projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mm: page_mkwrite change prototype to match fault
[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>
5103e947 37#include <linux/xattr.h>
33268eaf 38#include <linux/posix_acl.h>
d899e052 39#include <linux/falloc.h>
4b4e25f2 40#include "compat.h"
39279cc3
CM		 41#include "ctree.h"
42#include "disk-io.h"
43#include "transaction.h"
44#include "btrfs_inode.h"
45#include "ioctl.h"
46#include "print-tree.h"
0b86a832 47#include "volumes.h"
e6dcd2dc 48#include "ordered-data.h"
95819c05 49#include "xattr.h"
e02119d5 50#include "tree-log.h"
5b84e8d6 51#include "ref-cache.h"
c8b97818 52#include "compression.h"
b4ce94de 53#include "locking.h"
39279cc3
CM		 54
55struct btrfs_iget_args {
56 u64 ino;
57 struct btrfs_root *root;
58};
59
60static struct inode_operations btrfs_dir_inode_operations;
61static struct inode_operations btrfs_symlink_inode_operations;
62static struct inode_operations btrfs_dir_ro_inode_operations;
618e21d5 63static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM		 64static struct inode_operations btrfs_file_inode_operations;
65static struct address_space_operations btrfs_aops;
66static struct address_space_operations btrfs_symlink_aops;
67static struct file_operations btrfs_dir_file_operations;
d1310b2e 68static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 69
70static struct kmem_cache *btrfs_inode_cachep;
71struct kmem_cache *btrfs_trans_handle_cachep;
72struct kmem_cache *btrfs_transaction_cachep;
73struct kmem_cache *btrfs_bit_radix_cachep;
74struct kmem_cache *btrfs_path_cachep;
75
76#define S_SHIFT 12
77static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85};
86
7b128766 87static void btrfs_truncate(struct inode *inode);
c8b97818 88static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 89static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
7b128766 93
0279b4cd
JO		 94static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
95{
96 int err;
97
98 err = btrfs_init_acl(inode, dir);
99 if (!err)
100 err = btrfs_xattr_security_init(inode, dir);
101 return err;
102}
103
c8b97818
CM		 104/*
105 * this does all the hard work for inserting an inline extent into
106 * the btree. The caller should have done a btrfs_drop_extents so that
107 * no overlapping inline items exist in the btree
108 */
d397712b 109static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 110 struct btrfs_root *root, struct inode *inode,
111 u64 start, size_t size, size_t compressed_size,
112 struct page **compressed_pages)
113{
114 struct btrfs_key key;
115 struct btrfs_path *path;
116 struct extent_buffer *leaf;
117 struct page *page = NULL;
118 char *kaddr;
119 unsigned long ptr;
120 struct btrfs_file_extent_item *ei;
121 int err = 0;
122 int ret;
123 size_t cur_size = size;
124 size_t datasize;
125 unsigned long offset;
126 int use_compress = 0;
127
128 if (compressed_size && compressed_pages) {
129 use_compress = 1;
130 cur_size = compressed_size;
131 }
132
d397712b
CM		 133 path = btrfs_alloc_path();
134 if (!path)
c8b97818
CM		 135 return -ENOMEM;
136
137 btrfs_set_trans_block_group(trans, inode);
138
139 key.objectid = inode->i_ino;
140 key.offset = start;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
146 datasize);
147 BUG_ON(ret);
148 if (ret) {
149 err = ret;
c8b97818
CM		 150 goto fail;
151 }
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
161
162 if (use_compress) {
163 struct page *cpage;
164 int i = 0;
d397712b 165 while (compressed_size > 0) {
c8b97818 166 cpage = compressed_pages[i];
5b050f04 167 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 168 PAGE_CACHE_SIZE);
169
170 kaddr = kmap(cpage);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap(cpage);
173
174 i++;
175 ptr += cur_size;
176 compressed_size -= cur_size;
177 }
178 btrfs_set_file_extent_compression(leaf, ei,
179 BTRFS_COMPRESS_ZLIB);
180 } else {
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
189 }
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
192
193 BTRFS_I(inode)->disk_i_size = inode->i_size;
194 btrfs_update_inode(trans, root, inode);
195 return 0;
196fail:
197 btrfs_free_path(path);
198 return err;
199}
200
201
202/*
203 * conditionally insert an inline extent into the file. This
204 * does the checks required to make sure the data is small enough
205 * to fit as an inline extent.
206 */
207static int cow_file_range_inline(struct btrfs_trans_handle *trans,
208 struct btrfs_root *root,
209 struct inode *inode, u64 start, u64 end,
210 size_t compressed_size,
211 struct page **compressed_pages)
212{
213 u64 isize = i_size_read(inode);
214 u64 actual_end = min(end + 1, isize);
215 u64 inline_len = actual_end - start;
216 u64 aligned_end = (end + root->sectorsize - 1) &
217 ~((u64)root->sectorsize - 1);
218 u64 hint_byte;
219 u64 data_len = inline_len;
220 int ret;
221
222 if (compressed_size)
223 data_len = compressed_size;
224
225 if (start > 0 ||
70b99e69 226 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 227 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
228 (!compressed_size &&
229 (actual_end & (root->sectorsize - 1)) == 0) ||
230 end + 1 < isize ||
231 data_len > root->fs_info->max_inline) {
232 return 1;
233 }
234
c8b97818 235 ret = btrfs_drop_extents(trans, root, inode, start,
70b99e69 236 aligned_end, start, &hint_byte);
c8b97818
CM		 237 BUG_ON(ret);
238
239 if (isize > actual_end)
240 inline_len = min_t(u64, isize, actual_end);
241 ret = insert_inline_extent(trans, root, inode, start,
242 inline_len, compressed_size,
243 compressed_pages);
244 BUG_ON(ret);
245 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 246 return 0;
247}
248
771ed689
CM		 249struct async_extent {
250 u64 start;
251 u64 ram_size;
252 u64 compressed_size;
253 struct page **pages;
254 unsigned long nr_pages;
255 struct list_head list;
256};
257
258struct async_cow {
259 struct inode *inode;
260 struct btrfs_root *root;
261 struct page *locked_page;
262 u64 start;
263 u64 end;
264 struct list_head extents;
265 struct btrfs_work work;
266};
267
268static noinline int add_async_extent(struct async_cow *cow,
269 u64 start, u64 ram_size,
270 u64 compressed_size,
271 struct page **pages,
272 unsigned long nr_pages)
273{
274 struct async_extent *async_extent;
275
276 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
277 async_extent->start = start;
278 async_extent->ram_size = ram_size;
279 async_extent->compressed_size = compressed_size;
280 async_extent->pages = pages;
281 async_extent->nr_pages = nr_pages;
282 list_add_tail(&async_extent->list, &cow->extents);
283 return 0;
284}
285
d352ac68 286/*
771ed689
CM		 287 * we create compressed extents in two phases. The first
288 * phase compresses a range of pages that have already been
289 * locked (both pages and state bits are locked).
c8b97818 290 *
771ed689
CM		 291 * This is done inside an ordered work queue, and the compression
292 * is spread across many cpus. The actual IO submission is step
293 * two, and the ordered work queue takes care of making sure that
294 * happens in the same order things were put onto the queue by
295 * writepages and friends.
c8b97818 296 *
771ed689
CM		 297 * If this code finds it can't get good compression, it puts an
298 * entry onto the work queue to write the uncompressed bytes. This
299 * makes sure that both compressed inodes and uncompressed inodes
300 * are written in the same order that pdflush sent them down.
d352ac68 301 */
771ed689
CM		 302static noinline int compress_file_range(struct inode *inode,
303 struct page *locked_page,
304 u64 start, u64 end,
305 struct async_cow *async_cow,
306 int *num_added)
b888db2b
CM		 307{
308 struct btrfs_root *root = BTRFS_I(inode)->root;
309 struct btrfs_trans_handle *trans;
db94535d 310 u64 num_bytes;
c8b97818
CM		 311 u64 orig_start;
312 u64 disk_num_bytes;
db94535d 313 u64 blocksize = root->sectorsize;
c8b97818 314 u64 actual_end;
42dc7bab 315 u64 isize = i_size_read(inode);
e6dcd2dc 316 int ret = 0;
c8b97818
CM		 317 struct page **pages = NULL;
318 unsigned long nr_pages;
319 unsigned long nr_pages_ret = 0;
320 unsigned long total_compressed = 0;
321 unsigned long total_in = 0;
322 unsigned long max_compressed = 128 * 1024;
771ed689 323 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 324 int i;
325 int will_compress;
b888db2b 326
c8b97818
CM		 327 orig_start = start;
328
42dc7bab 329 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 330again:
331 will_compress = 0;
332 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
333 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 334
f03d9301
CM		 335 /*
336 * we don't want to send crud past the end of i_size through
337 * compression, that's just a waste of CPU time. So, if the
338 * end of the file is before the start of our current
339 * requested range of bytes, we bail out to the uncompressed
340 * cleanup code that can deal with all of this.
341 *
342 * It isn't really the fastest way to fix things, but this is a
343 * very uncommon corner.
344 */
345 if (actual_end <= start)
346 goto cleanup_and_bail_uncompressed;
347
c8b97818
CM		 348 total_compressed = actual_end - start;
349
350 /* we want to make sure that amount of ram required to uncompress
351 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 352 * of a compressed extent to 128k. This is a crucial number
353 * because it also controls how easily we can spread reads across
354 * cpus for decompression.
355 *
356 * We also want to make sure the amount of IO required to do
357 * a random read is reasonably small, so we limit the size of
358 * a compressed extent to 128k.
c8b97818
CM		 359 */
360 total_compressed = min(total_compressed, max_uncompressed);
db94535d 361 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 362 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 363 disk_num_bytes = num_bytes;
364 total_in = 0;
365 ret = 0;
db94535d 366
771ed689
CM		 367 /*
368 * we do compression for mount -o compress and when the
369 * inode has not been flagged as nocompress. This flag can
370 * change at any time if we discover bad compression ratios.
c8b97818
CM		 371 */
372 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
373 btrfs_test_opt(root, COMPRESS)) {
374 WARN_ON(pages);
cfbc246e 375 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 376
c8b97818
CM		 377 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
378 total_compressed, pages,
379 nr_pages, &nr_pages_ret,
380 &total_in,
381 &total_compressed,
382 max_compressed);
383
384 if (!ret) {
385 unsigned long offset = total_compressed &
386 (PAGE_CACHE_SIZE - 1);
387 struct page *page = pages[nr_pages_ret - 1];
388 char *kaddr;
389
390 /* zero the tail end of the last page, we might be
391 * sending it down to disk
392 */
393 if (offset) {
394 kaddr = kmap_atomic(page, KM_USER0);
395 memset(kaddr + offset, 0,
396 PAGE_CACHE_SIZE - offset);
397 kunmap_atomic(kaddr, KM_USER0);
398 }
399 will_compress = 1;
400 }
401 }
402 if (start == 0) {
771ed689
CM		 403 trans = btrfs_join_transaction(root, 1);
404 BUG_ON(!trans);
405 btrfs_set_trans_block_group(trans, inode);
406
c8b97818 407 /* lets try to make an inline extent */
771ed689 408 if (ret || total_in < (actual_end - start)) {
c8b97818 409 /* we didn't compress the entire range, try
771ed689 410 * to make an uncompressed inline extent.
c8b97818
CM		 411 */
412 ret = cow_file_range_inline(trans, root, inode,
413 start, end, 0, NULL);
414 } else {
771ed689 415 /* try making a compressed inline extent */
c8b97818
CM		 416 ret = cow_file_range_inline(trans, root, inode,
417 start, end,
418 total_compressed, pages);
419 }
771ed689 420 btrfs_end_transaction(trans, root);
c8b97818 421 if (ret == 0) {
771ed689
CM		 422 /*
423 * inline extent creation worked, we don't need
424 * to create any more async work items. Unlock
425 * and free up our temp pages.
426 */
c8b97818
CM		 427 extent_clear_unlock_delalloc(inode,
428 &BTRFS_I(inode)->io_tree,
771ed689
CM		 429 start, end, NULL, 1, 0,
430 0, 1, 1, 1);
c8b97818
CM		 431 ret = 0;
432 goto free_pages_out;
433 }
434 }
435
436 if (will_compress) {
437 /*
438 * we aren't doing an inline extent round the compressed size
439 * up to a block size boundary so the allocator does sane
440 * things
441 */
442 total_compressed = (total_compressed + blocksize - 1) &
443 ~(blocksize - 1);
444
445 /*
446 * one last check to make sure the compression is really a
447 * win, compare the page count read with the blocks on disk
448 */
449 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
450 ~(PAGE_CACHE_SIZE - 1);
451 if (total_compressed >= total_in) {
452 will_compress = 0;
453 } else {
454 disk_num_bytes = total_compressed;
455 num_bytes = total_in;
456 }
457 }
458 if (!will_compress && pages) {
459 /*
460 * the compression code ran but failed to make things smaller,
461 * free any pages it allocated and our page pointer array
462 */
463 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 464 WARN_ON(pages[i]->mapping);
c8b97818
CM		 465 page_cache_release(pages[i]);
466 }
467 kfree(pages);
468 pages = NULL;
469 total_compressed = 0;
470 nr_pages_ret = 0;
471
472 /* flag the file so we don't compress in the future */
473 btrfs_set_flag(inode, NOCOMPRESS);
474 }
771ed689
CM		 475 if (will_compress) {
476 *num_added += 1;
c8b97818 477
771ed689
CM		 478 /* the async work queues will take care of doing actual
479 * allocation on disk for these compressed pages,
480 * and will submit them to the elevator.
481 */
482 add_async_extent(async_cow, start, num_bytes,
483 total_compressed, pages, nr_pages_ret);
179e29e4 484
42dc7bab 485 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM		 486 start += num_bytes;
487 pages = NULL;
488 cond_resched();
489 goto again;
490 }
491 } else {
f03d9301 492cleanup_and_bail_uncompressed:
771ed689
CM		 493 /*
494 * No compression, but we still need to write the pages in
495 * the file we've been given so far. redirty the locked
496 * page if it corresponds to our extent and set things up
497 * for the async work queue to run cow_file_range to do
498 * the normal delalloc dance
499 */
500 if (page_offset(locked_page) >= start &&
501 page_offset(locked_page) <= end) {
502 __set_page_dirty_nobuffers(locked_page);
503 /* unlocked later on in the async handlers */
504 }
505 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
506 *num_added += 1;
507 }
3b951516 508
771ed689
CM		 509out:
510 return 0;
511
512free_pages_out:
513 for (i = 0; i < nr_pages_ret; i++) {
514 WARN_ON(pages[i]->mapping);
515 page_cache_release(pages[i]);
516 }
d397712b 517 kfree(pages);
771ed689
CM		 518
519 goto out;
520}
521
522/*
523 * phase two of compressed writeback. This is the ordered portion
524 * of the code, which only gets called in the order the work was
525 * queued. We walk all the async extents created by compress_file_range
526 * and send them down to the disk.
527 */
528static noinline int submit_compressed_extents(struct inode *inode,
529 struct async_cow *async_cow)
530{
531 struct async_extent *async_extent;
532 u64 alloc_hint = 0;
533 struct btrfs_trans_handle *trans;
534 struct btrfs_key ins;
535 struct extent_map *em;
536 struct btrfs_root *root = BTRFS_I(inode)->root;
537 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538 struct extent_io_tree *io_tree;
539 int ret;
540
541 if (list_empty(&async_cow->extents))
542 return 0;
543
544 trans = btrfs_join_transaction(root, 1);
545
d397712b 546 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 547 async_extent = list_entry(async_cow->extents.next,
548 struct async_extent, list);
549 list_del(&async_extent->list);
c8b97818 550
771ed689
CM		 551 io_tree = &BTRFS_I(inode)->io_tree;
552
553 /* did the compression code fall back to uncompressed IO? */
554 if (!async_extent->pages) {
555 int page_started = 0;
556 unsigned long nr_written = 0;
557
558 lock_extent(io_tree, async_extent->start,
d397712b
CM		 559 async_extent->start +
560 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 561
562 /* allocate blocks */
563 cow_file_range(inode, async_cow->locked_page,
564 async_extent->start,
565 async_extent->start +
566 async_extent->ram_size - 1,
567 &page_started, &nr_written, 0);
568
569 /*
570 * if page_started, cow_file_range inserted an
571 * inline extent and took care of all the unlocking
572 * and IO for us. Otherwise, we need to submit
573 * all those pages down to the drive.
574 */
575 if (!page_started)
576 extent_write_locked_range(io_tree,
577 inode, async_extent->start,
d397712b 578 async_extent->start +
771ed689
CM		 579 async_extent->ram_size - 1,
580 btrfs_get_extent,
581 WB_SYNC_ALL);
582 kfree(async_extent);
583 cond_resched();
584 continue;
585 }
586
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start + async_extent->ram_size - 1,
589 GFP_NOFS);
c8b97818 590 /*
771ed689
CM		 591 * here we're doing allocation and writeback of the
592 * compressed pages
c8b97818 593 */
771ed689
CM		 594 btrfs_drop_extent_cache(inode, async_extent->start,
595 async_extent->start +
596 async_extent->ram_size - 1, 0);
597
598 ret = btrfs_reserve_extent(trans, root,
599 async_extent->compressed_size,
600 async_extent->compressed_size,
601 0, alloc_hint,
602 (u64)-1, &ins, 1);
603 BUG_ON(ret);
604 em = alloc_extent_map(GFP_NOFS);
605 em->start = async_extent->start;
606 em->len = async_extent->ram_size;
445a6944 607 em->orig_start = em->start;
c8b97818 608
771ed689
CM		 609 em->block_start = ins.objectid;
610 em->block_len = ins.offset;
611 em->bdev = root->fs_info->fs_devices->latest_bdev;
612 set_bit(EXTENT_FLAG_PINNED, &em->flags);
613 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
614
d397712b 615 while (1) {
771ed689
CM		 616 spin_lock(&em_tree->lock);
617 ret = add_extent_mapping(em_tree, em);
618 spin_unlock(&em_tree->lock);
619 if (ret != -EEXIST) {
620 free_extent_map(em);
621 break;
622 }
623 btrfs_drop_extent_cache(inode, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1, 0);
626 }
627
628 ret = btrfs_add_ordered_extent(inode, async_extent->start,
629 ins.objectid,
630 async_extent->ram_size,
631 ins.offset,
632 BTRFS_ORDERED_COMPRESSED);
633 BUG_ON(ret);
634
635 btrfs_end_transaction(trans, root);
636
637 /*
638 * clear dirty, set writeback and unlock the pages.
639 */
640 extent_clear_unlock_delalloc(inode,
641 &BTRFS_I(inode)->io_tree,
642 async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1,
645 NULL, 1, 1, 0, 1, 1, 0);
646
647 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 648 async_extent->start,
649 async_extent->ram_size,
650 ins.objectid,
651 ins.offset, async_extent->pages,
652 async_extent->nr_pages);
771ed689
CM		 653
654 BUG_ON(ret);
655 trans = btrfs_join_transaction(root, 1);
656 alloc_hint = ins.objectid + ins.offset;
657 kfree(async_extent);
658 cond_resched();
659 }
660
661 btrfs_end_transaction(trans, root);
662 return 0;
663}
664
665/*
666 * when extent_io.c finds a delayed allocation range in the file,
667 * the call backs end up in this code. The basic idea is to
668 * allocate extents on disk for the range, and create ordered data structs
669 * in ram to track those extents.
670 *
671 * locked_page is the page that writepage had locked already. We use
672 * it to make sure we don't do extra locks or unlocks.
673 *
674 * *page_started is set to one if we unlock locked_page and do everything
675 * required to start IO on it. It may be clean and already done with
676 * IO when we return.
677 */
678static noinline int cow_file_range(struct inode *inode,
679 struct page *locked_page,
680 u64 start, u64 end, int *page_started,
681 unsigned long *nr_written,
682 int unlock)
683{
684 struct btrfs_root *root = BTRFS_I(inode)->root;
685 struct btrfs_trans_handle *trans;
686 u64 alloc_hint = 0;
687 u64 num_bytes;
688 unsigned long ram_size;
689 u64 disk_num_bytes;
690 u64 cur_alloc_size;
691 u64 blocksize = root->sectorsize;
692 u64 actual_end;
42dc7bab 693 u64 isize = i_size_read(inode);
771ed689
CM		 694 struct btrfs_key ins;
695 struct extent_map *em;
696 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
697 int ret = 0;
698
699 trans = btrfs_join_transaction(root, 1);
700 BUG_ON(!trans);
701 btrfs_set_trans_block_group(trans, inode);
702
42dc7bab 703 actual_end = min_t(u64, isize, end + 1);
771ed689
CM		 704
705 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
706 num_bytes = max(blocksize, num_bytes);
707 disk_num_bytes = num_bytes;
708 ret = 0;
709
710 if (start == 0) {
711 /* lets try to make an inline extent */
712 ret = cow_file_range_inline(trans, root, inode,
713 start, end, 0, NULL);
714 if (ret == 0) {
715 extent_clear_unlock_delalloc(inode,
716 &BTRFS_I(inode)->io_tree,
717 start, end, NULL, 1, 1,
718 1, 1, 1, 1);
719 *nr_written = *nr_written +
720 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
721 *page_started = 1;
722 ret = 0;
723 goto out;
724 }
725 }
726
727 BUG_ON(disk_num_bytes >
728 btrfs_super_total_bytes(&root->fs_info->super_copy));
729
730 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
731
d397712b 732 while (disk_num_bytes > 0) {
c8b97818 733 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 734 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 735 root->sectorsize, 0, alloc_hint,
e6dcd2dc 736 (u64)-1, &ins, 1);
d397712b
CM		 737 BUG_ON(ret);
738
e6dcd2dc
CM		 739 em = alloc_extent_map(GFP_NOFS);
740 em->start = start;
445a6944 741 em->orig_start = em->start;
c8b97818 742
771ed689
CM		 743 ram_size = ins.offset;
744 em->len = ins.offset;
c8b97818 745
e6dcd2dc 746 em->block_start = ins.objectid;
c8b97818 747 em->block_len = ins.offset;
e6dcd2dc 748 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 749 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 750
d397712b 751 while (1) {
e6dcd2dc
CM		 752 spin_lock(&em_tree->lock);
753 ret = add_extent_mapping(em_tree, em);
754 spin_unlock(&em_tree->lock);
755 if (ret != -EEXIST) {
756 free_extent_map(em);
757 break;
758 }
759 btrfs_drop_extent_cache(inode, start,
c8b97818 760 start + ram_size - 1, 0);
e6dcd2dc
CM		 761 }
762
98d20f67 763 cur_alloc_size = ins.offset;
e6dcd2dc 764 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 765 ram_size, cur_alloc_size, 0);
e6dcd2dc 766 BUG_ON(ret);
c8b97818 767
17d217fe
YZ		 768 if (root->root_key.objectid ==
769 BTRFS_DATA_RELOC_TREE_OBJECTID) {
770 ret = btrfs_reloc_clone_csums(inode, start,
771 cur_alloc_size);
772 BUG_ON(ret);
773 }
774
d397712b 775 if (disk_num_bytes < cur_alloc_size)
3b951516 776 break;
d397712b 777
c8b97818
CM		 778 /* we're not doing compressed IO, don't unlock the first
779 * page (which the caller expects to stay locked), don't
780 * clear any dirty bits and don't set any writeback bits
781 */
782 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
783 start, start + ram_size - 1,
771ed689
CM		 784 locked_page, unlock, 1,
785 1, 0, 0, 0);
c8b97818 786 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 787 num_bytes -= cur_alloc_size;
788 alloc_hint = ins.objectid + ins.offset;
789 start += cur_alloc_size;
b888db2b 790 }
b888db2b 791out:
771ed689 792 ret = 0;
b888db2b 793 btrfs_end_transaction(trans, root);
c8b97818 794
be20aa9d 795 return ret;
771ed689 796}
c8b97818 797
771ed689
CM		 798/*
799 * work queue call back to started compression on a file and pages
800 */
801static noinline void async_cow_start(struct btrfs_work *work)
802{
803 struct async_cow *async_cow;
804 int num_added = 0;
805 async_cow = container_of(work, struct async_cow, work);
806
807 compress_file_range(async_cow->inode, async_cow->locked_page,
808 async_cow->start, async_cow->end, async_cow,
809 &num_added);
810 if (num_added == 0)
811 async_cow->inode = NULL;
812}
813
814/*
815 * work queue call back to submit previously compressed pages
816 */
817static noinline void async_cow_submit(struct btrfs_work *work)
818{
819 struct async_cow *async_cow;
820 struct btrfs_root *root;
821 unsigned long nr_pages;
822
823 async_cow = container_of(work, struct async_cow, work);
824
825 root = async_cow->root;
826 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
827 PAGE_CACHE_SHIFT;
828
829 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
830
831 if (atomic_read(&root->fs_info->async_delalloc_pages) <
832 5 * 1042 * 1024 &&
833 waitqueue_active(&root->fs_info->async_submit_wait))
834 wake_up(&root->fs_info->async_submit_wait);
835
d397712b 836 if (async_cow->inode)
771ed689 837 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 838}
c8b97818 839
771ed689
CM		 840static noinline void async_cow_free(struct btrfs_work *work)
841{
842 struct async_cow *async_cow;
843 async_cow = container_of(work, struct async_cow, work);
844 kfree(async_cow);
845}
846
847static int cow_file_range_async(struct inode *inode, struct page *locked_page,
848 u64 start, u64 end, int *page_started,
849 unsigned long *nr_written)
850{
851 struct async_cow *async_cow;
852 struct btrfs_root *root = BTRFS_I(inode)->root;
853 unsigned long nr_pages;
854 u64 cur_end;
855 int limit = 10 * 1024 * 1042;
856
857 if (!btrfs_test_opt(root, COMPRESS)) {
858 return cow_file_range(inode, locked_page, start, end,
859 page_started, nr_written, 1);
860 }
861
862 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
863 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
d397712b 864 while (start < end) {
771ed689
CM		 865 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
866 async_cow->inode = inode;
867 async_cow->root = root;
868 async_cow->locked_page = locked_page;
869 async_cow->start = start;
870
871 if (btrfs_test_flag(inode, NOCOMPRESS))
872 cur_end = end;
873 else
874 cur_end = min(end, start + 512 * 1024 - 1);
875
876 async_cow->end = cur_end;
877 INIT_LIST_HEAD(&async_cow->extents);
878
879 async_cow->work.func = async_cow_start;
880 async_cow->work.ordered_func = async_cow_submit;
881 async_cow->work.ordered_free = async_cow_free;
882 async_cow->work.flags = 0;
883
771ed689
CM		 884 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
885 PAGE_CACHE_SHIFT;
886 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
887
888 btrfs_queue_worker(&root->fs_info->delalloc_workers,
889 &async_cow->work);
890
891 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
892 wait_event(root->fs_info->async_submit_wait,
893 (atomic_read(&root->fs_info->async_delalloc_pages) <
894 limit));
895 }
896
d397712b 897 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 898 atomic_read(&root->fs_info->async_delalloc_pages)) {
899 wait_event(root->fs_info->async_submit_wait,
900 (atomic_read(&root->fs_info->async_delalloc_pages) ==
901 0));
902 }
903
904 *nr_written += nr_pages;
905 start = cur_end + 1;
906 }
907 *page_started = 1;
908 return 0;
be20aa9d
CM		 909}
910
d397712b 911static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 912 u64 bytenr, u64 num_bytes)
913{
914 int ret;
915 struct btrfs_ordered_sum *sums;
916 LIST_HEAD(list);
917
07d400a6
YZ		 918 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
919 bytenr + num_bytes - 1, &list);
17d217fe
YZ		 920 if (ret == 0 && list_empty(&list))
921 return 0;
922
923 while (!list_empty(&list)) {
924 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
925 list_del(&sums->list);
926 kfree(sums);
927 }
928 return 1;
929}
930
d352ac68
CM		 931/*
932 * when nowcow writeback call back. This checks for snapshots or COW copies
933 * of the extents that exist in the file, and COWs the file as required.
934 *
935 * If no cow copies or snapshots exist, we write directly to the existing
936 * blocks on disk
937 */
c8b97818 938static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
771ed689
CM		 939 u64 start, u64 end, int *page_started, int force,
940 unsigned long *nr_written)
be20aa9d 941{
be20aa9d 942 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 943 struct btrfs_trans_handle *trans;
be20aa9d 944 struct extent_buffer *leaf;
be20aa9d 945 struct btrfs_path *path;
80ff3856 946 struct btrfs_file_extent_item *fi;
be20aa9d 947 struct btrfs_key found_key;
80ff3856
YZ		 948 u64 cow_start;
949 u64 cur_offset;
950 u64 extent_end;
951 u64 disk_bytenr;
952 u64 num_bytes;
953 int extent_type;
954 int ret;
d899e052 955 int type;
80ff3856
YZ		 956 int nocow;
957 int check_prev = 1;
be20aa9d
CM		 958
959 path = btrfs_alloc_path();
960 BUG_ON(!path);
7ea394f1
YZ		 961 trans = btrfs_join_transaction(root, 1);
962 BUG_ON(!trans);
be20aa9d 963
80ff3856
YZ		 964 cow_start = (u64)-1;
965 cur_offset = start;
966 while (1) {
967 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
968 cur_offset, 0);
969 BUG_ON(ret < 0);
970 if (ret > 0 && path->slots[0] > 0 && check_prev) {
971 leaf = path->nodes[0];
972 btrfs_item_key_to_cpu(leaf, &found_key,
973 path->slots[0] - 1);
974 if (found_key.objectid == inode->i_ino &&
975 found_key.type == BTRFS_EXTENT_DATA_KEY)
976 path->slots[0]--;
977 }
978 check_prev = 0;
979next_slot:
980 leaf = path->nodes[0];
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
983 if (ret < 0)
984 BUG_ON(1);
985 if (ret > 0)
986 break;
987 leaf = path->nodes[0];
988 }
be20aa9d 989
80ff3856
YZ		 990 nocow = 0;
991 disk_bytenr = 0;
17d217fe 992 num_bytes = 0;
80ff3856
YZ		 993 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
994
995 if (found_key.objectid > inode->i_ino ||
996 found_key.type > BTRFS_EXTENT_DATA_KEY ||
997 found_key.offset > end)
998 break;
999
1000 if (found_key.offset > cur_offset) {
1001 extent_end = found_key.offset;
1002 goto out_check;
1003 }
1004
1005 fi = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_file_extent_item);
1007 extent_type = btrfs_file_extent_type(leaf, fi);
1008
d899e052
YZ		 1009 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1010 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 1011 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1012 extent_end = found_key.offset +
1013 btrfs_file_extent_num_bytes(leaf, fi);
1014 if (extent_end <= start) {
1015 path->slots[0]++;
1016 goto next_slot;
1017 }
17d217fe
YZ		 1018 if (disk_bytenr == 0)
1019 goto out_check;
80ff3856
YZ		 1020 if (btrfs_file_extent_compression(leaf, fi) ||
1021 btrfs_file_extent_encryption(leaf, fi) ||
1022 btrfs_file_extent_other_encoding(leaf, fi))
1023 goto out_check;
d899e052
YZ		 1024 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1025 goto out_check;
d2fb3437 1026 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1027 goto out_check;
17d217fe
YZ		 1028 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1029 disk_bytenr))
1030 goto out_check;
80ff3856 1031 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
17d217fe
YZ		 1032 disk_bytenr += cur_offset - found_key.offset;
1033 num_bytes = min(end + 1, extent_end) - cur_offset;
1034 /*
1035 * force cow if csum exists in the range.
1036 * this ensure that csum for a given extent are
1037 * either valid or do not exist.
1038 */
1039 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1040 goto out_check;
80ff3856
YZ		 1041 nocow = 1;
1042 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1043 extent_end = found_key.offset +
1044 btrfs_file_extent_inline_len(leaf, fi);
1045 extent_end = ALIGN(extent_end, root->sectorsize);
1046 } else {
1047 BUG_ON(1);
1048 }
1049out_check:
1050 if (extent_end <= start) {
1051 path->slots[0]++;
1052 goto next_slot;
1053 }
1054 if (!nocow) {
1055 if (cow_start == (u64)-1)
1056 cow_start = cur_offset;
1057 cur_offset = extent_end;
1058 if (cur_offset > end)
1059 break;
1060 path->slots[0]++;
1061 goto next_slot;
7ea394f1
YZ		 1062 }
1063
1064 btrfs_release_path(root, path);
80ff3856
YZ		 1065 if (cow_start != (u64)-1) {
1066 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1067 found_key.offset - 1, page_started,
1068 nr_written, 1);
80ff3856
YZ		 1069 BUG_ON(ret);
1070 cow_start = (u64)-1;
7ea394f1 1071 }
80ff3856 1072
d899e052
YZ		 1073 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1074 struct extent_map *em;
1075 struct extent_map_tree *em_tree;
1076 em_tree = &BTRFS_I(inode)->extent_tree;
1077 em = alloc_extent_map(GFP_NOFS);
1078 em->start = cur_offset;
445a6944 1079 em->orig_start = em->start;
d899e052
YZ		 1080 em->len = num_bytes;
1081 em->block_len = num_bytes;
1082 em->block_start = disk_bytenr;
1083 em->bdev = root->fs_info->fs_devices->latest_bdev;
1084 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1085 while (1) {
1086 spin_lock(&em_tree->lock);
1087 ret = add_extent_mapping(em_tree, em);
1088 spin_unlock(&em_tree->lock);
1089 if (ret != -EEXIST) {
1090 free_extent_map(em);
1091 break;
1092 }
1093 btrfs_drop_extent_cache(inode, em->start,
1094 em->start + em->len - 1, 0);
1095 }
1096 type = BTRFS_ORDERED_PREALLOC;
1097 } else {
1098 type = BTRFS_ORDERED_NOCOW;
1099 }
80ff3856
YZ		 1100
1101 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1102 num_bytes, num_bytes, type);
1103 BUG_ON(ret);
771ed689 1104
d899e052
YZ		 1105 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1106 cur_offset, cur_offset + num_bytes - 1,
771ed689 1107 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1108 cur_offset = extent_end;
1109 if (cur_offset > end)
1110 break;
be20aa9d 1111 }
80ff3856
YZ		 1112 btrfs_release_path(root, path);
1113
1114 if (cur_offset <= end && cow_start == (u64)-1)
1115 cow_start = cur_offset;
1116 if (cow_start != (u64)-1) {
1117 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1118 page_started, nr_written, 1);
80ff3856
YZ		 1119 BUG_ON(ret);
1120 }
1121
1122 ret = btrfs_end_transaction(trans, root);
1123 BUG_ON(ret);
7ea394f1 1124 btrfs_free_path(path);
80ff3856 1125 return 0;
be20aa9d
CM		 1126}
1127
d352ac68
CM		 1128/*
1129 * extent_io.c call back to do delayed allocation processing
1130 */
c8b97818 1131static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1132 u64 start, u64 end, int *page_started,
1133 unsigned long *nr_written)
be20aa9d 1134{
be20aa9d 1135 int ret;
a2135011 1136
17d217fe 1137 if (btrfs_test_flag(inode, NODATACOW))
c8b97818 1138 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1139 page_started, 1, nr_written);
d899e052
YZ		 1140 else if (btrfs_test_flag(inode, PREALLOC))
1141 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1142 page_started, 0, nr_written);
be20aa9d 1143 else
771ed689 1144 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1145 page_started, nr_written);
1832a6d5 1146
b888db2b
CM		 1147 return ret;
1148}
1149
d352ac68
CM		 1150/*
1151 * extent_io.c set_bit_hook, used to track delayed allocation
1152 * bytes in this file, and to maintain the list of inodes that
1153 * have pending delalloc work to be done.
1154 */
b2950863 1155static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1156 unsigned long old, unsigned long bits)
291d673e 1157{
75eff68e
CM		 1158 /*
1159 * set_bit and clear bit hooks normally require _irqsave/restore
1160 * but in this case, we are only testeing for the DELALLOC
1161 * bit, which is only set or cleared with irqs on
1162 */
b0c68f8b 1163 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1164 struct btrfs_root *root = BTRFS_I(inode)->root;
6a63209f 1165 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
75eff68e 1166 spin_lock(&root->fs_info->delalloc_lock);
9069218d 1167 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1168 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1169 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171 &root->fs_info->delalloc_inodes);
1172 }
75eff68e 1173 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1174 }
1175 return 0;
1176}
1177
d352ac68
CM		 1178/*
1179 * extent_io.c clear_bit_hook, see set_bit_hook for why
1180 */
b2950863 1181static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1182 unsigned long old, unsigned long bits)
291d673e 1183{
75eff68e
CM		 1184 /*
1185 * set_bit and clear bit hooks normally require _irqsave/restore
1186 * but in this case, we are only testeing for the DELALLOC
1187 * bit, which is only set or cleared with irqs on
1188 */
b0c68f8b 1189 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1190 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1191
75eff68e 1192 spin_lock(&root->fs_info->delalloc_lock);
b0c68f8b 1193 if (end - start + 1 > root->fs_info->delalloc_bytes) {
d397712b
CM		 1194 printk(KERN_INFO "btrfs warning: delalloc account "
1195 "%llu %llu\n",
1196 (unsigned long long)end - start + 1,
1197 (unsigned long long)
1198 root->fs_info->delalloc_bytes);
6a63209f 1199 btrfs_delalloc_free_space(root, inode, (u64)-1);
b0c68f8b 1200 root->fs_info->delalloc_bytes = 0;
9069218d 1201 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b 1202 } else {
6a63209f
JB		 1203 btrfs_delalloc_free_space(root, inode,
1204 end - start + 1);
b0c68f8b 1205 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1206 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1207 }
ea8c2819
CM		 1208 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211 }
75eff68e 1212 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1213 }
1214 return 0;
1215}
1216
d352ac68
CM		 1217/*
1218 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219 * we don't create bios that span stripes or chunks
1220 */
239b14b3 1221int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1222 size_t size, struct bio *bio,
1223 unsigned long bio_flags)
239b14b3
CM		 1224{
1225 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226 struct btrfs_mapping_tree *map_tree;
a62b9401 1227 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1228 u64 length = 0;
1229 u64 map_length;
239b14b3
CM		 1230 int ret;
1231
771ed689
CM		 1232 if (bio_flags & EXTENT_BIO_COMPRESSED)
1233 return 0;
1234
f2d8d74d 1235 length = bio->bi_size;
239b14b3
CM		 1236 map_tree = &root->fs_info->mapping_tree;
1237 map_length = length;
cea9e445 1238 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1239 &map_length, NULL, 0);
cea9e445 1240
d397712b 1241 if (map_length < length + size)
239b14b3 1242 return 1;
239b14b3
CM		 1243 return 0;
1244}
1245
d352ac68
CM		 1246/*
1247 * in order to insert checksums into the metadata in large chunks,
1248 * we wait until bio submission time. All the pages in the bio are
1249 * checksummed and sums are attached onto the ordered extent record.
1250 *
1251 * At IO completion time the cums attached on the ordered extent record
1252 * are inserted into the btree
1253 */
d397712b
CM		 1254static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255 struct bio *bio, int mirror_num,
1256 unsigned long bio_flags)
065631f6 1257{
065631f6 1258 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1259 int ret = 0;
e015640f 1260
d20f7043 1261 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1262 BUG_ON(ret);
4a69a410
CM		 1263 return 0;
1264}
e015640f 1265
4a69a410
CM		 1266/*
1267 * in order to insert checksums into the metadata in large chunks,
1268 * we wait until bio submission time. All the pages in the bio are
1269 * checksummed and sums are attached onto the ordered extent record.
1270 *
1271 * At IO completion time the cums attached on the ordered extent record
1272 * are inserted into the btree
1273 */
b2950863 1274static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM		 1275 int mirror_num, unsigned long bio_flags)
1276{
1277 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1278 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1279}
1280
d352ac68 1281/*
cad321ad
CM		 1282 * extent_io.c submission hook. This does the right thing for csum calculation
1283 * on write, or reading the csums from the tree before a read
d352ac68 1284 */
b2950863 1285static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1286 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1287{
1288 struct btrfs_root *root = BTRFS_I(inode)->root;
1289 int ret = 0;
19b9bdb0 1290 int skip_sum;
44b8bd7e 1291
cad321ad
CM		 1292 skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
e6dcd2dc
CM		 1294 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295 BUG_ON(ret);
065631f6 1296
4d1b5fb4 1297 if (!(rw & (1 << BIO_RW))) {
d20f7043 1298 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1299 return btrfs_submit_compressed_read(inode, bio,
1300 mirror_num, bio_flags);
d20f7043
CM		 1301 } else if (!skip_sum)
1302 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1303 goto mapit;
19b9bdb0 1304 } else if (!skip_sum) {
17d217fe
YZ		 1305 /* csum items have already been cloned */
1306 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307 goto mapit;
19b9bdb0
CM		 1308 /* we're doing a write, do the async checksumming */
1309 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1310 inode, rw, bio, mirror_num,
4a69a410
CM		 1311 bio_flags, __btrfs_submit_bio_start,
1312 __btrfs_submit_bio_done);
19b9bdb0
CM		 1313 }
1314
0b86a832 1315mapit:
8b712842 1316 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1317}
6885f308 1318
d352ac68
CM		 1319/*
1320 * given a list of ordered sums record them in the inode. This happens
1321 * at IO completion time based on sums calculated at bio submission time.
1322 */
ba1da2f4 1323static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1324 struct inode *inode, u64 file_offset,
1325 struct list_head *list)
1326{
e6dcd2dc
CM		 1327 struct btrfs_ordered_sum *sum;
1328
1329 btrfs_set_trans_block_group(trans, inode);
c6e30871
QF		 1330
1331 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1332 btrfs_csum_file_blocks(trans,
1333 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1334 }
1335 return 0;
1336}
1337
ea8c2819
CM		 1338int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339{
d397712b 1340 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1341 WARN_ON(1);
ea8c2819
CM		 1342 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343 GFP_NOFS);
1344}
1345
d352ac68 1346/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1347struct btrfs_writepage_fixup {
1348 struct page *page;
1349 struct btrfs_work work;
1350};
1351
b2950863 1352static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1353{
1354 struct btrfs_writepage_fixup *fixup;
1355 struct btrfs_ordered_extent *ordered;
1356 struct page *page;
1357 struct inode *inode;
1358 u64 page_start;
1359 u64 page_end;
1360
1361 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362 page = fixup->page;
4a096752 1363again:
247e743c
CM		 1364 lock_page(page);
1365 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366 ClearPageChecked(page);
1367 goto out_page;
1368 }
1369
1370 inode = page->mapping->host;
1371 page_start = page_offset(page);
1372 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1375
1376 /* already ordered? We're done */
1377 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378 EXTENT_ORDERED, 0)) {
247e743c 1379 goto out;
4a096752
CM		 1380 }
1381
1382 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383 if (ordered) {
1384 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385 page_end, GFP_NOFS);
1386 unlock_page(page);
1387 btrfs_start_ordered_extent(inode, ordered, 1);
1388 goto again;
1389 }
247e743c 1390
ea8c2819 1391 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1392 ClearPageChecked(page);
1393out:
1394 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395out_page:
1396 unlock_page(page);
1397 page_cache_release(page);
1398}
1399
1400/*
1401 * There are a few paths in the higher layers of the kernel that directly
1402 * set the page dirty bit without asking the filesystem if it is a
1403 * good idea. This causes problems because we want to make sure COW
1404 * properly happens and the data=ordered rules are followed.
1405 *
c8b97818 1406 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1407 * hasn't been properly setup for IO. We kick off an async process
1408 * to fix it up. The async helper will wait for ordered extents, set
1409 * the delalloc bit and make it safe to write the page.
1410 */
b2950863 1411static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1412{
1413 struct inode *inode = page->mapping->host;
1414 struct btrfs_writepage_fixup *fixup;
1415 struct btrfs_root *root = BTRFS_I(inode)->root;
1416 int ret;
1417
1418 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419 EXTENT_ORDERED, 0);
1420 if (ret)
1421 return 0;
1422
1423 if (PageChecked(page))
1424 return -EAGAIN;
1425
1426 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427 if (!fixup)
1428 return -EAGAIN;
f421950f 1429
247e743c
CM		 1430 SetPageChecked(page);
1431 page_cache_get(page);
1432 fixup->work.func = btrfs_writepage_fixup_worker;
1433 fixup->page = page;
1434 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435 return -EAGAIN;
1436}
1437
d899e052
YZ		 1438static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439 struct inode *inode, u64 file_pos,
1440 u64 disk_bytenr, u64 disk_num_bytes,
1441 u64 num_bytes, u64 ram_bytes,
1442 u8 compression, u8 encryption,
1443 u16 other_encoding, int extent_type)
1444{
1445 struct btrfs_root *root = BTRFS_I(inode)->root;
1446 struct btrfs_file_extent_item *fi;
1447 struct btrfs_path *path;
1448 struct extent_buffer *leaf;
1449 struct btrfs_key ins;
1450 u64 hint;
1451 int ret;
1452
1453 path = btrfs_alloc_path();
1454 BUG_ON(!path);
1455
1456 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1457 file_pos + num_bytes, file_pos, &hint);
1458 BUG_ON(ret);
1459
1460 ins.objectid = inode->i_ino;
1461 ins.offset = file_pos;
1462 ins.type = BTRFS_EXTENT_DATA_KEY;
1463 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1464 BUG_ON(ret);
1465 leaf = path->nodes[0];
1466 fi = btrfs_item_ptr(leaf, path->slots[0],
1467 struct btrfs_file_extent_item);
1468 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1469 btrfs_set_file_extent_type(leaf, fi, extent_type);
1470 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1471 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1472 btrfs_set_file_extent_offset(leaf, fi, 0);
1473 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1474 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1475 btrfs_set_file_extent_compression(leaf, fi, compression);
1476 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1477 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1478 btrfs_mark_buffer_dirty(leaf);
1479
1480 inode_add_bytes(inode, num_bytes);
1481 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1482
1483 ins.objectid = disk_bytenr;
1484 ins.offset = disk_num_bytes;
1485 ins.type = BTRFS_EXTENT_ITEM_KEY;
1486 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1487 root->root_key.objectid,
1488 trans->transid, inode->i_ino, &ins);
1489 BUG_ON(ret);
1490
1491 btrfs_free_path(path);
1492 return 0;
1493}
1494
d352ac68
CM		 1495/* as ordered data IO finishes, this gets called so we can finish
1496 * an ordered extent if the range of bytes in the file it covers are
1497 * fully written.
1498 */
211f90e6 1499static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1500{
e6dcd2dc
CM		 1501 struct btrfs_root *root = BTRFS_I(inode)->root;
1502 struct btrfs_trans_handle *trans;
1503 struct btrfs_ordered_extent *ordered_extent;
1504 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
d899e052 1505 int compressed = 0;
e6dcd2dc
CM		 1506 int ret;
1507
1508 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1509 if (!ret)
e6dcd2dc 1510 return 0;
e6dcd2dc 1511
f9295749 1512 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1513
1514 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1515 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1516 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1517 goto nocow;
e6dcd2dc
CM		 1518
1519 lock_extent(io_tree, ordered_extent->file_offset,
1520 ordered_extent->file_offset + ordered_extent->len - 1,
1521 GFP_NOFS);
1522
c8b97818 1523 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1524 compressed = 1;
1525 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1526 BUG_ON(compressed);
1527 ret = btrfs_mark_extent_written(trans, root, inode,
1528 ordered_extent->file_offset,
1529 ordered_extent->file_offset +
1530 ordered_extent->len);
1531 BUG_ON(ret);
1532 } else {
1533 ret = insert_reserved_file_extent(trans, inode,
1534 ordered_extent->file_offset,
1535 ordered_extent->start,
1536 ordered_extent->disk_len,
1537 ordered_extent->len,
1538 ordered_extent->len,
1539 compressed, 0, 0,
1540 BTRFS_FILE_EXTENT_REG);
1541 BUG_ON(ret);
1542 }
e6dcd2dc
CM		 1543 unlock_extent(io_tree, ordered_extent->file_offset,
1544 ordered_extent->file_offset + ordered_extent->len - 1,
1545 GFP_NOFS);
7ea394f1 1546nocow:
e6dcd2dc
CM		 1547 add_pending_csums(trans, inode, ordered_extent->file_offset,
1548 &ordered_extent->list);
1549
34353029 1550 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1551 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1552 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1553 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1554 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1555
e6dcd2dc
CM		 1556 /* once for us */
1557 btrfs_put_ordered_extent(ordered_extent);
1558 /* once for the tree */
1559 btrfs_put_ordered_extent(ordered_extent);
1560
e6dcd2dc
CM		 1561 btrfs_end_transaction(trans, root);
1562 return 0;
1563}
1564
b2950863 1565static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1566 struct extent_state *state, int uptodate)
1567{
1568 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1569}
1570
d352ac68
CM		 1571/*
1572 * When IO fails, either with EIO or csum verification fails, we
1573 * try other mirrors that might have a good copy of the data. This
1574 * io_failure_record is used to record state as we go through all the
1575 * mirrors. If another mirror has good data, the page is set up to date
1576 * and things continue. If a good mirror can't be found, the original
1577 * bio end_io callback is called to indicate things have failed.
1578 */
7e38326f
CM		 1579struct io_failure_record {
1580 struct page *page;
1581 u64 start;
1582 u64 len;
1583 u64 logical;
d20f7043 1584 unsigned long bio_flags;
7e38326f
CM		 1585 int last_mirror;
1586};
1587
b2950863 1588static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1589 struct page *page, u64 start, u64 end,
1590 struct extent_state *state)
7e38326f
CM		 1591{
1592 struct io_failure_record *failrec = NULL;
1593 u64 private;
1594 struct extent_map *em;
1595 struct inode *inode = page->mapping->host;
1596 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1597 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1598 struct bio *bio;
1599 int num_copies;
1600 int ret;
1259ab75 1601 int rw;
7e38326f
CM		 1602 u64 logical;
1603
1604 ret = get_state_private(failure_tree, start, &private);
1605 if (ret) {
7e38326f
CM		 1606 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1607 if (!failrec)
1608 return -ENOMEM;
1609 failrec->start = start;
1610 failrec->len = end - start + 1;
1611 failrec->last_mirror = 0;
d20f7043 1612 failrec->bio_flags = 0;
7e38326f 1613
3b951516
CM		 1614 spin_lock(&em_tree->lock);
1615 em = lookup_extent_mapping(em_tree, start, failrec->len);
1616 if (em->start > start || em->start + em->len < start) {
1617 free_extent_map(em);
1618 em = NULL;
1619 }
1620 spin_unlock(&em_tree->lock);
7e38326f
CM		 1621
1622 if (!em || IS_ERR(em)) {
1623 kfree(failrec);
1624 return -EIO;
1625 }
1626 logical = start - em->start;
1627 logical = em->block_start + logical;
d20f7043
CM		 1628 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1629 logical = em->block_start;
1630 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1631 }
7e38326f
CM		 1632 failrec->logical = logical;
1633 free_extent_map(em);
1634 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1635 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1636 set_state_private(failure_tree, start,
1637 (u64)(unsigned long)failrec);
7e38326f 1638 } else {
587f7704 1639 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1640 }
1641 num_copies = btrfs_num_copies(
1642 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1643 failrec->logical, failrec->len);
1644 failrec->last_mirror++;
1645 if (!state) {
cad321ad 1646 spin_lock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1647 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1648 failrec->start,
1649 EXTENT_LOCKED);
1650 if (state && state->start != failrec->start)
1651 state = NULL;
cad321ad 1652 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1653 }
1654 if (!state || failrec->last_mirror > num_copies) {
1655 set_state_private(failure_tree, failrec->start, 0);
1656 clear_extent_bits(failure_tree, failrec->start,
1657 failrec->start + failrec->len - 1,
1658 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1659 kfree(failrec);
1660 return -EIO;
1661 }
1662 bio = bio_alloc(GFP_NOFS, 1);
1663 bio->bi_private = state;
1664 bio->bi_end_io = failed_bio->bi_end_io;
1665 bio->bi_sector = failrec->logical >> 9;
1666 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1667 bio->bi_size = 0;
d20f7043 1668
7e38326f 1669 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1670 if (failed_bio->bi_rw & (1 << BIO_RW))
1671 rw = WRITE;
1672 else
1673 rw = READ;
1674
1675 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1676 failrec->last_mirror,
d20f7043 1677 failrec->bio_flags);
1259ab75
CM		 1678 return 0;
1679}
1680
d352ac68
CM		 1681/*
1682 * each time an IO finishes, we do a fast check in the IO failure tree
1683 * to see if we need to process or clean up an io_failure_record
1684 */
b2950863 1685static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1686{
1687 u64 private;
1688 u64 private_failure;
1689 struct io_failure_record *failure;
1690 int ret;
1691
1692 private = 0;
1693 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1694 (u64)-1, 1, EXTENT_DIRTY)) {
1695 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1696 start, &private_failure);
1697 if (ret == 0) {
1698 failure = (struct io_failure_record *)(unsigned long)
1699 private_failure;
1700 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1701 failure->start, 0);
1702 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1703 failure->start,
1704 failure->start + failure->len - 1,
1705 EXTENT_DIRTY | EXTENT_LOCKED,
1706 GFP_NOFS);
1707 kfree(failure);
1708 }
1709 }
7e38326f
CM		 1710 return 0;
1711}
1712
d352ac68
CM		 1713/*
1714 * when reads are done, we need to check csums to verify the data is correct
1715 * if there's a match, we allow the bio to finish. If not, we go through
1716 * the io_failure_record routines to find good copies
1717 */
b2950863 1718static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1719 struct extent_state *state)
07157aac 1720{
35ebb934 1721 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1722 struct inode *inode = page->mapping->host;
d1310b2e 1723 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1724 char *kaddr;
aadfeb6e 1725 u64 private = ~(u32)0;
07157aac 1726 int ret;
ff79f819
CM		 1727 struct btrfs_root *root = BTRFS_I(inode)->root;
1728 u32 csum = ~(u32)0;
d1310b2e 1729
d20f7043
CM		 1730 if (PageChecked(page)) {
1731 ClearPageChecked(page);
1732 goto good;
1733 }
17d217fe
YZ		 1734 if (btrfs_test_flag(inode, NODATASUM))
1735 return 0;
1736
1737 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1738 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1739 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1740 GFP_NOFS);
b6cda9bc 1741 return 0;
17d217fe 1742 }
d20f7043 1743
c2e639f0 1744 if (state && state->start == start) {
70dec807
CM		 1745 private = state->private;
1746 ret = 0;
1747 } else {
1748 ret = get_state_private(io_tree, start, &private);
1749 }
9ab86c8e 1750 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1751 if (ret)
07157aac 1752 goto zeroit;
d397712b 1753
ff79f819
CM		 1754 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1755 btrfs_csum_final(csum, (char *)&csum);
d397712b 1756 if (csum != private)
07157aac 1757 goto zeroit;
d397712b 1758
9ab86c8e 1759 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1760good:
7e38326f
CM		 1761 /* if the io failure tree for this inode is non-empty,
1762 * check to see if we've recovered from a failed IO
1763 */
1259ab75 1764 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1765 return 0;
1766
1767zeroit:
d397712b
CM		 1768 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1769 "private %llu\n", page->mapping->host->i_ino,
1770 (unsigned long long)start, csum,
1771 (unsigned long long)private);
db94535d
CM		 1772 memset(kaddr + offset, 1, end - start + 1);
1773 flush_dcache_page(page);
9ab86c8e 1774 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1775 if (private == 0)
1776 return 0;
7e38326f 1777 return -EIO;
07157aac 1778}
b888db2b 1779
7b128766
JB		 1780/*
1781 * This creates an orphan entry for the given inode in case something goes
1782 * wrong in the middle of an unlink/truncate.
1783 */
1784int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1785{
1786 struct btrfs_root *root = BTRFS_I(inode)->root;
1787 int ret = 0;
1788
bcc63abb 1789 spin_lock(&root->list_lock);
7b128766
JB		 1790
1791 /* already on the orphan list, we're good */
1792 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1793 spin_unlock(&root->list_lock);
7b128766
JB		 1794 return 0;
1795 }
1796
1797 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1798
bcc63abb 1799 spin_unlock(&root->list_lock);
7b128766
JB		 1800
1801 /*
1802 * insert an orphan item to track this unlinked/truncated file
1803 */
1804 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1805
1806 return ret;
1807}
1808
1809/*
1810 * We have done the truncate/delete so we can go ahead and remove the orphan
1811 * item for this particular inode.
1812 */
1813int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1814{
1815 struct btrfs_root *root = BTRFS_I(inode)->root;
1816 int ret = 0;
1817
bcc63abb 1818 spin_lock(&root->list_lock);
7b128766
JB		 1819
1820 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1821 spin_unlock(&root->list_lock);
7b128766
JB		 1822 return 0;
1823 }
1824
1825 list_del_init(&BTRFS_I(inode)->i_orphan);
1826 if (!trans) {
bcc63abb 1827 spin_unlock(&root->list_lock);
7b128766
JB		 1828 return 0;
1829 }
1830
bcc63abb 1831 spin_unlock(&root->list_lock);
7b128766
JB		 1832
1833 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1834
1835 return ret;
1836}
1837
1838/*
1839 * this cleans up any orphans that may be left on the list from the last use
1840 * of this root.
1841 */
1842void btrfs_orphan_cleanup(struct btrfs_root *root)
1843{
1844 struct btrfs_path *path;
1845 struct extent_buffer *leaf;
1846 struct btrfs_item *item;
1847 struct btrfs_key key, found_key;
1848 struct btrfs_trans_handle *trans;
1849 struct inode *inode;
1850 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1851
7b128766
JB		 1852 path = btrfs_alloc_path();
1853 if (!path)
1854 return;
1855 path->reada = -1;
1856
1857 key.objectid = BTRFS_ORPHAN_OBJECTID;
1858 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1859 key.offset = (u64)-1;
1860
7b128766
JB		 1861
1862 while (1) {
1863 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1864 if (ret < 0) {
1865 printk(KERN_ERR "Error searching slot for orphan: %d"
1866 "\n", ret);
1867 break;
1868 }
1869
1870 /*
1871 * if ret == 0 means we found what we were searching for, which
1872 * is weird, but possible, so only screw with path if we didnt
1873 * find the key and see if we have stuff that matches
1874 */
1875 if (ret > 0) {
1876 if (path->slots[0] == 0)
1877 break;
1878 path->slots[0]--;
1879 }
1880
1881 /* pull out the item */
1882 leaf = path->nodes[0];
1883 item = btrfs_item_nr(leaf, path->slots[0]);
1884 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1885
1886 /* make sure the item matches what we want */
1887 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1888 break;
1889 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1890 break;
1891
1892 /* release the path since we're done with it */
1893 btrfs_release_path(root, path);
1894
1895 /*
1896 * this is where we are basically btrfs_lookup, without the
1897 * crossing root thing. we store the inode number in the
1898 * offset of the orphan item.
1899 */
5b21f2ed 1900 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1901 found_key.offset, root);
1902 if (!inode)
1903 break;
1904
1905 if (inode->i_state & I_NEW) {
1906 BTRFS_I(inode)->root = root;
1907
1908 /* have to set the location manually */
1909 BTRFS_I(inode)->location.objectid = inode->i_ino;
1910 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1911 BTRFS_I(inode)->location.offset = 0;
1912
1913 btrfs_read_locked_inode(inode);
1914 unlock_new_inode(inode);
1915 }
1916
1917 /*
1918 * add this inode to the orphan list so btrfs_orphan_del does
1919 * the proper thing when we hit it
1920 */
bcc63abb 1921 spin_lock(&root->list_lock);
7b128766 1922 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1923 spin_unlock(&root->list_lock);
7b128766
JB		 1924
1925 /*
1926 * if this is a bad inode, means we actually succeeded in
1927 * removing the inode, but not the orphan record, which means
1928 * we need to manually delete the orphan since iput will just
1929 * do a destroy_inode
1930 */
1931 if (is_bad_inode(inode)) {
5b21f2ed 1932 trans = btrfs_start_transaction(root, 1);
7b128766 1933 btrfs_orphan_del(trans, inode);
5b21f2ed 1934 btrfs_end_transaction(trans, root);
7b128766
JB		 1935 iput(inode);
1936 continue;
1937 }
1938
1939 /* if we have links, this was a truncate, lets do that */
1940 if (inode->i_nlink) {
1941 nr_truncate++;
1942 btrfs_truncate(inode);
1943 } else {
1944 nr_unlink++;
1945 }
1946
1947 /* this will do delete_inode and everything for us */
1948 iput(inode);
1949 }
1950
1951 if (nr_unlink)
1952 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1953 if (nr_truncate)
1954 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1955
1956 btrfs_free_path(path);
7b128766
JB		 1957}
1958
d352ac68
CM		 1959/*
1960 * read an inode from the btree into the in-memory inode
1961 */
39279cc3
CM		 1962void btrfs_read_locked_inode(struct inode *inode)
1963{
1964 struct btrfs_path *path;
5f39d397 1965 struct extent_buffer *leaf;
39279cc3 1966 struct btrfs_inode_item *inode_item;
0b86a832 1967 struct btrfs_timespec *tspec;
39279cc3
CM		 1968 struct btrfs_root *root = BTRFS_I(inode)->root;
1969 struct btrfs_key location;
1970 u64 alloc_group_block;
618e21d5 1971 u32 rdev;
39279cc3
CM		 1972 int ret;
1973
1974 path = btrfs_alloc_path();
1975 BUG_ON(!path);
39279cc3 1976 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 1977
39279cc3 1978 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 1979 if (ret)
39279cc3 1980 goto make_bad;
39279cc3 1981
5f39d397
CM		 1982 leaf = path->nodes[0];
1983 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1984 struct btrfs_inode_item);
1985
1986 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1987 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1988 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1989 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 1990 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 1991
1992 tspec = btrfs_inode_atime(inode_item);
1993 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1994 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1995
1996 tspec = btrfs_inode_mtime(inode_item);
1997 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1998 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1999
2000 tspec = btrfs_inode_ctime(inode_item);
2001 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2002 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2003
a76a3cd4 2004 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2005 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2006 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2007 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2008 inode->i_rdev = 0;
5f39d397
CM		 2009 rdev = btrfs_inode_rdev(leaf, inode_item);
2010
aec7477b 2011 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2012 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2013
5f39d397 2014 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
b4ce94de 2015
d2fb3437
YZ		 2016 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2017 alloc_group_block, 0);
39279cc3
CM		 2018 btrfs_free_path(path);
2019 inode_item = NULL;
2020
39279cc3 2021 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2022 case S_IFREG:
2023 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2024 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2025 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2026 inode->i_fop = &btrfs_file_operations;
2027 inode->i_op = &btrfs_file_inode_operations;
2028 break;
2029 case S_IFDIR:
2030 inode->i_fop = &btrfs_dir_file_operations;
2031 if (root == root->fs_info->tree_root)
2032 inode->i_op = &btrfs_dir_ro_inode_operations;
2033 else
2034 inode->i_op = &btrfs_dir_inode_operations;
2035 break;
2036 case S_IFLNK:
2037 inode->i_op = &btrfs_symlink_inode_operations;
2038 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2039 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2040 break;
618e21d5 2041 default:
0279b4cd 2042 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2043 init_special_inode(inode, inode->i_mode, rdev);
2044 break;
39279cc3
CM		 2045 }
2046 return;
2047
2048make_bad:
39279cc3 2049 btrfs_free_path(path);
39279cc3
CM		 2050 make_bad_inode(inode);
2051}
2052
d352ac68
CM		 2053/*
2054 * given a leaf and an inode, copy the inode fields into the leaf
2055 */
e02119d5
CM		 2056static void fill_inode_item(struct btrfs_trans_handle *trans,
2057 struct extent_buffer *leaf,
5f39d397 2058 struct btrfs_inode_item *item,
39279cc3
CM		 2059 struct inode *inode)
2060{
5f39d397
CM		 2061 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2062 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2063 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2064 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2065 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2066
2067 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2068 inode->i_atime.tv_sec);
2069 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2070 inode->i_atime.tv_nsec);
2071
2072 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2073 inode->i_mtime.tv_sec);
2074 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2075 inode->i_mtime.tv_nsec);
2076
2077 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2078 inode->i_ctime.tv_sec);
2079 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2080 inode->i_ctime.tv_nsec);
2081
a76a3cd4 2082 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2083 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2084 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2085 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2086 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2087 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2088 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2089}
2090
d352ac68
CM		 2091/*
2092 * copy everything in the in-memory inode into the btree.
2093 */
d397712b
CM		 2094noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2096{
2097 struct btrfs_inode_item *inode_item;
2098 struct btrfs_path *path;
5f39d397 2099 struct extent_buffer *leaf;
39279cc3
CM		 2100 int ret;
2101
2102 path = btrfs_alloc_path();
2103 BUG_ON(!path);
39279cc3
CM		 2104 ret = btrfs_lookup_inode(trans, root, path,
2105 &BTRFS_I(inode)->location, 1);
2106 if (ret) {
2107 if (ret > 0)
2108 ret = -ENOENT;
2109 goto failed;
2110 }
2111
b4ce94de 2112 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2113 leaf = path->nodes[0];
2114 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2115 struct btrfs_inode_item);
2116
e02119d5 2117 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2118 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2119 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2120 ret = 0;
2121failed:
39279cc3
CM		 2122 btrfs_free_path(path);
2123 return ret;
2124}
2125
2126
d352ac68
CM		 2127/*
2128 * unlink helper that gets used here in inode.c and in the tree logging
2129 * recovery code. It remove a link in a directory with a given name, and
2130 * also drops the back refs in the inode to the directory
2131 */
e02119d5
CM		 2132int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct inode *dir, struct inode *inode,
2135 const char *name, int name_len)
39279cc3
CM		 2136{
2137 struct btrfs_path *path;
39279cc3 2138 int ret = 0;
5f39d397 2139 struct extent_buffer *leaf;
39279cc3 2140 struct btrfs_dir_item *di;
5f39d397 2141 struct btrfs_key key;
aec7477b 2142 u64 index;
39279cc3
CM		 2143
2144 path = btrfs_alloc_path();
54aa1f4d
CM		 2145 if (!path) {
2146 ret = -ENOMEM;
2147 goto err;
2148 }
2149
39279cc3
CM		 2150 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2151 name, name_len, -1);
2152 if (IS_ERR(di)) {
2153 ret = PTR_ERR(di);
2154 goto err;
2155 }
2156 if (!di) {
2157 ret = -ENOENT;
2158 goto err;
2159 }
5f39d397
CM		 2160 leaf = path->nodes[0];
2161 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2162 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2163 if (ret)
2164 goto err;
39279cc3
CM		 2165 btrfs_release_path(root, path);
2166
aec7477b 2167 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2168 inode->i_ino,
2169 dir->i_ino, &index);
aec7477b 2170 if (ret) {
d397712b 2171 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
aec7477b 2172 "inode %lu parent %lu\n", name_len, name,
e02119d5 2173 inode->i_ino, dir->i_ino);
aec7477b
JB		 2174 goto err;
2175 }
2176
39279cc3 2177 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2178 index, name, name_len, -1);
39279cc3
CM		 2179 if (IS_ERR(di)) {
2180 ret = PTR_ERR(di);
2181 goto err;
2182 }
2183 if (!di) {
2184 ret = -ENOENT;
2185 goto err;
2186 }
2187 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2188 btrfs_release_path(root, path);
39279cc3 2189
e02119d5
CM		 2190 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2191 inode, dir->i_ino);
49eb7e46
CM		 2192 BUG_ON(ret != 0 && ret != -ENOENT);
2193 if (ret != -ENOENT)
2194 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2195
2196 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2197 dir, index);
2198 BUG_ON(ret);
39279cc3
CM		 2199err:
2200 btrfs_free_path(path);
e02119d5
CM		 2201 if (ret)
2202 goto out;
2203
2204 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2205 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2206 btrfs_update_inode(trans, root, dir);
2207 btrfs_drop_nlink(inode);
2208 ret = btrfs_update_inode(trans, root, inode);
2209 dir->i_sb->s_dirt = 1;
2210out:
39279cc3
CM		 2211 return ret;
2212}
2213
2214static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2215{
2216 struct btrfs_root *root;
2217 struct btrfs_trans_handle *trans;
7b128766 2218 struct inode *inode = dentry->d_inode;
39279cc3 2219 int ret;
1832a6d5 2220 unsigned long nr = 0;
39279cc3
CM		 2221
2222 root = BTRFS_I(dir)->root;
1832a6d5 2223
39279cc3 2224 trans = btrfs_start_transaction(root, 1);
5f39d397 2225
39279cc3 2226 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2227 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2228 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2229
2230 if (inode->i_nlink == 0)
2231 ret = btrfs_orphan_add(trans, inode);
2232
d3c2fdcf 2233 nr = trans->blocks_used;
5f39d397 2234
89ce8a63 2235 btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2236 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2237 return ret;
2238}
2239
2240static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2241{
2242 struct inode *inode = dentry->d_inode;
1832a6d5 2243 int err = 0;
39279cc3
CM		 2244 int ret;
2245 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2246 struct btrfs_trans_handle *trans;
1832a6d5 2247 unsigned long nr = 0;
39279cc3 2248
3394e160
CM		 2249 /*
2250 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2251 * the root of a subvolume or snapshot
2252 */
2253 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2254 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2255 return -ENOTEMPTY;
925baedd 2256 }
134d4512 2257
39279cc3
CM		 2258 trans = btrfs_start_transaction(root, 1);
2259 btrfs_set_trans_block_group(trans, dir);
39279cc3 2260
7b128766
JB		 2261 err = btrfs_orphan_add(trans, inode);
2262 if (err)
2263 goto fail_trans;
2264
39279cc3 2265 /* now the directory is empty */
e02119d5
CM		 2266 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2267 dentry->d_name.name, dentry->d_name.len);
d397712b 2268 if (!err)
dbe674a9 2269 btrfs_i_size_write(inode, 0);
3954401f 2270
7b128766 2271fail_trans:
d3c2fdcf 2272 nr = trans->blocks_used;
89ce8a63 2273 ret = btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2274 btrfs_btree_balance_dirty(root, nr);
3954401f 2275
39279cc3
CM		 2276 if (ret && !err)
2277 err = ret;
2278 return err;
2279}
2280
d20f7043 2281#if 0
323ac95b
CM		 2282/*
2283 * when truncating bytes in a file, it is possible to avoid reading
2284 * the leaves that contain only checksum items. This can be the
2285 * majority of the IO required to delete a large file, but it must
2286 * be done carefully.
2287 *
2288 * The keys in the level just above the leaves are checked to make sure
2289 * the lowest key in a given leaf is a csum key, and starts at an offset
2290 * after the new size.
2291 *
2292 * Then the key for the next leaf is checked to make sure it also has
2293 * a checksum item for the same file. If it does, we know our target leaf
2294 * contains only checksum items, and it can be safely freed without reading
2295 * it.
2296 *
2297 * This is just an optimization targeted at large files. It may do
2298 * nothing. It will return 0 unless things went badly.
2299 */
2300static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2301 struct btrfs_root *root,
2302 struct btrfs_path *path,
2303 struct inode *inode, u64 new_size)
2304{
2305 struct btrfs_key key;
2306 int ret;
2307 int nritems;
2308 struct btrfs_key found_key;
2309 struct btrfs_key other_key;
5b84e8d6
YZ		 2310 struct btrfs_leaf_ref *ref;
2311 u64 leaf_gen;
2312 u64 leaf_start;
323ac95b
CM		 2313
2314 path->lowest_level = 1;
2315 key.objectid = inode->i_ino;
2316 key.type = BTRFS_CSUM_ITEM_KEY;
2317 key.offset = new_size;
2318again:
2319 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2320 if (ret < 0)
2321 goto out;
2322
2323 if (path->nodes[1] == NULL) {
2324 ret = 0;
2325 goto out;
2326 }
2327 ret = 0;
2328 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2329 nritems = btrfs_header_nritems(path->nodes[1]);
2330
2331 if (!nritems)
2332 goto out;
2333
2334 if (path->slots[1] >= nritems)
2335 goto next_node;
2336
2337 /* did we find a key greater than anything we want to delete? */
2338 if (found_key.objectid > inode->i_ino ||
2339 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2340 goto out;
2341
2342 /* we check the next key in the node to make sure the leave contains
2343 * only checksum items. This comparison doesn't work if our
2344 * leaf is the last one in the node
2345 */
2346 if (path->slots[1] + 1 >= nritems) {
2347next_node:
2348 /* search forward from the last key in the node, this
2349 * will bring us into the next node in the tree
2350 */
2351 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2352
2353 /* unlikely, but we inc below, so check to be safe */
2354 if (found_key.offset == (u64)-1)
2355 goto out;
2356
2357 /* search_forward needs a path with locks held, do the
2358 * search again for the original key. It is possible
2359 * this will race with a balance and return a path that
2360 * we could modify, but this drop is just an optimization
2361 * and is allowed to miss some leaves.
2362 */
2363 btrfs_release_path(root, path);
2364 found_key.offset++;
2365
2366 /* setup a max key for search_forward */
2367 other_key.offset = (u64)-1;
2368 other_key.type = key.type;
2369 other_key.objectid = key.objectid;
2370
2371 path->keep_locks = 1;
2372 ret = btrfs_search_forward(root, &found_key, &other_key,
2373 path, 0, 0);
2374 path->keep_locks = 0;
2375 if (ret || found_key.objectid != key.objectid ||
2376 found_key.type != key.type) {
2377 ret = 0;
2378 goto out;
2379 }
2380
2381 key.offset = found_key.offset;
2382 btrfs_release_path(root, path);
2383 cond_resched();
2384 goto again;
2385 }
2386
2387 /* we know there's one more slot after us in the tree,
2388 * read that key so we can verify it is also a checksum item
2389 */
2390 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2391
2392 if (found_key.objectid < inode->i_ino)
2393 goto next_key;
2394
2395 if (found_key.type != key.type || found_key.offset < new_size)
2396 goto next_key;
2397
2398 /*
2399 * if the key for the next leaf isn't a csum key from this objectid,
2400 * we can't be sure there aren't good items inside this leaf.
2401 * Bail out
2402 */
2403 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2404 goto out;
2405
5b84e8d6
YZ		 2406 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2407 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2408 /*
2409 * it is safe to delete this leaf, it contains only
2410 * csum items from this inode at an offset >= new_size
2411 */
5b84e8d6 2412 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2413 BUG_ON(ret);
2414
5b84e8d6
YZ		 2415 if (root->ref_cows && leaf_gen < trans->transid) {
2416 ref = btrfs_alloc_leaf_ref(root, 0);
2417 if (ref) {
2418 ref->root_gen = root->root_key.offset;
2419 ref->bytenr = leaf_start;
2420 ref->owner = 0;
2421 ref->generation = leaf_gen;
2422 ref->nritems = 0;
2423
bd56b302
CM		 2424 btrfs_sort_leaf_ref(ref);
2425
5b84e8d6
YZ		 2426 ret = btrfs_add_leaf_ref(root, ref, 0);
2427 WARN_ON(ret);
2428 btrfs_free_leaf_ref(root, ref);
2429 } else {
2430 WARN_ON(1);
2431 }
2432 }
323ac95b
CM		 2433next_key:
2434 btrfs_release_path(root, path);
2435
2436 if (other_key.objectid == inode->i_ino &&
2437 other_key.type == key.type && other_key.offset > key.offset) {
2438 key.offset = other_key.offset;
2439 cond_resched();
2440 goto again;
2441 }
2442 ret = 0;
2443out:
2444 /* fixup any changes we've made to the path */
2445 path->lowest_level = 0;
2446 path->keep_locks = 0;
2447 btrfs_release_path(root, path);
2448 return ret;
2449}
2450
d20f7043
CM		 2451#endif
2452
39279cc3
CM		 2453/*
2454 * this can truncate away extent items, csum items and directory items.
2455 * It starts at a high offset and removes keys until it can't find
d352ac68 2456 * any higher than new_size
39279cc3
CM		 2457 *
2458 * csum items that cross the new i_size are truncated to the new size
2459 * as well.
7b128766
JB		 2460 *
2461 * min_type is the minimum key type to truncate down to. If set to 0, this
2462 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2463 */
e02119d5
CM		 2464noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2465 struct btrfs_root *root,
2466 struct inode *inode,
2467 u64 new_size, u32 min_type)
39279cc3
CM		 2468{
2469 int ret;
2470 struct btrfs_path *path;
2471 struct btrfs_key key;
5f39d397 2472 struct btrfs_key found_key;
06d9a8d7 2473 u32 found_type = (u8)-1;
5f39d397 2474 struct extent_buffer *leaf;
39279cc3
CM		 2475 struct btrfs_file_extent_item *fi;
2476 u64 extent_start = 0;
db94535d 2477 u64 extent_num_bytes = 0;
39279cc3 2478 u64 item_end = 0;
7bb86316 2479 u64 root_gen = 0;
d8d5f3e1 2480 u64 root_owner = 0;
39279cc3
CM		 2481 int found_extent;
2482 int del_item;
85e21bac
CM		 2483 int pending_del_nr = 0;
2484 int pending_del_slot = 0;
179e29e4 2485 int extent_type = -1;
771ed689 2486 int encoding;
3b951516 2487 u64 mask = root->sectorsize - 1;
39279cc3 2488
e02119d5 2489 if (root->ref_cows)
5b21f2ed 2490 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2491 path = btrfs_alloc_path();
3c69faec 2492 path->reada = -1;
39279cc3 2493 BUG_ON(!path);
5f39d397 2494
39279cc3
CM		 2495 /* FIXME, add redo link to tree so we don't leak on crash */
2496 key.objectid = inode->i_ino;
2497 key.offset = (u64)-1;
5f39d397
CM		 2498 key.type = (u8)-1;
2499
85e21bac
CM		 2500search_again:
2501 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
d397712b 2502 if (ret < 0)
85e21bac 2503 goto error;
d397712b 2504
85e21bac 2505 if (ret > 0) {
e02119d5
CM		 2506 /* there are no items in the tree for us to truncate, we're
2507 * done
2508 */
2509 if (path->slots[0] == 0) {
2510 ret = 0;
2511 goto error;
2512 }
85e21bac
CM		 2513 path->slots[0]--;
2514 }
2515
d397712b 2516 while (1) {
39279cc3 2517 fi = NULL;
5f39d397
CM		 2518 leaf = path->nodes[0];
2519 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2520 found_type = btrfs_key_type(&found_key);
771ed689 2521 encoding = 0;
39279cc3 2522
5f39d397 2523 if (found_key.objectid != inode->i_ino)
39279cc3 2524 break;
5f39d397 2525
85e21bac 2526 if (found_type < min_type)
39279cc3
CM		 2527 break;
2528
5f39d397 2529 item_end = found_key.offset;
39279cc3 2530 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2531 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2532 struct btrfs_file_extent_item);
179e29e4 2533 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2534 encoding = btrfs_file_extent_compression(leaf, fi);
2535 encoding |= btrfs_file_extent_encryption(leaf, fi);
2536 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2537
179e29e4 2538 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2539 item_end +=
db94535d 2540 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2541 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2542 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2543 fi);
39279cc3 2544 }
008630c1 2545 item_end--;
39279cc3 2546 }
e02119d5 2547 if (item_end < new_size) {
d397712b 2548 if (found_type == BTRFS_DIR_ITEM_KEY)
b888db2b 2549 found_type = BTRFS_INODE_ITEM_KEY;
d397712b 2550 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
d20f7043 2551 found_type = BTRFS_EXTENT_DATA_KEY;
d397712b 2552 else if (found_type == BTRFS_EXTENT_DATA_KEY)
85e21bac 2553 found_type = BTRFS_XATTR_ITEM_KEY;
d397712b 2554 else if (found_type == BTRFS_XATTR_ITEM_KEY)
85e21bac 2555 found_type = BTRFS_INODE_REF_KEY;
d397712b 2556 else if (found_type)
b888db2b 2557 found_type--;
d397712b 2558 else
b888db2b 2559 break;
a61721d5 2560 btrfs_set_key_type(&key, found_type);
85e21bac 2561 goto next;
39279cc3 2562 }
e02119d5 2563 if (found_key.offset >= new_size)
39279cc3
CM		 2564 del_item = 1;
2565 else
2566 del_item = 0;
2567 found_extent = 0;
2568
2569 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2570 if (found_type != BTRFS_EXTENT_DATA_KEY)
2571 goto delete;
2572
2573 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2574 u64 num_dec;
db94535d 2575 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2576 if (!del_item && !encoding) {
db94535d
CM		 2577 u64 orig_num_bytes =
2578 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2579 extent_num_bytes = new_size -
5f39d397 2580 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2581 extent_num_bytes = extent_num_bytes &
2582 ~((u64)root->sectorsize - 1);
db94535d
CM		 2583 btrfs_set_file_extent_num_bytes(leaf, fi,
2584 extent_num_bytes);
2585 num_dec = (orig_num_bytes -
9069218d 2586 extent_num_bytes);
e02119d5 2587 if (root->ref_cows && extent_start != 0)
a76a3cd4 2588 inode_sub_bytes(inode, num_dec);
5f39d397 2589 btrfs_mark_buffer_dirty(leaf);
39279cc3 2590 } else {
db94535d
CM		 2591 extent_num_bytes =
2592 btrfs_file_extent_disk_num_bytes(leaf,
2593 fi);
39279cc3 2594 /* FIXME blocksize != 4096 */
9069218d 2595 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2596 if (extent_start != 0) {
2597 found_extent = 1;
e02119d5 2598 if (root->ref_cows)
a76a3cd4 2599 inode_sub_bytes(inode, num_dec);
e02119d5 2600 }
31840ae1 2601 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2602 root_owner = btrfs_header_owner(leaf);
39279cc3 2603 }
9069218d 2604 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2605 /*
2606 * we can't truncate inline items that have had
2607 * special encodings
2608 */
2609 if (!del_item &&
2610 btrfs_file_extent_compression(leaf, fi) == 0 &&
2611 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2612 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2613 u32 size = new_size - found_key.offset;
2614
2615 if (root->ref_cows) {
a76a3cd4
YZ		 2616 inode_sub_bytes(inode, item_end + 1 -
2617 new_size);
e02119d5
CM		 2618 }
2619 size =
2620 btrfs_file_extent_calc_inline_size(size);
9069218d 2621 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2622 size, 1);
9069218d 2623 BUG_ON(ret);
e02119d5 2624 } else if (root->ref_cows) {
a76a3cd4
YZ		 2625 inode_sub_bytes(inode, item_end + 1 -
2626 found_key.offset);
9069218d 2627 }
39279cc3 2628 }
179e29e4 2629delete:
39279cc3 2630 if (del_item) {
85e21bac
CM		 2631 if (!pending_del_nr) {
2632 /* no pending yet, add ourselves */
2633 pending_del_slot = path->slots[0];
2634 pending_del_nr = 1;
2635 } else if (pending_del_nr &&
2636 path->slots[0] + 1 == pending_del_slot) {
2637 /* hop on the pending chunk */
2638 pending_del_nr++;
2639 pending_del_slot = path->slots[0];
2640 } else {
d397712b 2641 BUG();
85e21bac 2642 }
39279cc3
CM		 2643 } else {
2644 break;
2645 }
39279cc3
CM		 2646 if (found_extent) {
2647 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2648 extent_num_bytes,
31840ae1 2649 leaf->start, root_owner,
3bb1a1bc 2650 root_gen, inode->i_ino, 0);
39279cc3
CM		 2651 BUG_ON(ret);
2652 }
85e21bac
CM		 2653next:
2654 if (path->slots[0] == 0) {
2655 if (pending_del_nr)
2656 goto del_pending;
2657 btrfs_release_path(root, path);
06d9a8d7
CM		 2658 if (found_type == BTRFS_INODE_ITEM_KEY)
2659 break;
85e21bac
CM		 2660 goto search_again;
2661 }
2662
2663 path->slots[0]--;
2664 if (pending_del_nr &&
2665 path->slots[0] + 1 != pending_del_slot) {
2666 struct btrfs_key debug;
2667del_pending:
2668 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2669 pending_del_slot);
2670 ret = btrfs_del_items(trans, root, path,
2671 pending_del_slot,
2672 pending_del_nr);
2673 BUG_ON(ret);
2674 pending_del_nr = 0;
2675 btrfs_release_path(root, path);
06d9a8d7
CM		 2676 if (found_type == BTRFS_INODE_ITEM_KEY)
2677 break;
85e21bac
CM		 2678 goto search_again;
2679 }
39279cc3
CM		 2680 }
2681 ret = 0;
2682error:
85e21bac
CM		 2683 if (pending_del_nr) {
2684 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2685 pending_del_nr);
2686 }
39279cc3
CM		 2687 btrfs_free_path(path);
2688 inode->i_sb->s_dirt = 1;
2689 return ret;
2690}
2691
2692/*
2693 * taken from block_truncate_page, but does cow as it zeros out
2694 * any bytes left in the last page in the file.
2695 */
2696static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2697{
2698 struct inode *inode = mapping->host;
db94535d 2699 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2700 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2701 struct btrfs_ordered_extent *ordered;
2702 char *kaddr;
db94535d 2703 u32 blocksize = root->sectorsize;
39279cc3
CM		 2704 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2705 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2706 struct page *page;
39279cc3 2707 int ret = 0;
a52d9a80 2708 u64 page_start;
e6dcd2dc 2709 u64 page_end;
39279cc3
CM		 2710
2711 if ((offset & (blocksize - 1)) == 0)
2712 goto out;
2713
2714 ret = -ENOMEM;
211c17f5 2715again:
39279cc3
CM		 2716 page = grab_cache_page(mapping, index);
2717 if (!page)
2718 goto out;
e6dcd2dc
CM		 2719
2720 page_start = page_offset(page);
2721 page_end = page_start + PAGE_CACHE_SIZE - 1;
2722
39279cc3 2723 if (!PageUptodate(page)) {
9ebefb18 2724 ret = btrfs_readpage(NULL, page);
39279cc3 2725 lock_page(page);
211c17f5
CM		 2726 if (page->mapping != mapping) {
2727 unlock_page(page);
2728 page_cache_release(page);
2729 goto again;
2730 }
39279cc3
CM		 2731 if (!PageUptodate(page)) {
2732 ret = -EIO;
89642229 2733 goto out_unlock;
39279cc3
CM		 2734 }
2735 }
211c17f5 2736 wait_on_page_writeback(page);
e6dcd2dc
CM		 2737
2738 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2739 set_page_extent_mapped(page);
2740
2741 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2742 if (ordered) {
2743 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2744 unlock_page(page);
2745 page_cache_release(page);
eb84ae03 2746 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2747 btrfs_put_ordered_extent(ordered);
2748 goto again;
2749 }
2750
ea8c2819 2751 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2752 ret = 0;
2753 if (offset != PAGE_CACHE_SIZE) {
2754 kaddr = kmap(page);
2755 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2756 flush_dcache_page(page);
2757 kunmap(page);
2758 }
247e743c 2759 ClearPageChecked(page);
e6dcd2dc
CM		 2760 set_page_dirty(page);
2761 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2762
89642229 2763out_unlock:
39279cc3
CM		 2764 unlock_page(page);
2765 page_cache_release(page);
2766out:
2767 return ret;
2768}
2769
9036c102 2770int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2771{
9036c102
YZ		 2772 struct btrfs_trans_handle *trans;
2773 struct btrfs_root *root = BTRFS_I(inode)->root;
2774 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2775 struct extent_map *em;
2776 u64 mask = root->sectorsize - 1;
2777 u64 hole_start = (inode->i_size + mask) & ~mask;
2778 u64 block_end = (size + mask) & ~mask;
2779 u64 last_byte;
2780 u64 cur_offset;
2781 u64 hole_size;
39279cc3
CM		 2782 int err;
2783
9036c102
YZ		 2784 if (size <= hole_start)
2785 return 0;
2786
6a63209f 2787 err = btrfs_check_metadata_free_space(root);
39279cc3
CM		 2788 if (err)
2789 return err;
2790
9036c102 2791 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2792
9036c102
YZ		 2793 while (1) {
2794 struct btrfs_ordered_extent *ordered;
2795 btrfs_wait_ordered_range(inode, hole_start,
2796 block_end - hole_start);
2797 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2798 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2799 if (!ordered)
2800 break;
2801 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2802 btrfs_put_ordered_extent(ordered);
2803 }
39279cc3 2804
9036c102
YZ		 2805 trans = btrfs_start_transaction(root, 1);
2806 btrfs_set_trans_block_group(trans, inode);
39279cc3 2807
9036c102
YZ		 2808 cur_offset = hole_start;
2809 while (1) {
2810 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2811 block_end - cur_offset, 0);
2812 BUG_ON(IS_ERR(em) || !em);
2813 last_byte = min(extent_map_end(em), block_end);
2814 last_byte = (last_byte + mask) & ~mask;
2815 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2816 u64 hint_byte = 0;
9036c102 2817 hole_size = last_byte - cur_offset;
771ed689
CM		 2818 err = btrfs_drop_extents(trans, root, inode,
2819 cur_offset,
2820 cur_offset + hole_size,
2821 cur_offset, &hint_byte);
2822 if (err)
2823 break;
9036c102
YZ		 2824 err = btrfs_insert_file_extent(trans, root,
2825 inode->i_ino, cur_offset, 0,
2826 0, hole_size, 0, hole_size,
2827 0, 0, 0);
2828 btrfs_drop_extent_cache(inode, hole_start,
2829 last_byte - 1, 0);
2830 }
2831 free_extent_map(em);
2832 cur_offset = last_byte;
2833 if (err || cur_offset >= block_end)
2834 break;
2835 }
1832a6d5 2836
9036c102
YZ		 2837 btrfs_end_transaction(trans, root);
2838 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2839 return err;
2840}
39279cc3 2841
9036c102
YZ		 2842static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2843{
2844 struct inode *inode = dentry->d_inode;
2845 int err;
39279cc3 2846
9036c102
YZ		 2847 err = inode_change_ok(inode, attr);
2848 if (err)
2849 return err;
2bf5a725 2850
9036c102
YZ		 2851 if (S_ISREG(inode->i_mode) &&
2852 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2853 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2854 if (err)
2855 return err;
39279cc3 2856 }
9036c102 2857
39279cc3 2858 err = inode_setattr(inode, attr);
33268eaf
JB		 2859
2860 if (!err && ((attr->ia_valid & ATTR_MODE)))
2861 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2862 return err;
2863}
61295eb8 2864
39279cc3
CM		 2865void btrfs_delete_inode(struct inode *inode)
2866{
2867 struct btrfs_trans_handle *trans;
2868 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2869 unsigned long nr;
39279cc3
CM		 2870 int ret;
2871
2872 truncate_inode_pages(&inode->i_data, 0);
2873 if (is_bad_inode(inode)) {
7b128766 2874 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2875 goto no_delete;
2876 }
4a096752 2877 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2878
dbe674a9 2879 btrfs_i_size_write(inode, 0);
180591bc 2880 trans = btrfs_join_transaction(root, 1);
5f39d397 2881
39279cc3 2882 btrfs_set_trans_block_group(trans, inode);
e02119d5 2883 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2884 if (ret) {
2885 btrfs_orphan_del(NULL, inode);
54aa1f4d 2886 goto no_delete_lock;
7b128766
JB		 2887 }
2888
2889 btrfs_orphan_del(trans, inode);
85e21bac 2890
d3c2fdcf 2891 nr = trans->blocks_used;
85e21bac 2892 clear_inode(inode);
5f39d397 2893
39279cc3 2894 btrfs_end_transaction(trans, root);
d3c2fdcf 2895 btrfs_btree_balance_dirty(root, nr);
39279cc3 2896 return;
54aa1f4d
CM		 2897
2898no_delete_lock:
d3c2fdcf 2899 nr = trans->blocks_used;
54aa1f4d 2900 btrfs_end_transaction(trans, root);
d3c2fdcf 2901 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2902no_delete:
2903 clear_inode(inode);
2904}
2905
2906/*
2907 * this returns the key found in the dir entry in the location pointer.
2908 * If no dir entries were found, location->objectid is 0.
2909 */
2910static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2911 struct btrfs_key *location)
2912{
2913 const char *name = dentry->d_name.name;
2914 int namelen = dentry->d_name.len;
2915 struct btrfs_dir_item *di;
2916 struct btrfs_path *path;
2917 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2918 int ret = 0;
39279cc3
CM		 2919
2920 path = btrfs_alloc_path();
2921 BUG_ON(!path);
3954401f 2922
39279cc3
CM		 2923 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2924 namelen, 0);
0d9f7f3e
Y		 2925 if (IS_ERR(di))
2926 ret = PTR_ERR(di);
d397712b
CM		 2927
2928 if (!di || IS_ERR(di))
3954401f 2929 goto out_err;
d397712b 2930
5f39d397 2931 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2932out:
39279cc3
CM		 2933 btrfs_free_path(path);
2934 return ret;
3954401f
CM		 2935out_err:
2936 location->objectid = 0;
2937 goto out;
39279cc3
CM		 2938}
2939
2940/*
2941 * when we hit a tree root in a directory, the btrfs part of the inode
2942 * needs to be changed to reflect the root directory of the tree root. This
2943 * is kind of like crossing a mount point.
2944 */
2945static int fixup_tree_root_location(struct btrfs_root *root,
2946 struct btrfs_key *location,
58176a96
JB		 2947 struct btrfs_root **sub_root,
2948 struct dentry *dentry)
39279cc3 2949{
39279cc3
CM		 2950 struct btrfs_root_item *ri;
2951
2952 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2953 return 0;
2954 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2955 return 0;
2956
58176a96
JB		 2957 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2958 dentry->d_name.name,
2959 dentry->d_name.len);
39279cc3
CM		 2960 if (IS_ERR(*sub_root))
2961 return PTR_ERR(*sub_root);
2962
2963 ri = &(*sub_root)->root_item;
2964 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2965 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2966 location->offset = 0;
2967
39279cc3
CM		 2968 return 0;
2969}
2970
e02119d5 2971static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2972{
e02119d5
CM		 2973 struct btrfs_inode *bi = BTRFS_I(inode);
2974
2975 bi->i_acl = NULL;
2976 bi->i_default_acl = NULL;
2977
2978 bi->generation = 0;
c3027eb5 2979 bi->sequence = 0;
e02119d5
CM		 2980 bi->last_trans = 0;
2981 bi->logged_trans = 0;
2982 bi->delalloc_bytes = 0;
6a63209f 2983 bi->reserved_bytes = 0;
e02119d5
CM		 2984 bi->disk_i_size = 0;
2985 bi->flags = 0;
2986 bi->index_cnt = (u64)-1;
49eb7e46 2987 bi->log_dirty_trans = 0;
d1310b2e
CM		 2988 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2989 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 2990 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 2991 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2992 inode->i_mapping, GFP_NOFS);
ea8c2819 2993 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 2994 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 2995 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 2996 mutex_init(&BTRFS_I(inode)->log_mutex);
2997}
2998
2999static int btrfs_init_locked_inode(struct inode *inode, void *p)
3000{
3001 struct btrfs_iget_args *args = p;
3002 inode->i_ino = args->ino;
3003 init_btrfs_i(inode);
3004 BTRFS_I(inode)->root = args->root;
6a63209f 3005 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3006 return 0;
3007}
3008
3009static int btrfs_find_actor(struct inode *inode, void *opaque)
3010{
3011 struct btrfs_iget_args *args = opaque;
d397712b
CM		 3012 return args->ino == inode->i_ino &&
3013 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3014}
3015
5b21f2ed
ZY		 3016struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3017 struct btrfs_root *root, int wait)
3018{
3019 struct inode *inode;
3020 struct btrfs_iget_args args;
3021 args.ino = objectid;
3022 args.root = root;
3023
3024 if (wait) {
3025 inode = ilookup5(s, objectid, btrfs_find_actor,
3026 (void *)&args);
3027 } else {
3028 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3029 (void *)&args);
3030 }
3031 return inode;
3032}
3033
39279cc3
CM		 3034struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3035 struct btrfs_root *root)
3036{
3037 struct inode *inode;
3038 struct btrfs_iget_args args;
3039 args.ino = objectid;
3040 args.root = root;
3041
3042 inode = iget5_locked(s, objectid, btrfs_find_actor,
3043 btrfs_init_locked_inode,
3044 (void *)&args);
3045 return inode;
3046}
3047
1a54ef8c
BR		 3048/* Get an inode object given its location and corresponding root.
3049 * Returns in *is_new if the inode was read from disk
3050 */
3051struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3052 struct btrfs_root *root, int *is_new)
3053{
3054 struct inode *inode;
3055
3056 inode = btrfs_iget_locked(s, location->objectid, root);
3057 if (!inode)
3058 return ERR_PTR(-EACCES);
3059
3060 if (inode->i_state & I_NEW) {
3061 BTRFS_I(inode)->root = root;
3062 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3063 btrfs_read_locked_inode(inode);
3064 unlock_new_inode(inode);
3065 if (is_new)
3066 *is_new = 1;
3067 } else {
3068 if (is_new)
3069 *is_new = 0;
3070 }
3071
3072 return inode;
3073}
3074
3de4586c 3075struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3076{
d397712b 3077 struct inode *inode;
39279cc3
CM		 3078 struct btrfs_inode *bi = BTRFS_I(dir);
3079 struct btrfs_root *root = bi->root;
3080 struct btrfs_root *sub_root = root;
3081 struct btrfs_key location;
c146afad 3082 int ret, new;
39279cc3
CM		 3083
3084 if (dentry->d_name.len > BTRFS_NAME_LEN)
3085 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3086
39279cc3 3087 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3088
39279cc3
CM		 3089 if (ret < 0)
3090 return ERR_PTR(ret);
5f39d397 3091
39279cc3
CM		 3092 inode = NULL;
3093 if (location.objectid) {
58176a96
JB		 3094 ret = fixup_tree_root_location(root, &location, &sub_root,
3095 dentry);
39279cc3
CM		 3096 if (ret < 0)
3097 return ERR_PTR(ret);
3098 if (ret > 0)
3099 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3100 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3101 if (IS_ERR(inode))
3102 return ERR_CAST(inode);
39279cc3 3103 }
3de4586c
CM		 3104 return inode;
3105}
3106
3107static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3108 struct nameidata *nd)
3109{
3110 struct inode *inode;
3111
3112 if (dentry->d_name.len > BTRFS_NAME_LEN)
3113 return ERR_PTR(-ENAMETOOLONG);
3114
3115 inode = btrfs_lookup_dentry(dir, dentry);
3116 if (IS_ERR(inode))
3117 return ERR_CAST(inode);
7b128766 3118
39279cc3
CM		 3119 return d_splice_alias(inode, dentry);
3120}
3121
39279cc3
CM		 3122static unsigned char btrfs_filetype_table[] = {
3123 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3124};
3125
cbdf5a24
DW		 3126static int btrfs_real_readdir(struct file *filp, void *dirent,
3127 filldir_t filldir)
39279cc3 3128{
6da6abae 3129 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3130 struct btrfs_root *root = BTRFS_I(inode)->root;
3131 struct btrfs_item *item;
3132 struct btrfs_dir_item *di;
3133 struct btrfs_key key;
5f39d397 3134 struct btrfs_key found_key;
39279cc3
CM		 3135 struct btrfs_path *path;
3136 int ret;
3137 u32 nritems;
5f39d397 3138 struct extent_buffer *leaf;
39279cc3
CM		 3139 int slot;
3140 int advance;
3141 unsigned char d_type;
3142 int over = 0;
3143 u32 di_cur;
3144 u32 di_total;
3145 u32 di_len;
3146 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3147 char tmp_name[32];
3148 char *name_ptr;
3149 int name_len;
39279cc3
CM		 3150
3151 /* FIXME, use a real flag for deciding about the key type */
3152 if (root->fs_info->tree_root == root)
3153 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3154
3954401f
CM		 3155 /* special case for "." */
3156 if (filp->f_pos == 0) {
3157 over = filldir(dirent, ".", 1,
3158 1, inode->i_ino,
3159 DT_DIR);
3160 if (over)
3161 return 0;
3162 filp->f_pos = 1;
3163 }
3954401f
CM		 3164 /* special case for .., just use the back ref */
3165 if (filp->f_pos == 1) {
5ecc7e5d 3166 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3167 over = filldir(dirent, "..", 2,
5ecc7e5d 3168 2, pino, DT_DIR);
3954401f 3169 if (over)
49593bfa 3170 return 0;
3954401f
CM		 3171 filp->f_pos = 2;
3172 }
49593bfa
DW		 3173 path = btrfs_alloc_path();
3174 path->reada = 2;
3175
39279cc3
CM		 3176 btrfs_set_key_type(&key, key_type);
3177 key.offset = filp->f_pos;
49593bfa 3178 key.objectid = inode->i_ino;
5f39d397 3179
39279cc3
CM		 3180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3181 if (ret < 0)
3182 goto err;
3183 advance = 0;
49593bfa
DW		 3184
3185 while (1) {
5f39d397
CM		 3186 leaf = path->nodes[0];
3187 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3188 slot = path->slots[0];
3189 if (advance || slot >= nritems) {
49593bfa 3190 if (slot >= nritems - 1) {
39279cc3
CM		 3191 ret = btrfs_next_leaf(root, path);
3192 if (ret)
3193 break;
5f39d397
CM		 3194 leaf = path->nodes[0];
3195 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3196 slot = path->slots[0];
3197 } else {
3198 slot++;
3199 path->slots[0]++;
3200 }
3201 }
3de4586c 3202
39279cc3 3203 advance = 1;
5f39d397
CM		 3204 item = btrfs_item_nr(leaf, slot);
3205 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3206
3207 if (found_key.objectid != key.objectid)
39279cc3 3208 break;
5f39d397 3209 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3210 break;
5f39d397 3211 if (found_key.offset < filp->f_pos)
39279cc3 3212 continue;
5f39d397
CM		 3213
3214 filp->f_pos = found_key.offset;
49593bfa 3215
39279cc3
CM		 3216 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3217 di_cur = 0;
5f39d397 3218 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3219
3220 while (di_cur < di_total) {
5f39d397
CM		 3221 struct btrfs_key location;
3222
3223 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3224 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3225 name_ptr = tmp_name;
3226 } else {
3227 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3228 if (!name_ptr) {
3229 ret = -ENOMEM;
3230 goto err;
3231 }
5f39d397
CM		 3232 }
3233 read_extent_buffer(leaf, name_ptr,
3234 (unsigned long)(di + 1), name_len);
3235
3236 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3237 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3238
3239 /* is this a reference to our own snapshot? If so
3240 * skip it
3241 */
3242 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3243 location.objectid == root->root_key.objectid) {
3244 over = 0;
3245 goto skip;
3246 }
5f39d397 3247 over = filldir(dirent, name_ptr, name_len,
49593bfa 3248 found_key.offset, location.objectid,
39279cc3 3249 d_type);
5f39d397 3250
3de4586c 3251skip:
5f39d397
CM		 3252 if (name_ptr != tmp_name)
3253 kfree(name_ptr);
3254
39279cc3
CM		 3255 if (over)
3256 goto nopos;
5103e947 3257 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3258 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3259 di_cur += di_len;
3260 di = (struct btrfs_dir_item *)((char *)di + di_len);
3261 }
3262 }
49593bfa
DW		 3263
3264 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 3265 if (key_type == BTRFS_DIR_INDEX_KEY)
89f135d8 3266 filp->f_pos = INT_LIMIT(off_t);
5e591a07
YZ		 3267 else
3268 filp->f_pos++;
39279cc3
CM		 3269nopos:
3270 ret = 0;
3271err:
39279cc3 3272 btrfs_free_path(path);
39279cc3
CM		 3273 return ret;
3274}
3275
3276int btrfs_write_inode(struct inode *inode, int wait)
3277{
3278 struct btrfs_root *root = BTRFS_I(inode)->root;
3279 struct btrfs_trans_handle *trans;
3280 int ret = 0;
3281
c146afad 3282 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3283 return 0;
3284
39279cc3 3285 if (wait) {
f9295749 3286 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3287 btrfs_set_trans_block_group(trans, inode);
3288 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3289 }
3290 return ret;
3291}
3292
3293/*
54aa1f4d 3294 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3295 * inode changes. But, it is most likely to find the inode in cache.
3296 * FIXME, needs more benchmarking...there are no reasons other than performance
3297 * to keep or drop this code.
3298 */
3299void btrfs_dirty_inode(struct inode *inode)
3300{
3301 struct btrfs_root *root = BTRFS_I(inode)->root;
3302 struct btrfs_trans_handle *trans;
3303
f9295749 3304 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3305 btrfs_set_trans_block_group(trans, inode);
3306 btrfs_update_inode(trans, root, inode);
3307 btrfs_end_transaction(trans, root);
39279cc3
CM		 3308}
3309
d352ac68
CM		 3310/*
3311 * find the highest existing sequence number in a directory
3312 * and then set the in-memory index_cnt variable to reflect
3313 * free sequence numbers
3314 */
aec7477b
JB		 3315static int btrfs_set_inode_index_count(struct inode *inode)
3316{
3317 struct btrfs_root *root = BTRFS_I(inode)->root;
3318 struct btrfs_key key, found_key;
3319 struct btrfs_path *path;
3320 struct extent_buffer *leaf;
3321 int ret;
3322
3323 key.objectid = inode->i_ino;
3324 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3325 key.offset = (u64)-1;
3326
3327 path = btrfs_alloc_path();
3328 if (!path)
3329 return -ENOMEM;
3330
3331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3332 if (ret < 0)
3333 goto out;
3334 /* FIXME: we should be able to handle this */
3335 if (ret == 0)
3336 goto out;
3337 ret = 0;
3338
3339 /*
3340 * MAGIC NUMBER EXPLANATION:
3341 * since we search a directory based on f_pos we have to start at 2
3342 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3343 * else has to start at 2
3344 */
3345 if (path->slots[0] == 0) {
3346 BTRFS_I(inode)->index_cnt = 2;
3347 goto out;
3348 }
3349
3350 path->slots[0]--;
3351
3352 leaf = path->nodes[0];
3353 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3354
3355 if (found_key.objectid != inode->i_ino ||
3356 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3357 BTRFS_I(inode)->index_cnt = 2;
3358 goto out;
3359 }
3360
3361 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3362out:
3363 btrfs_free_path(path);
3364 return ret;
3365}
3366
d352ac68
CM		 3367/*
3368 * helper to find a free sequence number in a given directory. This current
3369 * code is very simple, later versions will do smarter things in the btree
3370 */
3de4586c 3371int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3372{
3373 int ret = 0;
3374
3375 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3376 ret = btrfs_set_inode_index_count(dir);
d397712b 3377 if (ret)
aec7477b
JB		 3378 return ret;
3379 }
3380
00e4e6b3 3381 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3382 BTRFS_I(dir)->index_cnt++;
3383
3384 return ret;
3385}
3386
39279cc3
CM		 3387static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3388 struct btrfs_root *root,
aec7477b 3389 struct inode *dir,
9c58309d 3390 const char *name, int name_len,
d2fb3437
YZ		 3391 u64 ref_objectid, u64 objectid,
3392 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3393{
3394 struct inode *inode;
5f39d397 3395 struct btrfs_inode_item *inode_item;
39279cc3 3396 struct btrfs_key *location;
5f39d397 3397 struct btrfs_path *path;
9c58309d
CM		 3398 struct btrfs_inode_ref *ref;
3399 struct btrfs_key key[2];
3400 u32 sizes[2];
3401 unsigned long ptr;
39279cc3
CM		 3402 int ret;
3403 int owner;
3404
5f39d397
CM		 3405 path = btrfs_alloc_path();
3406 BUG_ON(!path);
3407
39279cc3
CM		 3408 inode = new_inode(root->fs_info->sb);
3409 if (!inode)
3410 return ERR_PTR(-ENOMEM);
3411
aec7477b 3412 if (dir) {
3de4586c 3413 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB		 3414 if (ret)
3415 return ERR_PTR(ret);
aec7477b
JB		 3416 }
3417 /*
3418 * index_cnt is ignored for everything but a dir,
3419 * btrfs_get_inode_index_count has an explanation for the magic
3420 * number
3421 */
e02119d5 3422 init_btrfs_i(inode);
aec7477b 3423 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3424 BTRFS_I(inode)->root = root;
e02119d5 3425 BTRFS_I(inode)->generation = trans->transid;
6a63209f 3426 btrfs_set_inode_space_info(root, inode);
b888db2b 3427
39279cc3
CM		 3428 if (mode & S_IFDIR)
3429 owner = 0;
3430 else
3431 owner = 1;
d2fb3437
YZ		 3432 BTRFS_I(inode)->block_group =
3433 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ		 3434 if ((mode & S_IFREG)) {
3435 if (btrfs_test_opt(root, NODATASUM))
3436 btrfs_set_flag(inode, NODATASUM);
3437 if (btrfs_test_opt(root, NODATACOW))
3438 btrfs_set_flag(inode, NODATACOW);
3439 }
9c58309d
CM		 3440
3441 key[0].objectid = objectid;
3442 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3443 key[0].offset = 0;
3444
3445 key[1].objectid = objectid;
3446 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3447 key[1].offset = ref_objectid;
3448
3449 sizes[0] = sizeof(struct btrfs_inode_item);
3450 sizes[1] = name_len + sizeof(*ref);
3451
3452 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3453 if (ret != 0)
5f39d397
CM		 3454 goto fail;
3455
9c58309d
CM		 3456 if (objectid > root->highest_inode)
3457 root->highest_inode = objectid;
3458
79683f2d 3459 inode->i_uid = current_fsuid();
8c087b51 3460
42f15d77 3461 if (dir && (dir->i_mode & S_ISGID)) {
8c087b51
CB		 3462 inode->i_gid = dir->i_gid;
3463 if (S_ISDIR(mode))
3464 mode |= S_ISGID;
3465 } else
3466 inode->i_gid = current_fsgid();
3467
39279cc3
CM		 3468 inode->i_mode = mode;
3469 inode->i_ino = objectid;
a76a3cd4 3470 inode_set_bytes(inode, 0);
39279cc3 3471 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3472 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3473 struct btrfs_inode_item);
e02119d5 3474 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3475
3476 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3477 struct btrfs_inode_ref);
3478 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3479 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3480 ptr = (unsigned long)(ref + 1);
3481 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3482
5f39d397
CM		 3483 btrfs_mark_buffer_dirty(path->nodes[0]);
3484 btrfs_free_path(path);
3485
39279cc3
CM		 3486 location = &BTRFS_I(inode)->location;
3487 location->objectid = objectid;
39279cc3
CM		 3488 location->offset = 0;
3489 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3490
39279cc3
CM		 3491 insert_inode_hash(inode);
3492 return inode;
5f39d397 3493fail:
aec7477b
JB		 3494 if (dir)
3495 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3496 btrfs_free_path(path);
3497 return ERR_PTR(ret);
39279cc3
CM		 3498}
3499
3500static inline u8 btrfs_inode_type(struct inode *inode)
3501{
3502 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3503}
3504
d352ac68
CM		 3505/*
3506 * utility function to add 'inode' into 'parent_inode' with
3507 * a give name and a given sequence number.
3508 * if 'add_backref' is true, also insert a backref from the
3509 * inode to the parent directory.
3510 */
e02119d5
CM		 3511int btrfs_add_link(struct btrfs_trans_handle *trans,
3512 struct inode *parent_inode, struct inode *inode,
3513 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3514{
3515 int ret;
3516 struct btrfs_key key;
e02119d5 3517 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3518
39279cc3 3519 key.objectid = inode->i_ino;
39279cc3
CM		 3520 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3521 key.offset = 0;
3522
e02119d5
CM		 3523 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3524 parent_inode->i_ino,
aec7477b 3525 &key, btrfs_inode_type(inode),
00e4e6b3 3526 index);
39279cc3 3527 if (ret == 0) {
9c58309d
CM		 3528 if (add_backref) {
3529 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3530 name, name_len,
3531 inode->i_ino,
3532 parent_inode->i_ino,
3533 index);
9c58309d 3534 }
dbe674a9 3535 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3536 name_len * 2);
79c44584 3537 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3538 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3539 }
3540 return ret;
3541}
3542
3543static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3544 struct dentry *dentry, struct inode *inode,
00e4e6b3 3545 int backref, u64 index)
39279cc3 3546{
e02119d5
CM		 3547 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3548 inode, dentry->d_name.name,
3549 dentry->d_name.len, backref, index);
39279cc3
CM		 3550 if (!err) {
3551 d_instantiate(dentry, inode);
3552 return 0;
3553 }
3554 if (err > 0)
3555 err = -EEXIST;
3556 return err;
3557}
3558
618e21d5
JB		 3559static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3560 int mode, dev_t rdev)
3561{
3562 struct btrfs_trans_handle *trans;
3563 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3564 struct inode *inode = NULL;
618e21d5
JB		 3565 int err;
3566 int drop_inode = 0;
3567 u64 objectid;
1832a6d5 3568 unsigned long nr = 0;
00e4e6b3 3569 u64 index = 0;
618e21d5
JB		 3570
3571 if (!new_valid_dev(rdev))
3572 return -EINVAL;
3573
6a63209f 3574 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3575 if (err)
3576 goto fail;
3577
618e21d5
JB		 3578 trans = btrfs_start_transaction(root, 1);
3579 btrfs_set_trans_block_group(trans, dir);
3580
3581 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3582 if (err) {
3583 err = -ENOSPC;
3584 goto out_unlock;
3585 }
3586
aec7477b 3587 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3588 dentry->d_name.len,
3589 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3590 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3591 err = PTR_ERR(inode);
3592 if (IS_ERR(inode))
3593 goto out_unlock;
3594
0279b4cd 3595 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3596 if (err) {
3597 drop_inode = 1;
3598 goto out_unlock;
3599 }
3600
618e21d5 3601 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3602 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3603 if (err)
3604 drop_inode = 1;
3605 else {
3606 inode->i_op = &btrfs_special_inode_operations;
3607 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3608 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3609 }
3610 dir->i_sb->s_dirt = 1;
3611 btrfs_update_inode_block_group(trans, inode);
3612 btrfs_update_inode_block_group(trans, dir);
3613out_unlock:
d3c2fdcf 3614 nr = trans->blocks_used;
89ce8a63 3615 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3616fail:
618e21d5
JB		 3617 if (drop_inode) {
3618 inode_dec_link_count(inode);
3619 iput(inode);
3620 }
d3c2fdcf 3621 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3622 return err;
3623}
3624
39279cc3
CM		 3625static int btrfs_create(struct inode *dir, struct dentry *dentry,
3626 int mode, struct nameidata *nd)
3627{
3628 struct btrfs_trans_handle *trans;
3629 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3630 struct inode *inode = NULL;
39279cc3
CM		 3631 int err;
3632 int drop_inode = 0;
1832a6d5 3633 unsigned long nr = 0;
39279cc3 3634 u64 objectid;
00e4e6b3 3635 u64 index = 0;
39279cc3 3636
6a63209f 3637 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3638 if (err)
3639 goto fail;
39279cc3
CM		 3640 trans = btrfs_start_transaction(root, 1);
3641 btrfs_set_trans_block_group(trans, dir);
3642
3643 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3644 if (err) {
3645 err = -ENOSPC;
3646 goto out_unlock;
3647 }
3648
aec7477b 3649 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3650 dentry->d_name.len,
3651 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3652 objectid, BTRFS_I(dir)->block_group, mode,
3653 &index);
39279cc3
CM		 3654 err = PTR_ERR(inode);
3655 if (IS_ERR(inode))
3656 goto out_unlock;
3657
0279b4cd 3658 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3659 if (err) {
3660 drop_inode = 1;
3661 goto out_unlock;
3662 }
3663
39279cc3 3664 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3665 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3666 if (err)
3667 drop_inode = 1;
3668 else {
3669 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3670 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3671 inode->i_fop = &btrfs_file_operations;
3672 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3673 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3674 }
3675 dir->i_sb->s_dirt = 1;
3676 btrfs_update_inode_block_group(trans, inode);
3677 btrfs_update_inode_block_group(trans, dir);
3678out_unlock:
d3c2fdcf 3679 nr = trans->blocks_used;
ab78c84d 3680 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3681fail:
39279cc3
CM		 3682 if (drop_inode) {
3683 inode_dec_link_count(inode);
3684 iput(inode);
3685 }
d3c2fdcf 3686 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3687 return err;
3688}
3689
3690static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3691 struct dentry *dentry)
3692{
3693 struct btrfs_trans_handle *trans;
3694 struct btrfs_root *root = BTRFS_I(dir)->root;
3695 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3696 u64 index;
1832a6d5 3697 unsigned long nr = 0;
39279cc3
CM		 3698 int err;
3699 int drop_inode = 0;
3700
3701 if (inode->i_nlink == 0)
3702 return -ENOENT;
3703
e02119d5 3704 btrfs_inc_nlink(inode);
6a63209f 3705 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3706 if (err)
3707 goto fail;
3de4586c 3708 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3709 if (err)
3710 goto fail;
3711
39279cc3 3712 trans = btrfs_start_transaction(root, 1);
5f39d397 3713
39279cc3
CM		 3714 btrfs_set_trans_block_group(trans, dir);
3715 atomic_inc(&inode->i_count);
aec7477b 3716
00e4e6b3 3717 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3718
39279cc3
CM		 3719 if (err)
3720 drop_inode = 1;
5f39d397 3721
39279cc3
CM		 3722 dir->i_sb->s_dirt = 1;
3723 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3724 err = btrfs_update_inode(trans, root, inode);
5f39d397 3725
54aa1f4d
CM		 3726 if (err)
3727 drop_inode = 1;
39279cc3 3728
d3c2fdcf 3729 nr = trans->blocks_used;
ab78c84d 3730 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3731fail:
39279cc3
CM		 3732 if (drop_inode) {
3733 inode_dec_link_count(inode);
3734 iput(inode);
3735 }
d3c2fdcf 3736 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3737 return err;
3738}
3739
39279cc3
CM		 3740static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3741{
b9d86667 3742 struct inode *inode = NULL;
39279cc3
CM		 3743 struct btrfs_trans_handle *trans;
3744 struct btrfs_root *root = BTRFS_I(dir)->root;
3745 int err = 0;
3746 int drop_on_err = 0;
b9d86667 3747 u64 objectid = 0;
00e4e6b3 3748 u64 index = 0;
d3c2fdcf 3749 unsigned long nr = 1;
39279cc3 3750
6a63209f 3751 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3752 if (err)
3753 goto out_unlock;
3754
39279cc3
CM		 3755 trans = btrfs_start_transaction(root, 1);
3756 btrfs_set_trans_block_group(trans, dir);
5f39d397 3757
39279cc3
CM		 3758 if (IS_ERR(trans)) {
3759 err = PTR_ERR(trans);
3760 goto out_unlock;
3761 }
3762
3763 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3764 if (err) {
3765 err = -ENOSPC;
3766 goto out_unlock;
3767 }
3768
aec7477b 3769 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3770 dentry->d_name.len,
3771 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3772 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3773 &index);
39279cc3
CM		 3774 if (IS_ERR(inode)) {
3775 err = PTR_ERR(inode);
3776 goto out_fail;
3777 }
5f39d397 3778
39279cc3 3779 drop_on_err = 1;
33268eaf 3780
0279b4cd 3781 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3782 if (err)
3783 goto out_fail;
3784
39279cc3
CM		 3785 inode->i_op = &btrfs_dir_inode_operations;
3786 inode->i_fop = &btrfs_dir_file_operations;
3787 btrfs_set_trans_block_group(trans, inode);
3788
dbe674a9 3789 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3790 err = btrfs_update_inode(trans, root, inode);
3791 if (err)
3792 goto out_fail;
5f39d397 3793
e02119d5
CM		 3794 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3795 inode, dentry->d_name.name,
3796 dentry->d_name.len, 0, index);
39279cc3
CM		 3797 if (err)
3798 goto out_fail;
5f39d397 3799
39279cc3
CM		 3800 d_instantiate(dentry, inode);
3801 drop_on_err = 0;
3802 dir->i_sb->s_dirt = 1;
3803 btrfs_update_inode_block_group(trans, inode);
3804 btrfs_update_inode_block_group(trans, dir);
3805
3806out_fail:
d3c2fdcf 3807 nr = trans->blocks_used;
ab78c84d 3808 btrfs_end_transaction_throttle(trans, root);
5f39d397 3809
39279cc3 3810out_unlock:
39279cc3
CM		 3811 if (drop_on_err)
3812 iput(inode);
d3c2fdcf 3813 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3814 return err;
3815}
3816
d352ac68
CM		 3817/* helper for btfs_get_extent. Given an existing extent in the tree,
3818 * and an extent that you want to insert, deal with overlap and insert
3819 * the new extent into the tree.
3820 */
3b951516
CM		 3821static int merge_extent_mapping(struct extent_map_tree *em_tree,
3822 struct extent_map *existing,
e6dcd2dc
CM		 3823 struct extent_map *em,
3824 u64 map_start, u64 map_len)
3b951516
CM		 3825{
3826 u64 start_diff;
3b951516 3827
e6dcd2dc
CM		 3828 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3829 start_diff = map_start - em->start;
3830 em->start = map_start;
3831 em->len = map_len;
c8b97818
CM		 3832 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3833 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3834 em->block_start += start_diff;
c8b97818
CM		 3835 em->block_len -= start_diff;
3836 }
e6dcd2dc 3837 return add_extent_mapping(em_tree, em);
3b951516
CM		 3838}
3839
c8b97818
CM		 3840static noinline int uncompress_inline(struct btrfs_path *path,
3841 struct inode *inode, struct page *page,
3842 size_t pg_offset, u64 extent_offset,
3843 struct btrfs_file_extent_item *item)
3844{
3845 int ret;
3846 struct extent_buffer *leaf = path->nodes[0];
3847 char *tmp;
3848 size_t max_size;
3849 unsigned long inline_size;
3850 unsigned long ptr;
3851
3852 WARN_ON(pg_offset != 0);
3853 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3854 inline_size = btrfs_file_extent_inline_item_len(leaf,
3855 btrfs_item_nr(leaf, path->slots[0]));
3856 tmp = kmalloc(inline_size, GFP_NOFS);
3857 ptr = btrfs_file_extent_inline_start(item);
3858
3859 read_extent_buffer(leaf, tmp, ptr, inline_size);
3860
5b050f04 3861 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3862 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3863 inline_size, max_size);
3864 if (ret) {
3865 char *kaddr = kmap_atomic(page, KM_USER0);
3866 unsigned long copy_size = min_t(u64,
3867 PAGE_CACHE_SIZE - pg_offset,
3868 max_size - extent_offset);
3869 memset(kaddr + pg_offset, 0, copy_size);
3870 kunmap_atomic(kaddr, KM_USER0);
3871 }
3872 kfree(tmp);
3873 return 0;
3874}
3875
d352ac68
CM		 3876/*
3877 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 3878 * the ugly parts come from merging extents from the disk with the in-ram
3879 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 3880 * where the in-ram extents might be locked pending data=ordered completion.
3881 *
3882 * This also copies inline extents directly into the page.
3883 */
d397712b 3884
a52d9a80 3885struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3886 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3887 int create)
3888{
3889 int ret;
3890 int err = 0;
db94535d 3891 u64 bytenr;
a52d9a80
CM		 3892 u64 extent_start = 0;
3893 u64 extent_end = 0;
3894 u64 objectid = inode->i_ino;
3895 u32 found_type;
f421950f 3896 struct btrfs_path *path = NULL;
a52d9a80
CM		 3897 struct btrfs_root *root = BTRFS_I(inode)->root;
3898 struct btrfs_file_extent_item *item;
5f39d397
CM		 3899 struct extent_buffer *leaf;
3900 struct btrfs_key found_key;
a52d9a80
CM		 3901 struct extent_map *em = NULL;
3902 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3903 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3904 struct btrfs_trans_handle *trans = NULL;
c8b97818 3905 int compressed;
a52d9a80 3906
a52d9a80 3907again:
d1310b2e
CM		 3908 spin_lock(&em_tree->lock);
3909 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3910 if (em)
3911 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3912 spin_unlock(&em_tree->lock);
3913
a52d9a80 3914 if (em) {
e1c4b745
CM		 3915 if (em->start > start || em->start + em->len <= start)
3916 free_extent_map(em);
3917 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3918 free_extent_map(em);
3919 else
3920 goto out;
a52d9a80 3921 }
d1310b2e 3922 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3923 if (!em) {
d1310b2e
CM		 3924 err = -ENOMEM;
3925 goto out;
a52d9a80 3926 }
e6dcd2dc 3927 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3928 em->start = EXTENT_MAP_HOLE;
445a6944 3929 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3930 em->len = (u64)-1;
c8b97818 3931 em->block_len = (u64)-1;
f421950f
CM		 3932
3933 if (!path) {
3934 path = btrfs_alloc_path();
3935 BUG_ON(!path);
3936 }
3937
179e29e4
CM		 3938 ret = btrfs_lookup_file_extent(trans, root, path,
3939 objectid, start, trans != NULL);
a52d9a80
CM		 3940 if (ret < 0) {
3941 err = ret;
3942 goto out;
3943 }
3944
3945 if (ret != 0) {
3946 if (path->slots[0] == 0)
3947 goto not_found;
3948 path->slots[0]--;
3949 }
3950
5f39d397
CM		 3951 leaf = path->nodes[0];
3952 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3953 struct btrfs_file_extent_item);
a52d9a80 3954 /* are we inside the extent that was found? */
5f39d397
CM		 3955 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3956 found_type = btrfs_key_type(&found_key);
3957 if (found_key.objectid != objectid ||
a52d9a80
CM		 3958 found_type != BTRFS_EXTENT_DATA_KEY) {
3959 goto not_found;
3960 }
3961
5f39d397
CM		 3962 found_type = btrfs_file_extent_type(leaf, item);
3963 extent_start = found_key.offset;
c8b97818 3964 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3965 if (found_type == BTRFS_FILE_EXTENT_REG ||
3966 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3967 extent_end = extent_start +
db94535d 3968 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3969 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3970 size_t size;
3971 size = btrfs_file_extent_inline_len(leaf, item);
3972 extent_end = (extent_start + size + root->sectorsize - 1) &
3973 ~((u64)root->sectorsize - 1);
3974 }
3975
3976 if (start >= extent_end) {
3977 path->slots[0]++;
3978 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3979 ret = btrfs_next_leaf(root, path);
3980 if (ret < 0) {
3981 err = ret;
3982 goto out;
a52d9a80 3983 }
9036c102
YZ		 3984 if (ret > 0)
3985 goto not_found;
3986 leaf = path->nodes[0];
a52d9a80 3987 }
9036c102
YZ		 3988 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3989 if (found_key.objectid != objectid ||
3990 found_key.type != BTRFS_EXTENT_DATA_KEY)
3991 goto not_found;
3992 if (start + len <= found_key.offset)
3993 goto not_found;
3994 em->start = start;
3995 em->len = found_key.offset - start;
3996 goto not_found_em;
3997 }
3998
d899e052
YZ		 3999 if (found_type == BTRFS_FILE_EXTENT_REG ||
4000 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 4001 em->start = extent_start;
4002 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 4003 em->orig_start = extent_start -
4004 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 4005 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4006 if (bytenr == 0) {
5f39d397 4007 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4008 goto insert;
4009 }
c8b97818
CM		 4010 if (compressed) {
4011 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4012 em->block_start = bytenr;
4013 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4014 item);
4015 } else {
4016 bytenr += btrfs_file_extent_offset(leaf, item);
4017 em->block_start = bytenr;
4018 em->block_len = em->len;
d899e052
YZ		 4019 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4020 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4021 }
a52d9a80
CM		 4022 goto insert;
4023 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4024 unsigned long ptr;
a52d9a80 4025 char *map;
3326d1b0
CM		 4026 size_t size;
4027 size_t extent_offset;
4028 size_t copy_size;
a52d9a80 4029
689f9346 4030 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4031 if (!page || create) {
689f9346 4032 em->start = extent_start;
9036c102 4033 em->len = extent_end - extent_start;
689f9346
Y		 4034 goto out;
4035 }
5f39d397 4036
9036c102
YZ		 4037 size = btrfs_file_extent_inline_len(leaf, item);
4038 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4039 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4040 size - extent_offset);
3326d1b0 4041 em->start = extent_start + extent_offset;
70dec807
CM		 4042 em->len = (copy_size + root->sectorsize - 1) &
4043 ~((u64)root->sectorsize - 1);
ff5b7ee3 4044 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 4045 if (compressed)
4046 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4047 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4048 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 4049 if (btrfs_file_extent_compression(leaf, item) ==
4050 BTRFS_COMPRESS_ZLIB) {
4051 ret = uncompress_inline(path, inode, page,
4052 pg_offset,
4053 extent_offset, item);
4054 BUG_ON(ret);
4055 } else {
4056 map = kmap(page);
4057 read_extent_buffer(leaf, map + pg_offset, ptr,
4058 copy_size);
4059 kunmap(page);
4060 }
179e29e4
CM		 4061 flush_dcache_page(page);
4062 } else if (create && PageUptodate(page)) {
4063 if (!trans) {
4064 kunmap(page);
4065 free_extent_map(em);
4066 em = NULL;
4067 btrfs_release_path(root, path);
f9295749 4068 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4069 goto again;
4070 }
c8b97818 4071 map = kmap(page);
70dec807 4072 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4073 copy_size);
c8b97818 4074 kunmap(page);
179e29e4 4075 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4076 }
d1310b2e
CM		 4077 set_extent_uptodate(io_tree, em->start,
4078 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4079 goto insert;
4080 } else {
d397712b 4081 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 4082 WARN_ON(1);
4083 }
4084not_found:
4085 em->start = start;
d1310b2e 4086 em->len = len;
a52d9a80 4087not_found_em:
5f39d397 4088 em->block_start = EXTENT_MAP_HOLE;
9036c102 4089 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4090insert:
4091 btrfs_release_path(root, path);
d1310b2e 4092 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 4093 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4094 "[%llu %llu]\n", (unsigned long long)em->start,
4095 (unsigned long long)em->len,
4096 (unsigned long long)start,
4097 (unsigned long long)len);
a52d9a80
CM		 4098 err = -EIO;
4099 goto out;
4100 }
d1310b2e
CM		 4101
4102 err = 0;
4103 spin_lock(&em_tree->lock);
a52d9a80 4104 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4105 /* it is possible that someone inserted the extent into the tree
4106 * while we had the lock dropped. It is also possible that
4107 * an overlapping map exists in the tree
4108 */
a52d9a80 4109 if (ret == -EEXIST) {
3b951516 4110 struct extent_map *existing;
e6dcd2dc
CM		 4111
4112 ret = 0;
4113
3b951516 4114 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4115 if (existing && (existing->start > start ||
4116 existing->start + existing->len <= start)) {
4117 free_extent_map(existing);
4118 existing = NULL;
4119 }
3b951516
CM		 4120 if (!existing) {
4121 existing = lookup_extent_mapping(em_tree, em->start,
4122 em->len);
4123 if (existing) {
4124 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4125 em, start,
4126 root->sectorsize);
3b951516
CM		 4127 free_extent_map(existing);
4128 if (err) {
4129 free_extent_map(em);
4130 em = NULL;
4131 }
4132 } else {
4133 err = -EIO;
3b951516
CM		 4134 free_extent_map(em);
4135 em = NULL;
4136 }
4137 } else {
4138 free_extent_map(em);
4139 em = existing;
e6dcd2dc 4140 err = 0;
a52d9a80 4141 }
a52d9a80 4142 }
d1310b2e 4143 spin_unlock(&em_tree->lock);
a52d9a80 4144out:
f421950f
CM		 4145 if (path)
4146 btrfs_free_path(path);
a52d9a80
CM		 4147 if (trans) {
4148 ret = btrfs_end_transaction(trans, root);
d397712b 4149 if (!err)
a52d9a80
CM		 4150 err = ret;
4151 }
a52d9a80
CM		 4152 if (err) {
4153 free_extent_map(em);
4154 WARN_ON(1);
4155 return ERR_PTR(err);
4156 }
4157 return em;
4158}
4159
16432985
CM		 4160static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4161 const struct iovec *iov, loff_t offset,
4162 unsigned long nr_segs)
4163{
e1c4b745 4164 return -EINVAL;
16432985
CM		 4165}
4166
1506fcc8
YS		 4167static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4168 __u64 start, __u64 len)
4169{
4170 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4171}
4172
a52d9a80 4173int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4174{
d1310b2e
CM		 4175 struct extent_io_tree *tree;
4176 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4177 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4178}
1832a6d5 4179
a52d9a80 4180static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4181{
d1310b2e 4182 struct extent_io_tree *tree;
b888db2b
CM		 4183
4184
4185 if (current->flags & PF_MEMALLOC) {
4186 redirty_page_for_writepage(wbc, page);
4187 unlock_page(page);
4188 return 0;
4189 }
d1310b2e 4190 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4191 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4192}
4193
f421950f
CM		 4194int btrfs_writepages(struct address_space *mapping,
4195 struct writeback_control *wbc)
b293f02e 4196{
d1310b2e 4197 struct extent_io_tree *tree;
771ed689 4198
d1310b2e 4199 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4200 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4201}
4202
3ab2fb5a
CM		 4203static int
4204btrfs_readpages(struct file *file, struct address_space *mapping,
4205 struct list_head *pages, unsigned nr_pages)
4206{
d1310b2e
CM		 4207 struct extent_io_tree *tree;
4208 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4209 return extent_readpages(tree, mapping, pages, nr_pages,
4210 btrfs_get_extent);
4211}
e6dcd2dc 4212static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4213{
d1310b2e
CM		 4214 struct extent_io_tree *tree;
4215 struct extent_map_tree *map;
a52d9a80 4216 int ret;
8c2383c3 4217
d1310b2e
CM		 4218 tree = &BTRFS_I(page->mapping->host)->io_tree;
4219 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4220 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4221 if (ret == 1) {
4222 ClearPagePrivate(page);
4223 set_page_private(page, 0);
4224 page_cache_release(page);
39279cc3 4225 }
a52d9a80 4226 return ret;
39279cc3
CM		 4227}
4228
e6dcd2dc
CM		 4229static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4230{
98509cfc
CM		 4231 if (PageWriteback(page) || PageDirty(page))
4232 return 0;
b335b003 4233 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 4234}
4235
a52d9a80 4236static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4237{
d1310b2e 4238 struct extent_io_tree *tree;
e6dcd2dc
CM		 4239 struct btrfs_ordered_extent *ordered;
4240 u64 page_start = page_offset(page);
4241 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4242
e6dcd2dc 4243 wait_on_page_writeback(page);
d1310b2e 4244 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4245 if (offset) {
4246 btrfs_releasepage(page, GFP_NOFS);
4247 return;
4248 }
4249
4250 lock_extent(tree, page_start, page_end, GFP_NOFS);
4251 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4252 page_offset(page));
4253 if (ordered) {
eb84ae03
CM		 4254 /*
4255 * IO on this page will never be started, so we need
4256 * to account for any ordered extents now
4257 */
e6dcd2dc
CM		 4258 clear_extent_bit(tree, page_start, page_end,
4259 EXTENT_DIRTY | EXTENT_DELALLOC |
4260 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4261 btrfs_finish_ordered_io(page->mapping->host,
4262 page_start, page_end);
e6dcd2dc
CM		 4263 btrfs_put_ordered_extent(ordered);
4264 lock_extent(tree, page_start, page_end, GFP_NOFS);
4265 }
4266 clear_extent_bit(tree, page_start, page_end,
4267 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4268 EXTENT_ORDERED,
4269 1, 1, GFP_NOFS);
4270 __btrfs_releasepage(page, GFP_NOFS);
4271
4a096752 4272 ClearPageChecked(page);
9ad6b7bc 4273 if (PagePrivate(page)) {
9ad6b7bc
CM		 4274 ClearPagePrivate(page);
4275 set_page_private(page, 0);
4276 page_cache_release(page);
4277 }
39279cc3
CM		 4278}
4279
9ebefb18
CM		 4280/*
4281 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4282 * called from a page fault handler when a page is first dirtied. Hence we must
4283 * be careful to check for EOF conditions here. We set the page up correctly
4284 * for a written page which means we get ENOSPC checking when writing into
4285 * holes and correct delalloc and unwritten extent mapping on filesystems that
4286 * support these features.
4287 *
4288 * We are not allowed to take the i_mutex here so we have to play games to
4289 * protect against truncate races as the page could now be beyond EOF. Because
4290 * vmtruncate() writes the inode size before removing pages, once we have the
4291 * page lock we can determine safely if the page is beyond EOF. If it is not
4292 * beyond EOF, then the page is guaranteed safe against truncation until we
4293 * unlock the page.
4294 */
c2ec175c 4295int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 4296{
c2ec175c 4297 struct page *page = vmf->page;
6da6abae 4298 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4299 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4300 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4301 struct btrfs_ordered_extent *ordered;
4302 char *kaddr;
4303 unsigned long zero_start;
9ebefb18 4304 loff_t size;
1832a6d5 4305 int ret;
a52d9a80 4306 u64 page_start;
e6dcd2dc 4307 u64 page_end;
9ebefb18 4308
6a63209f 4309 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
1832a6d5
CM		 4310 if (ret)
4311 goto out;
4312
4313 ret = -EINVAL;
e6dcd2dc 4314again:
9ebefb18 4315 lock_page(page);
9ebefb18 4316 size = i_size_read(inode);
e6dcd2dc
CM		 4317 page_start = page_offset(page);
4318 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4319
9ebefb18 4320 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4321 (page_start >= size)) {
6a63209f 4322 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
9ebefb18
CM		 4323 /* page got truncated out from underneath us */
4324 goto out_unlock;
4325 }
e6dcd2dc
CM		 4326 wait_on_page_writeback(page);
4327
4328 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4329 set_page_extent_mapped(page);
4330
eb84ae03
CM		 4331 /*
4332 * we can't set the delalloc bits if there are pending ordered
4333 * extents. Drop our locks and wait for them to finish
4334 */
e6dcd2dc
CM		 4335 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4336 if (ordered) {
4337 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4338 unlock_page(page);
eb84ae03 4339 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4340 btrfs_put_ordered_extent(ordered);
4341 goto again;
4342 }
4343
ea8c2819 4344 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4345 ret = 0;
9ebefb18
CM		 4346
4347 /* page is wholly or partially inside EOF */
a52d9a80 4348 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4349 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4350 else
e6dcd2dc 4351 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4352
e6dcd2dc
CM		 4353 if (zero_start != PAGE_CACHE_SIZE) {
4354 kaddr = kmap(page);
4355 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4356 flush_dcache_page(page);
4357 kunmap(page);
4358 }
247e743c 4359 ClearPageChecked(page);
e6dcd2dc
CM		 4360 set_page_dirty(page);
4361 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4362
4363out_unlock:
4364 unlock_page(page);
1832a6d5 4365out:
c2ec175c
NP		 4366 if (ret)
4367 ret = VM_FAULT_SIGBUS;
9ebefb18
CM		 4368 return ret;
4369}
4370
39279cc3
CM		 4371static void btrfs_truncate(struct inode *inode)
4372{
4373 struct btrfs_root *root = BTRFS_I(inode)->root;
4374 int ret;
4375 struct btrfs_trans_handle *trans;
d3c2fdcf 4376 unsigned long nr;
dbe674a9 4377 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4378
4379 if (!S_ISREG(inode->i_mode))
4380 return;
4381 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4382 return;
4383
4384 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4385 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4386
39279cc3
CM		 4387 trans = btrfs_start_transaction(root, 1);
4388 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4389 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4390
7b128766
JB		 4391 ret = btrfs_orphan_add(trans, inode);
4392 if (ret)
4393 goto out;
39279cc3 4394 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4395 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4396 BTRFS_EXTENT_DATA_KEY);
39279cc3 4397 btrfs_update_inode(trans, root, inode);
5f39d397 4398
7b128766
JB		 4399 ret = btrfs_orphan_del(trans, inode);
4400 BUG_ON(ret);
4401
4402out:
4403 nr = trans->blocks_used;
89ce8a63 4404 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4405 BUG_ON(ret);
d3c2fdcf 4406 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4407}
4408
d352ac68
CM		 4409/*
4410 * create a new subvolume directory/inode (helper for the ioctl).
4411 */
d2fb3437
YZ		 4412int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4413 struct btrfs_root *new_root, struct dentry *dentry,
4414 u64 new_dirid, u64 alloc_hint)
39279cc3 4415{
39279cc3 4416 struct inode *inode;
cb8e7090 4417 int error;
00e4e6b3 4418 u64 index = 0;
39279cc3 4419
aec7477b 4420 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4421 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4422 if (IS_ERR(inode))
f46b5a66 4423 return PTR_ERR(inode);
39279cc3
CM		 4424 inode->i_op = &btrfs_dir_inode_operations;
4425 inode->i_fop = &btrfs_dir_file_operations;
4426
39279cc3 4427 inode->i_nlink = 1;
dbe674a9 4428 btrfs_i_size_write(inode, 0);
3b96362c 4429
cb8e7090
CH		 4430 error = btrfs_update_inode(trans, new_root, inode);
4431 if (error)
4432 return error;
4433
4434 d_instantiate(dentry, inode);
4435 return 0;
39279cc3
CM		 4436}
4437
d352ac68
CM		 4438/* helper function for file defrag and space balancing. This
4439 * forces readahead on a given range of bytes in an inode
4440 */
edbd8d4e 4441unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4442 struct file_ra_state *ra, struct file *file,
4443 pgoff_t offset, pgoff_t last_index)
4444{
8e7bf94f 4445 pgoff_t req_size = last_index - offset + 1;
86479a04 4446
86479a04
CM		 4447 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4448 return offset + req_size;
86479a04
CM		 4449}
4450
39279cc3
CM		 4451struct inode *btrfs_alloc_inode(struct super_block *sb)
4452{
4453 struct btrfs_inode *ei;
4454
4455 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4456 if (!ei)
4457 return NULL;
15ee9bc7 4458 ei->last_trans = 0;
e02119d5 4459 ei->logged_trans = 0;
e6dcd2dc 4460 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4461 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4462 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4463 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4464 return &ei->vfs_inode;
4465}
4466
4467void btrfs_destroy_inode(struct inode *inode)
4468{
e6dcd2dc 4469 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4470 WARN_ON(!list_empty(&inode->i_dentry));
4471 WARN_ON(inode->i_data.nrpages);
4472
33268eaf
JB		 4473 if (BTRFS_I(inode)->i_acl &&
4474 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4475 posix_acl_release(BTRFS_I(inode)->i_acl);
4476 if (BTRFS_I(inode)->i_default_acl &&
4477 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4478 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4479
bcc63abb 4480 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4481 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4482 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4483 " list\n", inode->i_ino);
4484 dump_stack();
4485 }
bcc63abb 4486 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4487
d397712b 4488 while (1) {
e6dcd2dc
CM		 4489 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4490 if (!ordered)
4491 break;
4492 else {
d397712b
CM		 4493 printk(KERN_ERR "btrfs found ordered "
4494 "extent %llu %llu on inode cleanup\n",
4495 (unsigned long long)ordered->file_offset,
4496 (unsigned long long)ordered->len);
e6dcd2dc
CM		 4497 btrfs_remove_ordered_extent(inode, ordered);
4498 btrfs_put_ordered_extent(ordered);
4499 btrfs_put_ordered_extent(ordered);
4500 }
4501 }
5b21f2ed 4502 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4503 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4504}
4505
0ee0fda0 4506static void init_once(void *foo)
39279cc3
CM		 4507{
4508 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4509
4510 inode_init_once(&ei->vfs_inode);
4511}
4512
4513void btrfs_destroy_cachep(void)
4514{
4515 if (btrfs_inode_cachep)
4516 kmem_cache_destroy(btrfs_inode_cachep);
4517 if (btrfs_trans_handle_cachep)
4518 kmem_cache_destroy(btrfs_trans_handle_cachep);
4519 if (btrfs_transaction_cachep)
4520 kmem_cache_destroy(btrfs_transaction_cachep);
4521 if (btrfs_bit_radix_cachep)
4522 kmem_cache_destroy(btrfs_bit_radix_cachep);
4523 if (btrfs_path_cachep)
4524 kmem_cache_destroy(btrfs_path_cachep);
4525}
4526
86479a04 4527struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4528 unsigned long extra_flags,
2b1f55b0 4529 void (*ctor)(void *))
92fee66d
CM		 4530{
4531 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4532 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4533}
4534
39279cc3
CM		 4535int btrfs_init_cachep(void)
4536{
86479a04 4537 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4538 sizeof(struct btrfs_inode),
4539 0, init_once);
39279cc3
CM		 4540 if (!btrfs_inode_cachep)
4541 goto fail;
86479a04
CM		 4542 btrfs_trans_handle_cachep =
4543 btrfs_cache_create("btrfs_trans_handle_cache",
4544 sizeof(struct btrfs_trans_handle),
4545 0, NULL);
39279cc3
CM		 4546 if (!btrfs_trans_handle_cachep)
4547 goto fail;
86479a04 4548 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4549 sizeof(struct btrfs_transaction),
92fee66d 4550 0, NULL);
39279cc3
CM		 4551 if (!btrfs_transaction_cachep)
4552 goto fail;
86479a04 4553 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4554 sizeof(struct btrfs_path),
92fee66d 4555 0, NULL);
39279cc3
CM		 4556 if (!btrfs_path_cachep)
4557 goto fail;
86479a04 4558 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4559 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4560 if (!btrfs_bit_radix_cachep)
4561 goto fail;
4562 return 0;
4563fail:
4564 btrfs_destroy_cachep();
4565 return -ENOMEM;
4566}
4567
4568static int btrfs_getattr(struct vfsmount *mnt,
4569 struct dentry *dentry, struct kstat *stat)
4570{
4571 struct inode *inode = dentry->d_inode;
4572 generic_fillattr(inode, stat);
3394e160 4573 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4574 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4575 stat->blocks = (inode_get_bytes(inode) +
4576 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4577 return 0;
4578}
4579
d397712b
CM		 4580static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4581 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 4582{
4583 struct btrfs_trans_handle *trans;
4584 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4585 struct inode *new_inode = new_dentry->d_inode;
4586 struct inode *old_inode = old_dentry->d_inode;
4587 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4588 u64 index = 0;
39279cc3
CM		 4589 int ret;
4590
3394e160
CM		 4591 /* we're not allowed to rename between subvolumes */
4592 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4593 BTRFS_I(new_dir)->root->root_key.objectid)
4594 return -EXDEV;
4595
39279cc3
CM		 4596 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4597 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4598 return -ENOTEMPTY;
4599 }
5f39d397 4600
0660b5af
CM		 4601 /* to rename a snapshot or subvolume, we need to juggle the
4602 * backrefs. This isn't coded yet
4603 */
4604 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4605 return -EXDEV;
4606
6a63209f 4607 ret = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4608 if (ret)
4609 goto out_unlock;
4610
39279cc3 4611 trans = btrfs_start_transaction(root, 1);
5f39d397 4612
39279cc3 4613 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4614
e02119d5 4615 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4616 old_dir->i_ctime = old_dir->i_mtime = ctime;
4617 new_dir->i_ctime = new_dir->i_mtime = ctime;
4618 old_inode->i_ctime = ctime;
5f39d397 4619
e02119d5
CM		 4620 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4621 old_dentry->d_name.name,
4622 old_dentry->d_name.len);
39279cc3
CM		 4623 if (ret)
4624 goto out_fail;
4625
4626 if (new_inode) {
4627 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4628 ret = btrfs_unlink_inode(trans, root, new_dir,
4629 new_dentry->d_inode,
4630 new_dentry->d_name.name,
4631 new_dentry->d_name.len);
39279cc3
CM		 4632 if (ret)
4633 goto out_fail;
7b128766 4634 if (new_inode->i_nlink == 0) {
e02119d5 4635 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4636 if (ret)
4637 goto out_fail;
4638 }
e02119d5 4639
39279cc3 4640 }
3de4586c 4641 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4642 if (ret)
4643 goto out_fail;
4644
e02119d5
CM		 4645 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4646 old_inode, new_dentry->d_name.name,
4647 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4648 if (ret)
4649 goto out_fail;
4650
4651out_fail:
ab78c84d 4652 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4653out_unlock:
39279cc3
CM		 4654 return ret;
4655}
4656
d352ac68
CM		 4657/*
4658 * some fairly slow code that needs optimization. This walks the list
4659 * of all the inodes with pending delalloc and forces them to disk.
4660 */
ea8c2819
CM		 4661int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4662{
4663 struct list_head *head = &root->fs_info->delalloc_inodes;
4664 struct btrfs_inode *binode;
5b21f2ed 4665 struct inode *inode;
ea8c2819 4666
c146afad
YZ		 4667 if (root->fs_info->sb->s_flags & MS_RDONLY)
4668 return -EROFS;
4669
75eff68e 4670 spin_lock(&root->fs_info->delalloc_lock);
d397712b 4671 while (!list_empty(head)) {
ea8c2819
CM		 4672 binode = list_entry(head->next, struct btrfs_inode,
4673 delalloc_inodes);
5b21f2ed
ZY		 4674 inode = igrab(&binode->vfs_inode);
4675 if (!inode)
4676 list_del_init(&binode->delalloc_inodes);
75eff68e 4677 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 4678 if (inode) {
8c8bee1d 4679 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4680 iput(inode);
4681 }
4682 cond_resched();
75eff68e 4683 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 4684 }
75eff68e 4685 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 4686
4687 /* the filemap_flush will queue IO into the worker threads, but
4688 * we have to make sure the IO is actually started and that
4689 * ordered extents get created before we return
4690 */
4691 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 4692 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 4693 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4694 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4695 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4696 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4697 }
4698 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4699 return 0;
4700}
4701
39279cc3
CM		 4702static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4703 const char *symname)
4704{
4705 struct btrfs_trans_handle *trans;
4706 struct btrfs_root *root = BTRFS_I(dir)->root;
4707 struct btrfs_path *path;
4708 struct btrfs_key key;
1832a6d5 4709 struct inode *inode = NULL;
39279cc3
CM		 4710 int err;
4711 int drop_inode = 0;
4712 u64 objectid;
00e4e6b3 4713 u64 index = 0 ;
39279cc3
CM		 4714 int name_len;
4715 int datasize;
5f39d397 4716 unsigned long ptr;
39279cc3 4717 struct btrfs_file_extent_item *ei;
5f39d397 4718 struct extent_buffer *leaf;
1832a6d5 4719 unsigned long nr = 0;
39279cc3
CM		 4720
4721 name_len = strlen(symname) + 1;
4722 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4723 return -ENAMETOOLONG;
1832a6d5 4724
6a63209f 4725 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4726 if (err)
4727 goto out_fail;
4728
39279cc3
CM		 4729 trans = btrfs_start_transaction(root, 1);
4730 btrfs_set_trans_block_group(trans, dir);
4731
4732 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4733 if (err) {
4734 err = -ENOSPC;
4735 goto out_unlock;
4736 }
4737
aec7477b 4738 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4739 dentry->d_name.len,
4740 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4741 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4742 &index);
39279cc3
CM		 4743 err = PTR_ERR(inode);
4744 if (IS_ERR(inode))
4745 goto out_unlock;
4746
0279b4cd 4747 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 4748 if (err) {
4749 drop_inode = 1;
4750 goto out_unlock;
4751 }
4752
39279cc3 4753 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4754 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4755 if (err)
4756 drop_inode = 1;
4757 else {
4758 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4759 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4760 inode->i_fop = &btrfs_file_operations;
4761 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4762 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4763 }
4764 dir->i_sb->s_dirt = 1;
4765 btrfs_update_inode_block_group(trans, inode);
4766 btrfs_update_inode_block_group(trans, dir);
4767 if (drop_inode)
4768 goto out_unlock;
4769
4770 path = btrfs_alloc_path();
4771 BUG_ON(!path);
4772 key.objectid = inode->i_ino;
4773 key.offset = 0;
39279cc3
CM		 4774 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4775 datasize = btrfs_file_extent_calc_inline_size(name_len);
4776 err = btrfs_insert_empty_item(trans, root, path, &key,
4777 datasize);
54aa1f4d
CM		 4778 if (err) {
4779 drop_inode = 1;
4780 goto out_unlock;
4781 }
5f39d397
CM		 4782 leaf = path->nodes[0];
4783 ei = btrfs_item_ptr(leaf, path->slots[0],
4784 struct btrfs_file_extent_item);
4785 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4786 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4787 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4788 btrfs_set_file_extent_encryption(leaf, ei, 0);
4789 btrfs_set_file_extent_compression(leaf, ei, 0);
4790 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4791 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4792
39279cc3 4793 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4794 write_extent_buffer(leaf, symname, ptr, name_len);
4795 btrfs_mark_buffer_dirty(leaf);
39279cc3 4796 btrfs_free_path(path);
5f39d397 4797
39279cc3
CM		 4798 inode->i_op = &btrfs_symlink_inode_operations;
4799 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4800 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4801 inode_set_bytes(inode, name_len);
dbe674a9 4802 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4803 err = btrfs_update_inode(trans, root, inode);
4804 if (err)
4805 drop_inode = 1;
39279cc3
CM		 4806
4807out_unlock:
d3c2fdcf 4808 nr = trans->blocks_used;
ab78c84d 4809 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4810out_fail:
39279cc3
CM		 4811 if (drop_inode) {
4812 inode_dec_link_count(inode);
4813 iput(inode);
4814 }
d3c2fdcf 4815 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4816 return err;
4817}
16432985 4818
d899e052
YZ		 4819static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4820 u64 alloc_hint, int mode)
4821{
4822 struct btrfs_trans_handle *trans;
4823 struct btrfs_root *root = BTRFS_I(inode)->root;
4824 struct btrfs_key ins;
4825 u64 alloc_size;
4826 u64 cur_offset = start;
4827 u64 num_bytes = end - start;
4828 int ret = 0;
4829
4830 trans = btrfs_join_transaction(root, 1);
4831 BUG_ON(!trans);
4832 btrfs_set_trans_block_group(trans, inode);
4833
4834 while (num_bytes > 0) {
4835 alloc_size = min(num_bytes, root->fs_info->max_extent);
4836 ret = btrfs_reserve_extent(trans, root, alloc_size,
4837 root->sectorsize, 0, alloc_hint,
4838 (u64)-1, &ins, 1);
4839 if (ret) {
4840 WARN_ON(1);
4841 goto out;
4842 }
4843 ret = insert_reserved_file_extent(trans, inode,
4844 cur_offset, ins.objectid,
4845 ins.offset, ins.offset,
4846 ins.offset, 0, 0, 0,
4847 BTRFS_FILE_EXTENT_PREALLOC);
4848 BUG_ON(ret);
4849 num_bytes -= ins.offset;
4850 cur_offset += ins.offset;
4851 alloc_hint = ins.objectid + ins.offset;
4852 }
4853out:
4854 if (cur_offset > start) {
4855 inode->i_ctime = CURRENT_TIME;
4856 btrfs_set_flag(inode, PREALLOC);
4857 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4858 cur_offset > i_size_read(inode))
4859 btrfs_i_size_write(inode, cur_offset);
4860 ret = btrfs_update_inode(trans, root, inode);
4861 BUG_ON(ret);
4862 }
4863
4864 btrfs_end_transaction(trans, root);
4865 return ret;
4866}
4867
4868static long btrfs_fallocate(struct inode *inode, int mode,
4869 loff_t offset, loff_t len)
4870{
4871 u64 cur_offset;
4872 u64 last_byte;
4873 u64 alloc_start;
4874 u64 alloc_end;
4875 u64 alloc_hint = 0;
4876 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4877 struct extent_map *em;
4878 int ret;
4879
4880 alloc_start = offset & ~mask;
4881 alloc_end = (offset + len + mask) & ~mask;
4882
4883 mutex_lock(&inode->i_mutex);
4884 if (alloc_start > inode->i_size) {
4885 ret = btrfs_cont_expand(inode, alloc_start);
4886 if (ret)
4887 goto out;
4888 }
4889
4890 while (1) {
4891 struct btrfs_ordered_extent *ordered;
4892 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4893 alloc_end - 1, GFP_NOFS);
4894 ordered = btrfs_lookup_first_ordered_extent(inode,
4895 alloc_end - 1);
4896 if (ordered &&
4897 ordered->file_offset + ordered->len > alloc_start &&
4898 ordered->file_offset < alloc_end) {
4899 btrfs_put_ordered_extent(ordered);
4900 unlock_extent(&BTRFS_I(inode)->io_tree,
4901 alloc_start, alloc_end - 1, GFP_NOFS);
4902 btrfs_wait_ordered_range(inode, alloc_start,
4903 alloc_end - alloc_start);
4904 } else {
4905 if (ordered)
4906 btrfs_put_ordered_extent(ordered);
4907 break;
4908 }
4909 }
4910
4911 cur_offset = alloc_start;
4912 while (1) {
4913 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4914 alloc_end - cur_offset, 0);
4915 BUG_ON(IS_ERR(em) || !em);
4916 last_byte = min(extent_map_end(em), alloc_end);
4917 last_byte = (last_byte + mask) & ~mask;
4918 if (em->block_start == EXTENT_MAP_HOLE) {
4919 ret = prealloc_file_range(inode, cur_offset,
4920 last_byte, alloc_hint, mode);
4921 if (ret < 0) {
4922 free_extent_map(em);
4923 break;
4924 }
4925 }
4926 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4927 alloc_hint = em->block_start;
4928 free_extent_map(em);
4929
4930 cur_offset = last_byte;
4931 if (cur_offset >= alloc_end) {
4932 ret = 0;
4933 break;
4934 }
4935 }
4936 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4937 GFP_NOFS);
4938out:
4939 mutex_unlock(&inode->i_mutex);
4940 return ret;
4941}
4942
e6dcd2dc
CM		 4943static int btrfs_set_page_dirty(struct page *page)
4944{
e6dcd2dc
CM		 4945 return __set_page_dirty_nobuffers(page);
4946}
4947
0ee0fda0 4948static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4949{
4950 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4951 return -EACCES;
33268eaf 4952 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4953}
39279cc3
CM		 4954
4955static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4956 .getattr = btrfs_getattr,
39279cc3
CM		 4957 .lookup = btrfs_lookup,
4958 .create = btrfs_create,
4959 .unlink = btrfs_unlink,
4960 .link = btrfs_link,
4961 .mkdir = btrfs_mkdir,
4962 .rmdir = btrfs_rmdir,
4963 .rename = btrfs_rename,
4964 .symlink = btrfs_symlink,
4965 .setattr = btrfs_setattr,
618e21d5 4966 .mknod = btrfs_mknod,
95819c05
CH		 4967 .setxattr = btrfs_setxattr,
4968 .getxattr = btrfs_getxattr,
5103e947 4969 .listxattr = btrfs_listxattr,
95819c05 4970 .removexattr = btrfs_removexattr,
fdebe2bd 4971 .permission = btrfs_permission,
39279cc3 4972};
39279cc3
CM		 4973static struct inode_operations btrfs_dir_ro_inode_operations = {
4974 .lookup = btrfs_lookup,
fdebe2bd 4975 .permission = btrfs_permission,
39279cc3 4976};
39279cc3
CM		 4977static struct file_operations btrfs_dir_file_operations = {
4978 .llseek = generic_file_llseek,
4979 .read = generic_read_dir,
cbdf5a24 4980 .readdir = btrfs_real_readdir,
34287aa3 4981 .unlocked_ioctl = btrfs_ioctl,
39279cc3 4982#ifdef CONFIG_COMPAT
34287aa3 4983 .compat_ioctl = btrfs_ioctl,
39279cc3 4984#endif
6bf13c0c 4985 .release = btrfs_release_file,
e02119d5 4986 .fsync = btrfs_sync_file,
39279cc3
CM		 4987};
4988
d1310b2e 4989static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 4990 .fill_delalloc = run_delalloc_range,
065631f6 4991 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 4992 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 4993 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 4994 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 4995 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 4996 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 4997 .set_bit_hook = btrfs_set_bit_hook,
4998 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 4999};
5000
35054394
CM		 5001/*
5002 * btrfs doesn't support the bmap operation because swapfiles
5003 * use bmap to make a mapping of extents in the file. They assume
5004 * these extents won't change over the life of the file and they
5005 * use the bmap result to do IO directly to the drive.
5006 *
5007 * the btrfs bmap call would return logical addresses that aren't
5008 * suitable for IO and they also will change frequently as COW
5009 * operations happen. So, swapfile + btrfs == corruption.
5010 *
5011 * For now we're avoiding this by dropping bmap.
5012 */
39279cc3
CM		 5013static struct address_space_operations btrfs_aops = {
5014 .readpage = btrfs_readpage,
5015 .writepage = btrfs_writepage,
b293f02e 5016 .writepages = btrfs_writepages,
3ab2fb5a 5017 .readpages = btrfs_readpages,
39279cc3 5018 .sync_page = block_sync_page,
16432985 5019 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 5020 .invalidatepage = btrfs_invalidatepage,
5021 .releasepage = btrfs_releasepage,
e6dcd2dc 5022 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 5023};
5024
5025static struct address_space_operations btrfs_symlink_aops = {
5026 .readpage = btrfs_readpage,
5027 .writepage = btrfs_writepage,
2bf5a725
CM		 5028 .invalidatepage = btrfs_invalidatepage,
5029 .releasepage = btrfs_releasepage,
39279cc3
CM		 5030};
5031
5032static struct inode_operations btrfs_file_inode_operations = {
5033 .truncate = btrfs_truncate,
5034 .getattr = btrfs_getattr,
5035 .setattr = btrfs_setattr,
95819c05
CH		 5036 .setxattr = btrfs_setxattr,
5037 .getxattr = btrfs_getxattr,
5103e947 5038 .listxattr = btrfs_listxattr,
95819c05 5039 .removexattr = btrfs_removexattr,
fdebe2bd 5040 .permission = btrfs_permission,
d899e052 5041 .fallocate = btrfs_fallocate,
1506fcc8 5042 .fiemap = btrfs_fiemap,
39279cc3 5043};
618e21d5
JB		 5044static struct inode_operations btrfs_special_inode_operations = {
5045 .getattr = btrfs_getattr,
5046 .setattr = btrfs_setattr,
fdebe2bd 5047 .permission = btrfs_permission,
95819c05
CH		 5048 .setxattr = btrfs_setxattr,
5049 .getxattr = btrfs_getxattr,
33268eaf 5050 .listxattr = btrfs_listxattr,
95819c05 5051 .removexattr = btrfs_removexattr,
618e21d5 5052};
39279cc3
CM		 5053static struct inode_operations btrfs_symlink_inode_operations = {
5054 .readlink = generic_readlink,
5055 .follow_link = page_follow_link_light,
5056 .put_link = page_put_link,
fdebe2bd 5057 .permission = btrfs_permission,
0279b4cd
JO		 5058 .setxattr = btrfs_setxattr,
5059 .getxattr = btrfs_getxattr,
5060 .listxattr = btrfs_listxattr,
5061 .removexattr = btrfs_removexattr,
39279cc3 5062};
Linux 6.x block layer and io_uring tree(s)