projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: update nodatacow code 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>
39279cc3
CM		 40#include "ctree.h"
41#include "disk-io.h"
42#include "transaction.h"
43#include "btrfs_inode.h"
44#include "ioctl.h"
45#include "print-tree.h"
0b86a832 46#include "volumes.h"
e6dcd2dc 47#include "ordered-data.h"
95819c05 48#include "xattr.h"
e02119d5
CM		 49#include "compat.h"
50#include "tree-log.h"
5b84e8d6 51#include "ref-cache.h"
c8b97818 52#include "compression.h"
39279cc3
CM		 53
54struct btrfs_iget_args {
55 u64 ino;
56 struct btrfs_root *root;
57};
58
59static struct inode_operations btrfs_dir_inode_operations;
60static struct inode_operations btrfs_symlink_inode_operations;
61static struct inode_operations btrfs_dir_ro_inode_operations;
618e21d5 62static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM		 63static struct inode_operations btrfs_file_inode_operations;
64static struct address_space_operations btrfs_aops;
65static struct address_space_operations btrfs_symlink_aops;
66static struct file_operations btrfs_dir_file_operations;
d1310b2e 67static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 68
69static struct kmem_cache *btrfs_inode_cachep;
70struct kmem_cache *btrfs_trans_handle_cachep;
71struct kmem_cache *btrfs_transaction_cachep;
72struct kmem_cache *btrfs_bit_radix_cachep;
73struct kmem_cache *btrfs_path_cachep;
74
75#define S_SHIFT 12
76static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
78 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
79 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
80 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
81 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
82 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
83 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84};
85
7b128766 86static void btrfs_truncate(struct inode *inode);
c8b97818 87static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
7b128766 88
d352ac68
CM		 89/*
90 * a very lame attempt at stopping writes when the FS is 85% full. There
91 * are countless ways this is incorrect, but it is better than nothing.
92 */
1832a6d5
CM		 93int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
94 int for_del)
95{
a2135011
CM		 96 u64 total;
97 u64 used;
1832a6d5 98 u64 thresh;
bcbfce8a 99 unsigned long flags;
1832a6d5
CM		 100 int ret = 0;
101
a2135011
CM		 102 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
103 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
104 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
1832a6d5 105 if (for_del)
f9ef6604 106 thresh = total * 90;
1832a6d5 107 else
f9ef6604
CM		 108 thresh = total * 85;
109
110 do_div(thresh, 100);
1832a6d5 111
1832a6d5
CM		 112 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
113 ret = -ENOSPC;
bcbfce8a 114 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1832a6d5
CM		 115 return ret;
116}
117
c8b97818
CM		 118/*
119 * this does all the hard work for inserting an inline extent into
120 * the btree. The caller should have done a btrfs_drop_extents so that
121 * no overlapping inline items exist in the btree
122 */
123static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
124 struct btrfs_root *root, struct inode *inode,
125 u64 start, size_t size, size_t compressed_size,
126 struct page **compressed_pages)
127{
128 struct btrfs_key key;
129 struct btrfs_path *path;
130 struct extent_buffer *leaf;
131 struct page *page = NULL;
132 char *kaddr;
133 unsigned long ptr;
134 struct btrfs_file_extent_item *ei;
135 int err = 0;
136 int ret;
137 size_t cur_size = size;
138 size_t datasize;
139 unsigned long offset;
140 int use_compress = 0;
141
142 if (compressed_size && compressed_pages) {
143 use_compress = 1;
144 cur_size = compressed_size;
145 }
146
147 path = btrfs_alloc_path(); if (!path)
148 return -ENOMEM;
149
150 btrfs_set_trans_block_group(trans, inode);
151
152 key.objectid = inode->i_ino;
153 key.offset = start;
154 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
155 inode_add_bytes(inode, size);
156 datasize = btrfs_file_extent_calc_inline_size(cur_size);
157
158 inode_add_bytes(inode, size);
159 ret = btrfs_insert_empty_item(trans, root, path, &key,
160 datasize);
161 BUG_ON(ret);
162 if (ret) {
163 err = ret;
164 printk("got bad ret %d\n", ret);
165 goto fail;
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
177 if (use_compress) {
178 struct page *cpage;
179 int i = 0;
180 while(compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min(compressed_size,
183 PAGE_CACHE_SIZE);
184
185 kaddr = kmap(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap(cpage);
188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
194 BTRFS_COMPRESS_ZLIB);
195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page, KM_USER0);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr, KM_USER0);
203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
207
208 BTRFS_I(inode)->disk_i_size = inode->i_size;
209 btrfs_update_inode(trans, root, inode);
210 return 0;
211fail:
212 btrfs_free_path(path);
213 return err;
214}
215
216
217/*
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
221 */
222static int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size,
226 struct page **compressed_pages)
227{
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
233 u64 hint_byte;
234 u64 data_len = inline_len;
235 int ret;
236
237 if (compressed_size)
238 data_len = compressed_size;
239
240 if (start > 0 ||
241 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (!compressed_size &&
243 (actual_end & (root->sectorsize - 1)) == 0) ||
244 end + 1 < isize ||
245 data_len > root->fs_info->max_inline) {
246 return 1;
247 }
248
c8b97818
CM		 249 ret = btrfs_drop_extents(trans, root, inode, start,
250 aligned_end, aligned_end, &hint_byte);
251 BUG_ON(ret);
252
253 if (isize > actual_end)
254 inline_len = min_t(u64, isize, actual_end);
255 ret = insert_inline_extent(trans, root, inode, start,
256 inline_len, compressed_size,
257 compressed_pages);
258 BUG_ON(ret);
259 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 260 return 0;
261}
262
d352ac68
CM		 263/*
264 * when extent_io.c finds a delayed allocation range in the file,
265 * the call backs end up in this code. The basic idea is to
266 * allocate extents on disk for the range, and create ordered data structs
267 * in ram to track those extents.
c8b97818
CM		 268 *
269 * locked_page is the page that writepage had locked already. We use
270 * it to make sure we don't do extra locks or unlocks.
271 *
272 * *page_started is set to one if we unlock locked_page and do everything
273 * required to start IO on it. It may be clean and already done with
274 * IO when we return.
d352ac68 275 */
c8b97818
CM		 276static int cow_file_range(struct inode *inode, struct page *locked_page,
277 u64 start, u64 end, int *page_started)
b888db2b
CM		 278{
279 struct btrfs_root *root = BTRFS_I(inode)->root;
280 struct btrfs_trans_handle *trans;
b888db2b 281 u64 alloc_hint = 0;
db94535d 282 u64 num_bytes;
c8b97818
CM		 283 unsigned long ram_size;
284 u64 orig_start;
285 u64 disk_num_bytes;
c59f8951 286 u64 cur_alloc_size;
db94535d 287 u64 blocksize = root->sectorsize;
c8b97818 288 u64 actual_end;
be20aa9d 289 struct btrfs_key ins;
e6dcd2dc
CM		 290 struct extent_map *em;
291 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
292 int ret = 0;
c8b97818
CM		 293 struct page **pages = NULL;
294 unsigned long nr_pages;
295 unsigned long nr_pages_ret = 0;
296 unsigned long total_compressed = 0;
297 unsigned long total_in = 0;
298 unsigned long max_compressed = 128 * 1024;
299 unsigned long max_uncompressed = 256 * 1024;
300 int i;
80ff3856 301 int ordered_type;
c8b97818 302 int will_compress;
b888db2b 303
f9295749 304 trans = btrfs_join_transaction(root, 1);
b888db2b 305 BUG_ON(!trans);
be20aa9d 306 btrfs_set_trans_block_group(trans, inode);
c8b97818
CM		 307 orig_start = start;
308
309 /*
310 * compression made this loop a bit ugly, but the basic idea is to
311 * compress some pages but keep the total size of the compressed
312 * extent relatively small. If compression is off, this goto target
313 * is never used.
314 */
315again:
316 will_compress = 0;
317 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
318 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 319
c8b97818
CM		 320 actual_end = min_t(u64, i_size_read(inode), end + 1);
321 total_compressed = actual_end - start;
322
323 /* we want to make sure that amount of ram required to uncompress
324 * an extent is reasonable, so we limit the total size in ram
325 * of a compressed extent to 256k
326 */
327 total_compressed = min(total_compressed, max_uncompressed);
db94535d 328 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 329 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 330 disk_num_bytes = num_bytes;
331 total_in = 0;
332 ret = 0;
db94535d 333
c8b97818
CM		 334 /* we do compression for mount -o compress and when the
335 * inode has not been flagged as nocompress
336 */
337 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
338 btrfs_test_opt(root, COMPRESS)) {
339 WARN_ON(pages);
cfbc246e 340 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818
CM		 341
342 /* we want to make sure the amount of IO required to satisfy
343 * a random read is reasonably small, so we limit the size
344 * of a compressed extent to 128k
345 */
346 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
347 total_compressed, pages,
348 nr_pages, &nr_pages_ret,
349 &total_in,
350 &total_compressed,
351 max_compressed);
352
353 if (!ret) {
354 unsigned long offset = total_compressed &
355 (PAGE_CACHE_SIZE - 1);
356 struct page *page = pages[nr_pages_ret - 1];
357 char *kaddr;
358
359 /* zero the tail end of the last page, we might be
360 * sending it down to disk
361 */
362 if (offset) {
363 kaddr = kmap_atomic(page, KM_USER0);
364 memset(kaddr + offset, 0,
365 PAGE_CACHE_SIZE - offset);
366 kunmap_atomic(kaddr, KM_USER0);
367 }
368 will_compress = 1;
369 }
370 }
371 if (start == 0) {
372 /* lets try to make an inline extent */
373 if (ret || total_in < (end - start + 1)) {
374 /* we didn't compress the entire range, try
375 * to make an uncompressed inline extent. This
376 * is almost sure to fail, but maybe inline sizes
377 * will get bigger later
378 */
379 ret = cow_file_range_inline(trans, root, inode,
380 start, end, 0, NULL);
381 } else {
382 ret = cow_file_range_inline(trans, root, inode,
383 start, end,
384 total_compressed, pages);
385 }
386 if (ret == 0) {
387 extent_clear_unlock_delalloc(inode,
388 &BTRFS_I(inode)->io_tree,
389 start, end, NULL,
390 1, 1, 1);
391 *page_started = 1;
392 ret = 0;
393 goto free_pages_out;
394 }
395 }
396
397 if (will_compress) {
398 /*
399 * we aren't doing an inline extent round the compressed size
400 * up to a block size boundary so the allocator does sane
401 * things
402 */
403 total_compressed = (total_compressed + blocksize - 1) &
404 ~(blocksize - 1);
405
406 /*
407 * one last check to make sure the compression is really a
408 * win, compare the page count read with the blocks on disk
409 */
410 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
411 ~(PAGE_CACHE_SIZE - 1);
412 if (total_compressed >= total_in) {
413 will_compress = 0;
414 } else {
415 disk_num_bytes = total_compressed;
416 num_bytes = total_in;
417 }
418 }
419 if (!will_compress && pages) {
420 /*
421 * the compression code ran but failed to make things smaller,
422 * free any pages it allocated and our page pointer array
423 */
424 for (i = 0; i < nr_pages_ret; i++) {
425 page_cache_release(pages[i]);
426 }
427 kfree(pages);
428 pages = NULL;
429 total_compressed = 0;
430 nr_pages_ret = 0;
431
432 /* flag the file so we don't compress in the future */
433 btrfs_set_flag(inode, NOCOMPRESS);
434 }
435
436 BUG_ON(disk_num_bytes >
437 btrfs_super_total_bytes(&root->fs_info->super_copy));
179e29e4 438
5b21f2ed 439 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
3b951516 440
c8b97818
CM		 441 while(disk_num_bytes > 0) {
442 unsigned long min_bytes;
443
444 /*
445 * the max size of a compressed extent is pretty small,
446 * make the code a little less complex by forcing
447 * the allocator to find a whole compressed extent at once
448 */
449 if (will_compress)
450 min_bytes = disk_num_bytes;
451 else
452 min_bytes = root->sectorsize;
453
454 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 455 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
c8b97818 456 min_bytes, 0, alloc_hint,
e6dcd2dc 457 (u64)-1, &ins, 1);
c59f8951
CM		 458 if (ret) {
459 WARN_ON(1);
c8b97818 460 goto free_pages_out_fail;
c59f8951 461 }
e6dcd2dc
CM		 462 em = alloc_extent_map(GFP_NOFS);
463 em->start = start;
c8b97818
CM		 464
465 if (will_compress) {
466 ram_size = num_bytes;
467 em->len = num_bytes;
468 } else {
469 /* ramsize == disk size */
470 ram_size = ins.offset;
471 em->len = ins.offset;
472 }
473
e6dcd2dc 474 em->block_start = ins.objectid;
c8b97818 475 em->block_len = ins.offset;
e6dcd2dc 476 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 477 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818
CM		 478
479 if (will_compress)
480 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
481
e6dcd2dc
CM		 482 while(1) {
483 spin_lock(&em_tree->lock);
484 ret = add_extent_mapping(em_tree, em);
485 spin_unlock(&em_tree->lock);
486 if (ret != -EEXIST) {
487 free_extent_map(em);
488 break;
489 }
490 btrfs_drop_extent_cache(inode, start,
c8b97818 491 start + ram_size - 1, 0);
e6dcd2dc
CM		 492 }
493
98d20f67 494 cur_alloc_size = ins.offset;
80ff3856 495 ordered_type = will_compress ? BTRFS_ORDERED_COMPRESSED : 0;
e6dcd2dc 496 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
80ff3856
YZ		 497 ram_size, cur_alloc_size,
498 ordered_type);
e6dcd2dc 499 BUG_ON(ret);
c8b97818
CM		 500
501 if (disk_num_bytes < cur_alloc_size) {
502 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
3b951516
CM		 503 cur_alloc_size);
504 break;
505 }
c8b97818
CM		 506
507 if (will_compress) {
508 /*
509 * we're doing compression, we and we need to
510 * submit the compressed extents down to the device.
511 *
512 * We lock down all the file pages, clearing their
513 * dirty bits and setting them writeback. Everyone
514 * that wants to modify the page will wait on the
515 * ordered extent above.
516 *
517 * The writeback bits on the file pages are
518 * cleared when the compressed pages are on disk
519 */
520 btrfs_end_transaction(trans, root);
521
522 if (start <= page_offset(locked_page) &&
523 page_offset(locked_page) < start + ram_size) {
524 *page_started = 1;
525 }
526
527 extent_clear_unlock_delalloc(inode,
528 &BTRFS_I(inode)->io_tree,
529 start,
530 start + ram_size - 1,
531 NULL, 1, 1, 0);
532
533 ret = btrfs_submit_compressed_write(inode, start,
534 ram_size, ins.objectid,
535 cur_alloc_size, pages,
536 nr_pages_ret);
537
538 BUG_ON(ret);
539 trans = btrfs_join_transaction(root, 1);
540 if (start + ram_size < end) {
541 start += ram_size;
542 alloc_hint = ins.objectid + ins.offset;
543 /* pages will be freed at end_bio time */
544 pages = NULL;
545 goto again;
546 } else {
547 /* we've written everything, time to go */
548 break;
549 }
550 }
551 /* we're not doing compressed IO, don't unlock the first
552 * page (which the caller expects to stay locked), don't
553 * clear any dirty bits and don't set any writeback bits
554 */
555 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
556 start, start + ram_size - 1,
557 locked_page, 0, 0, 0);
558 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 559 num_bytes -= cur_alloc_size;
560 alloc_hint = ins.objectid + ins.offset;
561 start += cur_alloc_size;
b888db2b 562 }
c8b97818
CM		 563
564 ret = 0;
b888db2b
CM		 565out:
566 btrfs_end_transaction(trans, root);
c8b97818 567
be20aa9d 568 return ret;
c8b97818
CM		 569
570free_pages_out_fail:
571 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
572 start, end, locked_page, 0, 0, 0);
573free_pages_out:
574 for (i = 0; i < nr_pages_ret; i++)
575 page_cache_release(pages[i]);
576 if (pages)
577 kfree(pages);
578
579 goto out;
be20aa9d
CM		 580}
581
d352ac68
CM		 582/*
583 * when nowcow writeback call back. This checks for snapshots or COW copies
584 * of the extents that exist in the file, and COWs the file as required.
585 *
586 * If no cow copies or snapshots exist, we write directly to the existing
587 * blocks on disk
588 */
c8b97818
CM		 589static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
590 u64 start, u64 end, int *page_started)
be20aa9d 591{
be20aa9d 592 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 593 struct btrfs_trans_handle *trans;
be20aa9d 594 struct extent_buffer *leaf;
be20aa9d 595 struct btrfs_path *path;
80ff3856 596 struct btrfs_file_extent_item *fi;
be20aa9d 597 struct btrfs_key found_key;
80ff3856
YZ		 598 u64 cow_start;
599 u64 cur_offset;
600 u64 extent_end;
601 u64 disk_bytenr;
602 u64 num_bytes;
603 int extent_type;
604 int ret;
605 int nocow;
606 int check_prev = 1;
be20aa9d
CM		 607
608 path = btrfs_alloc_path();
609 BUG_ON(!path);
7ea394f1
YZ		 610 trans = btrfs_join_transaction(root, 1);
611 BUG_ON(!trans);
be20aa9d 612
80ff3856
YZ		 613 cow_start = (u64)-1;
614 cur_offset = start;
615 while (1) {
616 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
617 cur_offset, 0);
618 BUG_ON(ret < 0);
619 if (ret > 0 && path->slots[0] > 0 && check_prev) {
620 leaf = path->nodes[0];
621 btrfs_item_key_to_cpu(leaf, &found_key,
622 path->slots[0] - 1);
623 if (found_key.objectid == inode->i_ino &&
624 found_key.type == BTRFS_EXTENT_DATA_KEY)
625 path->slots[0]--;
626 }
627 check_prev = 0;
628next_slot:
629 leaf = path->nodes[0];
630 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
631 ret = btrfs_next_leaf(root, path);
632 if (ret < 0)
633 BUG_ON(1);
634 if (ret > 0)
635 break;
636 leaf = path->nodes[0];
637 }
be20aa9d 638
80ff3856
YZ		 639 nocow = 0;
640 disk_bytenr = 0;
641 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
642
643 if (found_key.objectid > inode->i_ino ||
644 found_key.type > BTRFS_EXTENT_DATA_KEY ||
645 found_key.offset > end)
646 break;
647
648 if (found_key.offset > cur_offset) {
649 extent_end = found_key.offset;
650 goto out_check;
651 }
652
653 fi = btrfs_item_ptr(leaf, path->slots[0],
654 struct btrfs_file_extent_item);
655 extent_type = btrfs_file_extent_type(leaf, fi);
656
657 if (extent_type == BTRFS_FILE_EXTENT_REG) {
658 struct btrfs_block_group_cache *block_group;
659 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
660 extent_end = found_key.offset +
661 btrfs_file_extent_num_bytes(leaf, fi);
662 if (extent_end <= start) {
663 path->slots[0]++;
664 goto next_slot;
665 }
666 if (btrfs_file_extent_compression(leaf, fi) ||
667 btrfs_file_extent_encryption(leaf, fi) ||
668 btrfs_file_extent_other_encoding(leaf, fi))
669 goto out_check;
670 if (disk_bytenr == 0)
671 goto out_check;
672 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
673 goto out_check;
674 block_group = btrfs_lookup_block_group(root->fs_info,
675 disk_bytenr);
676 if (!block_group || block_group->ro)
677 goto out_check;
678 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
679 nocow = 1;
680 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
681 extent_end = found_key.offset +
682 btrfs_file_extent_inline_len(leaf, fi);
683 extent_end = ALIGN(extent_end, root->sectorsize);
684 } else {
685 BUG_ON(1);
686 }
687out_check:
688 if (extent_end <= start) {
689 path->slots[0]++;
690 goto next_slot;
691 }
692 if (!nocow) {
693 if (cow_start == (u64)-1)
694 cow_start = cur_offset;
695 cur_offset = extent_end;
696 if (cur_offset > end)
697 break;
698 path->slots[0]++;
699 goto next_slot;
7ea394f1
YZ		 700 }
701
702 btrfs_release_path(root, path);
80ff3856
YZ		 703 if (cow_start != (u64)-1) {
704 ret = cow_file_range(inode, locked_page, cow_start,
705 found_key.offset - 1, page_started);
706 BUG_ON(ret);
707 cow_start = (u64)-1;
7ea394f1 708 }
80ff3856
YZ		 709
710 disk_bytenr += cur_offset - found_key.offset;
711 num_bytes = min(end + 1, extent_end) - cur_offset;
712
713 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
714 num_bytes, num_bytes,
715 BTRFS_ORDERED_NOCOW);
716 cur_offset = extent_end;
717 if (cur_offset > end)
718 break;
be20aa9d 719 }
80ff3856
YZ		 720 btrfs_release_path(root, path);
721
722 if (cur_offset <= end && cow_start == (u64)-1)
723 cow_start = cur_offset;
724 if (cow_start != (u64)-1) {
725 ret = cow_file_range(inode, locked_page, cow_start, end,
726 page_started);
727 BUG_ON(ret);
728 }
729
730 ret = btrfs_end_transaction(trans, root);
731 BUG_ON(ret);
7ea394f1 732 btrfs_free_path(path);
80ff3856 733 return 0;
be20aa9d
CM		 734}
735
d352ac68
CM		 736/*
737 * extent_io.c call back to do delayed allocation processing
738 */
c8b97818
CM		 739static int run_delalloc_range(struct inode *inode, struct page *locked_page,
740 u64 start, u64 end, int *page_started)
be20aa9d
CM		 741{
742 struct btrfs_root *root = BTRFS_I(inode)->root;
743 int ret;
a2135011 744
b98b6767
Y		 745 if (btrfs_test_opt(root, NODATACOW) ||
746 btrfs_test_flag(inode, NODATACOW))
c8b97818
CM		 747 ret = run_delalloc_nocow(inode, locked_page, start, end,
748 page_started);
be20aa9d 749 else
c8b97818
CM		 750 ret = cow_file_range(inode, locked_page, start, end,
751 page_started);
1832a6d5 752
b888db2b
CM		 753 return ret;
754}
755
d352ac68
CM		 756/*
757 * extent_io.c set_bit_hook, used to track delayed allocation
758 * bytes in this file, and to maintain the list of inodes that
759 * have pending delalloc work to be done.
760 */
291d673e 761int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 762 unsigned long old, unsigned long bits)
291d673e 763{
bcbfce8a 764 unsigned long flags;
b0c68f8b 765 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 766 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 767 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
9069218d 768 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 769 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 770 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
771 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
772 &root->fs_info->delalloc_inodes);
773 }
bcbfce8a 774 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 775 }
776 return 0;
777}
778
d352ac68
CM		 779/*
780 * extent_io.c clear_bit_hook, see set_bit_hook for why
781 */
291d673e 782int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 783 unsigned long old, unsigned long bits)
291d673e 784{
b0c68f8b 785 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 786 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a
CM		 787 unsigned long flags;
788
789 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
b0c68f8b
CM		 790 if (end - start + 1 > root->fs_info->delalloc_bytes) {
791 printk("warning: delalloc account %Lu %Lu\n",
792 end - start + 1, root->fs_info->delalloc_bytes);
793 root->fs_info->delalloc_bytes = 0;
9069218d 794 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b
CM		 795 } else {
796 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 797 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 798 }
ea8c2819
CM		 799 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
800 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
801 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
802 }
bcbfce8a 803 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
291d673e
CM		 804 }
805 return 0;
806}
807
d352ac68
CM		 808/*
809 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
810 * we don't create bios that span stripes or chunks
811 */
239b14b3 812int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 813 size_t size, struct bio *bio,
814 unsigned long bio_flags)
239b14b3
CM		 815{
816 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
817 struct btrfs_mapping_tree *map_tree;
a62b9401 818 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 819 u64 length = 0;
820 u64 map_length;
239b14b3
CM		 821 int ret;
822
f2d8d74d 823 length = bio->bi_size;
239b14b3
CM		 824 map_tree = &root->fs_info->mapping_tree;
825 map_length = length;
cea9e445 826 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 827 &map_length, NULL, 0);
cea9e445 828
239b14b3 829 if (map_length < length + size) {
239b14b3
CM		 830 return 1;
831 }
832 return 0;
833}
834
d352ac68
CM		 835/*
836 * in order to insert checksums into the metadata in large chunks,
837 * we wait until bio submission time. All the pages in the bio are
838 * checksummed and sums are attached onto the ordered extent record.
839 *
840 * At IO completion time the cums attached on the ordered extent record
841 * are inserted into the btree
842 */
44b8bd7e 843int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 844 int mirror_num, unsigned long bio_flags)
065631f6 845{
065631f6 846 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 847 int ret = 0;
e015640f 848
3edf7d33 849 ret = btrfs_csum_one_bio(root, inode, bio);
44b8bd7e 850 BUG_ON(ret);
e015640f 851
8b712842 852 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 853}
854
d352ac68
CM		 855/*
856 * extent_io.c submission hook. This does the right thing for csum calculation on write,
857 * or reading the csums from the tree before a read
858 */
44b8bd7e 859int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 860 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 861{
862 struct btrfs_root *root = BTRFS_I(inode)->root;
863 int ret = 0;
19b9bdb0 864 int skip_sum;
44b8bd7e 865
e6dcd2dc
CM		 866 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
867 BUG_ON(ret);
065631f6 868
19b9bdb0
CM		 869 skip_sum = btrfs_test_opt(root, NODATASUM) ||
870 btrfs_test_flag(inode, NODATASUM);
7ea394f1 871
4d1b5fb4 872 if (!(rw & (1 << BIO_RW))) {
19b9bdb0
CM		 873 if (!skip_sum)
874 btrfs_lookup_bio_sums(root, inode, bio);
c8b97818 875
19b9bdb0 876 if (bio_flags & EXTENT_BIO_COMPRESSED)
c8b97818
CM		 877 return btrfs_submit_compressed_read(inode, bio,
878 mirror_num, bio_flags);
4d1b5fb4 879 goto mapit;
19b9bdb0
CM		 880 } else if (!skip_sum) {
881 /* we're doing a write, do the async checksumming */
882 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 883 inode, rw, bio, mirror_num,
c8b97818 884 bio_flags, __btrfs_submit_bio_hook);
19b9bdb0
CM		 885 }
886
0b86a832 887mapit:
8b712842 888 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 889}
6885f308 890
d352ac68
CM		 891/*
892 * given a list of ordered sums record them in the inode. This happens
893 * at IO completion time based on sums calculated at bio submission time.
894 */
ba1da2f4 895static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 896 struct inode *inode, u64 file_offset,
897 struct list_head *list)
898{
899 struct list_head *cur;
900 struct btrfs_ordered_sum *sum;
901
902 btrfs_set_trans_block_group(trans, inode);
ba1da2f4 903 list_for_each(cur, list) {
e6dcd2dc 904 sum = list_entry(cur, struct btrfs_ordered_sum, list);
e6dcd2dc
CM		 905 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
906 inode, sum);
e6dcd2dc
CM		 907 }
908 return 0;
909}
910
ea8c2819
CM		 911int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
912{
913 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
914 GFP_NOFS);
915}
916
d352ac68 917/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 918struct btrfs_writepage_fixup {
919 struct page *page;
920 struct btrfs_work work;
921};
922
247e743c
CM		 923void btrfs_writepage_fixup_worker(struct btrfs_work *work)
924{
925 struct btrfs_writepage_fixup *fixup;
926 struct btrfs_ordered_extent *ordered;
927 struct page *page;
928 struct inode *inode;
929 u64 page_start;
930 u64 page_end;
931
932 fixup = container_of(work, struct btrfs_writepage_fixup, work);
933 page = fixup->page;
4a096752 934again:
247e743c
CM		 935 lock_page(page);
936 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
937 ClearPageChecked(page);
938 goto out_page;
939 }
940
941 inode = page->mapping->host;
942 page_start = page_offset(page);
943 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
944
945 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 946
947 /* already ordered? We're done */
948 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
949 EXTENT_ORDERED, 0)) {
247e743c 950 goto out;
4a096752
CM		 951 }
952
953 ordered = btrfs_lookup_ordered_extent(inode, page_start);
954 if (ordered) {
955 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
956 page_end, GFP_NOFS);
957 unlock_page(page);
958 btrfs_start_ordered_extent(inode, ordered, 1);
959 goto again;
960 }
247e743c 961
ea8c2819 962 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 963 ClearPageChecked(page);
964out:
965 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
966out_page:
967 unlock_page(page);
968 page_cache_release(page);
969}
970
971/*
972 * There are a few paths in the higher layers of the kernel that directly
973 * set the page dirty bit without asking the filesystem if it is a
974 * good idea. This causes problems because we want to make sure COW
975 * properly happens and the data=ordered rules are followed.
976 *
c8b97818 977 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 978 * hasn't been properly setup for IO. We kick off an async process
979 * to fix it up. The async helper will wait for ordered extents, set
980 * the delalloc bit and make it safe to write the page.
981 */
982int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
983{
984 struct inode *inode = page->mapping->host;
985 struct btrfs_writepage_fixup *fixup;
986 struct btrfs_root *root = BTRFS_I(inode)->root;
987 int ret;
988
989 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
990 EXTENT_ORDERED, 0);
991 if (ret)
992 return 0;
993
994 if (PageChecked(page))
995 return -EAGAIN;
996
997 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
998 if (!fixup)
999 return -EAGAIN;
f421950f 1000
247e743c
CM		 1001 SetPageChecked(page);
1002 page_cache_get(page);
1003 fixup->work.func = btrfs_writepage_fixup_worker;
1004 fixup->page = page;
1005 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1006 return -EAGAIN;
1007}
1008
d352ac68
CM		 1009/* as ordered data IO finishes, this gets called so we can finish
1010 * an ordered extent if the range of bytes in the file it covers are
1011 * fully written.
1012 */
211f90e6 1013static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1014{
e6dcd2dc
CM		 1015 struct btrfs_root *root = BTRFS_I(inode)->root;
1016 struct btrfs_trans_handle *trans;
1017 struct btrfs_ordered_extent *ordered_extent;
1018 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
31840ae1
ZY		 1019 struct btrfs_file_extent_item *extent_item;
1020 struct btrfs_path *path = NULL;
1021 struct extent_buffer *leaf;
e6dcd2dc
CM		 1022 u64 alloc_hint = 0;
1023 struct list_head list;
1024 struct btrfs_key ins;
1025 int ret;
1026
1027 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1028 if (!ret)
e6dcd2dc 1029 return 0;
e6dcd2dc 1030
f9295749 1031 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1032
1033 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1034 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1035 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1036 goto nocow;
e6dcd2dc 1037
31840ae1
ZY		 1038 path = btrfs_alloc_path();
1039 BUG_ON(!path);
1040
e6dcd2dc
CM		 1041 lock_extent(io_tree, ordered_extent->file_offset,
1042 ordered_extent->file_offset + ordered_extent->len - 1,
1043 GFP_NOFS);
1044
1045 INIT_LIST_HEAD(&list);
1046
e6dcd2dc
CM		 1047 ret = btrfs_drop_extents(trans, root, inode,
1048 ordered_extent->file_offset,
1049 ordered_extent->file_offset +
1050 ordered_extent->len,
1051 ordered_extent->file_offset, &alloc_hint);
1052 BUG_ON(ret);
31840ae1
ZY		 1053
1054 ins.objectid = inode->i_ino;
1055 ins.offset = ordered_extent->file_offset;
1056 ins.type = BTRFS_EXTENT_DATA_KEY;
1057 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1058 sizeof(*extent_item));
e6dcd2dc 1059 BUG_ON(ret);
31840ae1
ZY		 1060 leaf = path->nodes[0];
1061 extent_item = btrfs_item_ptr(leaf, path->slots[0],
1062 struct btrfs_file_extent_item);
1063 btrfs_set_file_extent_generation(leaf, extent_item, trans->transid);
1064 btrfs_set_file_extent_type(leaf, extent_item, BTRFS_FILE_EXTENT_REG);
1065 btrfs_set_file_extent_disk_bytenr(leaf, extent_item,
1066 ordered_extent->start);
1067 btrfs_set_file_extent_disk_num_bytes(leaf, extent_item,
c8b97818 1068 ordered_extent->disk_len);
31840ae1 1069 btrfs_set_file_extent_offset(leaf, extent_item, 0);
c8b97818
CM		 1070
1071 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1072 btrfs_set_file_extent_compression(leaf, extent_item, 1);
1073 else
1074 btrfs_set_file_extent_compression(leaf, extent_item, 0);
1075 btrfs_set_file_extent_encryption(leaf, extent_item, 0);
1076 btrfs_set_file_extent_other_encoding(leaf, extent_item, 0);
1077
1078 /* ram bytes = extent_num_bytes for now */
31840ae1
ZY		 1079 btrfs_set_file_extent_num_bytes(leaf, extent_item,
1080 ordered_extent->len);
c8b97818
CM		 1081 btrfs_set_file_extent_ram_bytes(leaf, extent_item,
1082 ordered_extent->len);
31840ae1 1083 btrfs_mark_buffer_dirty(leaf);
7f3c74fb 1084
e6dcd2dc
CM		 1085 btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
1086 ordered_extent->file_offset +
5b21f2ed 1087 ordered_extent->len - 1, 0);
ee6e6504 1088
31840ae1 1089 ins.objectid = ordered_extent->start;
c8b97818 1090 ins.offset = ordered_extent->disk_len;
31840ae1
ZY		 1091 ins.type = BTRFS_EXTENT_ITEM_KEY;
1092 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1093 root->root_key.objectid,
3bb1a1bc 1094 trans->transid, inode->i_ino, &ins);
31840ae1
ZY		 1095 BUG_ON(ret);
1096 btrfs_release_path(root, path);
1097
a76a3cd4 1098 inode_add_bytes(inode, ordered_extent->len);
e6dcd2dc
CM		 1099 unlock_extent(io_tree, ordered_extent->file_offset,
1100 ordered_extent->file_offset + ordered_extent->len - 1,
1101 GFP_NOFS);
7ea394f1 1102nocow:
e6dcd2dc
CM		 1103 add_pending_csums(trans, inode, ordered_extent->file_offset,
1104 &ordered_extent->list);
1105
34353029 1106 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1107 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1108 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1109 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1110 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1111
e6dcd2dc
CM		 1112 /* once for us */
1113 btrfs_put_ordered_extent(ordered_extent);
1114 /* once for the tree */
1115 btrfs_put_ordered_extent(ordered_extent);
1116
e6dcd2dc 1117 btrfs_end_transaction(trans, root);
31840ae1
ZY		 1118 if (path)
1119 btrfs_free_path(path);
e6dcd2dc
CM		 1120 return 0;
1121}
1122
211f90e6
CM		 1123int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1124 struct extent_state *state, int uptodate)
1125{
1126 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1127}
1128
d352ac68
CM		 1129/*
1130 * When IO fails, either with EIO or csum verification fails, we
1131 * try other mirrors that might have a good copy of the data. This
1132 * io_failure_record is used to record state as we go through all the
1133 * mirrors. If another mirror has good data, the page is set up to date
1134 * and things continue. If a good mirror can't be found, the original
1135 * bio end_io callback is called to indicate things have failed.
1136 */
7e38326f
CM		 1137struct io_failure_record {
1138 struct page *page;
1139 u64 start;
1140 u64 len;
1141 u64 logical;
1142 int last_mirror;
1143};
1144
1259ab75
CM		 1145int btrfs_io_failed_hook(struct bio *failed_bio,
1146 struct page *page, u64 start, u64 end,
1147 struct extent_state *state)
7e38326f
CM		 1148{
1149 struct io_failure_record *failrec = NULL;
1150 u64 private;
1151 struct extent_map *em;
1152 struct inode *inode = page->mapping->host;
1153 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1154 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1155 struct bio *bio;
1156 int num_copies;
1157 int ret;
1259ab75 1158 int rw;
7e38326f 1159 u64 logical;
c8b97818 1160 unsigned long bio_flags = 0;
7e38326f
CM		 1161
1162 ret = get_state_private(failure_tree, start, &private);
1163 if (ret) {
7e38326f
CM		 1164 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1165 if (!failrec)
1166 return -ENOMEM;
1167 failrec->start = start;
1168 failrec->len = end - start + 1;
1169 failrec->last_mirror = 0;
1170
3b951516
CM		 1171 spin_lock(&em_tree->lock);
1172 em = lookup_extent_mapping(em_tree, start, failrec->len);
1173 if (em->start > start || em->start + em->len < start) {
1174 free_extent_map(em);
1175 em = NULL;
1176 }
1177 spin_unlock(&em_tree->lock);
7e38326f
CM		 1178
1179 if (!em || IS_ERR(em)) {
1180 kfree(failrec);
1181 return -EIO;
1182 }
1183 logical = start - em->start;
1184 logical = em->block_start + logical;
c8b97818
CM		 1185 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1186 bio_flags = EXTENT_BIO_COMPRESSED;
7e38326f
CM		 1187 failrec->logical = logical;
1188 free_extent_map(em);
1189 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1190 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1191 set_state_private(failure_tree, start,
1192 (u64)(unsigned long)failrec);
7e38326f 1193 } else {
587f7704 1194 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1195 }
1196 num_copies = btrfs_num_copies(
1197 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1198 failrec->logical, failrec->len);
1199 failrec->last_mirror++;
1200 if (!state) {
1201 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1202 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1203 failrec->start,
1204 EXTENT_LOCKED);
1205 if (state && state->start != failrec->start)
1206 state = NULL;
1207 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1208 }
1209 if (!state || failrec->last_mirror > num_copies) {
1210 set_state_private(failure_tree, failrec->start, 0);
1211 clear_extent_bits(failure_tree, failrec->start,
1212 failrec->start + failrec->len - 1,
1213 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1214 kfree(failrec);
1215 return -EIO;
1216 }
1217 bio = bio_alloc(GFP_NOFS, 1);
1218 bio->bi_private = state;
1219 bio->bi_end_io = failed_bio->bi_end_io;
1220 bio->bi_sector = failrec->logical >> 9;
1221 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1222 bio->bi_size = 0;
7e38326f 1223 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1224 if (failed_bio->bi_rw & (1 << BIO_RW))
1225 rw = WRITE;
1226 else
1227 rw = READ;
1228
1229 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818
CM		 1230 failrec->last_mirror,
1231 bio_flags);
1259ab75
CM		 1232 return 0;
1233}
1234
d352ac68
CM		 1235/*
1236 * each time an IO finishes, we do a fast check in the IO failure tree
1237 * to see if we need to process or clean up an io_failure_record
1238 */
1259ab75
CM		 1239int btrfs_clean_io_failures(struct inode *inode, u64 start)
1240{
1241 u64 private;
1242 u64 private_failure;
1243 struct io_failure_record *failure;
1244 int ret;
1245
1246 private = 0;
1247 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1248 (u64)-1, 1, EXTENT_DIRTY)) {
1249 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1250 start, &private_failure);
1251 if (ret == 0) {
1252 failure = (struct io_failure_record *)(unsigned long)
1253 private_failure;
1254 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1255 failure->start, 0);
1256 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1257 failure->start,
1258 failure->start + failure->len - 1,
1259 EXTENT_DIRTY | EXTENT_LOCKED,
1260 GFP_NOFS);
1261 kfree(failure);
1262 }
1263 }
7e38326f
CM		 1264 return 0;
1265}
1266
d352ac68
CM		 1267/*
1268 * when reads are done, we need to check csums to verify the data is correct
1269 * if there's a match, we allow the bio to finish. If not, we go through
1270 * the io_failure_record routines to find good copies
1271 */
70dec807
CM		 1272int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1273 struct extent_state *state)
07157aac 1274{
35ebb934 1275 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1276 struct inode *inode = page->mapping->host;
d1310b2e 1277 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1278 char *kaddr;
aadfeb6e 1279 u64 private = ~(u32)0;
07157aac 1280 int ret;
ff79f819
CM		 1281 struct btrfs_root *root = BTRFS_I(inode)->root;
1282 u32 csum = ~(u32)0;
bbf0d006 1283 unsigned long flags;
d1310b2e 1284
b98b6767
Y		 1285 if (btrfs_test_opt(root, NODATASUM) ||
1286 btrfs_test_flag(inode, NODATASUM))
b6cda9bc 1287 return 0;
c2e639f0 1288 if (state && state->start == start) {
70dec807
CM		 1289 private = state->private;
1290 ret = 0;
1291 } else {
1292 ret = get_state_private(io_tree, start, &private);
1293 }
bbf0d006 1294 local_irq_save(flags);
07157aac
CM		 1295 kaddr = kmap_atomic(page, KM_IRQ0);
1296 if (ret) {
1297 goto zeroit;
1298 }
ff79f819
CM		 1299 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1300 btrfs_csum_final(csum, (char *)&csum);
1301 if (csum != private) {
07157aac
CM		 1302 goto zeroit;
1303 }
1304 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1305 local_irq_restore(flags);
7e38326f
CM		 1306
1307 /* if the io failure tree for this inode is non-empty,
1308 * check to see if we've recovered from a failed IO
1309 */
1259ab75 1310 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1311 return 0;
1312
1313zeroit:
aadfeb6e
CM		 1314 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1315 page->mapping->host->i_ino, (unsigned long long)start, csum,
1316 private);
db94535d
CM		 1317 memset(kaddr + offset, 1, end - start + 1);
1318 flush_dcache_page(page);
07157aac 1319 kunmap_atomic(kaddr, KM_IRQ0);
bbf0d006 1320 local_irq_restore(flags);
3b951516
CM		 1321 if (private == 0)
1322 return 0;
7e38326f 1323 return -EIO;
07157aac 1324}
b888db2b 1325
7b128766
JB		 1326/*
1327 * This creates an orphan entry for the given inode in case something goes
1328 * wrong in the middle of an unlink/truncate.
1329 */
1330int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1331{
1332 struct btrfs_root *root = BTRFS_I(inode)->root;
1333 int ret = 0;
1334
bcc63abb 1335 spin_lock(&root->list_lock);
7b128766
JB		 1336
1337 /* already on the orphan list, we're good */
1338 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1339 spin_unlock(&root->list_lock);
7b128766
JB		 1340 return 0;
1341 }
1342
1343 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1344
bcc63abb 1345 spin_unlock(&root->list_lock);
7b128766
JB		 1346
1347 /*
1348 * insert an orphan item to track this unlinked/truncated file
1349 */
1350 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1351
1352 return ret;
1353}
1354
1355/*
1356 * We have done the truncate/delete so we can go ahead and remove the orphan
1357 * item for this particular inode.
1358 */
1359int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1360{
1361 struct btrfs_root *root = BTRFS_I(inode)->root;
1362 int ret = 0;
1363
bcc63abb 1364 spin_lock(&root->list_lock);
7b128766
JB		 1365
1366 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1367 spin_unlock(&root->list_lock);
7b128766
JB		 1368 return 0;
1369 }
1370
1371 list_del_init(&BTRFS_I(inode)->i_orphan);
1372 if (!trans) {
bcc63abb 1373 spin_unlock(&root->list_lock);
7b128766
JB		 1374 return 0;
1375 }
1376
bcc63abb 1377 spin_unlock(&root->list_lock);
7b128766
JB		 1378
1379 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1380
1381 return ret;
1382}
1383
1384/*
1385 * this cleans up any orphans that may be left on the list from the last use
1386 * of this root.
1387 */
1388void btrfs_orphan_cleanup(struct btrfs_root *root)
1389{
1390 struct btrfs_path *path;
1391 struct extent_buffer *leaf;
1392 struct btrfs_item *item;
1393 struct btrfs_key key, found_key;
1394 struct btrfs_trans_handle *trans;
1395 struct inode *inode;
1396 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1397
1398 /* don't do orphan cleanup if the fs is readonly. */
5b21f2ed 1399 if (root->fs_info->sb->s_flags & MS_RDONLY)
7b128766
JB		 1400 return;
1401
1402 path = btrfs_alloc_path();
1403 if (!path)
1404 return;
1405 path->reada = -1;
1406
1407 key.objectid = BTRFS_ORPHAN_OBJECTID;
1408 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1409 key.offset = (u64)-1;
1410
7b128766
JB		 1411
1412 while (1) {
1413 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1414 if (ret < 0) {
1415 printk(KERN_ERR "Error searching slot for orphan: %d"
1416 "\n", ret);
1417 break;
1418 }
1419
1420 /*
1421 * if ret == 0 means we found what we were searching for, which
1422 * is weird, but possible, so only screw with path if we didnt
1423 * find the key and see if we have stuff that matches
1424 */
1425 if (ret > 0) {
1426 if (path->slots[0] == 0)
1427 break;
1428 path->slots[0]--;
1429 }
1430
1431 /* pull out the item */
1432 leaf = path->nodes[0];
1433 item = btrfs_item_nr(leaf, path->slots[0]);
1434 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1435
1436 /* make sure the item matches what we want */
1437 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1438 break;
1439 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1440 break;
1441
1442 /* release the path since we're done with it */
1443 btrfs_release_path(root, path);
1444
1445 /*
1446 * this is where we are basically btrfs_lookup, without the
1447 * crossing root thing. we store the inode number in the
1448 * offset of the orphan item.
1449 */
5b21f2ed 1450 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1451 found_key.offset, root);
1452 if (!inode)
1453 break;
1454
1455 if (inode->i_state & I_NEW) {
1456 BTRFS_I(inode)->root = root;
1457
1458 /* have to set the location manually */
1459 BTRFS_I(inode)->location.objectid = inode->i_ino;
1460 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1461 BTRFS_I(inode)->location.offset = 0;
1462
1463 btrfs_read_locked_inode(inode);
1464 unlock_new_inode(inode);
1465 }
1466
1467 /*
1468 * add this inode to the orphan list so btrfs_orphan_del does
1469 * the proper thing when we hit it
1470 */
bcc63abb 1471 spin_lock(&root->list_lock);
7b128766 1472 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1473 spin_unlock(&root->list_lock);
7b128766
JB		 1474
1475 /*
1476 * if this is a bad inode, means we actually succeeded in
1477 * removing the inode, but not the orphan record, which means
1478 * we need to manually delete the orphan since iput will just
1479 * do a destroy_inode
1480 */
1481 if (is_bad_inode(inode)) {
5b21f2ed 1482 trans = btrfs_start_transaction(root, 1);
7b128766 1483 btrfs_orphan_del(trans, inode);
5b21f2ed 1484 btrfs_end_transaction(trans, root);
7b128766
JB		 1485 iput(inode);
1486 continue;
1487 }
1488
1489 /* if we have links, this was a truncate, lets do that */
1490 if (inode->i_nlink) {
1491 nr_truncate++;
1492 btrfs_truncate(inode);
1493 } else {
1494 nr_unlink++;
1495 }
1496
1497 /* this will do delete_inode and everything for us */
1498 iput(inode);
1499 }
1500
1501 if (nr_unlink)
1502 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1503 if (nr_truncate)
1504 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1505
1506 btrfs_free_path(path);
7b128766
JB		 1507}
1508
d352ac68
CM		 1509/*
1510 * read an inode from the btree into the in-memory inode
1511 */
39279cc3
CM		 1512void btrfs_read_locked_inode(struct inode *inode)
1513{
1514 struct btrfs_path *path;
5f39d397 1515 struct extent_buffer *leaf;
39279cc3 1516 struct btrfs_inode_item *inode_item;
0b86a832 1517 struct btrfs_timespec *tspec;
39279cc3
CM		 1518 struct btrfs_root *root = BTRFS_I(inode)->root;
1519 struct btrfs_key location;
1520 u64 alloc_group_block;
618e21d5 1521 u32 rdev;
39279cc3
CM		 1522 int ret;
1523
1524 path = btrfs_alloc_path();
1525 BUG_ON(!path);
39279cc3 1526 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1527
39279cc3 1528 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1529 if (ret)
39279cc3 1530 goto make_bad;
39279cc3 1531
5f39d397
CM		 1532 leaf = path->nodes[0];
1533 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1534 struct btrfs_inode_item);
1535
1536 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1537 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1538 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1539 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1540 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1541
1542 tspec = btrfs_inode_atime(inode_item);
1543 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1544 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1545
1546 tspec = btrfs_inode_mtime(inode_item);
1547 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1548 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1549
1550 tspec = btrfs_inode_ctime(inode_item);
1551 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1552 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1553
a76a3cd4 1554 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5
CM		 1555 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1556 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 1557 inode->i_rdev = 0;
5f39d397
CM		 1558 rdev = btrfs_inode_rdev(leaf, inode_item);
1559
aec7477b
JB		 1560 BTRFS_I(inode)->index_cnt = (u64)-1;
1561
5f39d397 1562 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
39279cc3
CM		 1563 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1564 alloc_group_block);
b98b6767 1565 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
e52ec0eb
CM		 1566 if (!BTRFS_I(inode)->block_group) {
1567 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
0b86a832
CM		 1568 NULL, 0,
1569 BTRFS_BLOCK_GROUP_METADATA, 0);
e52ec0eb 1570 }
39279cc3
CM		 1571 btrfs_free_path(path);
1572 inode_item = NULL;
1573
39279cc3 1574 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 1575 case S_IFREG:
1576 inode->i_mapping->a_ops = &btrfs_aops;
04160088 1577 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 1578 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 1579 inode->i_fop = &btrfs_file_operations;
1580 inode->i_op = &btrfs_file_inode_operations;
1581 break;
1582 case S_IFDIR:
1583 inode->i_fop = &btrfs_dir_file_operations;
1584 if (root == root->fs_info->tree_root)
1585 inode->i_op = &btrfs_dir_ro_inode_operations;
1586 else
1587 inode->i_op = &btrfs_dir_inode_operations;
1588 break;
1589 case S_IFLNK:
1590 inode->i_op = &btrfs_symlink_inode_operations;
1591 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 1592 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 1593 break;
618e21d5
JB		 1594 default:
1595 init_special_inode(inode, inode->i_mode, rdev);
1596 break;
39279cc3
CM		 1597 }
1598 return;
1599
1600make_bad:
39279cc3 1601 btrfs_free_path(path);
39279cc3
CM		 1602 make_bad_inode(inode);
1603}
1604
d352ac68
CM		 1605/*
1606 * given a leaf and an inode, copy the inode fields into the leaf
1607 */
e02119d5
CM		 1608static void fill_inode_item(struct btrfs_trans_handle *trans,
1609 struct extent_buffer *leaf,
5f39d397 1610 struct btrfs_inode_item *item,
39279cc3
CM		 1611 struct inode *inode)
1612{
5f39d397
CM		 1613 btrfs_set_inode_uid(leaf, item, inode->i_uid);
1614 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 1615 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 1616 btrfs_set_inode_mode(leaf, item, inode->i_mode);
1617 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1618
1619 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1620 inode->i_atime.tv_sec);
1621 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1622 inode->i_atime.tv_nsec);
1623
1624 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1625 inode->i_mtime.tv_sec);
1626 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1627 inode->i_mtime.tv_nsec);
1628
1629 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1630 inode->i_ctime.tv_sec);
1631 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1632 inode->i_ctime.tv_nsec);
1633
a76a3cd4 1634 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5
CM		 1635 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1636 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 1637 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 1638 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
5f39d397 1639 btrfs_set_inode_block_group(leaf, item,
39279cc3
CM		 1640 BTRFS_I(inode)->block_group->key.objectid);
1641}
1642
d352ac68
CM		 1643/*
1644 * copy everything in the in-memory inode into the btree.
1645 */
ba1da2f4 1646int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
39279cc3
CM		 1647 struct btrfs_root *root,
1648 struct inode *inode)
1649{
1650 struct btrfs_inode_item *inode_item;
1651 struct btrfs_path *path;
5f39d397 1652 struct extent_buffer *leaf;
39279cc3
CM		 1653 int ret;
1654
1655 path = btrfs_alloc_path();
1656 BUG_ON(!path);
39279cc3
CM		 1657 ret = btrfs_lookup_inode(trans, root, path,
1658 &BTRFS_I(inode)->location, 1);
1659 if (ret) {
1660 if (ret > 0)
1661 ret = -ENOENT;
1662 goto failed;
1663 }
1664
5f39d397
CM		 1665 leaf = path->nodes[0];
1666 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 1667 struct btrfs_inode_item);
1668
e02119d5 1669 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 1670 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 1671 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 1672 ret = 0;
1673failed:
39279cc3
CM		 1674 btrfs_free_path(path);
1675 return ret;
1676}
1677
1678
d352ac68
CM		 1679/*
1680 * unlink helper that gets used here in inode.c and in the tree logging
1681 * recovery code. It remove a link in a directory with a given name, and
1682 * also drops the back refs in the inode to the directory
1683 */
e02119d5
CM		 1684int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1685 struct btrfs_root *root,
1686 struct inode *dir, struct inode *inode,
1687 const char *name, int name_len)
39279cc3
CM		 1688{
1689 struct btrfs_path *path;
39279cc3 1690 int ret = 0;
5f39d397 1691 struct extent_buffer *leaf;
39279cc3 1692 struct btrfs_dir_item *di;
5f39d397 1693 struct btrfs_key key;
aec7477b 1694 u64 index;
39279cc3
CM		 1695
1696 path = btrfs_alloc_path();
54aa1f4d
CM		 1697 if (!path) {
1698 ret = -ENOMEM;
1699 goto err;
1700 }
1701
39279cc3
CM		 1702 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1703 name, name_len, -1);
1704 if (IS_ERR(di)) {
1705 ret = PTR_ERR(di);
1706 goto err;
1707 }
1708 if (!di) {
1709 ret = -ENOENT;
1710 goto err;
1711 }
5f39d397
CM		 1712 leaf = path->nodes[0];
1713 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 1714 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 1715 if (ret)
1716 goto err;
39279cc3
CM		 1717 btrfs_release_path(root, path);
1718
aec7477b 1719 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 1720 inode->i_ino,
1721 dir->i_ino, &index);
aec7477b
JB		 1722 if (ret) {
1723 printk("failed to delete reference to %.*s, "
1724 "inode %lu parent %lu\n", name_len, name,
e02119d5 1725 inode->i_ino, dir->i_ino);
aec7477b
JB		 1726 goto err;
1727 }
1728
39279cc3 1729 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 1730 index, name, name_len, -1);
39279cc3
CM		 1731 if (IS_ERR(di)) {
1732 ret = PTR_ERR(di);
1733 goto err;
1734 }
1735 if (!di) {
1736 ret = -ENOENT;
1737 goto err;
1738 }
1739 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 1740 btrfs_release_path(root, path);
39279cc3 1741
e02119d5
CM		 1742 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1743 inode, dir->i_ino);
49eb7e46
CM		 1744 BUG_ON(ret != 0 && ret != -ENOENT);
1745 if (ret != -ENOENT)
1746 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 1747
1748 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1749 dir, index);
1750 BUG_ON(ret);
39279cc3
CM		 1751err:
1752 btrfs_free_path(path);
e02119d5
CM		 1753 if (ret)
1754 goto out;
1755
1756 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1757 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1758 btrfs_update_inode(trans, root, dir);
1759 btrfs_drop_nlink(inode);
1760 ret = btrfs_update_inode(trans, root, inode);
1761 dir->i_sb->s_dirt = 1;
1762out:
39279cc3
CM		 1763 return ret;
1764}
1765
1766static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1767{
1768 struct btrfs_root *root;
1769 struct btrfs_trans_handle *trans;
7b128766 1770 struct inode *inode = dentry->d_inode;
39279cc3 1771 int ret;
1832a6d5 1772 unsigned long nr = 0;
39279cc3
CM		 1773
1774 root = BTRFS_I(dir)->root;
1832a6d5
CM		 1775
1776 ret = btrfs_check_free_space(root, 1, 1);
1777 if (ret)
1778 goto fail;
1779
39279cc3 1780 trans = btrfs_start_transaction(root, 1);
5f39d397 1781
39279cc3 1782 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 1783 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1784 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 1785
1786 if (inode->i_nlink == 0)
1787 ret = btrfs_orphan_add(trans, inode);
1788
d3c2fdcf 1789 nr = trans->blocks_used;
5f39d397 1790
89ce8a63 1791 btrfs_end_transaction_throttle(trans, root);
1832a6d5 1792fail:
d3c2fdcf 1793 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 1794 return ret;
1795}
1796
1797static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1798{
1799 struct inode *inode = dentry->d_inode;
1832a6d5 1800 int err = 0;
39279cc3
CM		 1801 int ret;
1802 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 1803 struct btrfs_trans_handle *trans;
1832a6d5 1804 unsigned long nr = 0;
39279cc3 1805
925baedd 1806 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
134d4512 1807 return -ENOTEMPTY;
925baedd 1808 }
134d4512 1809
1832a6d5
CM		 1810 ret = btrfs_check_free_space(root, 1, 1);
1811 if (ret)
1812 goto fail;
1813
39279cc3
CM		 1814 trans = btrfs_start_transaction(root, 1);
1815 btrfs_set_trans_block_group(trans, dir);
39279cc3 1816
7b128766
JB		 1817 err = btrfs_orphan_add(trans, inode);
1818 if (err)
1819 goto fail_trans;
1820
39279cc3 1821 /* now the directory is empty */
e02119d5
CM		 1822 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1823 dentry->d_name.name, dentry->d_name.len);
39279cc3 1824 if (!err) {
dbe674a9 1825 btrfs_i_size_write(inode, 0);
39279cc3 1826 }
3954401f 1827
7b128766 1828fail_trans:
d3c2fdcf 1829 nr = trans->blocks_used;
89ce8a63 1830 ret = btrfs_end_transaction_throttle(trans, root);
1832a6d5 1831fail:
d3c2fdcf 1832 btrfs_btree_balance_dirty(root, nr);
3954401f 1833
39279cc3
CM		 1834 if (ret && !err)
1835 err = ret;
1836 return err;
1837}
1838
323ac95b
CM		 1839/*
1840 * when truncating bytes in a file, it is possible to avoid reading
1841 * the leaves that contain only checksum items. This can be the
1842 * majority of the IO required to delete a large file, but it must
1843 * be done carefully.
1844 *
1845 * The keys in the level just above the leaves are checked to make sure
1846 * the lowest key in a given leaf is a csum key, and starts at an offset
1847 * after the new size.
1848 *
1849 * Then the key for the next leaf is checked to make sure it also has
1850 * a checksum item for the same file. If it does, we know our target leaf
1851 * contains only checksum items, and it can be safely freed without reading
1852 * it.
1853 *
1854 * This is just an optimization targeted at large files. It may do
1855 * nothing. It will return 0 unless things went badly.
1856 */
1857static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
1858 struct btrfs_root *root,
1859 struct btrfs_path *path,
1860 struct inode *inode, u64 new_size)
1861{
1862 struct btrfs_key key;
1863 int ret;
1864 int nritems;
1865 struct btrfs_key found_key;
1866 struct btrfs_key other_key;
5b84e8d6
YZ		 1867 struct btrfs_leaf_ref *ref;
1868 u64 leaf_gen;
1869 u64 leaf_start;
323ac95b
CM		 1870
1871 path->lowest_level = 1;
1872 key.objectid = inode->i_ino;
1873 key.type = BTRFS_CSUM_ITEM_KEY;
1874 key.offset = new_size;
1875again:
1876 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1877 if (ret < 0)
1878 goto out;
1879
1880 if (path->nodes[1] == NULL) {
1881 ret = 0;
1882 goto out;
1883 }
1884 ret = 0;
1885 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
1886 nritems = btrfs_header_nritems(path->nodes[1]);
1887
1888 if (!nritems)
1889 goto out;
1890
1891 if (path->slots[1] >= nritems)
1892 goto next_node;
1893
1894 /* did we find a key greater than anything we want to delete? */
1895 if (found_key.objectid > inode->i_ino ||
1896 (found_key.objectid == inode->i_ino && found_key.type > key.type))
1897 goto out;
1898
1899 /* we check the next key in the node to make sure the leave contains
1900 * only checksum items. This comparison doesn't work if our
1901 * leaf is the last one in the node
1902 */
1903 if (path->slots[1] + 1 >= nritems) {
1904next_node:
1905 /* search forward from the last key in the node, this
1906 * will bring us into the next node in the tree
1907 */
1908 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
1909
1910 /* unlikely, but we inc below, so check to be safe */
1911 if (found_key.offset == (u64)-1)
1912 goto out;
1913
1914 /* search_forward needs a path with locks held, do the
1915 * search again for the original key. It is possible
1916 * this will race with a balance and return a path that
1917 * we could modify, but this drop is just an optimization
1918 * and is allowed to miss some leaves.
1919 */
1920 btrfs_release_path(root, path);
1921 found_key.offset++;
1922
1923 /* setup a max key for search_forward */
1924 other_key.offset = (u64)-1;
1925 other_key.type = key.type;
1926 other_key.objectid = key.objectid;
1927
1928 path->keep_locks = 1;
1929 ret = btrfs_search_forward(root, &found_key, &other_key,
1930 path, 0, 0);
1931 path->keep_locks = 0;
1932 if (ret || found_key.objectid != key.objectid ||
1933 found_key.type != key.type) {
1934 ret = 0;
1935 goto out;
1936 }
1937
1938 key.offset = found_key.offset;
1939 btrfs_release_path(root, path);
1940 cond_resched();
1941 goto again;
1942 }
1943
1944 /* we know there's one more slot after us in the tree,
1945 * read that key so we can verify it is also a checksum item
1946 */
1947 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
1948
1949 if (found_key.objectid < inode->i_ino)
1950 goto next_key;
1951
1952 if (found_key.type != key.type || found_key.offset < new_size)
1953 goto next_key;
1954
1955 /*
1956 * if the key for the next leaf isn't a csum key from this objectid,
1957 * we can't be sure there aren't good items inside this leaf.
1958 * Bail out
1959 */
1960 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
1961 goto out;
1962
5b84e8d6
YZ		 1963 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
1964 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 1965 /*
1966 * it is safe to delete this leaf, it contains only
1967 * csum items from this inode at an offset >= new_size
1968 */
5b84e8d6 1969 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 1970 BUG_ON(ret);
1971
5b84e8d6
YZ		 1972 if (root->ref_cows && leaf_gen < trans->transid) {
1973 ref = btrfs_alloc_leaf_ref(root, 0);
1974 if (ref) {
1975 ref->root_gen = root->root_key.offset;
1976 ref->bytenr = leaf_start;
1977 ref->owner = 0;
1978 ref->generation = leaf_gen;
1979 ref->nritems = 0;
1980
1981 ret = btrfs_add_leaf_ref(root, ref, 0);
1982 WARN_ON(ret);
1983 btrfs_free_leaf_ref(root, ref);
1984 } else {
1985 WARN_ON(1);
1986 }
1987 }
323ac95b
CM		 1988next_key:
1989 btrfs_release_path(root, path);
1990
1991 if (other_key.objectid == inode->i_ino &&
1992 other_key.type == key.type && other_key.offset > key.offset) {
1993 key.offset = other_key.offset;
1994 cond_resched();
1995 goto again;
1996 }
1997 ret = 0;
1998out:
1999 /* fixup any changes we've made to the path */
2000 path->lowest_level = 0;
2001 path->keep_locks = 0;
2002 btrfs_release_path(root, path);
2003 return ret;
2004}
2005
39279cc3
CM		 2006/*
2007 * this can truncate away extent items, csum items and directory items.
2008 * It starts at a high offset and removes keys until it can't find
d352ac68 2009 * any higher than new_size
39279cc3
CM		 2010 *
2011 * csum items that cross the new i_size are truncated to the new size
2012 * as well.
7b128766
JB		 2013 *
2014 * min_type is the minimum key type to truncate down to. If set to 0, this
2015 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2016 */
e02119d5
CM		 2017noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2018 struct btrfs_root *root,
2019 struct inode *inode,
2020 u64 new_size, u32 min_type)
39279cc3
CM		 2021{
2022 int ret;
2023 struct btrfs_path *path;
2024 struct btrfs_key key;
5f39d397 2025 struct btrfs_key found_key;
39279cc3 2026 u32 found_type;
5f39d397 2027 struct extent_buffer *leaf;
39279cc3
CM		 2028 struct btrfs_file_extent_item *fi;
2029 u64 extent_start = 0;
db94535d 2030 u64 extent_num_bytes = 0;
39279cc3 2031 u64 item_end = 0;
7bb86316 2032 u64 root_gen = 0;
d8d5f3e1 2033 u64 root_owner = 0;
39279cc3
CM		 2034 int found_extent;
2035 int del_item;
85e21bac
CM		 2036 int pending_del_nr = 0;
2037 int pending_del_slot = 0;
179e29e4 2038 int extent_type = -1;
3b951516 2039 u64 mask = root->sectorsize - 1;
39279cc3 2040
e02119d5 2041 if (root->ref_cows)
5b21f2ed 2042 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2043 path = btrfs_alloc_path();
3c69faec 2044 path->reada = -1;
39279cc3 2045 BUG_ON(!path);
5f39d397 2046
39279cc3
CM		 2047 /* FIXME, add redo link to tree so we don't leak on crash */
2048 key.objectid = inode->i_ino;
2049 key.offset = (u64)-1;
5f39d397
CM		 2050 key.type = (u8)-1;
2051
85e21bac 2052 btrfs_init_path(path);
323ac95b
CM		 2053
2054 ret = drop_csum_leaves(trans, root, path, inode, new_size);
2055 BUG_ON(ret);
2056
85e21bac
CM		 2057search_again:
2058 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2059 if (ret < 0) {
2060 goto error;
2061 }
2062 if (ret > 0) {
e02119d5
CM		 2063 /* there are no items in the tree for us to truncate, we're
2064 * done
2065 */
2066 if (path->slots[0] == 0) {
2067 ret = 0;
2068 goto error;
2069 }
85e21bac
CM		 2070 path->slots[0]--;
2071 }
2072
39279cc3 2073 while(1) {
39279cc3 2074 fi = NULL;
5f39d397
CM		 2075 leaf = path->nodes[0];
2076 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2077 found_type = btrfs_key_type(&found_key);
39279cc3 2078
5f39d397 2079 if (found_key.objectid != inode->i_ino)
39279cc3 2080 break;
5f39d397 2081
85e21bac 2082 if (found_type < min_type)
39279cc3
CM		 2083 break;
2084
5f39d397 2085 item_end = found_key.offset;
39279cc3 2086 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2087 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2088 struct btrfs_file_extent_item);
179e29e4
CM		 2089 extent_type = btrfs_file_extent_type(leaf, fi);
2090 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2091 item_end +=
db94535d 2092 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2093 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2094 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2095 fi);
39279cc3 2096 }
008630c1 2097 item_end--;
39279cc3
CM		 2098 }
2099 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2100 ret = btrfs_csum_truncate(trans, root, path,
e02119d5 2101 new_size);
39279cc3
CM		 2102 BUG_ON(ret);
2103 }
e02119d5 2104 if (item_end < new_size) {
b888db2b
CM		 2105 if (found_type == BTRFS_DIR_ITEM_KEY) {
2106 found_type = BTRFS_INODE_ITEM_KEY;
2107 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2108 found_type = BTRFS_CSUM_ITEM_KEY;
85e21bac
CM		 2109 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2110 found_type = BTRFS_XATTR_ITEM_KEY;
2111 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2112 found_type = BTRFS_INODE_REF_KEY;
b888db2b
CM		 2113 } else if (found_type) {
2114 found_type--;
2115 } else {
2116 break;
39279cc3 2117 }
a61721d5 2118 btrfs_set_key_type(&key, found_type);
85e21bac 2119 goto next;
39279cc3 2120 }
e02119d5 2121 if (found_key.offset >= new_size)
39279cc3
CM		 2122 del_item = 1;
2123 else
2124 del_item = 0;
2125 found_extent = 0;
2126
2127 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2128 if (found_type != BTRFS_EXTENT_DATA_KEY)
2129 goto delete;
2130
2131 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2132 u64 num_dec;
db94535d 2133 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
39279cc3 2134 if (!del_item) {
db94535d
CM		 2135 u64 orig_num_bytes =
2136 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2137 extent_num_bytes = new_size -
5f39d397 2138 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2139 extent_num_bytes = extent_num_bytes &
2140 ~((u64)root->sectorsize - 1);
db94535d
CM		 2141 btrfs_set_file_extent_num_bytes(leaf, fi,
2142 extent_num_bytes);
2143 num_dec = (orig_num_bytes -
9069218d 2144 extent_num_bytes);
e02119d5 2145 if (root->ref_cows && extent_start != 0)
a76a3cd4 2146 inode_sub_bytes(inode, num_dec);
5f39d397 2147 btrfs_mark_buffer_dirty(leaf);
39279cc3 2148 } else {
db94535d
CM		 2149 extent_num_bytes =
2150 btrfs_file_extent_disk_num_bytes(leaf,
2151 fi);
39279cc3 2152 /* FIXME blocksize != 4096 */
9069218d 2153 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2154 if (extent_start != 0) {
2155 found_extent = 1;
e02119d5 2156 if (root->ref_cows)
a76a3cd4 2157 inode_sub_bytes(inode, num_dec);
e02119d5 2158 }
31840ae1 2159 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2160 root_owner = btrfs_header_owner(leaf);
39279cc3 2161 }
9069218d 2162 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2163 /*
2164 * we can't truncate inline items that have had
2165 * special encodings
2166 */
2167 if (!del_item &&
2168 btrfs_file_extent_compression(leaf, fi) == 0 &&
2169 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2170 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2171 u32 size = new_size - found_key.offset;
2172
2173 if (root->ref_cows) {
a76a3cd4
YZ		 2174 inode_sub_bytes(inode, item_end + 1 -
2175 new_size);
e02119d5
CM		 2176 }
2177 size =
2178 btrfs_file_extent_calc_inline_size(size);
9069218d 2179 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2180 size, 1);
9069218d 2181 BUG_ON(ret);
e02119d5 2182 } else if (root->ref_cows) {
a76a3cd4
YZ		 2183 inode_sub_bytes(inode, item_end + 1 -
2184 found_key.offset);
9069218d 2185 }
39279cc3 2186 }
179e29e4 2187delete:
39279cc3 2188 if (del_item) {
85e21bac
CM		 2189 if (!pending_del_nr) {
2190 /* no pending yet, add ourselves */
2191 pending_del_slot = path->slots[0];
2192 pending_del_nr = 1;
2193 } else if (pending_del_nr &&
2194 path->slots[0] + 1 == pending_del_slot) {
2195 /* hop on the pending chunk */
2196 pending_del_nr++;
2197 pending_del_slot = path->slots[0];
2198 } else {
2199 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2200 }
39279cc3
CM		 2201 } else {
2202 break;
2203 }
39279cc3
CM		 2204 if (found_extent) {
2205 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2206 extent_num_bytes,
31840ae1 2207 leaf->start, root_owner,
3bb1a1bc 2208 root_gen, inode->i_ino, 0);
39279cc3
CM		 2209 BUG_ON(ret);
2210 }
85e21bac
CM		 2211next:
2212 if (path->slots[0] == 0) {
2213 if (pending_del_nr)
2214 goto del_pending;
2215 btrfs_release_path(root, path);
2216 goto search_again;
2217 }
2218
2219 path->slots[0]--;
2220 if (pending_del_nr &&
2221 path->slots[0] + 1 != pending_del_slot) {
2222 struct btrfs_key debug;
2223del_pending:
2224 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2225 pending_del_slot);
2226 ret = btrfs_del_items(trans, root, path,
2227 pending_del_slot,
2228 pending_del_nr);
2229 BUG_ON(ret);
2230 pending_del_nr = 0;
2231 btrfs_release_path(root, path);
2232 goto search_again;
2233 }
39279cc3
CM		 2234 }
2235 ret = 0;
2236error:
85e21bac
CM		 2237 if (pending_del_nr) {
2238 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2239 pending_del_nr);
2240 }
39279cc3
CM		 2241 btrfs_free_path(path);
2242 inode->i_sb->s_dirt = 1;
2243 return ret;
2244}
2245
2246/*
2247 * taken from block_truncate_page, but does cow as it zeros out
2248 * any bytes left in the last page in the file.
2249 */
2250static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2251{
2252 struct inode *inode = mapping->host;
db94535d 2253 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2254 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2255 struct btrfs_ordered_extent *ordered;
2256 char *kaddr;
db94535d 2257 u32 blocksize = root->sectorsize;
39279cc3
CM		 2258 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2259 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2260 struct page *page;
39279cc3 2261 int ret = 0;
a52d9a80 2262 u64 page_start;
e6dcd2dc 2263 u64 page_end;
39279cc3
CM		 2264
2265 if ((offset & (blocksize - 1)) == 0)
2266 goto out;
2267
2268 ret = -ENOMEM;
211c17f5 2269again:
39279cc3
CM		 2270 page = grab_cache_page(mapping, index);
2271 if (!page)
2272 goto out;
e6dcd2dc
CM		 2273
2274 page_start = page_offset(page);
2275 page_end = page_start + PAGE_CACHE_SIZE - 1;
2276
39279cc3 2277 if (!PageUptodate(page)) {
9ebefb18 2278 ret = btrfs_readpage(NULL, page);
39279cc3 2279 lock_page(page);
211c17f5
CM		 2280 if (page->mapping != mapping) {
2281 unlock_page(page);
2282 page_cache_release(page);
2283 goto again;
2284 }
39279cc3
CM		 2285 if (!PageUptodate(page)) {
2286 ret = -EIO;
89642229 2287 goto out_unlock;
39279cc3
CM		 2288 }
2289 }
211c17f5 2290 wait_on_page_writeback(page);
e6dcd2dc
CM		 2291
2292 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2293 set_page_extent_mapped(page);
2294
2295 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2296 if (ordered) {
2297 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2298 unlock_page(page);
2299 page_cache_release(page);
eb84ae03 2300 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2301 btrfs_put_ordered_extent(ordered);
2302 goto again;
2303 }
2304
ea8c2819 2305 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2306 ret = 0;
2307 if (offset != PAGE_CACHE_SIZE) {
2308 kaddr = kmap(page);
2309 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2310 flush_dcache_page(page);
2311 kunmap(page);
2312 }
247e743c 2313 ClearPageChecked(page);
e6dcd2dc
CM		 2314 set_page_dirty(page);
2315 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2316
89642229 2317out_unlock:
39279cc3
CM		 2318 unlock_page(page);
2319 page_cache_release(page);
2320out:
2321 return ret;
2322}
2323
9036c102 2324int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2325{
9036c102
YZ		 2326 struct btrfs_trans_handle *trans;
2327 struct btrfs_root *root = BTRFS_I(inode)->root;
2328 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2329 struct extent_map *em;
2330 u64 mask = root->sectorsize - 1;
2331 u64 hole_start = (inode->i_size + mask) & ~mask;
2332 u64 block_end = (size + mask) & ~mask;
2333 u64 last_byte;
2334 u64 cur_offset;
2335 u64 hole_size;
39279cc3
CM		 2336 int err;
2337
9036c102
YZ		 2338 if (size <= hole_start)
2339 return 0;
2340
2341 err = btrfs_check_free_space(root, 1, 0);
39279cc3
CM		 2342 if (err)
2343 return err;
2344
9036c102 2345 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2346
9036c102
YZ		 2347 while (1) {
2348 struct btrfs_ordered_extent *ordered;
2349 btrfs_wait_ordered_range(inode, hole_start,
2350 block_end - hole_start);
2351 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2352 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2353 if (!ordered)
2354 break;
2355 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2356 btrfs_put_ordered_extent(ordered);
2357 }
39279cc3 2358
9036c102
YZ		 2359 trans = btrfs_start_transaction(root, 1);
2360 btrfs_set_trans_block_group(trans, inode);
39279cc3 2361
9036c102
YZ		 2362 cur_offset = hole_start;
2363 while (1) {
2364 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2365 block_end - cur_offset, 0);
2366 BUG_ON(IS_ERR(em) || !em);
2367 last_byte = min(extent_map_end(em), block_end);
2368 last_byte = (last_byte + mask) & ~mask;
2369 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2370 hole_size = last_byte - cur_offset;
2371 err = btrfs_insert_file_extent(trans, root,
2372 inode->i_ino, cur_offset, 0,
2373 0, hole_size, 0, hole_size,
2374 0, 0, 0);
2375 btrfs_drop_extent_cache(inode, hole_start,
2376 last_byte - 1, 0);
2377 }
2378 free_extent_map(em);
2379 cur_offset = last_byte;
2380 if (err || cur_offset >= block_end)
2381 break;
2382 }
1832a6d5 2383
9036c102
YZ		 2384 btrfs_end_transaction(trans, root);
2385 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2386 return err;
2387}
39279cc3 2388
9036c102
YZ		 2389static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2390{
2391 struct inode *inode = dentry->d_inode;
2392 int err;
39279cc3 2393
9036c102
YZ		 2394 err = inode_change_ok(inode, attr);
2395 if (err)
2396 return err;
2bf5a725 2397
9036c102
YZ		 2398 if (S_ISREG(inode->i_mode) &&
2399 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2400 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2401 if (err)
2402 return err;
39279cc3 2403 }
9036c102 2404
39279cc3 2405 err = inode_setattr(inode, attr);
33268eaf
JB		 2406
2407 if (!err && ((attr->ia_valid & ATTR_MODE)))
2408 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2409 return err;
2410}
61295eb8 2411
39279cc3
CM		 2412void btrfs_delete_inode(struct inode *inode)
2413{
2414 struct btrfs_trans_handle *trans;
2415 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2416 unsigned long nr;
39279cc3
CM		 2417 int ret;
2418
2419 truncate_inode_pages(&inode->i_data, 0);
2420 if (is_bad_inode(inode)) {
7b128766 2421 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2422 goto no_delete;
2423 }
4a096752 2424 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2425
dbe674a9 2426 btrfs_i_size_write(inode, 0);
39279cc3 2427 trans = btrfs_start_transaction(root, 1);
5f39d397 2428
39279cc3 2429 btrfs_set_trans_block_group(trans, inode);
e02119d5 2430 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2431 if (ret) {
2432 btrfs_orphan_del(NULL, inode);
54aa1f4d 2433 goto no_delete_lock;
7b128766
JB		 2434 }
2435
2436 btrfs_orphan_del(trans, inode);
85e21bac 2437
d3c2fdcf 2438 nr = trans->blocks_used;
85e21bac 2439 clear_inode(inode);
5f39d397 2440
39279cc3 2441 btrfs_end_transaction(trans, root);
d3c2fdcf 2442 btrfs_btree_balance_dirty(root, nr);
39279cc3 2443 return;
54aa1f4d
CM		 2444
2445no_delete_lock:
d3c2fdcf 2446 nr = trans->blocks_used;
54aa1f4d 2447 btrfs_end_transaction(trans, root);
d3c2fdcf 2448 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2449no_delete:
2450 clear_inode(inode);
2451}
2452
2453/*
2454 * this returns the key found in the dir entry in the location pointer.
2455 * If no dir entries were found, location->objectid is 0.
2456 */
2457static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2458 struct btrfs_key *location)
2459{
2460 const char *name = dentry->d_name.name;
2461 int namelen = dentry->d_name.len;
2462 struct btrfs_dir_item *di;
2463 struct btrfs_path *path;
2464 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2465 int ret = 0;
39279cc3
CM		 2466
2467 path = btrfs_alloc_path();
2468 BUG_ON(!path);
3954401f 2469
39279cc3
CM		 2470 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2471 namelen, 0);
0d9f7f3e
Y		 2472 if (IS_ERR(di))
2473 ret = PTR_ERR(di);
39279cc3 2474 if (!di || IS_ERR(di)) {
3954401f 2475 goto out_err;
39279cc3 2476 }
5f39d397 2477 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2478out:
39279cc3
CM		 2479 btrfs_free_path(path);
2480 return ret;
3954401f
CM		 2481out_err:
2482 location->objectid = 0;
2483 goto out;
39279cc3
CM		 2484}
2485
2486/*
2487 * when we hit a tree root in a directory, the btrfs part of the inode
2488 * needs to be changed to reflect the root directory of the tree root. This
2489 * is kind of like crossing a mount point.
2490 */
2491static int fixup_tree_root_location(struct btrfs_root *root,
2492 struct btrfs_key *location,
58176a96
JB		 2493 struct btrfs_root **sub_root,
2494 struct dentry *dentry)
39279cc3 2495{
39279cc3
CM		 2496 struct btrfs_root_item *ri;
2497
2498 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2499 return 0;
2500 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2501 return 0;
2502
58176a96
JB		 2503 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2504 dentry->d_name.name,
2505 dentry->d_name.len);
39279cc3
CM		 2506 if (IS_ERR(*sub_root))
2507 return PTR_ERR(*sub_root);
2508
2509 ri = &(*sub_root)->root_item;
2510 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2511 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2512 location->offset = 0;
2513
39279cc3
CM		 2514 return 0;
2515}
2516
e02119d5 2517static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2518{
e02119d5
CM		 2519 struct btrfs_inode *bi = BTRFS_I(inode);
2520
2521 bi->i_acl = NULL;
2522 bi->i_default_acl = NULL;
2523
2524 bi->generation = 0;
2525 bi->last_trans = 0;
2526 bi->logged_trans = 0;
2527 bi->delalloc_bytes = 0;
2528 bi->disk_i_size = 0;
2529 bi->flags = 0;
2530 bi->index_cnt = (u64)-1;
49eb7e46 2531 bi->log_dirty_trans = 0;
d1310b2e
CM		 2532 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2533 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2534 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2535 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2536 inode->i_mapping, GFP_NOFS);
ea8c2819 2537 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2538 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1b1e2135 2539 mutex_init(&BTRFS_I(inode)->csum_mutex);
ee6e6504 2540 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2541 mutex_init(&BTRFS_I(inode)->log_mutex);
2542}
2543
2544static int btrfs_init_locked_inode(struct inode *inode, void *p)
2545{
2546 struct btrfs_iget_args *args = p;
2547 inode->i_ino = args->ino;
2548 init_btrfs_i(inode);
2549 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 2550 return 0;
2551}
2552
2553static int btrfs_find_actor(struct inode *inode, void *opaque)
2554{
2555 struct btrfs_iget_args *args = opaque;
2556 return (args->ino == inode->i_ino &&
2557 args->root == BTRFS_I(inode)->root);
2558}
2559
5b21f2ed
ZY		 2560struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2561 struct btrfs_root *root, int wait)
2562{
2563 struct inode *inode;
2564 struct btrfs_iget_args args;
2565 args.ino = objectid;
2566 args.root = root;
2567
2568 if (wait) {
2569 inode = ilookup5(s, objectid, btrfs_find_actor,
2570 (void *)&args);
2571 } else {
2572 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2573 (void *)&args);
2574 }
2575 return inode;
2576}
2577
39279cc3
CM		 2578struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
2579 struct btrfs_root *root)
2580{
2581 struct inode *inode;
2582 struct btrfs_iget_args args;
2583 args.ino = objectid;
2584 args.root = root;
2585
2586 inode = iget5_locked(s, objectid, btrfs_find_actor,
2587 btrfs_init_locked_inode,
2588 (void *)&args);
2589 return inode;
2590}
2591
1a54ef8c
BR		 2592/* Get an inode object given its location and corresponding root.
2593 * Returns in *is_new if the inode was read from disk
2594 */
2595struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
2596 struct btrfs_root *root, int *is_new)
2597{
2598 struct inode *inode;
2599
2600 inode = btrfs_iget_locked(s, location->objectid, root);
2601 if (!inode)
2602 return ERR_PTR(-EACCES);
2603
2604 if (inode->i_state & I_NEW) {
2605 BTRFS_I(inode)->root = root;
2606 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
2607 btrfs_read_locked_inode(inode);
2608 unlock_new_inode(inode);
2609 if (is_new)
2610 *is_new = 1;
2611 } else {
2612 if (is_new)
2613 *is_new = 0;
2614 }
2615
2616 return inode;
2617}
2618
39279cc3
CM		 2619static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
2620 struct nameidata *nd)
2621{
2622 struct inode * inode;
2623 struct btrfs_inode *bi = BTRFS_I(dir);
2624 struct btrfs_root *root = bi->root;
2625 struct btrfs_root *sub_root = root;
2626 struct btrfs_key location;
1a54ef8c 2627 int ret, new, do_orphan = 0;
39279cc3
CM		 2628
2629 if (dentry->d_name.len > BTRFS_NAME_LEN)
2630 return ERR_PTR(-ENAMETOOLONG);
5f39d397 2631
39279cc3 2632 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 2633
39279cc3
CM		 2634 if (ret < 0)
2635 return ERR_PTR(ret);
5f39d397 2636
39279cc3
CM		 2637 inode = NULL;
2638 if (location.objectid) {
58176a96
JB		 2639 ret = fixup_tree_root_location(root, &location, &sub_root,
2640 dentry);
39279cc3
CM		 2641 if (ret < 0)
2642 return ERR_PTR(ret);
2643 if (ret > 0)
2644 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 2645 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
2646 if (IS_ERR(inode))
2647 return ERR_CAST(inode);
2648
2649 /* the inode and parent dir are two different roots */
2650 if (new && root != sub_root) {
2651 igrab(inode);
2652 sub_root->inode = inode;
2653 do_orphan = 1;
39279cc3
CM		 2654 }
2655 }
7b128766
JB		 2656
2657 if (unlikely(do_orphan))
2658 btrfs_orphan_cleanup(sub_root);
2659
39279cc3
CM		 2660 return d_splice_alias(inode, dentry);
2661}
2662
39279cc3
CM		 2663static unsigned char btrfs_filetype_table[] = {
2664 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
2665};
2666
cbdf5a24
DW		 2667static int btrfs_real_readdir(struct file *filp, void *dirent,
2668 filldir_t filldir)
39279cc3 2669{
6da6abae 2670 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 2671 struct btrfs_root *root = BTRFS_I(inode)->root;
2672 struct btrfs_item *item;
2673 struct btrfs_dir_item *di;
2674 struct btrfs_key key;
5f39d397 2675 struct btrfs_key found_key;
39279cc3
CM		 2676 struct btrfs_path *path;
2677 int ret;
2678 u32 nritems;
5f39d397 2679 struct extent_buffer *leaf;
39279cc3
CM		 2680 int slot;
2681 int advance;
2682 unsigned char d_type;
2683 int over = 0;
2684 u32 di_cur;
2685 u32 di_total;
2686 u32 di_len;
2687 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 2688 char tmp_name[32];
2689 char *name_ptr;
2690 int name_len;
39279cc3
CM		 2691
2692 /* FIXME, use a real flag for deciding about the key type */
2693 if (root->fs_info->tree_root == root)
2694 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 2695
3954401f
CM		 2696 /* special case for "." */
2697 if (filp->f_pos == 0) {
2698 over = filldir(dirent, ".", 1,
2699 1, inode->i_ino,
2700 DT_DIR);
2701 if (over)
2702 return 0;
2703 filp->f_pos = 1;
2704 }
3954401f
CM		 2705 /* special case for .., just use the back ref */
2706 if (filp->f_pos == 1) {
5ecc7e5d 2707 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 2708 over = filldir(dirent, "..", 2,
5ecc7e5d 2709 2, pino, DT_DIR);
3954401f 2710 if (over)
49593bfa 2711 return 0;
3954401f
CM		 2712 filp->f_pos = 2;
2713 }
2714
49593bfa
DW		 2715 path = btrfs_alloc_path();
2716 path->reada = 2;
2717
39279cc3
CM		 2718 btrfs_set_key_type(&key, key_type);
2719 key.offset = filp->f_pos;
49593bfa 2720 key.objectid = inode->i_ino;
5f39d397 2721
39279cc3
CM		 2722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2723 if (ret < 0)
2724 goto err;
2725 advance = 0;
49593bfa
DW		 2726
2727 while (1) {
5f39d397
CM		 2728 leaf = path->nodes[0];
2729 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 2730 slot = path->slots[0];
2731 if (advance || slot >= nritems) {
49593bfa 2732 if (slot >= nritems - 1) {
39279cc3
CM		 2733 ret = btrfs_next_leaf(root, path);
2734 if (ret)
2735 break;
5f39d397
CM		 2736 leaf = path->nodes[0];
2737 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 2738 slot = path->slots[0];
2739 } else {
2740 slot++;
2741 path->slots[0]++;
2742 }
2743 }
2744 advance = 1;
5f39d397
CM		 2745 item = btrfs_item_nr(leaf, slot);
2746 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2747
2748 if (found_key.objectid != key.objectid)
39279cc3 2749 break;
5f39d397 2750 if (btrfs_key_type(&found_key) != key_type)
39279cc3 2751 break;
5f39d397 2752 if (found_key.offset < filp->f_pos)
39279cc3 2753 continue;
5f39d397
CM		 2754
2755 filp->f_pos = found_key.offset;
49593bfa 2756
39279cc3
CM		 2757 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2758 di_cur = 0;
5f39d397 2759 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 2760
2761 while (di_cur < di_total) {
5f39d397
CM		 2762 struct btrfs_key location;
2763
2764 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 2765 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 2766 name_ptr = tmp_name;
2767 } else {
2768 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 2769 if (!name_ptr) {
2770 ret = -ENOMEM;
2771 goto err;
2772 }
5f39d397
CM		 2773 }
2774 read_extent_buffer(leaf, name_ptr,
2775 (unsigned long)(di + 1), name_len);
2776
2777 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2778 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5f39d397 2779 over = filldir(dirent, name_ptr, name_len,
49593bfa 2780 found_key.offset, location.objectid,
39279cc3 2781 d_type);
5f39d397
CM		 2782
2783 if (name_ptr != tmp_name)
2784 kfree(name_ptr);
2785
39279cc3
CM		 2786 if (over)
2787 goto nopos;
49593bfa 2788
5103e947 2789 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 2790 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 2791 di_cur += di_len;
2792 di = (struct btrfs_dir_item *)((char *)di + di_len);
2793 }
2794 }
49593bfa
DW		 2795
2796 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07
YZ		 2797 if (key_type == BTRFS_DIR_INDEX_KEY)
2798 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2799 else
2800 filp->f_pos++;
39279cc3
CM		 2801nopos:
2802 ret = 0;
2803err:
39279cc3 2804 btrfs_free_path(path);
39279cc3
CM		 2805 return ret;
2806}
2807
2808int btrfs_write_inode(struct inode *inode, int wait)
2809{
2810 struct btrfs_root *root = BTRFS_I(inode)->root;
2811 struct btrfs_trans_handle *trans;
2812 int ret = 0;
2813
4ca8b41e
CM		 2814 if (root->fs_info->closing > 1)
2815 return 0;
2816
39279cc3 2817 if (wait) {
f9295749 2818 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 2819 btrfs_set_trans_block_group(trans, inode);
2820 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 2821 }
2822 return ret;
2823}
2824
2825/*
54aa1f4d 2826 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 2827 * inode changes. But, it is most likely to find the inode in cache.
2828 * FIXME, needs more benchmarking...there are no reasons other than performance
2829 * to keep or drop this code.
2830 */
2831void btrfs_dirty_inode(struct inode *inode)
2832{
2833 struct btrfs_root *root = BTRFS_I(inode)->root;
2834 struct btrfs_trans_handle *trans;
2835
f9295749 2836 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 2837 btrfs_set_trans_block_group(trans, inode);
2838 btrfs_update_inode(trans, root, inode);
2839 btrfs_end_transaction(trans, root);
39279cc3
CM		 2840}
2841
d352ac68
CM		 2842/*
2843 * find the highest existing sequence number in a directory
2844 * and then set the in-memory index_cnt variable to reflect
2845 * free sequence numbers
2846 */
aec7477b
JB		 2847static int btrfs_set_inode_index_count(struct inode *inode)
2848{
2849 struct btrfs_root *root = BTRFS_I(inode)->root;
2850 struct btrfs_key key, found_key;
2851 struct btrfs_path *path;
2852 struct extent_buffer *leaf;
2853 int ret;
2854
2855 key.objectid = inode->i_ino;
2856 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2857 key.offset = (u64)-1;
2858
2859 path = btrfs_alloc_path();
2860 if (!path)
2861 return -ENOMEM;
2862
2863 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2864 if (ret < 0)
2865 goto out;
2866 /* FIXME: we should be able to handle this */
2867 if (ret == 0)
2868 goto out;
2869 ret = 0;
2870
2871 /*
2872 * MAGIC NUMBER EXPLANATION:
2873 * since we search a directory based on f_pos we have to start at 2
2874 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2875 * else has to start at 2
2876 */
2877 if (path->slots[0] == 0) {
2878 BTRFS_I(inode)->index_cnt = 2;
2879 goto out;
2880 }
2881
2882 path->slots[0]--;
2883
2884 leaf = path->nodes[0];
2885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2886
2887 if (found_key.objectid != inode->i_ino ||
2888 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2889 BTRFS_I(inode)->index_cnt = 2;
2890 goto out;
2891 }
2892
2893 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2894out:
2895 btrfs_free_path(path);
2896 return ret;
2897}
2898
d352ac68
CM		 2899/*
2900 * helper to find a free sequence number in a given directory. This current
2901 * code is very simple, later versions will do smarter things in the btree
2902 */
00e4e6b3
CM		 2903static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2904 u64 *index)
aec7477b
JB		 2905{
2906 int ret = 0;
2907
2908 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2909 ret = btrfs_set_inode_index_count(dir);
8d5bf1cb 2910 if (ret) {
aec7477b 2911 return ret;
8d5bf1cb 2912 }
aec7477b
JB		 2913 }
2914
00e4e6b3 2915 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 2916 BTRFS_I(dir)->index_cnt++;
2917
2918 return ret;
2919}
2920
39279cc3
CM		 2921static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2922 struct btrfs_root *root,
aec7477b 2923 struct inode *dir,
9c58309d
CM		 2924 const char *name, int name_len,
2925 u64 ref_objectid,
39279cc3
CM		 2926 u64 objectid,
2927 struct btrfs_block_group_cache *group,
00e4e6b3 2928 int mode, u64 *index)
39279cc3
CM		 2929{
2930 struct inode *inode;
5f39d397 2931 struct btrfs_inode_item *inode_item;
6324fbf3 2932 struct btrfs_block_group_cache *new_inode_group;
39279cc3 2933 struct btrfs_key *location;
5f39d397 2934 struct btrfs_path *path;
9c58309d
CM		 2935 struct btrfs_inode_ref *ref;
2936 struct btrfs_key key[2];
2937 u32 sizes[2];
2938 unsigned long ptr;
39279cc3
CM		 2939 int ret;
2940 int owner;
2941
5f39d397
CM		 2942 path = btrfs_alloc_path();
2943 BUG_ON(!path);
2944
39279cc3
CM		 2945 inode = new_inode(root->fs_info->sb);
2946 if (!inode)
2947 return ERR_PTR(-ENOMEM);
2948
aec7477b 2949 if (dir) {
00e4e6b3 2950 ret = btrfs_set_inode_index(dir, inode, index);
aec7477b
JB		 2951 if (ret)
2952 return ERR_PTR(ret);
aec7477b
JB		 2953 }
2954 /*
2955 * index_cnt is ignored for everything but a dir,
2956 * btrfs_get_inode_index_count has an explanation for the magic
2957 * number
2958 */
e02119d5 2959 init_btrfs_i(inode);
aec7477b 2960 BTRFS_I(inode)->index_cnt = 2;
39279cc3 2961 BTRFS_I(inode)->root = root;
e02119d5 2962 BTRFS_I(inode)->generation = trans->transid;
b888db2b 2963
39279cc3
CM		 2964 if (mode & S_IFDIR)
2965 owner = 0;
2966 else
2967 owner = 1;
6324fbf3 2968 new_inode_group = btrfs_find_block_group(root, group, 0,
0b86a832 2969 BTRFS_BLOCK_GROUP_METADATA, owner);
6324fbf3
CM		 2970 if (!new_inode_group) {
2971 printk("find_block group failed\n");
2972 new_inode_group = group;
2973 }
2974 BTRFS_I(inode)->block_group = new_inode_group;
9c58309d
CM		 2975
2976 key[0].objectid = objectid;
2977 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2978 key[0].offset = 0;
2979
2980 key[1].objectid = objectid;
2981 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2982 key[1].offset = ref_objectid;
2983
2984 sizes[0] = sizeof(struct btrfs_inode_item);
2985 sizes[1] = name_len + sizeof(*ref);
2986
2987 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2988 if (ret != 0)
5f39d397
CM		 2989 goto fail;
2990
9c58309d
CM		 2991 if (objectid > root->highest_inode)
2992 root->highest_inode = objectid;
2993
39279cc3
CM		 2994 inode->i_uid = current->fsuid;
2995 inode->i_gid = current->fsgid;
2996 inode->i_mode = mode;
2997 inode->i_ino = objectid;
a76a3cd4 2998 inode_set_bytes(inode, 0);
39279cc3 2999 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3000 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3001 struct btrfs_inode_item);
e02119d5 3002 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3003
3004 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3005 struct btrfs_inode_ref);
3006 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3007 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3008 ptr = (unsigned long)(ref + 1);
3009 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3010
5f39d397
CM		 3011 btrfs_mark_buffer_dirty(path->nodes[0]);
3012 btrfs_free_path(path);
3013
39279cc3
CM		 3014 location = &BTRFS_I(inode)->location;
3015 location->objectid = objectid;
39279cc3
CM		 3016 location->offset = 0;
3017 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3018
39279cc3
CM		 3019 insert_inode_hash(inode);
3020 return inode;
5f39d397 3021fail:
aec7477b
JB		 3022 if (dir)
3023 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3024 btrfs_free_path(path);
3025 return ERR_PTR(ret);
39279cc3
CM		 3026}
3027
3028static inline u8 btrfs_inode_type(struct inode *inode)
3029{
3030 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3031}
3032
d352ac68
CM		 3033/*
3034 * utility function to add 'inode' into 'parent_inode' with
3035 * a give name and a given sequence number.
3036 * if 'add_backref' is true, also insert a backref from the
3037 * inode to the parent directory.
3038 */
e02119d5
CM		 3039int btrfs_add_link(struct btrfs_trans_handle *trans,
3040 struct inode *parent_inode, struct inode *inode,
3041 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3042{
3043 int ret;
3044 struct btrfs_key key;
e02119d5 3045 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3046
39279cc3 3047 key.objectid = inode->i_ino;
39279cc3
CM		 3048 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3049 key.offset = 0;
3050
e02119d5
CM		 3051 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3052 parent_inode->i_ino,
aec7477b 3053 &key, btrfs_inode_type(inode),
00e4e6b3 3054 index);
39279cc3 3055 if (ret == 0) {
9c58309d
CM		 3056 if (add_backref) {
3057 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3058 name, name_len,
3059 inode->i_ino,
3060 parent_inode->i_ino,
3061 index);
9c58309d 3062 }
dbe674a9 3063 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3064 name_len * 2);
79c44584 3065 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3066 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3067 }
3068 return ret;
3069}
3070
3071static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3072 struct dentry *dentry, struct inode *inode,
00e4e6b3 3073 int backref, u64 index)
39279cc3 3074{
e02119d5
CM		 3075 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3076 inode, dentry->d_name.name,
3077 dentry->d_name.len, backref, index);
39279cc3
CM		 3078 if (!err) {
3079 d_instantiate(dentry, inode);
3080 return 0;
3081 }
3082 if (err > 0)
3083 err = -EEXIST;
3084 return err;
3085}
3086
618e21d5
JB		 3087static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3088 int mode, dev_t rdev)
3089{
3090 struct btrfs_trans_handle *trans;
3091 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3092 struct inode *inode = NULL;
618e21d5
JB		 3093 int err;
3094 int drop_inode = 0;
3095 u64 objectid;
1832a6d5 3096 unsigned long nr = 0;
00e4e6b3 3097 u64 index = 0;
618e21d5
JB		 3098
3099 if (!new_valid_dev(rdev))
3100 return -EINVAL;
3101
1832a6d5
CM		 3102 err = btrfs_check_free_space(root, 1, 0);
3103 if (err)
3104 goto fail;
3105
618e21d5
JB		 3106 trans = btrfs_start_transaction(root, 1);
3107 btrfs_set_trans_block_group(trans, dir);
3108
3109 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3110 if (err) {
3111 err = -ENOSPC;
3112 goto out_unlock;
3113 }
3114
aec7477b 3115 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3116 dentry->d_name.len,
3117 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3118 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3119 err = PTR_ERR(inode);
3120 if (IS_ERR(inode))
3121 goto out_unlock;
3122
33268eaf
JB		 3123 err = btrfs_init_acl(inode, dir);
3124 if (err) {
3125 drop_inode = 1;
3126 goto out_unlock;
3127 }
3128
618e21d5 3129 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3130 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3131 if (err)
3132 drop_inode = 1;
3133 else {
3134 inode->i_op = &btrfs_special_inode_operations;
3135 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3136 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3137 }
3138 dir->i_sb->s_dirt = 1;
3139 btrfs_update_inode_block_group(trans, inode);
3140 btrfs_update_inode_block_group(trans, dir);
3141out_unlock:
d3c2fdcf 3142 nr = trans->blocks_used;
89ce8a63 3143 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3144fail:
618e21d5
JB		 3145 if (drop_inode) {
3146 inode_dec_link_count(inode);
3147 iput(inode);
3148 }
d3c2fdcf 3149 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3150 return err;
3151}
3152
39279cc3
CM		 3153static int btrfs_create(struct inode *dir, struct dentry *dentry,
3154 int mode, struct nameidata *nd)
3155{
3156 struct btrfs_trans_handle *trans;
3157 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3158 struct inode *inode = NULL;
39279cc3
CM		 3159 int err;
3160 int drop_inode = 0;
1832a6d5 3161 unsigned long nr = 0;
39279cc3 3162 u64 objectid;
00e4e6b3 3163 u64 index = 0;
39279cc3 3164
1832a6d5
CM		 3165 err = btrfs_check_free_space(root, 1, 0);
3166 if (err)
3167 goto fail;
39279cc3
CM		 3168 trans = btrfs_start_transaction(root, 1);
3169 btrfs_set_trans_block_group(trans, dir);
3170
3171 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3172 if (err) {
3173 err = -ENOSPC;
3174 goto out_unlock;
3175 }
3176
aec7477b 3177 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3178 dentry->d_name.len,
3179 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3180 objectid, BTRFS_I(dir)->block_group, mode,
3181 &index);
39279cc3
CM		 3182 err = PTR_ERR(inode);
3183 if (IS_ERR(inode))
3184 goto out_unlock;
3185
33268eaf
JB		 3186 err = btrfs_init_acl(inode, dir);
3187 if (err) {
3188 drop_inode = 1;
3189 goto out_unlock;
3190 }
3191
39279cc3 3192 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3193 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3194 if (err)
3195 drop_inode = 1;
3196 else {
3197 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3198 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3199 inode->i_fop = &btrfs_file_operations;
3200 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3201 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3202 }
3203 dir->i_sb->s_dirt = 1;
3204 btrfs_update_inode_block_group(trans, inode);
3205 btrfs_update_inode_block_group(trans, dir);
3206out_unlock:
d3c2fdcf 3207 nr = trans->blocks_used;
ab78c84d 3208 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3209fail:
39279cc3
CM		 3210 if (drop_inode) {
3211 inode_dec_link_count(inode);
3212 iput(inode);
3213 }
d3c2fdcf 3214 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3215 return err;
3216}
3217
3218static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3219 struct dentry *dentry)
3220{
3221 struct btrfs_trans_handle *trans;
3222 struct btrfs_root *root = BTRFS_I(dir)->root;
3223 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3224 u64 index;
1832a6d5 3225 unsigned long nr = 0;
39279cc3
CM		 3226 int err;
3227 int drop_inode = 0;
3228
3229 if (inode->i_nlink == 0)
3230 return -ENOENT;
3231
e02119d5 3232 btrfs_inc_nlink(inode);
1832a6d5
CM		 3233 err = btrfs_check_free_space(root, 1, 0);
3234 if (err)
3235 goto fail;
00e4e6b3 3236 err = btrfs_set_inode_index(dir, inode, &index);
aec7477b
JB		 3237 if (err)
3238 goto fail;
3239
39279cc3 3240 trans = btrfs_start_transaction(root, 1);
5f39d397 3241
39279cc3
CM		 3242 btrfs_set_trans_block_group(trans, dir);
3243 atomic_inc(&inode->i_count);
aec7477b 3244
00e4e6b3 3245 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3246
39279cc3
CM		 3247 if (err)
3248 drop_inode = 1;
5f39d397 3249
39279cc3
CM		 3250 dir->i_sb->s_dirt = 1;
3251 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3252 err = btrfs_update_inode(trans, root, inode);
5f39d397 3253
54aa1f4d
CM		 3254 if (err)
3255 drop_inode = 1;
39279cc3 3256
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
39279cc3
CM		 3268static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3269{
b9d86667 3270 struct inode *inode = NULL;
39279cc3
CM		 3271 struct btrfs_trans_handle *trans;
3272 struct btrfs_root *root = BTRFS_I(dir)->root;
3273 int err = 0;
3274 int drop_on_err = 0;
b9d86667 3275 u64 objectid = 0;
00e4e6b3 3276 u64 index = 0;
d3c2fdcf 3277 unsigned long nr = 1;
39279cc3 3278
1832a6d5
CM		 3279 err = btrfs_check_free_space(root, 1, 0);
3280 if (err)
3281 goto out_unlock;
3282
39279cc3
CM		 3283 trans = btrfs_start_transaction(root, 1);
3284 btrfs_set_trans_block_group(trans, dir);
5f39d397 3285
39279cc3
CM		 3286 if (IS_ERR(trans)) {
3287 err = PTR_ERR(trans);
3288 goto out_unlock;
3289 }
3290
3291 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3292 if (err) {
3293 err = -ENOSPC;
3294 goto out_unlock;
3295 }
3296
aec7477b 3297 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3298 dentry->d_name.len,
3299 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3300 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3301 &index);
39279cc3
CM		 3302 if (IS_ERR(inode)) {
3303 err = PTR_ERR(inode);
3304 goto out_fail;
3305 }
5f39d397 3306
39279cc3 3307 drop_on_err = 1;
33268eaf
JB		 3308
3309 err = btrfs_init_acl(inode, dir);
3310 if (err)
3311 goto out_fail;
3312
39279cc3
CM		 3313 inode->i_op = &btrfs_dir_inode_operations;
3314 inode->i_fop = &btrfs_dir_file_operations;
3315 btrfs_set_trans_block_group(trans, inode);
3316
dbe674a9 3317 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3318 err = btrfs_update_inode(trans, root, inode);
3319 if (err)
3320 goto out_fail;
5f39d397 3321
e02119d5
CM		 3322 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3323 inode, dentry->d_name.name,
3324 dentry->d_name.len, 0, index);
39279cc3
CM		 3325 if (err)
3326 goto out_fail;
5f39d397 3327
39279cc3
CM		 3328 d_instantiate(dentry, inode);
3329 drop_on_err = 0;
3330 dir->i_sb->s_dirt = 1;
3331 btrfs_update_inode_block_group(trans, inode);
3332 btrfs_update_inode_block_group(trans, dir);
3333
3334out_fail:
d3c2fdcf 3335 nr = trans->blocks_used;
ab78c84d 3336 btrfs_end_transaction_throttle(trans, root);
5f39d397 3337
39279cc3 3338out_unlock:
39279cc3
CM		 3339 if (drop_on_err)
3340 iput(inode);
d3c2fdcf 3341 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3342 return err;
3343}
3344
d352ac68
CM		 3345/* helper for btfs_get_extent. Given an existing extent in the tree,
3346 * and an extent that you want to insert, deal with overlap and insert
3347 * the new extent into the tree.
3348 */
3b951516
CM		 3349static int merge_extent_mapping(struct extent_map_tree *em_tree,
3350 struct extent_map *existing,
e6dcd2dc
CM		 3351 struct extent_map *em,
3352 u64 map_start, u64 map_len)
3b951516
CM		 3353{
3354 u64 start_diff;
3b951516 3355
e6dcd2dc
CM		 3356 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3357 start_diff = map_start - em->start;
3358 em->start = map_start;
3359 em->len = map_len;
c8b97818
CM		 3360 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3361 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3362 em->block_start += start_diff;
c8b97818
CM		 3363 em->block_len -= start_diff;
3364 }
e6dcd2dc 3365 return add_extent_mapping(em_tree, em);
3b951516
CM		 3366}
3367
c8b97818
CM		 3368static noinline int uncompress_inline(struct btrfs_path *path,
3369 struct inode *inode, struct page *page,
3370 size_t pg_offset, u64 extent_offset,
3371 struct btrfs_file_extent_item *item)
3372{
3373 int ret;
3374 struct extent_buffer *leaf = path->nodes[0];
3375 char *tmp;
3376 size_t max_size;
3377 unsigned long inline_size;
3378 unsigned long ptr;
3379
3380 WARN_ON(pg_offset != 0);
3381 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3382 inline_size = btrfs_file_extent_inline_item_len(leaf,
3383 btrfs_item_nr(leaf, path->slots[0]));
3384 tmp = kmalloc(inline_size, GFP_NOFS);
3385 ptr = btrfs_file_extent_inline_start(item);
3386
3387 read_extent_buffer(leaf, tmp, ptr, inline_size);
3388
3389 max_size = min(PAGE_CACHE_SIZE, max_size);
3390 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3391 inline_size, max_size);
3392 if (ret) {
3393 char *kaddr = kmap_atomic(page, KM_USER0);
3394 unsigned long copy_size = min_t(u64,
3395 PAGE_CACHE_SIZE - pg_offset,
3396 max_size - extent_offset);
3397 memset(kaddr + pg_offset, 0, copy_size);
3398 kunmap_atomic(kaddr, KM_USER0);
3399 }
3400 kfree(tmp);
3401 return 0;
3402}
3403
d352ac68
CM		 3404/*
3405 * a bit scary, this does extent mapping from logical file offset to the disk.
3406 * the ugly parts come from merging extents from the disk with the
3407 * in-ram representation. This gets more complex because of the data=ordered code,
3408 * where the in-ram extents might be locked pending data=ordered completion.
3409 *
3410 * This also copies inline extents directly into the page.
3411 */
a52d9a80 3412struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3413 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3414 int create)
3415{
3416 int ret;
3417 int err = 0;
db94535d 3418 u64 bytenr;
a52d9a80
CM		 3419 u64 extent_start = 0;
3420 u64 extent_end = 0;
3421 u64 objectid = inode->i_ino;
3422 u32 found_type;
f421950f 3423 struct btrfs_path *path = NULL;
a52d9a80
CM		 3424 struct btrfs_root *root = BTRFS_I(inode)->root;
3425 struct btrfs_file_extent_item *item;
5f39d397
CM		 3426 struct extent_buffer *leaf;
3427 struct btrfs_key found_key;
a52d9a80
CM		 3428 struct extent_map *em = NULL;
3429 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3430 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3431 struct btrfs_trans_handle *trans = NULL;
c8b97818 3432 int compressed;
a52d9a80 3433
a52d9a80 3434again:
d1310b2e
CM		 3435 spin_lock(&em_tree->lock);
3436 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3437 if (em)
3438 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3439 spin_unlock(&em_tree->lock);
3440
a52d9a80 3441 if (em) {
e1c4b745
CM		 3442 if (em->start > start || em->start + em->len <= start)
3443 free_extent_map(em);
3444 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3445 free_extent_map(em);
3446 else
3447 goto out;
a52d9a80 3448 }
d1310b2e 3449 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3450 if (!em) {
d1310b2e
CM		 3451 err = -ENOMEM;
3452 goto out;
a52d9a80 3453 }
e6dcd2dc 3454 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3455 em->start = EXTENT_MAP_HOLE;
3456 em->len = (u64)-1;
c8b97818 3457 em->block_len = (u64)-1;
f421950f
CM		 3458
3459 if (!path) {
3460 path = btrfs_alloc_path();
3461 BUG_ON(!path);
3462 }
3463
179e29e4
CM		 3464 ret = btrfs_lookup_file_extent(trans, root, path,
3465 objectid, start, trans != NULL);
a52d9a80
CM		 3466 if (ret < 0) {
3467 err = ret;
3468 goto out;
3469 }
3470
3471 if (ret != 0) {
3472 if (path->slots[0] == 0)
3473 goto not_found;
3474 path->slots[0]--;
3475 }
3476
5f39d397
CM		 3477 leaf = path->nodes[0];
3478 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3479 struct btrfs_file_extent_item);
a52d9a80 3480 /* are we inside the extent that was found? */
5f39d397
CM		 3481 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3482 found_type = btrfs_key_type(&found_key);
3483 if (found_key.objectid != objectid ||
a52d9a80
CM		 3484 found_type != BTRFS_EXTENT_DATA_KEY) {
3485 goto not_found;
3486 }
3487
5f39d397
CM		 3488 found_type = btrfs_file_extent_type(leaf, item);
3489 extent_start = found_key.offset;
c8b97818 3490 compressed = btrfs_file_extent_compression(leaf, item);
a52d9a80
CM		 3491 if (found_type == BTRFS_FILE_EXTENT_REG) {
3492 extent_end = extent_start +
db94535d 3493 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3494 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3495 size_t size;
3496 size = btrfs_file_extent_inline_len(leaf, item);
3497 extent_end = (extent_start + size + root->sectorsize - 1) &
3498 ~((u64)root->sectorsize - 1);
3499 }
3500
3501 if (start >= extent_end) {
3502 path->slots[0]++;
3503 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3504 ret = btrfs_next_leaf(root, path);
3505 if (ret < 0) {
3506 err = ret;
3507 goto out;
a52d9a80 3508 }
9036c102
YZ		 3509 if (ret > 0)
3510 goto not_found;
3511 leaf = path->nodes[0];
a52d9a80 3512 }
9036c102
YZ		 3513 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3514 if (found_key.objectid != objectid ||
3515 found_key.type != BTRFS_EXTENT_DATA_KEY)
3516 goto not_found;
3517 if (start + len <= found_key.offset)
3518 goto not_found;
3519 em->start = start;
3520 em->len = found_key.offset - start;
3521 goto not_found_em;
3522 }
3523
3524 if (found_type == BTRFS_FILE_EXTENT_REG) {
3525 em->start = extent_start;
3526 em->len = extent_end - extent_start;
db94535d
CM		 3527 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3528 if (bytenr == 0) {
5f39d397 3529 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 3530 goto insert;
3531 }
c8b97818
CM		 3532 if (compressed) {
3533 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3534 em->block_start = bytenr;
3535 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3536 item);
3537 } else {
3538 bytenr += btrfs_file_extent_offset(leaf, item);
3539 em->block_start = bytenr;
3540 em->block_len = em->len;
3541 }
a52d9a80
CM		 3542 goto insert;
3543 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3544 unsigned long ptr;
a52d9a80 3545 char *map;
3326d1b0
CM		 3546 size_t size;
3547 size_t extent_offset;
3548 size_t copy_size;
a52d9a80 3549
689f9346 3550 em->block_start = EXTENT_MAP_INLINE;
c8b97818 3551 if (!page || create) {
689f9346 3552 em->start = extent_start;
9036c102 3553 em->len = extent_end - extent_start;
689f9346
Y		 3554 goto out;
3555 }
5f39d397 3556
9036c102
YZ		 3557 size = btrfs_file_extent_inline_len(leaf, item);
3558 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 3559 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 3560 size - extent_offset);
3326d1b0 3561 em->start = extent_start + extent_offset;
70dec807
CM		 3562 em->len = (copy_size + root->sectorsize - 1) &
3563 ~((u64)root->sectorsize - 1);
c8b97818
CM		 3564 if (compressed)
3565 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 3566 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 3567 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 3568 if (btrfs_file_extent_compression(leaf, item) ==
3569 BTRFS_COMPRESS_ZLIB) {
3570 ret = uncompress_inline(path, inode, page,
3571 pg_offset,
3572 extent_offset, item);
3573 BUG_ON(ret);
3574 } else {
3575 map = kmap(page);
3576 read_extent_buffer(leaf, map + pg_offset, ptr,
3577 copy_size);
3578 kunmap(page);
3579 }
179e29e4
CM		 3580 flush_dcache_page(page);
3581 } else if (create && PageUptodate(page)) {
3582 if (!trans) {
3583 kunmap(page);
3584 free_extent_map(em);
3585 em = NULL;
3586 btrfs_release_path(root, path);
f9295749 3587 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 3588 goto again;
3589 }
c8b97818 3590 map = kmap(page);
70dec807 3591 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 3592 copy_size);
c8b97818 3593 kunmap(page);
179e29e4 3594 btrfs_mark_buffer_dirty(leaf);
a52d9a80 3595 }
d1310b2e
CM		 3596 set_extent_uptodate(io_tree, em->start,
3597 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 3598 goto insert;
3599 } else {
3600 printk("unkknown found_type %d\n", found_type);
3601 WARN_ON(1);
3602 }
3603not_found:
3604 em->start = start;
d1310b2e 3605 em->len = len;
a52d9a80 3606not_found_em:
5f39d397 3607 em->block_start = EXTENT_MAP_HOLE;
9036c102 3608 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 3609insert:
3610 btrfs_release_path(root, path);
d1310b2e
CM		 3611 if (em->start > start || extent_map_end(em) <= start) {
3612 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
a52d9a80
CM		 3613 err = -EIO;
3614 goto out;
3615 }
d1310b2e
CM		 3616
3617 err = 0;
3618 spin_lock(&em_tree->lock);
a52d9a80 3619 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 3620 /* it is possible that someone inserted the extent into the tree
3621 * while we had the lock dropped. It is also possible that
3622 * an overlapping map exists in the tree
3623 */
a52d9a80 3624 if (ret == -EEXIST) {
3b951516 3625 struct extent_map *existing;
e6dcd2dc
CM		 3626
3627 ret = 0;
3628
3b951516 3629 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 3630 if (existing && (existing->start > start ||
3631 existing->start + existing->len <= start)) {
3632 free_extent_map(existing);
3633 existing = NULL;
3634 }
3b951516
CM		 3635 if (!existing) {
3636 existing = lookup_extent_mapping(em_tree, em->start,
3637 em->len);
3638 if (existing) {
3639 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 3640 em, start,
3641 root->sectorsize);
3b951516
CM		 3642 free_extent_map(existing);
3643 if (err) {
3644 free_extent_map(em);
3645 em = NULL;
3646 }
3647 } else {
3648 err = -EIO;
3649 printk("failing to insert %Lu %Lu\n",
3650 start, len);
3651 free_extent_map(em);
3652 em = NULL;
3653 }
3654 } else {
3655 free_extent_map(em);
3656 em = existing;
e6dcd2dc 3657 err = 0;
a52d9a80 3658 }
a52d9a80 3659 }
d1310b2e 3660 spin_unlock(&em_tree->lock);
a52d9a80 3661out:
f421950f
CM		 3662 if (path)
3663 btrfs_free_path(path);
a52d9a80
CM		 3664 if (trans) {
3665 ret = btrfs_end_transaction(trans, root);
e6dcd2dc 3666 if (!err) {
a52d9a80 3667 err = ret;
e6dcd2dc 3668 }
a52d9a80 3669 }
a52d9a80
CM		 3670 if (err) {
3671 free_extent_map(em);
3672 WARN_ON(1);
3673 return ERR_PTR(err);
3674 }
3675 return em;
3676}
3677
16432985
CM		 3678static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
3679 const struct iovec *iov, loff_t offset,
3680 unsigned long nr_segs)
3681{
e1c4b745 3682 return -EINVAL;
16432985
CM		 3683}
3684
d396c6f5 3685static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
39279cc3 3686{
d396c6f5 3687 return extent_bmap(mapping, iblock, btrfs_get_extent);
39279cc3
CM		 3688}
3689
a52d9a80 3690int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 3691{
d1310b2e
CM		 3692 struct extent_io_tree *tree;
3693 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 3694 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 3695}
1832a6d5 3696
a52d9a80 3697static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 3698{
d1310b2e 3699 struct extent_io_tree *tree;
b888db2b
CM		 3700
3701
3702 if (current->flags & PF_MEMALLOC) {
3703 redirty_page_for_writepage(wbc, page);
3704 unlock_page(page);
3705 return 0;
3706 }
d1310b2e 3707 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 3708 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 3709}
3710
f421950f
CM		 3711int btrfs_writepages(struct address_space *mapping,
3712 struct writeback_control *wbc)
b293f02e 3713{
d1310b2e
CM		 3714 struct extent_io_tree *tree;
3715 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 3716 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3717}
3718
3ab2fb5a
CM		 3719static int
3720btrfs_readpages(struct file *file, struct address_space *mapping,
3721 struct list_head *pages, unsigned nr_pages)
3722{
d1310b2e
CM		 3723 struct extent_io_tree *tree;
3724 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 3725 return extent_readpages(tree, mapping, pages, nr_pages,
3726 btrfs_get_extent);
3727}
e6dcd2dc 3728static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 3729{
d1310b2e
CM		 3730 struct extent_io_tree *tree;
3731 struct extent_map_tree *map;
a52d9a80 3732 int ret;
8c2383c3 3733
d1310b2e
CM		 3734 tree = &BTRFS_I(page->mapping->host)->io_tree;
3735 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 3736 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 3737 if (ret == 1) {
3738 ClearPagePrivate(page);
3739 set_page_private(page, 0);
3740 page_cache_release(page);
39279cc3 3741 }
a52d9a80 3742 return ret;
39279cc3
CM		 3743}
3744
e6dcd2dc
CM		 3745static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3746{
98509cfc
CM		 3747 if (PageWriteback(page) || PageDirty(page))
3748 return 0;
e6dcd2dc
CM		 3749 return __btrfs_releasepage(page, gfp_flags);
3750}
3751
a52d9a80 3752static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 3753{
d1310b2e 3754 struct extent_io_tree *tree;
e6dcd2dc
CM		 3755 struct btrfs_ordered_extent *ordered;
3756 u64 page_start = page_offset(page);
3757 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 3758
e6dcd2dc 3759 wait_on_page_writeback(page);
d1310b2e 3760 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 3761 if (offset) {
3762 btrfs_releasepage(page, GFP_NOFS);
3763 return;
3764 }
3765
3766 lock_extent(tree, page_start, page_end, GFP_NOFS);
3767 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3768 page_offset(page));
3769 if (ordered) {
eb84ae03
CM		 3770 /*
3771 * IO on this page will never be started, so we need
3772 * to account for any ordered extents now
3773 */
e6dcd2dc
CM		 3774 clear_extent_bit(tree, page_start, page_end,
3775 EXTENT_DIRTY | EXTENT_DELALLOC |
3776 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 3777 btrfs_finish_ordered_io(page->mapping->host,
3778 page_start, page_end);
e6dcd2dc
CM		 3779 btrfs_put_ordered_extent(ordered);
3780 lock_extent(tree, page_start, page_end, GFP_NOFS);
3781 }
3782 clear_extent_bit(tree, page_start, page_end,
3783 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3784 EXTENT_ORDERED,
3785 1, 1, GFP_NOFS);
3786 __btrfs_releasepage(page, GFP_NOFS);
3787
4a096752 3788 ClearPageChecked(page);
9ad6b7bc 3789 if (PagePrivate(page)) {
9ad6b7bc
CM		 3790 ClearPagePrivate(page);
3791 set_page_private(page, 0);
3792 page_cache_release(page);
3793 }
39279cc3
CM		 3794}
3795
9ebefb18
CM		 3796/*
3797 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3798 * called from a page fault handler when a page is first dirtied. Hence we must
3799 * be careful to check for EOF conditions here. We set the page up correctly
3800 * for a written page which means we get ENOSPC checking when writing into
3801 * holes and correct delalloc and unwritten extent mapping on filesystems that
3802 * support these features.
3803 *
3804 * We are not allowed to take the i_mutex here so we have to play games to
3805 * protect against truncate races as the page could now be beyond EOF. Because
3806 * vmtruncate() writes the inode size before removing pages, once we have the
3807 * page lock we can determine safely if the page is beyond EOF. If it is not
3808 * beyond EOF, then the page is guaranteed safe against truncation until we
3809 * unlock the page.
3810 */
3811int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3812{
6da6abae 3813 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 3814 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3815 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3816 struct btrfs_ordered_extent *ordered;
3817 char *kaddr;
3818 unsigned long zero_start;
9ebefb18 3819 loff_t size;
1832a6d5 3820 int ret;
a52d9a80 3821 u64 page_start;
e6dcd2dc 3822 u64 page_end;
9ebefb18 3823
1832a6d5 3824 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1832a6d5
CM		 3825 if (ret)
3826 goto out;
3827
3828 ret = -EINVAL;
e6dcd2dc 3829again:
9ebefb18 3830 lock_page(page);
9ebefb18 3831 size = i_size_read(inode);
e6dcd2dc
CM		 3832 page_start = page_offset(page);
3833 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 3834
9ebefb18 3835 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 3836 (page_start >= size)) {
9ebefb18
CM		 3837 /* page got truncated out from underneath us */
3838 goto out_unlock;
3839 }
e6dcd2dc
CM		 3840 wait_on_page_writeback(page);
3841
3842 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3843 set_page_extent_mapped(page);
3844
eb84ae03
CM		 3845 /*
3846 * we can't set the delalloc bits if there are pending ordered
3847 * extents. Drop our locks and wait for them to finish
3848 */
e6dcd2dc
CM		 3849 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3850 if (ordered) {
3851 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3852 unlock_page(page);
eb84ae03 3853 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3854 btrfs_put_ordered_extent(ordered);
3855 goto again;
3856 }
3857
ea8c2819 3858 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 3859 ret = 0;
9ebefb18
CM		 3860
3861 /* page is wholly or partially inside EOF */
a52d9a80 3862 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 3863 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 3864 else
e6dcd2dc 3865 zero_start = PAGE_CACHE_SIZE;
9ebefb18 3866
e6dcd2dc
CM		 3867 if (zero_start != PAGE_CACHE_SIZE) {
3868 kaddr = kmap(page);
3869 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3870 flush_dcache_page(page);
3871 kunmap(page);
3872 }
247e743c 3873 ClearPageChecked(page);
e6dcd2dc
CM		 3874 set_page_dirty(page);
3875 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 3876
3877out_unlock:
3878 unlock_page(page);
1832a6d5 3879out:
9ebefb18
CM		 3880 return ret;
3881}
3882
39279cc3
CM		 3883static void btrfs_truncate(struct inode *inode)
3884{
3885 struct btrfs_root *root = BTRFS_I(inode)->root;
3886 int ret;
3887 struct btrfs_trans_handle *trans;
d3c2fdcf 3888 unsigned long nr;
dbe674a9 3889 u64 mask = root->sectorsize - 1;
39279cc3
CM		 3890
3891 if (!S_ISREG(inode->i_mode))
3892 return;
3893 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3894 return;
3895
3896 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 3897 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 3898
39279cc3
CM		 3899 trans = btrfs_start_transaction(root, 1);
3900 btrfs_set_trans_block_group(trans, inode);
dbe674a9 3901 btrfs_i_size_write(inode, inode->i_size);
39279cc3 3902
7b128766
JB		 3903 ret = btrfs_orphan_add(trans, inode);
3904 if (ret)
3905 goto out;
39279cc3 3906 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 3907 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 3908 BTRFS_EXTENT_DATA_KEY);
39279cc3 3909 btrfs_update_inode(trans, root, inode);
5f39d397 3910
7b128766
JB		 3911 ret = btrfs_orphan_del(trans, inode);
3912 BUG_ON(ret);
3913
3914out:
3915 nr = trans->blocks_used;
89ce8a63 3916 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 3917 BUG_ON(ret);
d3c2fdcf 3918 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3919}
3920
3b96362c
SW		 3921/*
3922 * Invalidate a single dcache entry at the root of the filesystem.
3923 * Needed after creation of snapshot or subvolume.
3924 */
3925void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3926 int namelen)
3927{
3928 struct dentry *alias, *entry;
3929 struct qstr qstr;
3930
3931 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3932 if (alias) {
3933 qstr.name = name;
3934 qstr.len = namelen;
3935 /* change me if btrfs ever gets a d_hash operation */
3936 qstr.hash = full_name_hash(qstr.name, qstr.len);
3937 entry = d_lookup(alias, &qstr);
3938 dput(alias);
3939 if (entry) {
3940 d_invalidate(entry);
3941 dput(entry);
3942 }
3943 }
3944}
3945
d352ac68
CM		 3946/*
3947 * create a new subvolume directory/inode (helper for the ioctl).
3948 */
cb8e7090 3949int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
f46b5a66
CH		 3950 struct btrfs_trans_handle *trans, u64 new_dirid,
3951 struct btrfs_block_group_cache *block_group)
39279cc3 3952{
39279cc3 3953 struct inode *inode;
cb8e7090 3954 int error;
00e4e6b3 3955 u64 index = 0;
39279cc3 3956
aec7477b 3957 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
00e4e6b3 3958 new_dirid, block_group, S_IFDIR | 0700, &index);
54aa1f4d 3959 if (IS_ERR(inode))
f46b5a66 3960 return PTR_ERR(inode);
39279cc3
CM		 3961 inode->i_op = &btrfs_dir_inode_operations;
3962 inode->i_fop = &btrfs_dir_file_operations;
34088780 3963 new_root->inode = inode;
39279cc3 3964
39279cc3 3965 inode->i_nlink = 1;
dbe674a9 3966 btrfs_i_size_write(inode, 0);
3b96362c 3967
cb8e7090
CH		 3968 error = btrfs_update_inode(trans, new_root, inode);
3969 if (error)
3970 return error;
3971
3972 d_instantiate(dentry, inode);
3973 return 0;
39279cc3
CM		 3974}
3975
d352ac68
CM		 3976/* helper function for file defrag and space balancing. This
3977 * forces readahead on a given range of bytes in an inode
3978 */
edbd8d4e 3979unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 3980 struct file_ra_state *ra, struct file *file,
3981 pgoff_t offset, pgoff_t last_index)
3982{
8e7bf94f 3983 pgoff_t req_size = last_index - offset + 1;
86479a04 3984
86479a04
CM		 3985 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3986 return offset + req_size;
86479a04
CM		 3987}
3988
39279cc3
CM		 3989struct inode *btrfs_alloc_inode(struct super_block *sb)
3990{
3991 struct btrfs_inode *ei;
3992
3993 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3994 if (!ei)
3995 return NULL;
15ee9bc7 3996 ei->last_trans = 0;
e02119d5 3997 ei->logged_trans = 0;
e6dcd2dc 3998 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 3999 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4000 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4001 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4002 return &ei->vfs_inode;
4003}
4004
4005void btrfs_destroy_inode(struct inode *inode)
4006{
e6dcd2dc 4007 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4008 WARN_ON(!list_empty(&inode->i_dentry));
4009 WARN_ON(inode->i_data.nrpages);
4010
33268eaf
JB		 4011 if (BTRFS_I(inode)->i_acl &&
4012 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4013 posix_acl_release(BTRFS_I(inode)->i_acl);
4014 if (BTRFS_I(inode)->i_default_acl &&
4015 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4016 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4017
bcc63abb 4018 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4019 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4020 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4021 " list\n", inode->i_ino);
4022 dump_stack();
4023 }
bcc63abb 4024 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4025
e6dcd2dc
CM		 4026 while(1) {
4027 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4028 if (!ordered)
4029 break;
4030 else {
4031 printk("found ordered extent %Lu %Lu\n",
4032 ordered->file_offset, ordered->len);
4033 btrfs_remove_ordered_extent(inode, ordered);
4034 btrfs_put_ordered_extent(ordered);
4035 btrfs_put_ordered_extent(ordered);
4036 }
4037 }
5b21f2ed 4038 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4039 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4040}
4041
0ee0fda0 4042static void init_once(void *foo)
39279cc3
CM		 4043{
4044 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4045
4046 inode_init_once(&ei->vfs_inode);
4047}
4048
4049void btrfs_destroy_cachep(void)
4050{
4051 if (btrfs_inode_cachep)
4052 kmem_cache_destroy(btrfs_inode_cachep);
4053 if (btrfs_trans_handle_cachep)
4054 kmem_cache_destroy(btrfs_trans_handle_cachep);
4055 if (btrfs_transaction_cachep)
4056 kmem_cache_destroy(btrfs_transaction_cachep);
4057 if (btrfs_bit_radix_cachep)
4058 kmem_cache_destroy(btrfs_bit_radix_cachep);
4059 if (btrfs_path_cachep)
4060 kmem_cache_destroy(btrfs_path_cachep);
4061}
4062
86479a04 4063struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4064 unsigned long extra_flags,
2b1f55b0 4065 void (*ctor)(void *))
92fee66d
CM		 4066{
4067 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4068 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4069}
4070
39279cc3
CM		 4071int btrfs_init_cachep(void)
4072{
86479a04 4073 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4074 sizeof(struct btrfs_inode),
4075 0, init_once);
39279cc3
CM		 4076 if (!btrfs_inode_cachep)
4077 goto fail;
86479a04
CM		 4078 btrfs_trans_handle_cachep =
4079 btrfs_cache_create("btrfs_trans_handle_cache",
4080 sizeof(struct btrfs_trans_handle),
4081 0, NULL);
39279cc3
CM		 4082 if (!btrfs_trans_handle_cachep)
4083 goto fail;
86479a04 4084 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4085 sizeof(struct btrfs_transaction),
92fee66d 4086 0, NULL);
39279cc3
CM		 4087 if (!btrfs_transaction_cachep)
4088 goto fail;
86479a04 4089 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4090 sizeof(struct btrfs_path),
92fee66d 4091 0, NULL);
39279cc3
CM		 4092 if (!btrfs_path_cachep)
4093 goto fail;
86479a04 4094 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4095 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4096 if (!btrfs_bit_radix_cachep)
4097 goto fail;
4098 return 0;
4099fail:
4100 btrfs_destroy_cachep();
4101 return -ENOMEM;
4102}
4103
4104static int btrfs_getattr(struct vfsmount *mnt,
4105 struct dentry *dentry, struct kstat *stat)
4106{
4107 struct inode *inode = dentry->d_inode;
4108 generic_fillattr(inode, stat);
d6667462 4109 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4110 stat->blocks = (inode_get_bytes(inode) +
4111 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4112 return 0;
4113}
4114
4115static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4116 struct inode * new_dir,struct dentry *new_dentry)
4117{
4118 struct btrfs_trans_handle *trans;
4119 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4120 struct inode *new_inode = new_dentry->d_inode;
4121 struct inode *old_inode = old_dentry->d_inode;
4122 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4123 u64 index = 0;
39279cc3
CM		 4124 int ret;
4125
4126 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4127 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4128 return -ENOTEMPTY;
4129 }
5f39d397 4130
1832a6d5
CM		 4131 ret = btrfs_check_free_space(root, 1, 0);
4132 if (ret)
4133 goto out_unlock;
4134
39279cc3 4135 trans = btrfs_start_transaction(root, 1);
5f39d397 4136
39279cc3 4137 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4138
e02119d5 4139 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4140 old_dir->i_ctime = old_dir->i_mtime = ctime;
4141 new_dir->i_ctime = new_dir->i_mtime = ctime;
4142 old_inode->i_ctime = ctime;
5f39d397 4143
e02119d5
CM		 4144 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4145 old_dentry->d_name.name,
4146 old_dentry->d_name.len);
39279cc3
CM		 4147 if (ret)
4148 goto out_fail;
4149
4150 if (new_inode) {
4151 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4152 ret = btrfs_unlink_inode(trans, root, new_dir,
4153 new_dentry->d_inode,
4154 new_dentry->d_name.name,
4155 new_dentry->d_name.len);
39279cc3
CM		 4156 if (ret)
4157 goto out_fail;
7b128766 4158 if (new_inode->i_nlink == 0) {
e02119d5 4159 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4160 if (ret)
4161 goto out_fail;
4162 }
e02119d5 4163
39279cc3 4164 }
00e4e6b3 4165 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
aec7477b
JB		 4166 if (ret)
4167 goto out_fail;
4168
e02119d5
CM		 4169 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4170 old_inode, new_dentry->d_name.name,
4171 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4172 if (ret)
4173 goto out_fail;
4174
4175out_fail:
ab78c84d 4176 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4177out_unlock:
39279cc3
CM		 4178 return ret;
4179}
4180
d352ac68
CM		 4181/*
4182 * some fairly slow code that needs optimization. This walks the list
4183 * of all the inodes with pending delalloc and forces them to disk.
4184 */
ea8c2819
CM		 4185int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4186{
4187 struct list_head *head = &root->fs_info->delalloc_inodes;
4188 struct btrfs_inode *binode;
5b21f2ed 4189 struct inode *inode;
ea8c2819
CM		 4190 unsigned long flags;
4191
4192 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4193 while(!list_empty(head)) {
4194 binode = list_entry(head->next, struct btrfs_inode,
4195 delalloc_inodes);
5b21f2ed
ZY		 4196 inode = igrab(&binode->vfs_inode);
4197 if (!inode)
4198 list_del_init(&binode->delalloc_inodes);
ea8c2819 4199 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
5b21f2ed 4200 if (inode) {
8c8bee1d 4201 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4202 iput(inode);
4203 }
4204 cond_resched();
ea8c2819
CM		 4205 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4206 }
4207 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
8c8bee1d
CM		 4208
4209 /* the filemap_flush will queue IO into the worker threads, but
4210 * we have to make sure the IO is actually started and that
4211 * ordered extents get created before we return
4212 */
4213 atomic_inc(&root->fs_info->async_submit_draining);
4214 while(atomic_read(&root->fs_info->nr_async_submits)) {
4215 wait_event(root->fs_info->async_submit_wait,
4216 (atomic_read(&root->fs_info->nr_async_submits) == 0));
4217 }
4218 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4219 return 0;
4220}
4221
39279cc3
CM		 4222static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4223 const char *symname)
4224{
4225 struct btrfs_trans_handle *trans;
4226 struct btrfs_root *root = BTRFS_I(dir)->root;
4227 struct btrfs_path *path;
4228 struct btrfs_key key;
1832a6d5 4229 struct inode *inode = NULL;
39279cc3
CM		 4230 int err;
4231 int drop_inode = 0;
4232 u64 objectid;
00e4e6b3 4233 u64 index = 0 ;
39279cc3
CM		 4234 int name_len;
4235 int datasize;
5f39d397 4236 unsigned long ptr;
39279cc3 4237 struct btrfs_file_extent_item *ei;
5f39d397 4238 struct extent_buffer *leaf;
1832a6d5 4239 unsigned long nr = 0;
39279cc3
CM		 4240
4241 name_len = strlen(symname) + 1;
4242 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4243 return -ENAMETOOLONG;
1832a6d5 4244
1832a6d5
CM		 4245 err = btrfs_check_free_space(root, 1, 0);
4246 if (err)
4247 goto out_fail;
4248
39279cc3
CM		 4249 trans = btrfs_start_transaction(root, 1);
4250 btrfs_set_trans_block_group(trans, dir);
4251
4252 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4253 if (err) {
4254 err = -ENOSPC;
4255 goto out_unlock;
4256 }
4257
aec7477b 4258 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4259 dentry->d_name.len,
4260 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4261 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4262 &index);
39279cc3
CM		 4263 err = PTR_ERR(inode);
4264 if (IS_ERR(inode))
4265 goto out_unlock;
4266
33268eaf
JB		 4267 err = btrfs_init_acl(inode, dir);
4268 if (err) {
4269 drop_inode = 1;
4270 goto out_unlock;
4271 }
4272
39279cc3 4273 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4274 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4275 if (err)
4276 drop_inode = 1;
4277 else {
4278 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4279 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4280 inode->i_fop = &btrfs_file_operations;
4281 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4282 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4283 }
4284 dir->i_sb->s_dirt = 1;
4285 btrfs_update_inode_block_group(trans, inode);
4286 btrfs_update_inode_block_group(trans, dir);
4287 if (drop_inode)
4288 goto out_unlock;
4289
4290 path = btrfs_alloc_path();
4291 BUG_ON(!path);
4292 key.objectid = inode->i_ino;
4293 key.offset = 0;
39279cc3
CM		 4294 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4295 datasize = btrfs_file_extent_calc_inline_size(name_len);
4296 err = btrfs_insert_empty_item(trans, root, path, &key,
4297 datasize);
54aa1f4d
CM		 4298 if (err) {
4299 drop_inode = 1;
4300 goto out_unlock;
4301 }
5f39d397
CM		 4302 leaf = path->nodes[0];
4303 ei = btrfs_item_ptr(leaf, path->slots[0],
4304 struct btrfs_file_extent_item);
4305 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4306 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4307 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4308 btrfs_set_file_extent_encryption(leaf, ei, 0);
4309 btrfs_set_file_extent_compression(leaf, ei, 0);
4310 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4311 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4312
39279cc3 4313 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4314 write_extent_buffer(leaf, symname, ptr, name_len);
4315 btrfs_mark_buffer_dirty(leaf);
39279cc3 4316 btrfs_free_path(path);
5f39d397 4317
39279cc3
CM		 4318 inode->i_op = &btrfs_symlink_inode_operations;
4319 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4320 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
dbe674a9 4321 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4322 err = btrfs_update_inode(trans, root, inode);
4323 if (err)
4324 drop_inode = 1;
39279cc3
CM		 4325
4326out_unlock:
d3c2fdcf 4327 nr = trans->blocks_used;
ab78c84d 4328 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4329out_fail:
39279cc3
CM		 4330 if (drop_inode) {
4331 inode_dec_link_count(inode);
4332 iput(inode);
4333 }
d3c2fdcf 4334 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4335 return err;
4336}
16432985 4337
e6dcd2dc
CM		 4338static int btrfs_set_page_dirty(struct page *page)
4339{
e6dcd2dc
CM		 4340 return __set_page_dirty_nobuffers(page);
4341}
4342
0ee0fda0 4343static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4344{
4345 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4346 return -EACCES;
33268eaf 4347 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4348}
39279cc3
CM		 4349
4350static struct inode_operations btrfs_dir_inode_operations = {
4351 .lookup = btrfs_lookup,
4352 .create = btrfs_create,
4353 .unlink = btrfs_unlink,
4354 .link = btrfs_link,
4355 .mkdir = btrfs_mkdir,
4356 .rmdir = btrfs_rmdir,
4357 .rename = btrfs_rename,
4358 .symlink = btrfs_symlink,
4359 .setattr = btrfs_setattr,
618e21d5 4360 .mknod = btrfs_mknod,
95819c05
CH		 4361 .setxattr = btrfs_setxattr,
4362 .getxattr = btrfs_getxattr,
5103e947 4363 .listxattr = btrfs_listxattr,
95819c05 4364 .removexattr = btrfs_removexattr,
fdebe2bd 4365 .permission = btrfs_permission,
39279cc3 4366};
39279cc3
CM		 4367static struct inode_operations btrfs_dir_ro_inode_operations = {
4368 .lookup = btrfs_lookup,
fdebe2bd 4369 .permission = btrfs_permission,
39279cc3 4370};
39279cc3
CM		 4371static struct file_operations btrfs_dir_file_operations = {
4372 .llseek = generic_file_llseek,
4373 .read = generic_read_dir,
cbdf5a24 4374 .readdir = btrfs_real_readdir,
34287aa3 4375 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4376#ifdef CONFIG_COMPAT
34287aa3 4377 .compat_ioctl = btrfs_ioctl,
39279cc3 4378#endif
6bf13c0c 4379 .release = btrfs_release_file,
e02119d5 4380 .fsync = btrfs_sync_file,
39279cc3
CM		 4381};
4382
d1310b2e 4383static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4384 .fill_delalloc = run_delalloc_range,
065631f6 4385 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4386 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4387 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4388 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4389 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4390 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4391 .set_bit_hook = btrfs_set_bit_hook,
4392 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4393};
4394
39279cc3
CM		 4395static struct address_space_operations btrfs_aops = {
4396 .readpage = btrfs_readpage,
4397 .writepage = btrfs_writepage,
b293f02e 4398 .writepages = btrfs_writepages,
3ab2fb5a 4399 .readpages = btrfs_readpages,
39279cc3 4400 .sync_page = block_sync_page,
39279cc3 4401 .bmap = btrfs_bmap,
16432985 4402 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 4403 .invalidatepage = btrfs_invalidatepage,
4404 .releasepage = btrfs_releasepage,
e6dcd2dc 4405 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 4406};
4407
4408static struct address_space_operations btrfs_symlink_aops = {
4409 .readpage = btrfs_readpage,
4410 .writepage = btrfs_writepage,
2bf5a725
CM		 4411 .invalidatepage = btrfs_invalidatepage,
4412 .releasepage = btrfs_releasepage,
39279cc3
CM		 4413};
4414
4415static struct inode_operations btrfs_file_inode_operations = {
4416 .truncate = btrfs_truncate,
4417 .getattr = btrfs_getattr,
4418 .setattr = btrfs_setattr,
95819c05
CH		 4419 .setxattr = btrfs_setxattr,
4420 .getxattr = btrfs_getxattr,
5103e947 4421 .listxattr = btrfs_listxattr,
95819c05 4422 .removexattr = btrfs_removexattr,
fdebe2bd 4423 .permission = btrfs_permission,
39279cc3 4424};
618e21d5
JB		 4425static struct inode_operations btrfs_special_inode_operations = {
4426 .getattr = btrfs_getattr,
4427 .setattr = btrfs_setattr,
fdebe2bd 4428 .permission = btrfs_permission,
95819c05
CH		 4429 .setxattr = btrfs_setxattr,
4430 .getxattr = btrfs_getxattr,
33268eaf 4431 .listxattr = btrfs_listxattr,
95819c05 4432 .removexattr = btrfs_removexattr,
618e21d5 4433};
39279cc3
CM		 4434static struct inode_operations btrfs_symlink_inode_operations = {
4435 .readlink = generic_readlink,
4436 .follow_link = page_follow_link_light,
4437 .put_link = page_put_link,
fdebe2bd 4438 .permission = btrfs_permission,
39279cc3 4439};
Linux 6.x block layer and io_uring tree(s)