cc3 2801 break;
5f39d397 2802 if (found_key.offset < filp->f_pos)
39279cc3 2803 continue;
5f39d397
CM		 2804
2805 filp->f_pos = found_key.offset;
49593bfa 2806
39279cc3
CM		 2807 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2808 di_cur = 0;
5f39d397 2809 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 2810
2811 while (di_cur < di_total) {
5f39d397
CM		 2812 struct btrfs_key location;
2813
2814 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 2815 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 2816 name_ptr = tmp_name;
2817 } else {
2818 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 2819 if (!name_ptr) {
2820 ret = -ENOMEM;
2821 goto err;
2822 }
5f39d397
CM		 2823 }
2824 read_extent_buffer(leaf, name_ptr,
2825 (unsigned long)(di + 1), name_len);
2826
2827 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2828 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5f39d397 2829 over = filldir(dirent, name_ptr, name_len,
49593bfa 2830 found_key.offset, location.objectid,
39279cc3 2831 d_type);
5f39d397
CM		 2832
2833 if (name_ptr != tmp_name)
2834 kfree(name_ptr);
2835
39279cc3
CM		 2836 if (over)
2837 goto nopos;
49593bfa 2838
5103e947 2839 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 2840 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 2841 di_cur += di_len;
2842 di = (struct btrfs_dir_item *)((char *)di + di_len);
2843 }
2844 }
49593bfa
DW		 2845
2846 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 2847 if (key_type == BTRFS_DIR_INDEX_KEY)
2848 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2849 else
2850 filp->f_pos++;
39279cc3
CM		 2851nopos:
2852 ret = 0;
2853err:
39279cc3 2854 btrfs_free_path(path);
39279cc3
CM		 2855 return ret;
2856}
2857
2858int btrfs_write_inode(struct inode *inode, int wait)
2859{
2860 struct btrfs_root *root = BTRFS_I(inode)->root;
2861 struct btrfs_trans_handle *trans;
2862 int ret = 0;
2863
4ca8b41e
CM		 2864 if (root->fs_info->closing > 1)
2865 return 0;
2866
39279cc3 2867 if (wait) {
f9295749 2868 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 2869 btrfs_set_trans_block_group(trans, inode);
2870 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 2871 }
2872 return ret;
2873}
2874
2875/*
54aa1f4d 2876 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 2877 * inode changes. But, it is most likely to find the inode in cache.
2878 * FIXME, needs more benchmarking...there are no reasons other than performance
2879 * to keep or drop this code.
2880 */
2881void btrfs_dirty_inode(struct inode *inode)
2882{
2883 struct btrfs_root *root = BTRFS_I(inode)->root;
2884 struct btrfs_trans_handle *trans;
2885
f9295749 2886 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 2887 btrfs_set_trans_block_group(trans, inode);
2888 btrfs_update_inode(trans, root, inode);
2889 btrfs_end_transaction(trans, root);
39279cc3
CM		 2890}
2891
d352ac68
CM		 2892/*
2893 * find the highest existing sequence number in a directory
2894 * and then set the in-memory index_cnt variable to reflect
2895 * free sequence numbers
2896 */
aec7477b
JB		 2897static int btrfs_set_inode_index_count(struct inode *inode)
2898{
2899 struct btrfs_root *root = BTRFS_I(inode)->root;
2900 struct btrfs_key key, found_key;
2901 struct btrfs_path *path;
2902 struct extent_buffer *leaf;
2903 int ret;
2904
2905 key.objectid = inode->i_ino;
2906 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2907 key.offset = (u64)-1;
2908
2909 path = btrfs_alloc_path();
2910 if (!path)
2911 return -ENOMEM;
2912
2913 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2914 if (ret < 0)
2915 goto out;
2916 /* FIXME: we should be able to handle this */
2917 if (ret == 0)
2918 goto out;
2919 ret = 0;
2920
2921 /*
2922 * MAGIC NUMBER EXPLANATION:
2923 * since we search a directory based on f_pos we have to start at 2
2924 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2925 * else has to start at 2
2926 */
2927 if (path->slots[0] == 0) {
2928 BTRFS_I(inode)->index_cnt = 2;
2929 goto out;
2930 }
2931
2932 path->slots[0]--;
2933
2934 leaf = path->nodes[0];
2935 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2936
2937 if (found_key.objectid != inode->i_ino ||
2938 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2939 BTRFS_I(inode)->index_cnt = 2;
2940 goto out;
2941 }
2942
2943 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2944out:
2945 btrfs_free_path(path);
2946 return ret;
2947}
2948
d352ac68
CM		 2949/*
2950 * helper to find a free sequence number in a given directory. This current
2951 * code is very simple, later versions will do smarter things in the btree
2952 */
00e4e6b3
CM		 2953static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2954 u64 *index)
aec7477b
JB		 2955{
2956 int ret = 0;
2957
2958 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2959 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 2960 if (ret) {
aec7477b 2961 return ret;
8d5bf1cb 2962 }
aec7477b
JB		 2963 }
2964
00e4e6b3 2965 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 2966 BTRFS_I(dir)->index_cnt++;
2967
2968 return ret;
2969}
2970
39279cc3
CM		 2971static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2972 struct btrfs_root *root,
aec7477b 2973 struct inode *dir,
9c58309d
CM		 2974 const char *name, int name_len,
2975 u64 ref_objectid,
39279cc3
CM		 2976 u64 objectid,
2977 struct btrfs_block_group_cache *group,
00e4e6b3 2978 int mode, u64 *index)
39279cc3
CM		 2979{
2980 struct inode *inode;
5f39d397 2981 struct btrfs_inode_item *inode_item;
6324fbf3 2982 struct btrfs_block_group_cache *new_inode_group;
39279cc3 2983 struct btrfs_key *location;
5f39d397 2984 struct btrfs_path *path;
9c58309d
CM		 2985 struct btrfs_inode_ref *ref;
2986 struct btrfs_key key[2];
2987 u32 sizes[2];
2988 unsigned long ptr;
39279cc3
CM		 2989 int ret;
2990 int owner;
2991
5f39d397
CM		 2992 path = btrfs_alloc_path();
2993 BUG_ON(!path);
2994
39279cc3
CM		 2995 inode = new_inode(root->fs_info->sb);
2996 if (!inode)
2997 return ERR_PTR(-ENOMEM);
2998
aec7477b 2999 if (dir) {
00e4e6b3 3000 ret = btrfs_set_inode_index(dir, inode, index);
aec7477b
JB		 3001 if (ret)
3002 return ERR_PTR(ret);
aec7477b
JB		 3003 }
3004 /*
3005 * index_cnt is ignored for everything but a dir,
3006 * btrfs_get_inode_index_count has an explanation for the magic
3007 * number
3008 */
e02119d5 3009 init_btrfs_i(inode);
aec7477b 3010 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3011 BTRFS_I(inode)->root = root;
e02119d5 3012 BTRFS_I(inode)->generation = trans->transid;
b888db2b 3013
39279cc3
CM		 3014 if (mode & S_IFDIR)
3015 owner = 0;
3016 else
3017 owner = 1;
6324fbf3 3018 new_inode_group = btrfs_find_block_group(root, group, 0,
0b86a832 3019 BTRFS_BLOCK_GROUP_METADATA, owner);
6324fbf3
CM		 3020 if (!new_inode_group) {
3021 printk("find_block group failed\n");
3022 new_inode_group = group;
3023 }
3024 BTRFS_I(inode)->block_group = new_inode_group;
9c58309d
CM		 3025
3026 key[0].objectid = objectid;
3027 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3028 key[0].offset = 0;
3029
3030 key[1].objectid = objectid;
3031 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3032 key[1].offset = ref_objectid;
3033
3034 sizes[0] = sizeof(struct btrfs_inode_item);
3035 sizes[1] = name_len + sizeof(*ref);
3036
3037 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3038 if (ret != 0)
5f39d397
CM		 3039 goto fail;
3040
9c58309d
CM		 3041 if (objectid > root->highest_inode)
3042 root->highest_inode = objectid;
3043
39279cc3
CM		 3044 inode->i_uid = current->fsuid;
3045 inode->i_gid = current->fsgid;
3046 inode->i_mode = mode;
3047 inode->i_ino = objectid;
a76a3cd4 3048 inode_set_bytes(inode, 0);
39279cc3 3049 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3050 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3051 struct btrfs_inode_item);
e02119d5 3052 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3053
3054 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3055 struct btrfs_inode_ref);
3056 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3057 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3058 ptr = (unsigned long)(ref + 1);
3059 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3060
5f39d397
CM		 3061 btrfs_mark_buffer_dirty(path->nodes[0]);
3062 btrfs_free_path(path);
3063
39279cc3
CM		 3064 location = &BTRFS_I(inode)->location;
3065 location->objectid = objectid;
39279cc3
CM		 3066 location->offset = 0;
3067 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3068
39279cc3
CM		 3069 insert_inode_hash(inode);
3070 return inode;
5f39d397 3071fail:
aec7477b
JB		 3072 if (dir)
3073 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3074 btrfs_free_path(path);
3075 return ERR_PTR(ret);
39279cc3
CM		 3076}
3077
3078static inline u8 btrfs_inode_type(struct inode *inode)
3079{
3080 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3081}
3082
d352ac68
CM		 3083/*
3084 * utility function to add 'inode' into 'parent_inode' with
3085 * a give name and a given sequence number.
3086 * if 'add_backref' is true, also insert a backref from the
3087 * inode to the parent directory.
3088 */
e02119d5
CM		 3089int btrfs_add_link(struct btrfs_trans_handle *trans,
3090 struct inode *parent_inode, struct inode *inode,
3091 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3092{
3093 int ret;
3094 struct btrfs_key key;
e02119d5 3095 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3096
39279cc3 3097 key.objectid = inode->i_ino;
39279cc3
CM		 3098 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3099 key.offset = 0;
3100
e02119d5
CM		 3101 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3102 parent_inode->i_ino,
aec7477b 3103 &key, btrfs_inode_type(inode),
00e4e6b3 3104 index);
39279cc3 3105 if (ret == 0) {
9c58309d
CM		 3106 if (add_backref) {
3107 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3108 name, name_len,
3109 inode->i_ino,
3110 parent_inode->i_ino,
3111 index);
9c58309d 3112 }
dbe674a9 3113 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3114 name_len * 2);
79c44584 3115 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3116 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3117 }
3118 return ret;
3119}
3120
3121static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3122 struct dentry *dentry, struct inode *inode,
00e4e6b3 3123 int backref, u64 index)
39279cc3 3124{
e02119d5
CM		 3125 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3126 inode, dentry->d_name.name,
3127 dentry->d_name.len, backref, index);
39279cc3
CM		 3128 if (!err) {
3129 d_instantiate(dentry, inode);
3130 return 0;
3131 }
3132 if (err > 0)
3133 err = -EEXIST;
3134 return err;
3135}
3136
618e21d5
JB		 3137static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3138 int mode, dev_t rdev)
3139{
3140 struct btrfs_trans_handle *trans;
3141 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3142 struct inode *inode = NULL;
618e21d5
JB		 3143 int err;
3144 int drop_inode = 0;
3145 u64 objectid;
1832a6d5 3146 unsigned long nr = 0;
00e4e6b3 3147 u64 index = 0;
618e21d5
JB		 3148
3149 if (!new_valid_dev(rdev))
3150 return -EINVAL;
3151
1832a6d5
CM		 3152 err = btrfs_check_free_space(root, 1, 0);
3153 if (err)
3154 goto fail;
3155
618e21d5
JB		 3156 trans = btrfs_start_transaction(root, 1);
3157 btrfs_set_trans_block_group(trans, dir);
3158
3159 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3160 if (err) {
3161 err = -ENOSPC;
3162 goto out_unlock;
3163 }
3164
aec7477b 3165 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3166 dentry->d_name.len,
3167 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3168 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3169 err = PTR_ERR(inode);
3170 if (IS_ERR(inode))
3171 goto out_unlock;
3172
33268eaf
JB		 3173 err = btrfs_init_acl(inode, dir);
3174 if (err) {
3175 drop_inode = 1;
3176 goto out_unlock;
3177 }
3178
618e21d5 3179 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3180 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3181 if (err)
3182 drop_inode = 1;
3183 else {
3184 inode->i_op = &btrfs_special_inode_operations;
3185 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3186 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3187 }
3188 dir->i_sb->s_dirt = 1;
3189 btrfs_update_inode_block_group(trans, inode);
3190 btrfs_update_inode_block_group(trans, dir);
3191out_unlock:
d3c2fdcf 3192 nr = trans->blocks_used;
89ce8a63 3193 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3194fail:
618e21d5
JB		 3195 if (drop_inode) {
3196 inode_dec_link_count(inode);
3197 iput(inode);
3198 }
d3c2fdcf 3199 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3200 return err;
3201}
3202
39279cc3
CM		 3203static int btrfs_create(struct inode *dir, struct dentry *dentry,
3204 int mode, struct nameidata *nd)
3205{
3206 struct btrfs_trans_handle *trans;
3207 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3208 struct inode *inode = NULL;
39279cc3
CM		 3209 int err;
3210 int drop_inode = 0;
1832a6d5 3211 unsigned long nr = 0;
39279cc3 3212 u64 objectid;
00e4e6b3 3213 u64 index = 0;
39279cc3 3214
1832a6d5
CM		 3215 err = btrfs_check_free_space(root, 1, 0);
3216 if (err)
3217 goto fail;
39279cc3
CM		 3218 trans = btrfs_start_transaction(root, 1);
3219 btrfs_set_trans_block_group(trans, dir);
3220
3221 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3222 if (err) {
3223 err = -ENOSPC;
3224 goto out_unlock;
3225 }
3226
aec7477b 3227 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3228 dentry->d_name.len,
3229 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3230 objectid, BTRFS_I(dir)->block_group, mode,
3231 &index);
39279cc3
CM		 3232 err = PTR_ERR(inode);
3233 if (IS_ERR(inode))
3234 goto out_unlock;
3235
33268eaf
JB		 3236 err = btrfs_init_acl(inode, dir);
3237 if (err) {
3238 drop_inode = 1;
3239 goto out_unlock;
3240 }
3241
39279cc3 3242 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3243 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3244 if (err)
3245 drop_inode = 1;
3246 else {
3247 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3248 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3249 inode->i_fop = &btrfs_file_operations;
3250 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3251 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3252 }
3253 dir->i_sb->s_dirt = 1;
3254 btrfs_update_inode_block_group(trans, inode);
3255 btrfs_update_inode_block_group(trans, dir);
3256out_unlock:
d3c2fdcf 3257 nr = trans->blocks_used;
ab78c84d 3258 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3259fail:
39279cc3
CM		 3260 if (drop_inode) {
3261 inode_dec_link_count(inode);
3262 iput(inode);
3263 }
d3c2fdcf 3264 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3265 return err;
3266}
3267
3268static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3269 struct dentry *dentry)
3270{
3271 struct btrfs_trans_handle *trans;
3272 struct btrfs_root *root = BTRFS_I(dir)->root;
3273 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3274 u64 index;
1832a6d5 3275 unsigned long nr = 0;
39279cc3
CM		 3276 int err;
3277 int drop_inode = 0;
3278
3279 if (inode->i_nlink == 0)
3280 return -ENOENT;
3281
e02119d5 3282 btrfs_inc_nlink(inode);
1832a6d5
CM		 3283 err = btrfs_check_free_space(root, 1, 0);
3284 if (err)
3285 goto fail;
00e4e6b3 3286 err = btrfs_set_inode_index(dir, inode, &index);
aec7477b
JB		 3287 if (err)
3288 goto fail;
3289
39279cc3 3290 trans = btrfs_start_transaction(root, 1);
5f39d397 3291
39279cc3
CM		 3292 btrfs_set_trans_block_group(trans, dir);
3293 atomic_inc(&inode->i_count);
aec7477b 3294
00e4e6b3 3295 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3296
39279cc3
CM		 3297 if (err)
3298 drop_inode = 1;
5f39d397 3299
39279cc3
CM		 3300 dir->i_sb->s_dirt = 1;
3301 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3302 err = btrfs_update_inode(trans, root, inode);
5f39d397 3303
54aa1f4d
CM		 3304 if (err)
3305 drop_inode = 1;
39279cc3 3306
d3c2fdcf 3307 nr = trans->blocks_used;
ab78c84d 3308 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3309fail:
39279cc3
CM		 3310 if (drop_inode) {
3311 inode_dec_link_count(inode);
3312 iput(inode);
3313 }
d3c2fdcf 3314 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3315 return err;
3316}
3317
39279cc3
CM		 3318static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3319{
b9d86667 3320 struct inode *inode = NULL;
39279cc3
CM		 3321 struct btrfs_trans_handle *trans;
3322 struct btrfs_root *root = BTRFS_I(dir)->root;
3323 int err = 0;
3324 int drop_on_err = 0;
b9d86667 3325 u64 objectid = 0;
00e4e6b3 3326 u64 index = 0;
d3c2fdcf 3327 unsigned long nr = 1;
39279cc3 3328
1832a6d5
CM		 3329 err = btrfs_check_free_space(root, 1, 0);
3330 if (err)
3331 goto out_unlock;
3332
39279cc3
CM		 3333 trans = btrfs_start_transaction(root, 1);
3334 btrfs_set_trans_block_group(trans, dir);
5f39d397 3335
39279cc3
CM		 3336 if (IS_ERR(trans)) {
3337 err = PTR_ERR(trans);
3338 goto out_unlock;
3339 }
3340
3341 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3342 if (err) {
3343 err = -ENOSPC;
3344 goto out_unlock;
3345 }
3346
aec7477b 3347 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3348 dentry->d_name.len,
3349 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3350 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3351 &index);
39279cc3
CM		 3352 if (IS_ERR(inode)) {
3353 err = PTR_ERR(inode);
3354 goto out_fail;
3355 }
5f39d397 3356
39279cc3 3357 drop_on_err = 1;
33268eaf
JB		 3358
3359 err = btrfs_init_acl(inode, dir);
3360 if (err)
3361 goto out_fail;
3362
39279cc3
CM		 3363 inode->i_op = &btrfs_dir_inode_operations;
3364 inode->i_fop = &btrfs_dir_file_operations;
3365 btrfs_set_trans_block_group(trans, inode);
3366
dbe674a9 3367 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3368 err = btrfs_update_inode(trans, root, inode);
3369 if (err)
3370 goto out_fail;
5f39d397 3371
e02119d5
CM		 3372 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3373 inode, dentry->d_name.name,
3374 dentry->d_name.len, 0, index);
39279cc3
CM		 3375 if (err)
3376 goto out_fail;
5f39d397 3377
39279cc3
CM		 3378 d_instantiate(dentry, inode);
3379 drop_on_err = 0;
3380 dir->i_sb->s_dirt = 1;
3381 btrfs_update_inode_block_group(trans, inode);
3382 btrfs_update_inode_block_group(trans, dir);
3383
3384out_fail:
d3c2fdcf 3385 nr = trans->blocks_used;
ab78c84d 3386 btrfs_end_transaction_throttle(trans, root);
5f39d397 3387
39279cc3 3388out_unlock:
39279cc3
CM		 3389 if (drop_on_err)
3390 iput(inode);
d3c2fdcf 3391 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3392 return err;
3393}
3394
d352ac68
CM		 3395/* helper for btfs_get_extent. Given an existing extent in the tree,
3396 * and an extent that you want to insert, deal with overlap and insert
3397 * the new extent into the tree.
3398 */
3b951516
CM		 3399static int merge_extent_mapping(struct extent_map_tree *em_tree,
3400 struct extent_map *existing,
e6dcd2dc
CM		 3401 struct extent_map *em,
3402 u64 map_start, u64 map_len)
3b951516
CM		 3403{
3404 u64 start_diff;
3b951516 3405
e6dcd2dc
CM		 3406 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3407 start_diff = map_start - em->start;
3408 em->start = map_start;
3409 em->len = map_len;
c8b97818
CM		 3410 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3411 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3412 em->block_start += start_diff;
c8b97818
CM		 3413 em->block_len -= start_diff;
3414 }
e6dcd2dc 3415 return add_extent_mapping(em_tree, em);
3b951516
CM		 3416}
3417
c8b97818
CM		 3418static noinline int uncompress_inline(struct btrfs_path *path,
3419 struct inode *inode, struct page *page,
3420 size_t pg_offset, u64 extent_offset,
3421 struct btrfs_file_extent_item *item)
3422{
3423 int ret;
3424 struct extent_buffer *leaf = path->nodes[0];
3425 char *tmp;
3426 size_t max_size;
3427 unsigned long inline_size;
3428 unsigned long ptr;
3429
3430 WARN_ON(pg_offset != 0);
3431 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3432 inline_size = btrfs_file_extent_inline_item_len(leaf,
3433 btrfs_item_nr(leaf, path->slots[0]));
3434 tmp = kmalloc(inline_size, GFP_NOFS);
3435 ptr = btrfs_file_extent_inline_start(item);
3436
3437 read_extent_buffer(leaf, tmp, ptr, inline_size);
3438
3439 max_size = min(PAGE_CACHE_SIZE, max_size);
3440 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3441 inline_size, max_size);
3442 if (ret) {
3443 char *kaddr = kmap_atomic(page, KM_USER0);
3444 unsigned long copy_size = min_t(u64,
3445 PAGE_CACHE_SIZE - pg_offset,
3446 max_size - extent_offset);
3447 memset(kaddr + pg_offset, 0, copy_size);
3448 kunmap_atomic(kaddr, KM_USER0);
3449 }
3450 kfree(tmp);
3451 return 0;
3452}
3453
d352ac68
CM		 3454/*
3455 * a bit scary, this does extent mapping from logical file offset to the disk.
3456 * the ugly parts come from merging extents from the disk with the
3457 * in-ram representation. This gets more complex because of the data=ordered code,
3458 * where the in-ram extents might be locked pending data=ordered completion.
3459 *
3460 * This also copies inline extents directly into the page.
3461 */
a52d9a80 3462struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3463 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3464 int create)
3465{
3466 int ret;
3467 int err = 0;
db94535d 3468 u64 bytenr;
a52d9a80
CM		 3469 u64 extent_start = 0;
3470 u64 extent_end = 0;
3471 u64 objectid = inode->i_ino;
3472 u32 found_type;
f421950f 3473 struct btrfs_path *path = NULL;
a52d9a80
CM		 3474 struct btrfs_root *root = BTRFS_I(inode)->root;
3475 struct btrfs_file_extent_item *item;
5f39d397
CM		 3476 struct extent_buffer *leaf;
3477 struct btrfs_key found_key;
a52d9a80
CM		 3478 struct extent_map *em = NULL;
3479 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3480 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3481 struct btrfs_trans_handle *trans = NULL;
c8b97818 3482 int compressed;
a52d9a80 3483
a52d9a80 3484again:
d1310b2e
CM		 3485 spin_lock(&em_tree->lock);
3486 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3487 if (em)
3488 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3489 spin_unlock(&em_tree->lock);
3490
a52d9a80 3491 if (em) {
e1c4b745
CM		 3492 if (em->start > start || em->start + em->len <= start)
3493 free_extent_map(em);
3494 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3495 free_extent_map(em);
3496 else
3497 goto out;
a52d9a80 3498 }
d1310b2e 3499 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3500 if (!em) {
d1310b2e
CM		 3501 err = -ENOMEM;
3502 goto out;
a52d9a80 3503 }
e6dcd2dc 3504 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3505 em->start = EXTENT_MAP_HOLE;
3506 em->len = (u64)-1;
c8b97818 3507 em->block_len = (u64)-1;
f421950f
CM		 3508
3509 if (!path) {
3510 path = btrfs_alloc_path();
3511 BUG_ON(!path);
3512 }
3513
179e29e4
CM		 3514 ret = btrfs_lookup_file_extent(trans, root, path,
3515 objectid, start, trans != NULL);
a52d9a80
CM		 3516 if (ret < 0) {
3517 err = ret;
3518 goto out;
3519 }
3520
3521 if (ret != 0) {
3522 if (path->slots[0] == 0)
3523 goto not_found;
3524 path->slots[0]--;
3525 }
3526
5f39d397
CM		 3527 leaf = path->nodes[0];
3528 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3529 struct btrfs_file_extent_item);
a52d9a80 3530 /* are we inside the extent that was found? */
5f39d397
CM		 3531 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3532 found_type = btrfs_key_type(&found_key);
3533 if (found_key.objectid != objectid ||
a52d9a80
CM		 3534 found_type != BTRFS_EXTENT_DATA_KEY) {
3535 goto not_found;
3536 }
3537
5f39d397
CM		 3538 found_type = btrfs_file_extent_type(leaf, item);
3539 extent_start = found_key.offset;
c8b97818 3540 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3541 if (found_type == BTRFS_FILE_EXTENT_REG ||
3542 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3543 extent_end = extent_start +
db94535d 3544 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3545 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3546 size_t size;
3547 size = btrfs_file_extent_inline_len(leaf, item);
3548 extent_end = (extent_start + size + root->sectorsize - 1) &
3549 ~((u64)root->sectorsize - 1);
3550 }
3551
3552 if (start >= extent_end) {
3553 path->slots[0]++;
3554 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3555 ret = btrfs_next_leaf(root, path);
3556 if (ret < 0) {
3557 err = ret;
3558 goto out;
a52d9a80 3559 }
9036c102
YZ		 3560 if (ret > 0)
3561 goto not_found;
3562 leaf = path->nodes[0];
a52d9a80 3563 }
9036c102
YZ		 3564 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3565 if (found_key.objectid != objectid ||
3566 found_key.type != BTRFS_EXTENT_DATA_KEY)
3567 goto not_found;
3568 if (start + len <= found_key.offset)
3569 goto not_found;
3570 em->start = start;
3571 em->len = found_key.offset - start;
3572 goto not_found_em;
3573 }
3574
d899e052
YZ		 3575 if (found_type == BTRFS_FILE_EXTENT_REG ||
3576 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 3577 em->start = extent_start;
3578 em->len = extent_end - extent_start;
db94535d
CM		 3579 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3580 if (bytenr == 0) {
5f39d397 3581 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 3582 goto insert;
3583 }
c8b97818
CM		 3584 if (compressed) {
3585 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3586 em->block_start = bytenr;
3587 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3588 item);
3589 } else {
3590 bytenr += btrfs_file_extent_offset(leaf, item);
3591 em->block_start = bytenr;
3592 em->block_len = em->len;
d899e052
YZ		 3593 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3594 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 3595 }
a52d9a80
CM		 3596 goto insert;
3597 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3598 unsigned long ptr;
a52d9a80 3599 char *map;
3326d1b0
CM		 3600 size_t size;
3601 size_t extent_offset;
3602 size_t copy_size;
a52d9a80 3603
689f9346 3604 em->block_start = EXTENT_MAP_INLINE;
c8b97818 3605 if (!page || create) {
689f9346 3606 em->start = extent_start;
9036c102 3607 em->len = extent_end - extent_start;
689f9346
Y		 3608 goto out;
3609 }
5f39d397 3610
9036c102
YZ		 3611 size = btrfs_file_extent_inline_len(leaf, item);
3612 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 3613 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 3614 size - extent_offset);
3326d1b0 3615 em->start = extent_start + extent_offset;
70dec807
CM		 3616 em->len = (copy_size + root->sectorsize - 1) &
3617 ~((u64)root->sectorsize - 1);
c8b97818
CM		 3618 if (compressed)
3619 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 3620 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 3621 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 3622 if (btrfs_file_extent_compression(leaf, item) ==
3623 BTRFS_COMPRESS_ZLIB) {
3624 ret = uncompress_inline(path, inode, page,
3625 pg_offset,
3626 extent_offset, item);
3627 BUG_ON(ret);
3628 } else {
3629 map = kmap(page);
3630 read_extent_buffer(leaf, map + pg_offset, ptr,
3631 copy_size);
3632 kunmap(page);
3633 }
179e29e4
CM		 3634 flush_dcache_page(page);
3635 } else if (create && PageUptodate(page)) {
3636 if (!trans) {
3637 kunmap(page);
3638 free_extent_map(em);
3639 em = NULL;
3640 btrfs_release_path(root, path);
f9295749 3641 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 3642 goto again;
3643 }
c8b97818 3644 map = kmap(page);
70dec807 3645 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 3646 copy_size);
c8b97818 3647 kunmap(page);
179e29e4 3648 btrfs_mark_buffer_dirty(leaf);
a52d9a80 3649 }
d1310b2e
CM		 3650 set_extent_uptodate(io_tree, em->start,
3651 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 3652 goto insert;
3653 } else {
3654 printk("unkknown found_type %d\n", found_type);
3655 WARN_ON(1);
3656 }
3657not_found:
3658 em->start = start;
d1310b2e 3659 em->len = len;
a52d9a80 3660not_found_em:
5f39d397 3661 em->block_start = EXTENT_MAP_HOLE;
9036c102 3662 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 3663insert:
3664 btrfs_release_path(root, path);
d1310b2e
CM		 3665 if (em->start > start || extent_map_end(em) <= start) {
3666 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM		 3667 err = -EIO;
3668 goto out;
3669 }
d1310b2e
CM		 3670
3671 err = 0;
3672 spin_lock(&em_tree->lock);
a52d9a80 3673 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 3674 /* it is possible that someone inserted the extent into the tree
3675 * while we had the lock dropped. It is also possible that
3676 * an overlapping map exists in the tree
3677 */
a52d9a80 3678 if (ret == -EEXIST) {
3b951516 3679 struct extent_map *existing;
e6dcd2dc
CM		 3680
3681 ret = 0;
3682
3b951516 3683 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 3684 if (existing && (existing->start > start ||
3685 existing->start + existing->len <= start)) {
3686 free_extent_map(existing);
3687 existing = NULL;
3688 }
3b951516
CM		 3689 if (!existing) {
3690 existing = lookup_extent_mapping(em_tree, em->start,
3691 em->len);
3692 if (existing) {
3693 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 3694 em, start,
3695 root->sectorsize);
3b951516
CM		 3696 free_extent_map(existing);
3697 if (err) {
3698 free_extent_map(em);
3699 em = NULL;
3700 }
3701 } else {
3702 err = -EIO;
3703 printk("failing to insert %Lu %Lu\n",
3704 start, len);
3705 free_extent_map(em);
3706 em = NULL;
3707 }
3708 } else {
3709 free_extent_map(em);
3710 em = existing;
e6dcd2dc 3711 err = 0;
a52d9a80 3712 }
a52d9a80 3713 }
d1310b2e 3714 spin_unlock(&em_tree->lock);
a52d9a80 3715out:
f421950f
CM		 3716 if (path)
3717 btrfs_free_path(path);
a52d9a80
CM		 3718 if (trans) {
3719 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 3720 if (!err) {
a52d9a80 3721 err = ret;
e6dcd2dc 3722 }
a52d9a80 3723 }
a52d9a80
CM		 3724 if (err) {
3725 free_extent_map(em);
3726 WARN_ON(1);
3727 return ERR_PTR(err);
3728 }
3729 return em;
3730}
3731
16432985
CM		 3732static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
3733 const struct iovec *iov, loff_t offset,
3734 unsigned long nr_segs)
3735{
e1c4b745 3736 return -EINVAL;
16432985
CM		 3737}
3738
d396c6f5 3739static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 3740{
d396c6f5 3741 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 3742}
3743
a52d9a80 3744int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 3745{
d1310b2e
CM		 3746 struct extent_io_tree *tree;
3747 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 3748 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 3749}
1832a6d5 3750
a52d9a80 3751static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 3752{
d1310b2e 3753 struct extent_io_tree *tree;
b888db2b
CM		 3754
3755
3756 if (current->flags & PF_MEMALLOC) {
3757 redirty_page_for_writepage(wbc, page);
3758 unlock_page(page);
3759 return 0;
3760 }
d1310b2e 3761 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 3762 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 3763}
3764
f421950f
CM		 3765int btrfs_writepages(struct address_space *mapping,
3766 struct writeback_control *wbc)
b293f02e 3767{
d1310b2e
CM		 3768 struct extent_io_tree *tree;
3769 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 3770 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3771}
3772
3ab2fb5a
CM		 3773static int
3774btrfs_readpages(struct file *file, struct address_space *mapping,
3775 struct list_head *pages, unsigned nr_pages)
3776{
d1310b2e
CM		 3777 struct extent_io_tree *tree;
3778 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 3779 return extent_readpages(tree, mapping, pages, nr_pages,
3780 btrfs_get_extent);
3781}
e6dcd2dc 3782static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 3783{
d1310b2e
CM		 3784 struct extent_io_tree *tree;
3785 struct extent_map_tree *map;
a52d9a80 3786 int ret;
8c2383c3 3787
d1310b2e
CM		 3788 tree = &BTRFS_I(page->mapping->host)->io_tree;
3789 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 3790 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 3791 if (ret == 1) {
3792 ClearPagePrivate(page);
3793 set_page_private(page, 0);
3794 page_cache_release(page);
39279cc3 3795 }
a52d9a80 3796 return ret;
39279cc3
CM		 3797}
3798
e6dcd2dc
CM		 3799static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3800{
98509cfc
CM		 3801 if (PageWriteback(page) || PageDirty(page))
3802 return 0;
e6dcd2dc
CM		 3803 return __btrfs_releasepage(page, gfp_flags);
3804}
3805
a52d9a80 3806static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 3807{
d1310b2e 3808 struct extent_io_tree *tree;
e6dcd2dc
CM		 3809 struct btrfs_ordered_extent *ordered;
3810 u64 page_start = page_offset(page);
3811 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 3812
e6dcd2dc 3813 wait_on_page_writeback(page);
d1310b2e 3814 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 3815 if (offset) {
3816 btrfs_releasepage(page, GFP_NOFS);
3817 return;
3818 }
3819
3820 lock_extent(tree, page_start, page_end, GFP_NOFS);
3821 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3822 page_offset(page));
3823 if (ordered) {
eb84ae03
CM		 3824 /*
3825 * IO on this page will never be started, so we need
3826 * to account for any ordered extents now
3827 */
e6dcd2dc
CM		 3828 clear_extent_bit(tree, page_start, page_end,
3829 EXTENT_DIRTY | EXTENT_DELALLOC |
3830 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 3831 btrfs_finish_ordered_io(page->mapping->host,
3832 page_start, page_end);
e6dcd2dc
CM		 3833 btrfs_put_ordered_extent(ordered);
3834 lock_extent(tree, page_start, page_end, GFP_NOFS);
3835 }
3836 clear_extent_bit(tree, page_start, page_end,
3837 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3838 EXTENT_ORDERED,
3839 1, 1, GFP_NOFS);
3840 __btrfs_releasepage(page, GFP_NOFS);
3841
4a096752 3842 ClearPageChecked(page);
9ad6b7bc 3843 if (PagePrivate(page)) {
9ad6b7bc
CM		 3844 ClearPagePrivate(page);
3845 set_page_private(page, 0);
3846 page_cache_release(page);
3847 }
39279cc3
CM		 3848}
3849
9ebefb18
CM		 3850/*
3851 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3852 * called from a page fault handler when a page is first dirtied. Hence we must
3853 * be careful to check for EOF conditions here. We set the page up correctly
3854 * for a written page which means we get ENOSPC checking when writing into
3855 * holes and correct delalloc and unwritten extent mapping on filesystems that
3856 * support these features.
3857 *
3858 * We are not allowed to take the i_mutex here so we have to play games to
3859 * protect against truncate races as the page could now be beyond EOF. Because
3860 * vmtruncate() writes the inode size before removing pages, once we have the
3861 * page lock we can determine safely if the page is beyond EOF. If it is not
3862 * beyond EOF, then the page is guaranteed safe against truncation until we
3863 * unlock the page.
3864 */
3865int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3866{
6da6abae 3867 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 3868 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3869 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3870 struct btrfs_ordered_extent *ordered;
3871 char *kaddr;
3872 unsigned long zero_start;
9ebefb18 3873 loff_t size;
1832a6d5 3874 int ret;
a52d9a80 3875 u64 page_start;
e6dcd2dc 3876 u64 page_end;
9ebefb18 3877
1832a6d5 3878 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 3879 if (ret)
3880 goto out;
3881
3882 ret = -EINVAL;
e6dcd2dc 3883again:
9ebefb18 3884 lock_page(page);
9ebefb18 3885 size = i_size_read(inode);
e6dcd2dc
CM		 3886 page_start = page_offset(page);
3887 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 3888
9ebefb18 3889 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 3890 (page_start >= size)) {
9ebefb18
CM		 3891 /* page got truncated out from underneath us */
3892 goto out_unlock;
3893 }
e6dcd2dc
CM		 3894 wait_on_page_writeback(page);
3895
3896 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3897 set_page_extent_mapped(page);
3898
eb84ae03
CM		 3899 /*
3900 * we can't set the delalloc bits if there are pending ordered
3901 * extents. Drop our locks and wait for them to finish
3902 */
e6dcd2dc
CM		 3903 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3904 if (ordered) {
3905 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3906 unlock_page(page);
eb84ae03 3907 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3908 btrfs_put_ordered_extent(ordered);
3909 goto again;
3910 }
3911
ea8c2819 3912 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 3913 ret = 0;
9ebefb18
CM		 3914
3915 /* page is wholly or partially inside EOF */
a52d9a80 3916 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 3917 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 3918 else
e6dcd2dc 3919 zero_start = PAGE_CACHE_SIZE;
9ebefb18 3920
e6dcd2dc
CM		 3921 if (zero_start != PAGE_CACHE_SIZE) {
3922 kaddr = kmap(page);
3923 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3924 flush_dcache_page(page);
3925 kunmap(page);
3926 }
247e743c 3927 ClearPageChecked(page);
e6dcd2dc
CM		 3928 set_page_dirty(page);
3929 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 3930
3931out_unlock:
3932 unlock_page(page);
1832a6d5 3933out:
9ebefb18
CM		 3934 return ret;
3935}
3936
39279cc3
CM		 3937static void btrfs_truncate(struct inode *inode)
3938{
3939 struct btrfs_root *root = BTRFS_I(inode)->root;
3940 int ret;
3941 struct btrfs_trans_handle *trans;
d3c2fdcf 3942 unsigned long nr;
dbe674a9 3943 u64 mask = root->sectorsize - 1;
39279cc3
CM		 3944
3945 if (!S_ISREG(inode->i_mode))
3946 return;
3947 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3948 return;
3949
3950 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 3951 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 3952
39279cc3
CM		 3953 trans = btrfs_start_transaction(root, 1);
3954 btrfs_set_trans_block_group(trans, inode);
dbe674a9 3955 btrfs_i_size_write(inode, inode->i_size);
39279cc3 3956
7b128766
JB		 3957 ret = btrfs_orphan_add(trans, inode);
3958 if (ret)
3959 goto out;
39279cc3 3960 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 3961 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 3962 BTRFS_EXTENT_DATA_KEY);
39279cc3 3963 btrfs_update_inode(trans, root, inode);
5f39d397 3964
7b128766
JB		 3965 ret = btrfs_orphan_del(trans, inode);
3966 BUG_ON(ret);
3967
3968out:
3969 nr = trans->blocks_used;
89ce8a63 3970 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 3971 BUG_ON(ret);
d3c2fdcf 3972 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3973}
3974
3b96362c
SW		 3975/*
3976 * Invalidate a single dcache entry at the root of the filesystem.
3977 * Needed after creation of snapshot or subvolume.
3978 */
3979void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3980 int namelen)
3981{
3982 struct dentry *alias, *entry;
3983 struct qstr qstr;
3984
3985 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3986 if (alias) {
3987 qstr.name = name;
3988 qstr.len = namelen;
3989 /* change me if btrfs ever gets a d_hash operation */
3990 qstr.hash = full_name_hash(qstr.name, qstr.len);
3991 entry = d_lookup(alias, &qstr);
3992 dput(alias);
3993 if (entry) {
3994 d_invalidate(entry);
3995 dput(entry);
3996 }
3997 }
3998}
3999
d352ac68
CM		 4000/*
4001 * create a new subvolume directory/inode (helper for the ioctl).
4002 */
cb8e7090 4003int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
f46b5a66
CH		 4004 struct btrfs_trans_handle *trans, u64 new_dirid,
4005 struct btrfs_block_group_cache *block_group)
39279cc3 4006{
39279cc3 4007 struct inode *inode;
cb8e7090 4008 int error;
00e4e6b3 4009 u64 index = 0;
39279cc3 4010
aec7477b 4011 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
00e4e6b3 4012 new_dirid, block_group, S_IFDIR | 0700, &index);
54aa1f4d 4013 if (IS_ERR(inode))
f46b5a66 4014 return PTR_ERR(inode);
39279cc3
CM		 4015 inode->i_op = &btrfs_dir_inode_operations;
4016 inode->i_fop = &btrfs_dir_file_operations;
34088780 4017 new_root->inode = inode;
39279cc3 4018
39279cc3 4019 inode->i_nlink = 1;
dbe674a9 4020 btrfs_i_size_write(inode, 0);
3b96362c 4021
cb8e7090
CH		 4022 error = btrfs_update_inode(trans, new_root, inode);
4023 if (error)
4024 return error;
4025
d899e052 4026 atomic_inc(&inode->i_count);
cb8e7090
CH		 4027 d_instantiate(dentry, inode);
4028 return 0;
39279cc3
CM		 4029}
4030
d352ac68
CM		 4031/* helper function for file defrag and space balancing. This
4032 * forces readahead on a given range of bytes in an inode
4033 */
edbd8d4e 4034unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4035 struct file_ra_state *ra, struct file *file,
4036 pgoff_t offset, pgoff_t last_index)
4037{
8e7bf94f 4038 pgoff_t req_size = last_index - offset + 1;
86479a04 4039
86479a04
CM		 4040 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4041 return offset + req_size;
86479a04
CM		 4042}
4043
39279cc3
CM		 4044struct inode *btrfs_alloc_inode(struct super_block *sb)
4045{
4046 struct btrfs_inode *ei;
4047
4048 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4049 if (!ei)
4050 return NULL;
15ee9bc7 4051 ei->last_trans = 0;
e02119d5 4052 ei->logged_trans = 0;
e6dcd2dc 4053 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4054 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4055 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4056 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4057 return &ei->vfs_inode;
4058}
4059
4060void btrfs_destroy_inode(struct inode *inode)
4061{
e6dcd2dc 4062 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4063 WARN_ON(!list_empty(&inode->i_dentry));
4064 WARN_ON(inode->i_data.nrpages);
4065
33268eaf
JB		 4066 if (BTRFS_I(inode)->i_acl &&
4067 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4068 posix_acl_release(BTRFS_I(inode)->i_acl);
4069 if (BTRFS_I(inode)->i_default_acl &&
4070 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4071 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4072
bcc63abb 4073 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4074 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4075 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4076 " list\n", inode->i_ino);
4077 dump_stack();
4078 }
bcc63abb 4079 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4080
e6dcd2dc
CM		 4081 while(1) {
4082 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4083 if (!ordered)
4084 break;
4085 else {
4086 printk("found ordered extent %Lu %Lu\n",
4087 ordered->file_offset, ordered->len);
4088 btrfs_remove_ordered_extent(inode, ordered);
4089 btrfs_put_ordered_extent(ordered);
4090 btrfs_put_ordered_extent(ordered);
4091 }
4092 }
5b21f2ed 4093 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4094 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4095}
4096
0ee0fda0 4097static void init_once(void *foo)
39279cc3
CM		 4098{
4099 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4100
4101 inode_init_once(&ei->vfs_inode);
4102}
4103
4104void btrfs_destroy_cachep(void)
4105{
4106 if (btrfs_inode_cachep)
4107 kmem_cache_destroy(btrfs_inode_cachep);
4108 if (btrfs_trans_handle_cachep)
4109 kmem_cache_destroy(btrfs_trans_handle_cachep);
4110 if (btrfs_transaction_cachep)
4111 kmem_cache_destroy(btrfs_transaction_cachep);
4112 if (btrfs_bit_radix_cachep)
4113 kmem_cache_destroy(btrfs_bit_radix_cachep);
4114 if (btrfs_path_cachep)
4115 kmem_cache_destroy(btrfs_path_cachep);
4116}
4117
86479a04 4118struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4119 unsigned long extra_flags,
2b1f55b0 4120 void (*ctor)(void *))
92fee66d
CM		 4121{
4122 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4123 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4124}
4125
39279cc3
CM		 4126int btrfs_init_cachep(void)
4127{
86479a04 4128 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4129 sizeof(struct btrfs_inode),
4130 0, init_once);
39279cc3
CM		 4131 if (!btrfs_inode_cachep)
4132 goto fail;
86479a04
CM		 4133 btrfs_trans_handle_cachep =
4134 btrfs_cache_create("btrfs_trans_handle_cache",
4135 sizeof(struct btrfs_trans_handle),
4136 0, NULL);
39279cc3
CM		 4137 if (!btrfs_trans_handle_cachep)
4138 goto fail;
86479a04 4139 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4140 sizeof(struct btrfs_transaction),
92fee66d 4141 0, NULL);
39279cc3
CM		 4142 if (!btrfs_transaction_cachep)
4143 goto fail;
86479a04 4144 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4145 sizeof(struct btrfs_path),
92fee66d 4146 0, NULL);
39279cc3
CM		 4147 if (!btrfs_path_cachep)
4148 goto fail;
86479a04 4149 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4150 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4151 if (!btrfs_bit_radix_cachep)
4152 goto fail;
4153 return 0;
4154fail:
4155 btrfs_destroy_cachep();
4156 return -ENOMEM;
4157}
4158
4159static int btrfs_getattr(struct vfsmount *mnt,
4160 struct dentry *dentry, struct kstat *stat)
4161{
4162 struct inode *inode = dentry->d_inode;
4163 generic_fillattr(inode, stat);
d6667462 4164 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4165 stat->blocks = (inode_get_bytes(inode) +
4166 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4167 return 0;
4168}
4169
4170static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4171 struct inode * new_dir,struct dentry *new_dentry)
4172{
4173 struct btrfs_trans_handle *trans;
4174 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4175 struct inode *new_inode = new_dentry->d_inode;
4176 struct inode *old_inode = old_dentry->d_inode;
4177 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4178 u64 index = 0;
39279cc3
CM		 4179 int ret;
4180
4181 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4182 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4183 return -ENOTEMPTY;
4184 }
5f39d397 4185
1832a6d5
CM		 4186 ret = btrfs_check_free_space(root, 1, 0);
4187 if (ret)
4188 goto out_unlock;
4189
39279cc3 4190 trans = btrfs_start_transaction(root, 1);
5f39d397 4191
39279cc3 4192 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4193
e02119d5 4194 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4195 old_dir->i_ctime = old_dir->i_mtime = ctime;
4196 new_dir->i_ctime = new_dir->i_mtime = ctime;
4197 old_inode->i_ctime = ctime;
5f39d397 4198
e02119d5
CM		 4199 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4200 old_dentry->d_name.name,
4201 old_dentry->d_name.len);
39279cc3
CM		 4202 if (ret)
4203 goto out_fail;
4204
4205 if (new_inode) {
4206 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4207 ret = btrfs_unlink_inode(trans, root, new_dir,
4208 new_dentry->d_inode,
4209 new_dentry->d_name.name,
4210 new_dentry->d_name.len);
39279cc3
CM		 4211 if (ret)
4212 goto out_fail;
7b128766 4213 if (new_inode->i_nlink == 0) {
e02119d5 4214 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4215 if (ret)
4216 goto out_fail;
4217 }
e02119d5 4218
39279cc3 4219 }
00e4e6b3 4220 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
aec7477b
JB		 4221 if (ret)
4222 goto out_fail;
4223
e02119d5
CM		 4224 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4225 old_inode, new_dentry->d_name.name,
4226 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4227 if (ret)
4228 goto out_fail;
4229
4230out_fail:
ab78c84d 4231 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4232out_unlock:
39279cc3
CM		 4233 return ret;
4234}
4235
d352ac68
CM		 4236/*
4237 * some fairly slow code that needs optimization. This walks the list
4238 * of all the inodes with pending delalloc and forces them to disk.
4239 */
ea8c2819
CM		 4240int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4241{
4242 struct list_head *head = &root->fs_info->delalloc_inodes;
4243 struct btrfs_inode *binode;
5b21f2ed 4244 struct inode *inode;
ea8c2819
CM		 4245 unsigned long flags;
4246
4247 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4248 while(!list_empty(head)) {
4249 binode = list_entry(head->next, struct btrfs_inode,
4250 delalloc_inodes);
5b21f2ed
ZY		 4251 inode = igrab(&binode->vfs_inode);
4252 if (!inode)
4253 list_del_init(&binode->delalloc_inodes);
ea8c2819 4254 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
5b21f2ed 4255 if (inode) {
8c8bee1d 4256 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4257 iput(inode);
4258 }
4259 cond_resched();
ea8c2819
CM		 4260 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4261 }
4262 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
8c8bee1d
CM		 4263
4264 /* the filemap_flush will queue IO into the worker threads, but
4265 * we have to make sure the IO is actually started and that
4266 * ordered extents get created before we return
4267 */
4268 atomic_inc(&root->fs_info->async_submit_draining);
4269 while(atomic_read(&root->fs_info->nr_async_submits)) {
4270 wait_event(root->fs_info->async_submit_wait,
4271 (atomic_read(&root->fs_info->nr_async_submits) == 0));
4272 }
4273 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4274 return 0;
4275}
4276
39279cc3
CM		 4277static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4278 const char *symname)
4279{
4280 struct btrfs_trans_handle *trans;
4281 struct btrfs_root *root = BTRFS_I(dir)->root;
4282 struct btrfs_path *path;
4283 struct btrfs_key key;
1832a6d5 4284 struct inode *inode = NULL;
39279cc3
CM		 4285 int err;
4286 int drop_inode = 0;
4287 u64 objectid;
00e4e6b3 4288 u64 index = 0 ;
39279cc3
CM		 4289 int name_len;
4290 int datasize;
5f39d397 4291 unsigned long ptr;
39279cc3 4292 struct btrfs_file_extent_item *ei;
5f39d397 4293 struct extent_buffer *leaf;
1832a6d5 4294 unsigned long nr = 0;
39279cc3
CM		 4295
4296 name_len = strlen(symname) + 1;
4297 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4298 return -ENAMETOOLONG;
1832a6d5 4299
1832a6d5
CM		 4300 err = btrfs_check_free_space(root, 1, 0);
4301 if (err)
4302 goto out_fail;
4303
39279cc3
CM		 4304 trans = btrfs_start_transaction(root, 1);
4305 btrfs_set_trans_block_group(trans, dir);
4306
4307 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4308 if (err) {
4309 err = -ENOSPC;
4310 goto out_unlock;
4311 }
4312
aec7477b 4313 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4314 dentry->d_name.len,
4315 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4316 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4317 &index);
39279cc3
CM		 4318 err = PTR_ERR(inode);
4319 if (IS_ERR(inode))
4320 goto out_unlock;
4321
33268eaf
JB		 4322 err = btrfs_init_acl(inode, dir);
4323 if (err) {
4324 drop_inode = 1;
4325 goto out_unlock;
4326 }
4327
39279cc3 4328 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4329 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4330 if (err)
4331 drop_inode = 1;
4332 else {
4333 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4334 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4335 inode->i_fop = &btrfs_file_operations;
4336 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4337 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4338 }
4339 dir->i_sb->s_dirt = 1;
4340 btrfs_update_inode_block_group(trans, inode);
4341 btrfs_update_inode_block_group(trans, dir);
4342 if (drop_inode)
4343 goto out_unlock;
4344
4345 path = btrfs_alloc_path();
4346 BUG_ON(!path);
4347 key.objectid = inode->i_ino;
4348 key.offset = 0;
39279cc3
CM		 4349 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4350 datasize = btrfs_file_extent_calc_inline_size(name_len);
4351 err = btrfs_insert_empty_item(trans, root, path, &key,
4352 datasize);
54aa1f4d
CM		 4353 if (err) {
4354 drop_inode = 1;
4355 goto out_unlock;
4356 }
5f39d397
CM		 4357 leaf = path->nodes[0];
4358 ei = btrfs_item_ptr(leaf, path->slots[0],
4359 struct btrfs_file_extent_item);
4360 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4361 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4362 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4363 btrfs_set_file_extent_encryption(leaf, ei, 0);
4364 btrfs_set_file_extent_compression(leaf, ei, 0);
4365 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4366 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4367
39279cc3 4368 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4369 write_extent_buffer(leaf, symname, ptr, name_len);
4370 btrfs_mark_buffer_dirty(leaf);
39279cc3 4371 btrfs_free_path(path);
5f39d397 4372
39279cc3
CM		 4373 inode->i_op = &btrfs_symlink_inode_operations;
4374 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4375 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4376 inode_set_bytes(inode, name_len);
dbe674a9 4377 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4378 err = btrfs_update_inode(trans, root, inode);
4379 if (err)
4380 drop_inode = 1;
39279cc3
CM		 4381
4382out_unlock:
d3c2fdcf 4383 nr = trans->blocks_used;
ab78c84d 4384 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4385out_fail:
39279cc3
CM		 4386 if (drop_inode) {
4387 inode_dec_link_count(inode);
4388 iput(inode);
4389 }
d3c2fdcf 4390 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4391 return err;
4392}
16432985 4393
d899e052
YZ		 4394static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4395 u64 alloc_hint, int mode)
4396{
4397 struct btrfs_trans_handle *trans;
4398 struct btrfs_root *root = BTRFS_I(inode)->root;
4399 struct btrfs_key ins;
4400 u64 alloc_size;
4401 u64 cur_offset = start;
4402 u64 num_bytes = end - start;
4403 int ret = 0;
4404
4405 trans = btrfs_join_transaction(root, 1);
4406 BUG_ON(!trans);
4407 btrfs_set_trans_block_group(trans, inode);
4408
4409 while (num_bytes > 0) {
4410 alloc_size = min(num_bytes, root->fs_info->max_extent);
4411 ret = btrfs_reserve_extent(trans, root, alloc_size,
4412 root->sectorsize, 0, alloc_hint,
4413 (u64)-1, &ins, 1);
4414 if (ret) {
4415 WARN_ON(1);
4416 goto out;
4417 }
4418 ret = insert_reserved_file_extent(trans, inode,
4419 cur_offset, ins.objectid,
4420 ins.offset, ins.offset,
4421 ins.offset, 0, 0, 0,
4422 BTRFS_FILE_EXTENT_PREALLOC);
4423 BUG_ON(ret);
4424 num_bytes -= ins.offset;
4425 cur_offset += ins.offset;
4426 alloc_hint = ins.objectid + ins.offset;
4427 }
4428out:
4429 if (cur_offset > start) {
4430 inode->i_ctime = CURRENT_TIME;
4431 btrfs_set_flag(inode, PREALLOC);
4432 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4433 cur_offset > i_size_read(inode))
4434 btrfs_i_size_write(inode, cur_offset);
4435 ret = btrfs_update_inode(trans, root, inode);
4436 BUG_ON(ret);
4437 }
4438
4439 btrfs_end_transaction(trans, root);
4440 return ret;
4441}
4442
4443static long btrfs_fallocate(struct inode *inode, int mode,
4444 loff_t offset, loff_t len)
4445{
4446 u64 cur_offset;
4447 u64 last_byte;
4448 u64 alloc_start;
4449 u64 alloc_end;
4450 u64 alloc_hint = 0;
4451 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4452 struct extent_map *em;
4453 int ret;
4454
4455 alloc_start = offset & ~mask;
4456 alloc_end = (offset + len + mask) & ~mask;
4457
4458 mutex_lock(&inode->i_mutex);
4459 if (alloc_start > inode->i_size) {
4460 ret = btrfs_cont_expand(inode, alloc_start);
4461 if (ret)
4462 goto out;
4463 }
4464
4465 while (1) {
4466 struct btrfs_ordered_extent *ordered;
4467 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4468 alloc_end - 1, GFP_NOFS);
4469 ordered = btrfs_lookup_first_ordered_extent(inode,
4470 alloc_end - 1);
4471 if (ordered &&
4472 ordered->file_offset + ordered->len > alloc_start &&
4473 ordered->file_offset < alloc_end) {
4474 btrfs_put_ordered_extent(ordered);
4475 unlock_extent(&BTRFS_I(inode)->io_tree,
4476 alloc_start, alloc_end - 1, GFP_NOFS);
4477 btrfs_wait_ordered_range(inode, alloc_start,
4478 alloc_end - alloc_start);
4479 } else {
4480 if (ordered)
4481 btrfs_put_ordered_extent(ordered);
4482 break;
4483 }
4484 }
4485
4486 cur_offset = alloc_start;
4487 while (1) {
4488 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4489 alloc_end - cur_offset, 0);
4490 BUG_ON(IS_ERR(em) || !em);
4491 last_byte = min(extent_map_end(em), alloc_end);
4492 last_byte = (last_byte + mask) & ~mask;
4493 if (em->block_start == EXTENT_MAP_HOLE) {
4494 ret = prealloc_file_range(inode, cur_offset,
4495 last_byte, alloc_hint, mode);
4496 if (ret < 0) {
4497 free_extent_map(em);
4498 break;
4499 }
4500 }
4501 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4502 alloc_hint = em->block_start;
4503 free_extent_map(em);
4504
4505 cur_offset = last_byte;
4506 if (cur_offset >= alloc_end) {
4507 ret = 0;
4508 break;
4509 }
4510 }
4511 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4512 GFP_NOFS);
4513out:
4514 mutex_unlock(&inode->i_mutex);
4515 return ret;
4516}
4517
e6dcd2dc
CM		 4518static int btrfs_set_page_dirty(struct page *page)
4519{
e6dcd2dc
CM		 4520 return __set_page_dirty_nobuffers(page);
4521}
4522
0ee0fda0 4523static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4524{
4525 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4526 return -EACCES;
33268eaf 4527 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4528}
39279cc3
CM		 4529
4530static struct inode_operations btrfs_dir_inode_operations = {
4531 .lookup = btrfs_lookup,
4532 .create = btrfs_create,
4533 .unlink = btrfs_unlink,
4534 .link = btrfs_link,
4535 .mkdir = btrfs_mkdir,
4536 .rmdir = btrfs_rmdir,
4537 .rename = btrfs_rename,
4538 .symlink = btrfs_symlink,
4539 .setattr = btrfs_setattr,
618e21d5 4540 .mknod = btrfs_mknod,
95819c05
CH		 4541 .setxattr = btrfs_setxattr,
4542 .getxattr = btrfs_getxattr,
5103e947 4543 .listxattr = btrfs_listxattr,
95819c05 4544 .removexattr = btrfs_removexattr,
fdebe2bd 4545 .permission = btrfs_permission,
39279cc3 4546};
39279cc3
CM		 4547static struct inode_operations btrfs_dir_ro_inode_operations = {
4548 .lookup = btrfs_lookup,
fdebe2bd 4549 .permission = btrfs_permission,
39279cc3 4550};
39279cc3
CM		 4551static struct file_operations btrfs_dir_file_operations = {
4552 .llseek = generic_file_llseek,
4553 .read = generic_read_dir,
cbdf5a24 4554 .readdir = btrfs_real_readdir,
34287aa3 4555 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4556#ifdef CONFIG_COMPAT
34287aa3 4557 .compat_ioctl = btrfs_ioctl,
39279cc3 4558#endif
6bf13c0c 4559 .release = btrfs_release_file,
e02119d5 4560 .fsync = btrfs_sync_file,
39279cc3
CM		 4561};
4562
d1310b2e 4563static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4564 .fill_delalloc = run_delalloc_range,
065631f6 4565 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4566 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4567 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4568 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4569 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4570 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4571 .set_bit_hook = btrfs_set_bit_hook,
4572 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4573};
4574
39279cc3
CM		 4575static struct address_space_operations btrfs_aops = {
4576 .readpage = btrfs_readpage,
4577 .writepage = btrfs_writepage,
b293f02e 4578 .writepages = btrfs_writepages,
3ab2fb5a 4579 .readpages = btrfs_readpages,
39279cc3 4580 .sync_page = block_sync_page,
39279cc3 4581 .bmap = btrfs_bmap,
16432985 4582 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4583 .invalidatepage = btrfs_invalidatepage,
4584 .releasepage = btrfs_releasepage,
e6dcd2dc 4585 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 4586};
4587
4588static struct address_space_operations btrfs_symlink_aops = {
4589 .readpage = btrfs_readpage,
4590 .writepage = btrfs_writepage,
2bf5a725
CM		 4591 .invalidatepage = btrfs_invalidatepage,
4592 .releasepage = btrfs_releasepage,
39279cc3
CM		 4593};
4594
4595static struct inode_operations btrfs_file_inode_operations = {
4596 .truncate = btrfs_truncate,
4597 .getattr = btrfs_getattr,
4598 .setattr = btrfs_setattr,
95819c05
CH		 4599 .setxattr = btrfs_setxattr,
4600 .getxattr = btrfs_getxattr,
5103e947 4601 .listxattr = btrfs_listxattr,
95819c05 4602 .removexattr = btrfs_removexattr,
fdebe2bd 4603 .permission = btrfs_permission,
d899e052 4604 .fallocate = btrfs_fallocate,
39279cc3 4605};
618e21d5
JB		 4606static struct inode_operations btrfs_special_inode_operations = {
4607 .getattr = btrfs_getattr,
4608 .setattr = btrfs_setattr,
fdebe2bd 4609 .permission = btrfs_permission,
95819c05
CH		 4610 .setxattr = btrfs_setxattr,
4611 .getxattr = btrfs_getxattr,
33268eaf 4612 .listxattr = btrfs_listxattr,
95819c05 4613 .removexattr = btrfs_removexattr,
618e21d5 4614};
39279cc3
CM		 4615static struct inode_operations btrfs_symlink_inode_operations = {
4616 .readlink = generic_readlink,
4617 .follow_link = page_follow_link_light,
4618 .put_link = page_put_link,
fdebe2bd 4619 .permission = btrfs_permission,
39279cc3 4620};
Linux 6.x block layer and io_uring tree(s)