projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: leave btree locks spinning more often
[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
b9473439 137 path->leave_spinning = 1;
c8b97818
CM		 138 btrfs_set_trans_block_group(trans, inode);
139
140 key.objectid = inode->i_ino;
141 key.offset = start;
142 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 143 datasize = btrfs_file_extent_calc_inline_size(cur_size);
144
145 inode_add_bytes(inode, size);
146 ret = btrfs_insert_empty_item(trans, root, path, &key,
147 datasize);
148 BUG_ON(ret);
149 if (ret) {
150 err = ret;
c8b97818
CM		 151 goto fail;
152 }
153 leaf = path->nodes[0];
154 ei = btrfs_item_ptr(leaf, path->slots[0],
155 struct btrfs_file_extent_item);
156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158 btrfs_set_file_extent_encryption(leaf, ei, 0);
159 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161 ptr = btrfs_file_extent_inline_start(ei);
162
163 if (use_compress) {
164 struct page *cpage;
165 int i = 0;
d397712b 166 while (compressed_size > 0) {
c8b97818 167 cpage = compressed_pages[i];
5b050f04 168 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 169 PAGE_CACHE_SIZE);
170
b9473439 171 kaddr = kmap_atomic(cpage, KM_USER0);
c8b97818 172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
b9473439 173 kunmap_atomic(kaddr, KM_USER0);
c8b97818
CM		 174
175 i++;
176 ptr += cur_size;
177 compressed_size -= cur_size;
178 }
179 btrfs_set_file_extent_compression(leaf, ei,
180 BTRFS_COMPRESS_ZLIB);
181 } else {
182 page = find_get_page(inode->i_mapping,
183 start >> PAGE_CACHE_SHIFT);
184 btrfs_set_file_extent_compression(leaf, ei, 0);
185 kaddr = kmap_atomic(page, KM_USER0);
186 offset = start & (PAGE_CACHE_SIZE - 1);
187 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188 kunmap_atomic(kaddr, KM_USER0);
189 page_cache_release(page);
190 }
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_free_path(path);
193
194 BTRFS_I(inode)->disk_i_size = inode->i_size;
195 btrfs_update_inode(trans, root, inode);
196 return 0;
197fail:
198 btrfs_free_path(path);
199 return err;
200}
201
202
203/*
204 * conditionally insert an inline extent into the file. This
205 * does the checks required to make sure the data is small enough
206 * to fit as an inline extent.
207 */
7f366cfe 208static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 209 struct btrfs_root *root,
210 struct inode *inode, u64 start, u64 end,
211 size_t compressed_size,
212 struct page **compressed_pages)
213{
214 u64 isize = i_size_read(inode);
215 u64 actual_end = min(end + 1, isize);
216 u64 inline_len = actual_end - start;
217 u64 aligned_end = (end + root->sectorsize - 1) &
218 ~((u64)root->sectorsize - 1);
219 u64 hint_byte;
220 u64 data_len = inline_len;
221 int ret;
222
223 if (compressed_size)
224 data_len = compressed_size;
225
226 if (start > 0 ||
70b99e69 227 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 228 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
229 (!compressed_size &&
230 (actual_end & (root->sectorsize - 1)) == 0) ||
231 end + 1 < isize ||
232 data_len > root->fs_info->max_inline) {
233 return 1;
234 }
235
c8b97818 236 ret = btrfs_drop_extents(trans, root, inode, start,
70b99e69 237 aligned_end, start, &hint_byte);
c8b97818
CM		 238 BUG_ON(ret);
239
240 if (isize > actual_end)
241 inline_len = min_t(u64, isize, actual_end);
242 ret = insert_inline_extent(trans, root, inode, start,
243 inline_len, compressed_size,
244 compressed_pages);
245 BUG_ON(ret);
246 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 247 return 0;
248}
249
771ed689
CM		 250struct async_extent {
251 u64 start;
252 u64 ram_size;
253 u64 compressed_size;
254 struct page **pages;
255 unsigned long nr_pages;
256 struct list_head list;
257};
258
259struct async_cow {
260 struct inode *inode;
261 struct btrfs_root *root;
262 struct page *locked_page;
263 u64 start;
264 u64 end;
265 struct list_head extents;
266 struct btrfs_work work;
267};
268
269static noinline int add_async_extent(struct async_cow *cow,
270 u64 start, u64 ram_size,
271 u64 compressed_size,
272 struct page **pages,
273 unsigned long nr_pages)
274{
275 struct async_extent *async_extent;
276
277 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
278 async_extent->start = start;
279 async_extent->ram_size = ram_size;
280 async_extent->compressed_size = compressed_size;
281 async_extent->pages = pages;
282 async_extent->nr_pages = nr_pages;
283 list_add_tail(&async_extent->list, &cow->extents);
284 return 0;
285}
286
d352ac68 287/*
771ed689
CM		 288 * we create compressed extents in two phases. The first
289 * phase compresses a range of pages that have already been
290 * locked (both pages and state bits are locked).
c8b97818 291 *
771ed689
CM		 292 * This is done inside an ordered work queue, and the compression
293 * is spread across many cpus. The actual IO submission is step
294 * two, and the ordered work queue takes care of making sure that
295 * happens in the same order things were put onto the queue by
296 * writepages and friends.
c8b97818 297 *
771ed689
CM		 298 * If this code finds it can't get good compression, it puts an
299 * entry onto the work queue to write the uncompressed bytes. This
300 * makes sure that both compressed inodes and uncompressed inodes
301 * are written in the same order that pdflush sent them down.
d352ac68 302 */
771ed689
CM		 303static noinline int compress_file_range(struct inode *inode,
304 struct page *locked_page,
305 u64 start, u64 end,
306 struct async_cow *async_cow,
307 int *num_added)
b888db2b
CM		 308{
309 struct btrfs_root *root = BTRFS_I(inode)->root;
310 struct btrfs_trans_handle *trans;
db94535d 311 u64 num_bytes;
c8b97818
CM		 312 u64 orig_start;
313 u64 disk_num_bytes;
db94535d 314 u64 blocksize = root->sectorsize;
c8b97818 315 u64 actual_end;
42dc7bab 316 u64 isize = i_size_read(inode);
e6dcd2dc 317 int ret = 0;
c8b97818
CM		 318 struct page **pages = NULL;
319 unsigned long nr_pages;
320 unsigned long nr_pages_ret = 0;
321 unsigned long total_compressed = 0;
322 unsigned long total_in = 0;
323 unsigned long max_compressed = 128 * 1024;
771ed689 324 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 325 int i;
326 int will_compress;
b888db2b 327
c8b97818
CM		 328 orig_start = start;
329
42dc7bab 330 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 331again:
332 will_compress = 0;
333 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
334 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 335
f03d9301
CM		 336 /*
337 * we don't want to send crud past the end of i_size through
338 * compression, that's just a waste of CPU time. So, if the
339 * end of the file is before the start of our current
340 * requested range of bytes, we bail out to the uncompressed
341 * cleanup code that can deal with all of this.
342 *
343 * It isn't really the fastest way to fix things, but this is a
344 * very uncommon corner.
345 */
346 if (actual_end <= start)
347 goto cleanup_and_bail_uncompressed;
348
c8b97818
CM		 349 total_compressed = actual_end - start;
350
351 /* we want to make sure that amount of ram required to uncompress
352 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 353 * of a compressed extent to 128k. This is a crucial number
354 * because it also controls how easily we can spread reads across
355 * cpus for decompression.
356 *
357 * We also want to make sure the amount of IO required to do
358 * a random read is reasonably small, so we limit the size of
359 * a compressed extent to 128k.
c8b97818
CM		 360 */
361 total_compressed = min(total_compressed, max_uncompressed);
db94535d 362 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 363 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 364 disk_num_bytes = num_bytes;
365 total_in = 0;
366 ret = 0;
db94535d 367
771ed689
CM		 368 /*
369 * we do compression for mount -o compress and when the
370 * inode has not been flagged as nocompress. This flag can
371 * change at any time if we discover bad compression ratios.
c8b97818
CM		 372 */
373 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
374 btrfs_test_opt(root, COMPRESS)) {
375 WARN_ON(pages);
cfbc246e 376 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 377
c8b97818
CM		 378 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
379 total_compressed, pages,
380 nr_pages, &nr_pages_ret,
381 &total_in,
382 &total_compressed,
383 max_compressed);
384
385 if (!ret) {
386 unsigned long offset = total_compressed &
387 (PAGE_CACHE_SIZE - 1);
388 struct page *page = pages[nr_pages_ret - 1];
389 char *kaddr;
390
391 /* zero the tail end of the last page, we might be
392 * sending it down to disk
393 */
394 if (offset) {
395 kaddr = kmap_atomic(page, KM_USER0);
396 memset(kaddr + offset, 0,
397 PAGE_CACHE_SIZE - offset);
398 kunmap_atomic(kaddr, KM_USER0);
399 }
400 will_compress = 1;
401 }
402 }
403 if (start == 0) {
771ed689
CM		 404 trans = btrfs_join_transaction(root, 1);
405 BUG_ON(!trans);
406 btrfs_set_trans_block_group(trans, inode);
407
c8b97818 408 /* lets try to make an inline extent */
771ed689 409 if (ret || total_in < (actual_end - start)) {
c8b97818 410 /* we didn't compress the entire range, try
771ed689 411 * to make an uncompressed inline extent.
c8b97818
CM		 412 */
413 ret = cow_file_range_inline(trans, root, inode,
414 start, end, 0, NULL);
415 } else {
771ed689 416 /* try making a compressed inline extent */
c8b97818
CM		 417 ret = cow_file_range_inline(trans, root, inode,
418 start, end,
419 total_compressed, pages);
420 }
771ed689 421 btrfs_end_transaction(trans, root);
c8b97818 422 if (ret == 0) {
771ed689
CM		 423 /*
424 * inline extent creation worked, we don't need
425 * to create any more async work items. Unlock
426 * and free up our temp pages.
427 */
c8b97818
CM		 428 extent_clear_unlock_delalloc(inode,
429 &BTRFS_I(inode)->io_tree,
771ed689
CM		 430 start, end, NULL, 1, 0,
431 0, 1, 1, 1);
c8b97818
CM		 432 ret = 0;
433 goto free_pages_out;
434 }
435 }
436
437 if (will_compress) {
438 /*
439 * we aren't doing an inline extent round the compressed size
440 * up to a block size boundary so the allocator does sane
441 * things
442 */
443 total_compressed = (total_compressed + blocksize - 1) &
444 ~(blocksize - 1);
445
446 /*
447 * one last check to make sure the compression is really a
448 * win, compare the page count read with the blocks on disk
449 */
450 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
451 ~(PAGE_CACHE_SIZE - 1);
452 if (total_compressed >= total_in) {
453 will_compress = 0;
454 } else {
455 disk_num_bytes = total_compressed;
456 num_bytes = total_in;
457 }
458 }
459 if (!will_compress && pages) {
460 /*
461 * the compression code ran but failed to make things smaller,
462 * free any pages it allocated and our page pointer array
463 */
464 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 465 WARN_ON(pages[i]->mapping);
c8b97818
CM		 466 page_cache_release(pages[i]);
467 }
468 kfree(pages);
469 pages = NULL;
470 total_compressed = 0;
471 nr_pages_ret = 0;
472
473 /* flag the file so we don't compress in the future */
474 btrfs_set_flag(inode, NOCOMPRESS);
475 }
771ed689
CM		 476 if (will_compress) {
477 *num_added += 1;
c8b97818 478
771ed689
CM		 479 /* the async work queues will take care of doing actual
480 * allocation on disk for these compressed pages,
481 * and will submit them to the elevator.
482 */
483 add_async_extent(async_cow, start, num_bytes,
484 total_compressed, pages, nr_pages_ret);
179e29e4 485
42dc7bab 486 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM		 487 start += num_bytes;
488 pages = NULL;
489 cond_resched();
490 goto again;
491 }
492 } else {
f03d9301 493cleanup_and_bail_uncompressed:
771ed689
CM		 494 /*
495 * No compression, but we still need to write the pages in
496 * the file we've been given so far. redirty the locked
497 * page if it corresponds to our extent and set things up
498 * for the async work queue to run cow_file_range to do
499 * the normal delalloc dance
500 */
501 if (page_offset(locked_page) >= start &&
502 page_offset(locked_page) <= end) {
503 __set_page_dirty_nobuffers(locked_page);
504 /* unlocked later on in the async handlers */
505 }
506 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
507 *num_added += 1;
508 }
3b951516 509
771ed689
CM		 510out:
511 return 0;
512
513free_pages_out:
514 for (i = 0; i < nr_pages_ret; i++) {
515 WARN_ON(pages[i]->mapping);
516 page_cache_release(pages[i]);
517 }
d397712b 518 kfree(pages);
771ed689
CM		 519
520 goto out;
521}
522
523/*
524 * phase two of compressed writeback. This is the ordered portion
525 * of the code, which only gets called in the order the work was
526 * queued. We walk all the async extents created by compress_file_range
527 * and send them down to the disk.
528 */
529static noinline int submit_compressed_extents(struct inode *inode,
530 struct async_cow *async_cow)
531{
532 struct async_extent *async_extent;
533 u64 alloc_hint = 0;
534 struct btrfs_trans_handle *trans;
535 struct btrfs_key ins;
536 struct extent_map *em;
537 struct btrfs_root *root = BTRFS_I(inode)->root;
538 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
539 struct extent_io_tree *io_tree;
540 int ret;
541
542 if (list_empty(&async_cow->extents))
543 return 0;
544
545 trans = btrfs_join_transaction(root, 1);
546
d397712b 547 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 548 async_extent = list_entry(async_cow->extents.next,
549 struct async_extent, list);
550 list_del(&async_extent->list);
c8b97818 551
771ed689
CM		 552 io_tree = &BTRFS_I(inode)->io_tree;
553
554 /* did the compression code fall back to uncompressed IO? */
555 if (!async_extent->pages) {
556 int page_started = 0;
557 unsigned long nr_written = 0;
558
559 lock_extent(io_tree, async_extent->start,
d397712b
CM		 560 async_extent->start +
561 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 562
563 /* allocate blocks */
564 cow_file_range(inode, async_cow->locked_page,
565 async_extent->start,
566 async_extent->start +
567 async_extent->ram_size - 1,
568 &page_started, &nr_written, 0);
569
570 /*
571 * if page_started, cow_file_range inserted an
572 * inline extent and took care of all the unlocking
573 * and IO for us. Otherwise, we need to submit
574 * all those pages down to the drive.
575 */
576 if (!page_started)
577 extent_write_locked_range(io_tree,
578 inode, async_extent->start,
d397712b 579 async_extent->start +
771ed689
CM		 580 async_extent->ram_size - 1,
581 btrfs_get_extent,
582 WB_SYNC_ALL);
583 kfree(async_extent);
584 cond_resched();
585 continue;
586 }
587
588 lock_extent(io_tree, async_extent->start,
589 async_extent->start + async_extent->ram_size - 1,
590 GFP_NOFS);
c8b97818 591 /*
771ed689
CM		 592 * here we're doing allocation and writeback of the
593 * compressed pages
c8b97818 594 */
771ed689
CM		 595 btrfs_drop_extent_cache(inode, async_extent->start,
596 async_extent->start +
597 async_extent->ram_size - 1, 0);
598
599 ret = btrfs_reserve_extent(trans, root,
600 async_extent->compressed_size,
601 async_extent->compressed_size,
602 0, alloc_hint,
603 (u64)-1, &ins, 1);
604 BUG_ON(ret);
605 em = alloc_extent_map(GFP_NOFS);
606 em->start = async_extent->start;
607 em->len = async_extent->ram_size;
445a6944 608 em->orig_start = em->start;
c8b97818 609
771ed689
CM		 610 em->block_start = ins.objectid;
611 em->block_len = ins.offset;
612 em->bdev = root->fs_info->fs_devices->latest_bdev;
613 set_bit(EXTENT_FLAG_PINNED, &em->flags);
614 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
615
d397712b 616 while (1) {
771ed689
CM		 617 spin_lock(&em_tree->lock);
618 ret = add_extent_mapping(em_tree, em);
619 spin_unlock(&em_tree->lock);
620 if (ret != -EEXIST) {
621 free_extent_map(em);
622 break;
623 }
624 btrfs_drop_extent_cache(inode, async_extent->start,
625 async_extent->start +
626 async_extent->ram_size - 1, 0);
627 }
628
629 ret = btrfs_add_ordered_extent(inode, async_extent->start,
630 ins.objectid,
631 async_extent->ram_size,
632 ins.offset,
633 BTRFS_ORDERED_COMPRESSED);
634 BUG_ON(ret);
635
636 btrfs_end_transaction(trans, root);
637
638 /*
639 * clear dirty, set writeback and unlock the pages.
640 */
641 extent_clear_unlock_delalloc(inode,
642 &BTRFS_I(inode)->io_tree,
643 async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1,
646 NULL, 1, 1, 0, 1, 1, 0);
647
648 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 649 async_extent->start,
650 async_extent->ram_size,
651 ins.objectid,
652 ins.offset, async_extent->pages,
653 async_extent->nr_pages);
771ed689
CM		 654
655 BUG_ON(ret);
656 trans = btrfs_join_transaction(root, 1);
657 alloc_hint = ins.objectid + ins.offset;
658 kfree(async_extent);
659 cond_resched();
660 }
661
662 btrfs_end_transaction(trans, root);
663 return 0;
664}
665
666/*
667 * when extent_io.c finds a delayed allocation range in the file,
668 * the call backs end up in this code. The basic idea is to
669 * allocate extents on disk for the range, and create ordered data structs
670 * in ram to track those extents.
671 *
672 * locked_page is the page that writepage had locked already. We use
673 * it to make sure we don't do extra locks or unlocks.
674 *
675 * *page_started is set to one if we unlock locked_page and do everything
676 * required to start IO on it. It may be clean and already done with
677 * IO when we return.
678 */
679static noinline int cow_file_range(struct inode *inode,
680 struct page *locked_page,
681 u64 start, u64 end, int *page_started,
682 unsigned long *nr_written,
683 int unlock)
684{
685 struct btrfs_root *root = BTRFS_I(inode)->root;
686 struct btrfs_trans_handle *trans;
687 u64 alloc_hint = 0;
688 u64 num_bytes;
689 unsigned long ram_size;
690 u64 disk_num_bytes;
691 u64 cur_alloc_size;
692 u64 blocksize = root->sectorsize;
693 u64 actual_end;
42dc7bab 694 u64 isize = i_size_read(inode);
771ed689
CM		 695 struct btrfs_key ins;
696 struct extent_map *em;
697 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
698 int ret = 0;
699
700 trans = btrfs_join_transaction(root, 1);
701 BUG_ON(!trans);
702 btrfs_set_trans_block_group(trans, inode);
703
42dc7bab 704 actual_end = min_t(u64, isize, end + 1);
771ed689
CM		 705
706 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
707 num_bytes = max(blocksize, num_bytes);
708 disk_num_bytes = num_bytes;
709 ret = 0;
710
711 if (start == 0) {
712 /* lets try to make an inline extent */
713 ret = cow_file_range_inline(trans, root, inode,
714 start, end, 0, NULL);
715 if (ret == 0) {
716 extent_clear_unlock_delalloc(inode,
717 &BTRFS_I(inode)->io_tree,
718 start, end, NULL, 1, 1,
719 1, 1, 1, 1);
720 *nr_written = *nr_written +
721 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
722 *page_started = 1;
723 ret = 0;
724 goto out;
725 }
726 }
727
728 BUG_ON(disk_num_bytes >
729 btrfs_super_total_bytes(&root->fs_info->super_copy));
730
731 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
732
d397712b 733 while (disk_num_bytes > 0) {
c8b97818 734 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 735 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 736 root->sectorsize, 0, alloc_hint,
e6dcd2dc 737 (u64)-1, &ins, 1);
d397712b
CM		 738 BUG_ON(ret);
739
e6dcd2dc
CM		 740 em = alloc_extent_map(GFP_NOFS);
741 em->start = start;
445a6944 742 em->orig_start = em->start;
c8b97818 743
771ed689
CM		 744 ram_size = ins.offset;
745 em->len = ins.offset;
c8b97818 746
e6dcd2dc 747 em->block_start = ins.objectid;
c8b97818 748 em->block_len = ins.offset;
e6dcd2dc 749 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 750 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 751
d397712b 752 while (1) {
e6dcd2dc
CM		 753 spin_lock(&em_tree->lock);
754 ret = add_extent_mapping(em_tree, em);
755 spin_unlock(&em_tree->lock);
756 if (ret != -EEXIST) {
757 free_extent_map(em);
758 break;
759 }
760 btrfs_drop_extent_cache(inode, start,
c8b97818 761 start + ram_size - 1, 0);
e6dcd2dc
CM		 762 }
763
98d20f67 764 cur_alloc_size = ins.offset;
e6dcd2dc 765 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 766 ram_size, cur_alloc_size, 0);
e6dcd2dc 767 BUG_ON(ret);
c8b97818 768
17d217fe
YZ		 769 if (root->root_key.objectid ==
770 BTRFS_DATA_RELOC_TREE_OBJECTID) {
771 ret = btrfs_reloc_clone_csums(inode, start,
772 cur_alloc_size);
773 BUG_ON(ret);
774 }
775
d397712b 776 if (disk_num_bytes < cur_alloc_size)
3b951516 777 break;
d397712b 778
c8b97818
CM		 779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
782 */
783 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784 start, start + ram_size - 1,
771ed689
CM		 785 locked_page, unlock, 1,
786 1, 0, 0, 0);
c8b97818 787 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 788 num_bytes -= cur_alloc_size;
789 alloc_hint = ins.objectid + ins.offset;
790 start += cur_alloc_size;
b888db2b 791 }
b888db2b 792out:
771ed689 793 ret = 0;
b888db2b 794 btrfs_end_transaction(trans, root);
c8b97818 795
be20aa9d 796 return ret;
771ed689 797}
c8b97818 798
771ed689
CM		 799/*
800 * work queue call back to started compression on a file and pages
801 */
802static noinline void async_cow_start(struct btrfs_work *work)
803{
804 struct async_cow *async_cow;
805 int num_added = 0;
806 async_cow = container_of(work, struct async_cow, work);
807
808 compress_file_range(async_cow->inode, async_cow->locked_page,
809 async_cow->start, async_cow->end, async_cow,
810 &num_added);
811 if (num_added == 0)
812 async_cow->inode = NULL;
813}
814
815/*
816 * work queue call back to submit previously compressed pages
817 */
818static noinline void async_cow_submit(struct btrfs_work *work)
819{
820 struct async_cow *async_cow;
821 struct btrfs_root *root;
822 unsigned long nr_pages;
823
824 async_cow = container_of(work, struct async_cow, work);
825
826 root = async_cow->root;
827 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828 PAGE_CACHE_SHIFT;
829
830 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
831
832 if (atomic_read(&root->fs_info->async_delalloc_pages) <
833 5 * 1042 * 1024 &&
834 waitqueue_active(&root->fs_info->async_submit_wait))
835 wake_up(&root->fs_info->async_submit_wait);
836
d397712b 837 if (async_cow->inode)
771ed689 838 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 839}
c8b97818 840
771ed689
CM		 841static noinline void async_cow_free(struct btrfs_work *work)
842{
843 struct async_cow *async_cow;
844 async_cow = container_of(work, struct async_cow, work);
845 kfree(async_cow);
846}
847
848static int cow_file_range_async(struct inode *inode, struct page *locked_page,
849 u64 start, u64 end, int *page_started,
850 unsigned long *nr_written)
851{
852 struct async_cow *async_cow;
853 struct btrfs_root *root = BTRFS_I(inode)->root;
854 unsigned long nr_pages;
855 u64 cur_end;
856 int limit = 10 * 1024 * 1042;
857
771ed689
CM		 858 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
859 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
d397712b 860 while (start < end) {
771ed689
CM		 861 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
862 async_cow->inode = inode;
863 async_cow->root = root;
864 async_cow->locked_page = locked_page;
865 async_cow->start = start;
866
867 if (btrfs_test_flag(inode, NOCOMPRESS))
868 cur_end = end;
869 else
870 cur_end = min(end, start + 512 * 1024 - 1);
871
872 async_cow->end = cur_end;
873 INIT_LIST_HEAD(&async_cow->extents);
874
875 async_cow->work.func = async_cow_start;
876 async_cow->work.ordered_func = async_cow_submit;
877 async_cow->work.ordered_free = async_cow_free;
878 async_cow->work.flags = 0;
879
771ed689
CM		 880 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
881 PAGE_CACHE_SHIFT;
882 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
883
884 btrfs_queue_worker(&root->fs_info->delalloc_workers,
885 &async_cow->work);
886
887 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
888 wait_event(root->fs_info->async_submit_wait,
889 (atomic_read(&root->fs_info->async_delalloc_pages) <
890 limit));
891 }
892
d397712b 893 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 894 atomic_read(&root->fs_info->async_delalloc_pages)) {
895 wait_event(root->fs_info->async_submit_wait,
896 (atomic_read(&root->fs_info->async_delalloc_pages) ==
897 0));
898 }
899
900 *nr_written += nr_pages;
901 start = cur_end + 1;
902 }
903 *page_started = 1;
904 return 0;
be20aa9d
CM		 905}
906
d397712b 907static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 908 u64 bytenr, u64 num_bytes)
909{
910 int ret;
911 struct btrfs_ordered_sum *sums;
912 LIST_HEAD(list);
913
07d400a6
YZ		 914 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
915 bytenr + num_bytes - 1, &list);
17d217fe
YZ		 916 if (ret == 0 && list_empty(&list))
917 return 0;
918
919 while (!list_empty(&list)) {
920 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
921 list_del(&sums->list);
922 kfree(sums);
923 }
924 return 1;
925}
926
d352ac68
CM		 927/*
928 * when nowcow writeback call back. This checks for snapshots or COW copies
929 * of the extents that exist in the file, and COWs the file as required.
930 *
931 * If no cow copies or snapshots exist, we write directly to the existing
932 * blocks on disk
933 */
7f366cfe
CM		 934static noinline int run_delalloc_nocow(struct inode *inode,
935 struct page *locked_page,
771ed689
CM		 936 u64 start, u64 end, int *page_started, int force,
937 unsigned long *nr_written)
be20aa9d 938{
be20aa9d 939 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 940 struct btrfs_trans_handle *trans;
be20aa9d 941 struct extent_buffer *leaf;
be20aa9d 942 struct btrfs_path *path;
80ff3856 943 struct btrfs_file_extent_item *fi;
be20aa9d 944 struct btrfs_key found_key;
80ff3856
YZ		 945 u64 cow_start;
946 u64 cur_offset;
947 u64 extent_end;
948 u64 disk_bytenr;
949 u64 num_bytes;
950 int extent_type;
951 int ret;
d899e052 952 int type;
80ff3856
YZ		 953 int nocow;
954 int check_prev = 1;
be20aa9d
CM		 955
956 path = btrfs_alloc_path();
957 BUG_ON(!path);
7ea394f1
YZ		 958 trans = btrfs_join_transaction(root, 1);
959 BUG_ON(!trans);
be20aa9d 960
80ff3856
YZ		 961 cow_start = (u64)-1;
962 cur_offset = start;
963 while (1) {
964 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
965 cur_offset, 0);
966 BUG_ON(ret < 0);
967 if (ret > 0 && path->slots[0] > 0 && check_prev) {
968 leaf = path->nodes[0];
969 btrfs_item_key_to_cpu(leaf, &found_key,
970 path->slots[0] - 1);
971 if (found_key.objectid == inode->i_ino &&
972 found_key.type == BTRFS_EXTENT_DATA_KEY)
973 path->slots[0]--;
974 }
975 check_prev = 0;
976next_slot:
977 leaf = path->nodes[0];
978 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
979 ret = btrfs_next_leaf(root, path);
980 if (ret < 0)
981 BUG_ON(1);
982 if (ret > 0)
983 break;
984 leaf = path->nodes[0];
985 }
be20aa9d 986
80ff3856
YZ		 987 nocow = 0;
988 disk_bytenr = 0;
17d217fe 989 num_bytes = 0;
80ff3856
YZ		 990 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
991
992 if (found_key.objectid > inode->i_ino ||
993 found_key.type > BTRFS_EXTENT_DATA_KEY ||
994 found_key.offset > end)
995 break;
996
997 if (found_key.offset > cur_offset) {
998 extent_end = found_key.offset;
999 goto out_check;
1000 }
1001
1002 fi = btrfs_item_ptr(leaf, path->slots[0],
1003 struct btrfs_file_extent_item);
1004 extent_type = btrfs_file_extent_type(leaf, fi);
1005
d899e052
YZ		 1006 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1007 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856
YZ		 1008 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1009 extent_end = found_key.offset +
1010 btrfs_file_extent_num_bytes(leaf, fi);
1011 if (extent_end <= start) {
1012 path->slots[0]++;
1013 goto next_slot;
1014 }
17d217fe
YZ		 1015 if (disk_bytenr == 0)
1016 goto out_check;
80ff3856
YZ		 1017 if (btrfs_file_extent_compression(leaf, fi) ||
1018 btrfs_file_extent_encryption(leaf, fi) ||
1019 btrfs_file_extent_other_encoding(leaf, fi))
1020 goto out_check;
d899e052
YZ		 1021 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022 goto out_check;
d2fb3437 1023 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1024 goto out_check;
17d217fe
YZ		 1025 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026 disk_bytenr))
1027 goto out_check;
80ff3856 1028 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
17d217fe
YZ		 1029 disk_bytenr += cur_offset - found_key.offset;
1030 num_bytes = min(end + 1, extent_end) - cur_offset;
1031 /*
1032 * force cow if csum exists in the range.
1033 * this ensure that csum for a given extent are
1034 * either valid or do not exist.
1035 */
1036 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1037 goto out_check;
80ff3856
YZ		 1038 nocow = 1;
1039 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1040 extent_end = found_key.offset +
1041 btrfs_file_extent_inline_len(leaf, fi);
1042 extent_end = ALIGN(extent_end, root->sectorsize);
1043 } else {
1044 BUG_ON(1);
1045 }
1046out_check:
1047 if (extent_end <= start) {
1048 path->slots[0]++;
1049 goto next_slot;
1050 }
1051 if (!nocow) {
1052 if (cow_start == (u64)-1)
1053 cow_start = cur_offset;
1054 cur_offset = extent_end;
1055 if (cur_offset > end)
1056 break;
1057 path->slots[0]++;
1058 goto next_slot;
7ea394f1
YZ		 1059 }
1060
1061 btrfs_release_path(root, path);
80ff3856
YZ		 1062 if (cow_start != (u64)-1) {
1063 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1064 found_key.offset - 1, page_started,
1065 nr_written, 1);
80ff3856
YZ		 1066 BUG_ON(ret);
1067 cow_start = (u64)-1;
7ea394f1 1068 }
80ff3856 1069
d899e052
YZ		 1070 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1071 struct extent_map *em;
1072 struct extent_map_tree *em_tree;
1073 em_tree = &BTRFS_I(inode)->extent_tree;
1074 em = alloc_extent_map(GFP_NOFS);
1075 em->start = cur_offset;
445a6944 1076 em->orig_start = em->start;
d899e052
YZ		 1077 em->len = num_bytes;
1078 em->block_len = num_bytes;
1079 em->block_start = disk_bytenr;
1080 em->bdev = root->fs_info->fs_devices->latest_bdev;
1081 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082 while (1) {
1083 spin_lock(&em_tree->lock);
1084 ret = add_extent_mapping(em_tree, em);
1085 spin_unlock(&em_tree->lock);
1086 if (ret != -EEXIST) {
1087 free_extent_map(em);
1088 break;
1089 }
1090 btrfs_drop_extent_cache(inode, em->start,
1091 em->start + em->len - 1, 0);
1092 }
1093 type = BTRFS_ORDERED_PREALLOC;
1094 } else {
1095 type = BTRFS_ORDERED_NOCOW;
1096 }
80ff3856
YZ		 1097
1098 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1099 num_bytes, num_bytes, type);
1100 BUG_ON(ret);
771ed689 1101
d899e052
YZ		 1102 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103 cur_offset, cur_offset + num_bytes - 1,
771ed689 1104 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1105 cur_offset = extent_end;
1106 if (cur_offset > end)
1107 break;
be20aa9d 1108 }
80ff3856
YZ		 1109 btrfs_release_path(root, path);
1110
1111 if (cur_offset <= end && cow_start == (u64)-1)
1112 cow_start = cur_offset;
1113 if (cow_start != (u64)-1) {
1114 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1115 page_started, nr_written, 1);
80ff3856
YZ		 1116 BUG_ON(ret);
1117 }
1118
1119 ret = btrfs_end_transaction(trans, root);
1120 BUG_ON(ret);
7ea394f1 1121 btrfs_free_path(path);
80ff3856 1122 return 0;
be20aa9d
CM		 1123}
1124
d352ac68
CM		 1125/*
1126 * extent_io.c call back to do delayed allocation processing
1127 */
c8b97818 1128static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1129 u64 start, u64 end, int *page_started,
1130 unsigned long *nr_written)
be20aa9d 1131{
be20aa9d 1132 int ret;
7f366cfe 1133 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1134
17d217fe 1135 if (btrfs_test_flag(inode, NODATACOW))
c8b97818 1136 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1137 page_started, 1, nr_written);
d899e052
YZ		 1138 else if (btrfs_test_flag(inode, PREALLOC))
1139 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1140 page_started, 0, nr_written);
7f366cfe
CM		 1141 else if (!btrfs_test_opt(root, COMPRESS))
1142 ret = cow_file_range(inode, locked_page, start, end,
1143 page_started, nr_written, 1);
be20aa9d 1144 else
771ed689 1145 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1146 page_started, nr_written);
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
b9473439 1456 path->leave_spinning = 1;
d899e052
YZ		 1457 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458 file_pos + num_bytes, file_pos, &hint);
1459 BUG_ON(ret);
1460
1461 ins.objectid = inode->i_ino;
1462 ins.offset = file_pos;
1463 ins.type = BTRFS_EXTENT_DATA_KEY;
1464 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1465 BUG_ON(ret);
1466 leaf = path->nodes[0];
1467 fi = btrfs_item_ptr(leaf, path->slots[0],
1468 struct btrfs_file_extent_item);
1469 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1470 btrfs_set_file_extent_type(leaf, fi, extent_type);
1471 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1472 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1473 btrfs_set_file_extent_offset(leaf, fi, 0);
1474 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1475 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1476 btrfs_set_file_extent_compression(leaf, fi, compression);
1477 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1478 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1479
1480 btrfs_unlock_up_safe(path, 1);
1481 btrfs_set_lock_blocking(leaf);
1482
d899e052
YZ		 1483 btrfs_mark_buffer_dirty(leaf);
1484
1485 inode_add_bytes(inode, num_bytes);
1486 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1487
1488 ins.objectid = disk_bytenr;
1489 ins.offset = disk_num_bytes;
1490 ins.type = BTRFS_EXTENT_ITEM_KEY;
1491 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1492 root->root_key.objectid,
1493 trans->transid, inode->i_ino, &ins);
1494 BUG_ON(ret);
d899e052 1495 btrfs_free_path(path);
b9473439 1496
d899e052
YZ		 1497 return 0;
1498}
1499
d352ac68
CM		 1500/* as ordered data IO finishes, this gets called so we can finish
1501 * an ordered extent if the range of bytes in the file it covers are
1502 * fully written.
1503 */
211f90e6 1504static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1505{
e6dcd2dc
CM		 1506 struct btrfs_root *root = BTRFS_I(inode)->root;
1507 struct btrfs_trans_handle *trans;
1508 struct btrfs_ordered_extent *ordered_extent;
1509 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
b7ec40d7 1510 struct btrfs_path *path;
d899e052 1511 int compressed = 0;
e6dcd2dc
CM		 1512 int ret;
1513
1514 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1515 if (!ret)
e6dcd2dc 1516 return 0;
e6dcd2dc 1517
b7ec40d7
CM		 1518 /*
1519 * before we join the transaction, try to do some of our IO.
1520 * This will limit the amount of IO that we have to do with
1521 * the transaction running. We're unlikely to need to do any
1522 * IO if the file extents are new, the disk_i_size checks
1523 * covers the most common case.
1524 */
1525 if (start < BTRFS_I(inode)->disk_i_size) {
1526 path = btrfs_alloc_path();
1527 if (path) {
1528 ret = btrfs_lookup_file_extent(NULL, root, path,
1529 inode->i_ino,
1530 start, 0);
1531 btrfs_free_path(path);
1532 }
1533 }
1534
f9295749 1535 trans = btrfs_join_transaction(root, 1);
e6dcd2dc
CM		 1536
1537 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1538 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1539 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1540 goto nocow;
e6dcd2dc
CM		 1541
1542 lock_extent(io_tree, ordered_extent->file_offset,
1543 ordered_extent->file_offset + ordered_extent->len - 1,
1544 GFP_NOFS);
1545
c8b97818 1546 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1547 compressed = 1;
1548 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1549 BUG_ON(compressed);
1550 ret = btrfs_mark_extent_written(trans, root, inode,
1551 ordered_extent->file_offset,
1552 ordered_extent->file_offset +
1553 ordered_extent->len);
1554 BUG_ON(ret);
1555 } else {
1556 ret = insert_reserved_file_extent(trans, inode,
1557 ordered_extent->file_offset,
1558 ordered_extent->start,
1559 ordered_extent->disk_len,
1560 ordered_extent->len,
1561 ordered_extent->len,
1562 compressed, 0, 0,
1563 BTRFS_FILE_EXTENT_REG);
1564 BUG_ON(ret);
1565 }
e6dcd2dc
CM		 1566 unlock_extent(io_tree, ordered_extent->file_offset,
1567 ordered_extent->file_offset + ordered_extent->len - 1,
1568 GFP_NOFS);
7ea394f1 1569nocow:
e6dcd2dc
CM		 1570 add_pending_csums(trans, inode, ordered_extent->file_offset,
1571 &ordered_extent->list);
1572
34353029 1573 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1574 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1575 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1576 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1577 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1578
e6dcd2dc
CM		 1579 /* once for us */
1580 btrfs_put_ordered_extent(ordered_extent);
1581 /* once for the tree */
1582 btrfs_put_ordered_extent(ordered_extent);
1583
e6dcd2dc
CM		 1584 btrfs_end_transaction(trans, root);
1585 return 0;
1586}
1587
b2950863 1588static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1589 struct extent_state *state, int uptodate)
1590{
1591 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1592}
1593
d352ac68
CM		 1594/*
1595 * When IO fails, either with EIO or csum verification fails, we
1596 * try other mirrors that might have a good copy of the data. This
1597 * io_failure_record is used to record state as we go through all the
1598 * mirrors. If another mirror has good data, the page is set up to date
1599 * and things continue. If a good mirror can't be found, the original
1600 * bio end_io callback is called to indicate things have failed.
1601 */
7e38326f
CM		 1602struct io_failure_record {
1603 struct page *page;
1604 u64 start;
1605 u64 len;
1606 u64 logical;
d20f7043 1607 unsigned long bio_flags;
7e38326f
CM		 1608 int last_mirror;
1609};
1610
b2950863 1611static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1612 struct page *page, u64 start, u64 end,
1613 struct extent_state *state)
7e38326f
CM		 1614{
1615 struct io_failure_record *failrec = NULL;
1616 u64 private;
1617 struct extent_map *em;
1618 struct inode *inode = page->mapping->host;
1619 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1620 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1621 struct bio *bio;
1622 int num_copies;
1623 int ret;
1259ab75 1624 int rw;
7e38326f
CM		 1625 u64 logical;
1626
1627 ret = get_state_private(failure_tree, start, &private);
1628 if (ret) {
7e38326f
CM		 1629 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1630 if (!failrec)
1631 return -ENOMEM;
1632 failrec->start = start;
1633 failrec->len = end - start + 1;
1634 failrec->last_mirror = 0;
d20f7043 1635 failrec->bio_flags = 0;
7e38326f 1636
3b951516
CM		 1637 spin_lock(&em_tree->lock);
1638 em = lookup_extent_mapping(em_tree, start, failrec->len);
1639 if (em->start > start || em->start + em->len < start) {
1640 free_extent_map(em);
1641 em = NULL;
1642 }
1643 spin_unlock(&em_tree->lock);
7e38326f
CM		 1644
1645 if (!em || IS_ERR(em)) {
1646 kfree(failrec);
1647 return -EIO;
1648 }
1649 logical = start - em->start;
1650 logical = em->block_start + logical;
d20f7043
CM		 1651 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1652 logical = em->block_start;
1653 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1654 }
7e38326f
CM		 1655 failrec->logical = logical;
1656 free_extent_map(em);
1657 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1658 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1659 set_state_private(failure_tree, start,
1660 (u64)(unsigned long)failrec);
7e38326f 1661 } else {
587f7704 1662 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1663 }
1664 num_copies = btrfs_num_copies(
1665 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1666 failrec->logical, failrec->len);
1667 failrec->last_mirror++;
1668 if (!state) {
cad321ad 1669 spin_lock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1670 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1671 failrec->start,
1672 EXTENT_LOCKED);
1673 if (state && state->start != failrec->start)
1674 state = NULL;
cad321ad 1675 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1676 }
1677 if (!state || failrec->last_mirror > num_copies) {
1678 set_state_private(failure_tree, failrec->start, 0);
1679 clear_extent_bits(failure_tree, failrec->start,
1680 failrec->start + failrec->len - 1,
1681 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1682 kfree(failrec);
1683 return -EIO;
1684 }
1685 bio = bio_alloc(GFP_NOFS, 1);
1686 bio->bi_private = state;
1687 bio->bi_end_io = failed_bio->bi_end_io;
1688 bio->bi_sector = failrec->logical >> 9;
1689 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1690 bio->bi_size = 0;
d20f7043 1691
7e38326f 1692 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1693 if (failed_bio->bi_rw & (1 << BIO_RW))
1694 rw = WRITE;
1695 else
1696 rw = READ;
1697
1698 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1699 failrec->last_mirror,
d20f7043 1700 failrec->bio_flags);
1259ab75
CM		 1701 return 0;
1702}
1703
d352ac68
CM		 1704/*
1705 * each time an IO finishes, we do a fast check in the IO failure tree
1706 * to see if we need to process or clean up an io_failure_record
1707 */
b2950863 1708static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1709{
1710 u64 private;
1711 u64 private_failure;
1712 struct io_failure_record *failure;
1713 int ret;
1714
1715 private = 0;
1716 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1717 (u64)-1, 1, EXTENT_DIRTY)) {
1718 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1719 start, &private_failure);
1720 if (ret == 0) {
1721 failure = (struct io_failure_record *)(unsigned long)
1722 private_failure;
1723 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1724 failure->start, 0);
1725 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1726 failure->start,
1727 failure->start + failure->len - 1,
1728 EXTENT_DIRTY | EXTENT_LOCKED,
1729 GFP_NOFS);
1730 kfree(failure);
1731 }
1732 }
7e38326f
CM		 1733 return 0;
1734}
1735
d352ac68
CM		 1736/*
1737 * when reads are done, we need to check csums to verify the data is correct
1738 * if there's a match, we allow the bio to finish. If not, we go through
1739 * the io_failure_record routines to find good copies
1740 */
b2950863 1741static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1742 struct extent_state *state)
07157aac 1743{
35ebb934 1744 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1745 struct inode *inode = page->mapping->host;
d1310b2e 1746 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1747 char *kaddr;
aadfeb6e 1748 u64 private = ~(u32)0;
07157aac 1749 int ret;
ff79f819
CM		 1750 struct btrfs_root *root = BTRFS_I(inode)->root;
1751 u32 csum = ~(u32)0;
d1310b2e 1752
d20f7043
CM		 1753 if (PageChecked(page)) {
1754 ClearPageChecked(page);
1755 goto good;
1756 }
17d217fe
YZ		 1757 if (btrfs_test_flag(inode, NODATASUM))
1758 return 0;
1759
1760 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1761 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1762 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1763 GFP_NOFS);
b6cda9bc 1764 return 0;
17d217fe 1765 }
d20f7043 1766
c2e639f0 1767 if (state && state->start == start) {
70dec807
CM		 1768 private = state->private;
1769 ret = 0;
1770 } else {
1771 ret = get_state_private(io_tree, start, &private);
1772 }
9ab86c8e 1773 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1774 if (ret)
07157aac 1775 goto zeroit;
d397712b 1776
ff79f819
CM		 1777 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1778 btrfs_csum_final(csum, (char *)&csum);
d397712b 1779 if (csum != private)
07157aac 1780 goto zeroit;
d397712b 1781
9ab86c8e 1782 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1783good:
7e38326f
CM		 1784 /* if the io failure tree for this inode is non-empty,
1785 * check to see if we've recovered from a failed IO
1786 */
1259ab75 1787 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1788 return 0;
1789
1790zeroit:
d397712b
CM		 1791 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1792 "private %llu\n", page->mapping->host->i_ino,
1793 (unsigned long long)start, csum,
1794 (unsigned long long)private);
db94535d
CM		 1795 memset(kaddr + offset, 1, end - start + 1);
1796 flush_dcache_page(page);
9ab86c8e 1797 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1798 if (private == 0)
1799 return 0;
7e38326f 1800 return -EIO;
07157aac 1801}
b888db2b 1802
7b128766
JB		 1803/*
1804 * This creates an orphan entry for the given inode in case something goes
1805 * wrong in the middle of an unlink/truncate.
1806 */
1807int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1808{
1809 struct btrfs_root *root = BTRFS_I(inode)->root;
1810 int ret = 0;
1811
bcc63abb 1812 spin_lock(&root->list_lock);
7b128766
JB		 1813
1814 /* already on the orphan list, we're good */
1815 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1816 spin_unlock(&root->list_lock);
7b128766
JB		 1817 return 0;
1818 }
1819
1820 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1821
bcc63abb 1822 spin_unlock(&root->list_lock);
7b128766
JB		 1823
1824 /*
1825 * insert an orphan item to track this unlinked/truncated file
1826 */
1827 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1828
1829 return ret;
1830}
1831
1832/*
1833 * We have done the truncate/delete so we can go ahead and remove the orphan
1834 * item for this particular inode.
1835 */
1836int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1837{
1838 struct btrfs_root *root = BTRFS_I(inode)->root;
1839 int ret = 0;
1840
bcc63abb 1841 spin_lock(&root->list_lock);
7b128766
JB		 1842
1843 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1844 spin_unlock(&root->list_lock);
7b128766
JB		 1845 return 0;
1846 }
1847
1848 list_del_init(&BTRFS_I(inode)->i_orphan);
1849 if (!trans) {
bcc63abb 1850 spin_unlock(&root->list_lock);
7b128766
JB		 1851 return 0;
1852 }
1853
bcc63abb 1854 spin_unlock(&root->list_lock);
7b128766
JB		 1855
1856 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1857
1858 return ret;
1859}
1860
1861/*
1862 * this cleans up any orphans that may be left on the list from the last use
1863 * of this root.
1864 */
1865void btrfs_orphan_cleanup(struct btrfs_root *root)
1866{
1867 struct btrfs_path *path;
1868 struct extent_buffer *leaf;
1869 struct btrfs_item *item;
1870 struct btrfs_key key, found_key;
1871 struct btrfs_trans_handle *trans;
1872 struct inode *inode;
1873 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1874
7b128766
JB		 1875 path = btrfs_alloc_path();
1876 if (!path)
1877 return;
1878 path->reada = -1;
1879
1880 key.objectid = BTRFS_ORPHAN_OBJECTID;
1881 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1882 key.offset = (u64)-1;
1883
7b128766
JB		 1884
1885 while (1) {
1886 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1887 if (ret < 0) {
1888 printk(KERN_ERR "Error searching slot for orphan: %d"
1889 "\n", ret);
1890 break;
1891 }
1892
1893 /*
1894 * if ret == 0 means we found what we were searching for, which
1895 * is weird, but possible, so only screw with path if we didnt
1896 * find the key and see if we have stuff that matches
1897 */
1898 if (ret > 0) {
1899 if (path->slots[0] == 0)
1900 break;
1901 path->slots[0]--;
1902 }
1903
1904 /* pull out the item */
1905 leaf = path->nodes[0];
1906 item = btrfs_item_nr(leaf, path->slots[0]);
1907 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1908
1909 /* make sure the item matches what we want */
1910 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1911 break;
1912 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1913 break;
1914
1915 /* release the path since we're done with it */
1916 btrfs_release_path(root, path);
1917
1918 /*
1919 * this is where we are basically btrfs_lookup, without the
1920 * crossing root thing. we store the inode number in the
1921 * offset of the orphan item.
1922 */
5b21f2ed 1923 inode = btrfs_iget_locked(root->fs_info->sb,
7b128766
JB		 1924 found_key.offset, root);
1925 if (!inode)
1926 break;
1927
1928 if (inode->i_state & I_NEW) {
1929 BTRFS_I(inode)->root = root;
1930
1931 /* have to set the location manually */
1932 BTRFS_I(inode)->location.objectid = inode->i_ino;
1933 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1934 BTRFS_I(inode)->location.offset = 0;
1935
1936 btrfs_read_locked_inode(inode);
1937 unlock_new_inode(inode);
1938 }
1939
1940 /*
1941 * add this inode to the orphan list so btrfs_orphan_del does
1942 * the proper thing when we hit it
1943 */
bcc63abb 1944 spin_lock(&root->list_lock);
7b128766 1945 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1946 spin_unlock(&root->list_lock);
7b128766
JB		 1947
1948 /*
1949 * if this is a bad inode, means we actually succeeded in
1950 * removing the inode, but not the orphan record, which means
1951 * we need to manually delete the orphan since iput will just
1952 * do a destroy_inode
1953 */
1954 if (is_bad_inode(inode)) {
5b21f2ed 1955 trans = btrfs_start_transaction(root, 1);
7b128766 1956 btrfs_orphan_del(trans, inode);
5b21f2ed 1957 btrfs_end_transaction(trans, root);
7b128766
JB		 1958 iput(inode);
1959 continue;
1960 }
1961
1962 /* if we have links, this was a truncate, lets do that */
1963 if (inode->i_nlink) {
1964 nr_truncate++;
1965 btrfs_truncate(inode);
1966 } else {
1967 nr_unlink++;
1968 }
1969
1970 /* this will do delete_inode and everything for us */
1971 iput(inode);
1972 }
1973
1974 if (nr_unlink)
1975 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1976 if (nr_truncate)
1977 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1978
1979 btrfs_free_path(path);
7b128766
JB		 1980}
1981
d352ac68
CM		 1982/*
1983 * read an inode from the btree into the in-memory inode
1984 */
39279cc3
CM		 1985void btrfs_read_locked_inode(struct inode *inode)
1986{
1987 struct btrfs_path *path;
5f39d397 1988 struct extent_buffer *leaf;
39279cc3 1989 struct btrfs_inode_item *inode_item;
0b86a832 1990 struct btrfs_timespec *tspec;
39279cc3
CM		 1991 struct btrfs_root *root = BTRFS_I(inode)->root;
1992 struct btrfs_key location;
1993 u64 alloc_group_block;
618e21d5 1994 u32 rdev;
39279cc3
CM		 1995 int ret;
1996
1997 path = btrfs_alloc_path();
1998 BUG_ON(!path);
39279cc3 1999 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2000
39279cc3 2001 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2002 if (ret)
39279cc3 2003 goto make_bad;
39279cc3 2004
5f39d397
CM		 2005 leaf = path->nodes[0];
2006 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2007 struct btrfs_inode_item);
2008
2009 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2010 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2011 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2012 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2013 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2014
2015 tspec = btrfs_inode_atime(inode_item);
2016 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2017 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2018
2019 tspec = btrfs_inode_mtime(inode_item);
2020 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2021 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2022
2023 tspec = btrfs_inode_ctime(inode_item);
2024 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2025 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2026
a76a3cd4 2027 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2028 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2029 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2030 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2031 inode->i_rdev = 0;
5f39d397
CM		 2032 rdev = btrfs_inode_rdev(leaf, inode_item);
2033
aec7477b 2034 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2035 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2036
5f39d397 2037 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
b4ce94de 2038
d2fb3437
YZ		 2039 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2040 alloc_group_block, 0);
39279cc3
CM		 2041 btrfs_free_path(path);
2042 inode_item = NULL;
2043
39279cc3 2044 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2045 case S_IFREG:
2046 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2047 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2048 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2049 inode->i_fop = &btrfs_file_operations;
2050 inode->i_op = &btrfs_file_inode_operations;
2051 break;
2052 case S_IFDIR:
2053 inode->i_fop = &btrfs_dir_file_operations;
2054 if (root == root->fs_info->tree_root)
2055 inode->i_op = &btrfs_dir_ro_inode_operations;
2056 else
2057 inode->i_op = &btrfs_dir_inode_operations;
2058 break;
2059 case S_IFLNK:
2060 inode->i_op = &btrfs_symlink_inode_operations;
2061 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2062 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2063 break;
618e21d5 2064 default:
0279b4cd 2065 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2066 init_special_inode(inode, inode->i_mode, rdev);
2067 break;
39279cc3
CM		 2068 }
2069 return;
2070
2071make_bad:
39279cc3 2072 btrfs_free_path(path);
39279cc3
CM		 2073 make_bad_inode(inode);
2074}
2075
d352ac68
CM		 2076/*
2077 * given a leaf and an inode, copy the inode fields into the leaf
2078 */
e02119d5
CM		 2079static void fill_inode_item(struct btrfs_trans_handle *trans,
2080 struct extent_buffer *leaf,
5f39d397 2081 struct btrfs_inode_item *item,
39279cc3
CM		 2082 struct inode *inode)
2083{
5f39d397
CM		 2084 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2085 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2086 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2087 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2088 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2089
2090 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2091 inode->i_atime.tv_sec);
2092 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2093 inode->i_atime.tv_nsec);
2094
2095 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2096 inode->i_mtime.tv_sec);
2097 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2098 inode->i_mtime.tv_nsec);
2099
2100 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2101 inode->i_ctime.tv_sec);
2102 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2103 inode->i_ctime.tv_nsec);
2104
a76a3cd4 2105 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2106 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2107 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2108 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2109 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2110 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2111 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2112}
2113
d352ac68
CM		 2114/*
2115 * copy everything in the in-memory inode into the btree.
2116 */
d397712b
CM		 2117noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2118 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2119{
2120 struct btrfs_inode_item *inode_item;
2121 struct btrfs_path *path;
5f39d397 2122 struct extent_buffer *leaf;
39279cc3
CM		 2123 int ret;
2124
2125 path = btrfs_alloc_path();
2126 BUG_ON(!path);
b9473439 2127 path->leave_spinning = 1;
39279cc3
CM		 2128 ret = btrfs_lookup_inode(trans, root, path,
2129 &BTRFS_I(inode)->location, 1);
2130 if (ret) {
2131 if (ret > 0)
2132 ret = -ENOENT;
2133 goto failed;
2134 }
2135
b4ce94de 2136 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2137 leaf = path->nodes[0];
2138 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2139 struct btrfs_inode_item);
2140
e02119d5 2141 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2142 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2143 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2144 ret = 0;
2145failed:
39279cc3
CM		 2146 btrfs_free_path(path);
2147 return ret;
2148}
2149
2150
d352ac68
CM		 2151/*
2152 * unlink helper that gets used here in inode.c and in the tree logging
2153 * recovery code. It remove a link in a directory with a given name, and
2154 * also drops the back refs in the inode to the directory
2155 */
e02119d5
CM		 2156int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2157 struct btrfs_root *root,
2158 struct inode *dir, struct inode *inode,
2159 const char *name, int name_len)
39279cc3
CM		 2160{
2161 struct btrfs_path *path;
39279cc3 2162 int ret = 0;
5f39d397 2163 struct extent_buffer *leaf;
39279cc3 2164 struct btrfs_dir_item *di;
5f39d397 2165 struct btrfs_key key;
aec7477b 2166 u64 index;
39279cc3
CM		 2167
2168 path = btrfs_alloc_path();
54aa1f4d
CM		 2169 if (!path) {
2170 ret = -ENOMEM;
2171 goto err;
2172 }
2173
b9473439 2174 path->leave_spinning = 1;
39279cc3
CM		 2175 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2176 name, name_len, -1);
2177 if (IS_ERR(di)) {
2178 ret = PTR_ERR(di);
2179 goto err;
2180 }
2181 if (!di) {
2182 ret = -ENOENT;
2183 goto err;
2184 }
5f39d397
CM		 2185 leaf = path->nodes[0];
2186 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2187 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2188 if (ret)
2189 goto err;
39279cc3
CM		 2190 btrfs_release_path(root, path);
2191
aec7477b 2192 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2193 inode->i_ino,
2194 dir->i_ino, &index);
aec7477b 2195 if (ret) {
d397712b 2196 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
aec7477b 2197 "inode %lu parent %lu\n", name_len, name,
e02119d5 2198 inode->i_ino, dir->i_ino);
aec7477b
JB		 2199 goto err;
2200 }
2201
39279cc3 2202 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2203 index, name, name_len, -1);
39279cc3
CM		 2204 if (IS_ERR(di)) {
2205 ret = PTR_ERR(di);
2206 goto err;
2207 }
2208 if (!di) {
2209 ret = -ENOENT;
2210 goto err;
2211 }
2212 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2213 btrfs_release_path(root, path);
39279cc3 2214
e02119d5
CM		 2215 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2216 inode, dir->i_ino);
49eb7e46
CM		 2217 BUG_ON(ret != 0 && ret != -ENOENT);
2218 if (ret != -ENOENT)
2219 BTRFS_I(dir)->log_dirty_trans = trans->transid;
e02119d5
CM		 2220
2221 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2222 dir, index);
2223 BUG_ON(ret);
39279cc3
CM		 2224err:
2225 btrfs_free_path(path);
e02119d5
CM		 2226 if (ret)
2227 goto out;
2228
2229 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2230 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2231 btrfs_update_inode(trans, root, dir);
2232 btrfs_drop_nlink(inode);
2233 ret = btrfs_update_inode(trans, root, inode);
2234 dir->i_sb->s_dirt = 1;
2235out:
39279cc3
CM		 2236 return ret;
2237}
2238
2239static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2240{
2241 struct btrfs_root *root;
2242 struct btrfs_trans_handle *trans;
7b128766 2243 struct inode *inode = dentry->d_inode;
39279cc3 2244 int ret;
1832a6d5 2245 unsigned long nr = 0;
39279cc3
CM		 2246
2247 root = BTRFS_I(dir)->root;
1832a6d5 2248
39279cc3 2249 trans = btrfs_start_transaction(root, 1);
5f39d397 2250
39279cc3 2251 btrfs_set_trans_block_group(trans, dir);
e02119d5
CM		 2252 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2253 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2254
2255 if (inode->i_nlink == 0)
2256 ret = btrfs_orphan_add(trans, inode);
2257
d3c2fdcf 2258 nr = trans->blocks_used;
5f39d397 2259
89ce8a63 2260 btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2261 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2262 return ret;
2263}
2264
2265static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2266{
2267 struct inode *inode = dentry->d_inode;
1832a6d5 2268 int err = 0;
39279cc3
CM		 2269 int ret;
2270 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2271 struct btrfs_trans_handle *trans;
1832a6d5 2272 unsigned long nr = 0;
39279cc3 2273
3394e160
CM		 2274 /*
2275 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2276 * the root of a subvolume or snapshot
2277 */
2278 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2279 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2280 return -ENOTEMPTY;
925baedd 2281 }
134d4512 2282
39279cc3
CM		 2283 trans = btrfs_start_transaction(root, 1);
2284 btrfs_set_trans_block_group(trans, dir);
39279cc3 2285
7b128766
JB		 2286 err = btrfs_orphan_add(trans, inode);
2287 if (err)
2288 goto fail_trans;
2289
39279cc3 2290 /* now the directory is empty */
e02119d5
CM		 2291 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2292 dentry->d_name.name, dentry->d_name.len);
d397712b 2293 if (!err)
dbe674a9 2294 btrfs_i_size_write(inode, 0);
3954401f 2295
7b128766 2296fail_trans:
d3c2fdcf 2297 nr = trans->blocks_used;
89ce8a63 2298 ret = btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2299 btrfs_btree_balance_dirty(root, nr);
3954401f 2300
39279cc3
CM		 2301 if (ret && !err)
2302 err = ret;
2303 return err;
2304}
2305
d20f7043 2306#if 0
323ac95b
CM		 2307/*
2308 * when truncating bytes in a file, it is possible to avoid reading
2309 * the leaves that contain only checksum items. This can be the
2310 * majority of the IO required to delete a large file, but it must
2311 * be done carefully.
2312 *
2313 * The keys in the level just above the leaves are checked to make sure
2314 * the lowest key in a given leaf is a csum key, and starts at an offset
2315 * after the new size.
2316 *
2317 * Then the key for the next leaf is checked to make sure it also has
2318 * a checksum item for the same file. If it does, we know our target leaf
2319 * contains only checksum items, and it can be safely freed without reading
2320 * it.
2321 *
2322 * This is just an optimization targeted at large files. It may do
2323 * nothing. It will return 0 unless things went badly.
2324 */
2325static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2326 struct btrfs_root *root,
2327 struct btrfs_path *path,
2328 struct inode *inode, u64 new_size)
2329{
2330 struct btrfs_key key;
2331 int ret;
2332 int nritems;
2333 struct btrfs_key found_key;
2334 struct btrfs_key other_key;
5b84e8d6
YZ		 2335 struct btrfs_leaf_ref *ref;
2336 u64 leaf_gen;
2337 u64 leaf_start;
323ac95b
CM		 2338
2339 path->lowest_level = 1;
2340 key.objectid = inode->i_ino;
2341 key.type = BTRFS_CSUM_ITEM_KEY;
2342 key.offset = new_size;
2343again:
2344 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2345 if (ret < 0)
2346 goto out;
2347
2348 if (path->nodes[1] == NULL) {
2349 ret = 0;
2350 goto out;
2351 }
2352 ret = 0;
2353 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2354 nritems = btrfs_header_nritems(path->nodes[1]);
2355
2356 if (!nritems)
2357 goto out;
2358
2359 if (path->slots[1] >= nritems)
2360 goto next_node;
2361
2362 /* did we find a key greater than anything we want to delete? */
2363 if (found_key.objectid > inode->i_ino ||
2364 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2365 goto out;
2366
2367 /* we check the next key in the node to make sure the leave contains
2368 * only checksum items. This comparison doesn't work if our
2369 * leaf is the last one in the node
2370 */
2371 if (path->slots[1] + 1 >= nritems) {
2372next_node:
2373 /* search forward from the last key in the node, this
2374 * will bring us into the next node in the tree
2375 */
2376 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2377
2378 /* unlikely, but we inc below, so check to be safe */
2379 if (found_key.offset == (u64)-1)
2380 goto out;
2381
2382 /* search_forward needs a path with locks held, do the
2383 * search again for the original key. It is possible
2384 * this will race with a balance and return a path that
2385 * we could modify, but this drop is just an optimization
2386 * and is allowed to miss some leaves.
2387 */
2388 btrfs_release_path(root, path);
2389 found_key.offset++;
2390
2391 /* setup a max key for search_forward */
2392 other_key.offset = (u64)-1;
2393 other_key.type = key.type;
2394 other_key.objectid = key.objectid;
2395
2396 path->keep_locks = 1;
2397 ret = btrfs_search_forward(root, &found_key, &other_key,
2398 path, 0, 0);
2399 path->keep_locks = 0;
2400 if (ret || found_key.objectid != key.objectid ||
2401 found_key.type != key.type) {
2402 ret = 0;
2403 goto out;
2404 }
2405
2406 key.offset = found_key.offset;
2407 btrfs_release_path(root, path);
2408 cond_resched();
2409 goto again;
2410 }
2411
2412 /* we know there's one more slot after us in the tree,
2413 * read that key so we can verify it is also a checksum item
2414 */
2415 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2416
2417 if (found_key.objectid < inode->i_ino)
2418 goto next_key;
2419
2420 if (found_key.type != key.type || found_key.offset < new_size)
2421 goto next_key;
2422
2423 /*
2424 * if the key for the next leaf isn't a csum key from this objectid,
2425 * we can't be sure there aren't good items inside this leaf.
2426 * Bail out
2427 */
2428 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2429 goto out;
2430
5b84e8d6
YZ		 2431 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2432 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2433 /*
2434 * it is safe to delete this leaf, it contains only
2435 * csum items from this inode at an offset >= new_size
2436 */
5b84e8d6 2437 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2438 BUG_ON(ret);
2439
5b84e8d6
YZ		 2440 if (root->ref_cows && leaf_gen < trans->transid) {
2441 ref = btrfs_alloc_leaf_ref(root, 0);
2442 if (ref) {
2443 ref->root_gen = root->root_key.offset;
2444 ref->bytenr = leaf_start;
2445 ref->owner = 0;
2446 ref->generation = leaf_gen;
2447 ref->nritems = 0;
2448
bd56b302
CM		 2449 btrfs_sort_leaf_ref(ref);
2450
5b84e8d6
YZ		 2451 ret = btrfs_add_leaf_ref(root, ref, 0);
2452 WARN_ON(ret);
2453 btrfs_free_leaf_ref(root, ref);
2454 } else {
2455 WARN_ON(1);
2456 }
2457 }
323ac95b
CM		 2458next_key:
2459 btrfs_release_path(root, path);
2460
2461 if (other_key.objectid == inode->i_ino &&
2462 other_key.type == key.type && other_key.offset > key.offset) {
2463 key.offset = other_key.offset;
2464 cond_resched();
2465 goto again;
2466 }
2467 ret = 0;
2468out:
2469 /* fixup any changes we've made to the path */
2470 path->lowest_level = 0;
2471 path->keep_locks = 0;
2472 btrfs_release_path(root, path);
2473 return ret;
2474}
2475
d20f7043
CM		 2476#endif
2477
39279cc3
CM		 2478/*
2479 * this can truncate away extent items, csum items and directory items.
2480 * It starts at a high offset and removes keys until it can't find
d352ac68 2481 * any higher than new_size
39279cc3
CM		 2482 *
2483 * csum items that cross the new i_size are truncated to the new size
2484 * as well.
7b128766
JB		 2485 *
2486 * min_type is the minimum key type to truncate down to. If set to 0, this
2487 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2488 */
e02119d5
CM		 2489noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2490 struct btrfs_root *root,
2491 struct inode *inode,
2492 u64 new_size, u32 min_type)
39279cc3
CM		 2493{
2494 int ret;
2495 struct btrfs_path *path;
2496 struct btrfs_key key;
5f39d397 2497 struct btrfs_key found_key;
06d9a8d7 2498 u32 found_type = (u8)-1;
5f39d397 2499 struct extent_buffer *leaf;
39279cc3
CM		 2500 struct btrfs_file_extent_item *fi;
2501 u64 extent_start = 0;
db94535d 2502 u64 extent_num_bytes = 0;
39279cc3 2503 u64 item_end = 0;
7bb86316 2504 u64 root_gen = 0;
d8d5f3e1 2505 u64 root_owner = 0;
39279cc3
CM		 2506 int found_extent;
2507 int del_item;
85e21bac
CM		 2508 int pending_del_nr = 0;
2509 int pending_del_slot = 0;
179e29e4 2510 int extent_type = -1;
771ed689 2511 int encoding;
3b951516 2512 u64 mask = root->sectorsize - 1;
39279cc3 2513
e02119d5 2514 if (root->ref_cows)
5b21f2ed 2515 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2516 path = btrfs_alloc_path();
3c69faec 2517 path->reada = -1;
39279cc3 2518 BUG_ON(!path);
5f39d397 2519
39279cc3
CM		 2520 /* FIXME, add redo link to tree so we don't leak on crash */
2521 key.objectid = inode->i_ino;
2522 key.offset = (u64)-1;
5f39d397
CM		 2523 key.type = (u8)-1;
2524
85e21bac 2525search_again:
b9473439 2526 path->leave_spinning = 1;
85e21bac 2527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
d397712b 2528 if (ret < 0)
85e21bac 2529 goto error;
d397712b 2530
85e21bac 2531 if (ret > 0) {
e02119d5
CM		 2532 /* there are no items in the tree for us to truncate, we're
2533 * done
2534 */
2535 if (path->slots[0] == 0) {
2536 ret = 0;
2537 goto error;
2538 }
85e21bac
CM		 2539 path->slots[0]--;
2540 }
2541
d397712b 2542 while (1) {
39279cc3 2543 fi = NULL;
5f39d397
CM		 2544 leaf = path->nodes[0];
2545 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2546 found_type = btrfs_key_type(&found_key);
771ed689 2547 encoding = 0;
39279cc3 2548
5f39d397 2549 if (found_key.objectid != inode->i_ino)
39279cc3 2550 break;
5f39d397 2551
85e21bac 2552 if (found_type < min_type)
39279cc3
CM		 2553 break;
2554
5f39d397 2555 item_end = found_key.offset;
39279cc3 2556 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2557 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2558 struct btrfs_file_extent_item);
179e29e4 2559 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2560 encoding = btrfs_file_extent_compression(leaf, fi);
2561 encoding |= btrfs_file_extent_encryption(leaf, fi);
2562 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2563
179e29e4 2564 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2565 item_end +=
db94535d 2566 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2567 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2568 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2569 fi);
39279cc3 2570 }
008630c1 2571 item_end--;
39279cc3 2572 }
e02119d5 2573 if (item_end < new_size) {
d397712b 2574 if (found_type == BTRFS_DIR_ITEM_KEY)
b888db2b 2575 found_type = BTRFS_INODE_ITEM_KEY;
d397712b 2576 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
d20f7043 2577 found_type = BTRFS_EXTENT_DATA_KEY;
d397712b 2578 else if (found_type == BTRFS_EXTENT_DATA_KEY)
85e21bac 2579 found_type = BTRFS_XATTR_ITEM_KEY;
d397712b 2580 else if (found_type == BTRFS_XATTR_ITEM_KEY)
85e21bac 2581 found_type = BTRFS_INODE_REF_KEY;
d397712b 2582 else if (found_type)
b888db2b 2583 found_type--;
d397712b 2584 else
b888db2b 2585 break;
a61721d5 2586 btrfs_set_key_type(&key, found_type);
85e21bac 2587 goto next;
39279cc3 2588 }
e02119d5 2589 if (found_key.offset >= new_size)
39279cc3
CM		 2590 del_item = 1;
2591 else
2592 del_item = 0;
2593 found_extent = 0;
2594
2595 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2596 if (found_type != BTRFS_EXTENT_DATA_KEY)
2597 goto delete;
2598
2599 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2600 u64 num_dec;
db94535d 2601 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2602 if (!del_item && !encoding) {
db94535d
CM		 2603 u64 orig_num_bytes =
2604 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2605 extent_num_bytes = new_size -
5f39d397 2606 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2607 extent_num_bytes = extent_num_bytes &
2608 ~((u64)root->sectorsize - 1);
db94535d
CM		 2609 btrfs_set_file_extent_num_bytes(leaf, fi,
2610 extent_num_bytes);
2611 num_dec = (orig_num_bytes -
9069218d 2612 extent_num_bytes);
e02119d5 2613 if (root->ref_cows && extent_start != 0)
a76a3cd4 2614 inode_sub_bytes(inode, num_dec);
5f39d397 2615 btrfs_mark_buffer_dirty(leaf);
39279cc3 2616 } else {
db94535d
CM		 2617 extent_num_bytes =
2618 btrfs_file_extent_disk_num_bytes(leaf,
2619 fi);
39279cc3 2620 /* FIXME blocksize != 4096 */
9069218d 2621 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2622 if (extent_start != 0) {
2623 found_extent = 1;
e02119d5 2624 if (root->ref_cows)
a76a3cd4 2625 inode_sub_bytes(inode, num_dec);
e02119d5 2626 }
31840ae1 2627 root_gen = btrfs_header_generation(leaf);
d8d5f3e1 2628 root_owner = btrfs_header_owner(leaf);
39279cc3 2629 }
9069218d 2630 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2631 /*
2632 * we can't truncate inline items that have had
2633 * special encodings
2634 */
2635 if (!del_item &&
2636 btrfs_file_extent_compression(leaf, fi) == 0 &&
2637 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2638 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2639 u32 size = new_size - found_key.offset;
2640
2641 if (root->ref_cows) {
a76a3cd4
YZ		 2642 inode_sub_bytes(inode, item_end + 1 -
2643 new_size);
e02119d5
CM		 2644 }
2645 size =
2646 btrfs_file_extent_calc_inline_size(size);
9069218d 2647 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2648 size, 1);
9069218d 2649 BUG_ON(ret);
e02119d5 2650 } else if (root->ref_cows) {
a76a3cd4
YZ		 2651 inode_sub_bytes(inode, item_end + 1 -
2652 found_key.offset);
9069218d 2653 }
39279cc3 2654 }
179e29e4 2655delete:
39279cc3 2656 if (del_item) {
85e21bac
CM		 2657 if (!pending_del_nr) {
2658 /* no pending yet, add ourselves */
2659 pending_del_slot = path->slots[0];
2660 pending_del_nr = 1;
2661 } else if (pending_del_nr &&
2662 path->slots[0] + 1 == pending_del_slot) {
2663 /* hop on the pending chunk */
2664 pending_del_nr++;
2665 pending_del_slot = path->slots[0];
2666 } else {
d397712b 2667 BUG();
85e21bac 2668 }
39279cc3
CM		 2669 } else {
2670 break;
2671 }
39279cc3 2672 if (found_extent) {
b9473439 2673 btrfs_set_path_blocking(path);
39279cc3 2674 ret = btrfs_free_extent(trans, root, extent_start,
7bb86316 2675 extent_num_bytes,
31840ae1 2676 leaf->start, root_owner,
3bb1a1bc 2677 root_gen, inode->i_ino, 0);
39279cc3
CM		 2678 BUG_ON(ret);
2679 }
85e21bac
CM		 2680next:
2681 if (path->slots[0] == 0) {
2682 if (pending_del_nr)
2683 goto del_pending;
2684 btrfs_release_path(root, path);
06d9a8d7
CM		 2685 if (found_type == BTRFS_INODE_ITEM_KEY)
2686 break;
85e21bac
CM		 2687 goto search_again;
2688 }
2689
2690 path->slots[0]--;
2691 if (pending_del_nr &&
2692 path->slots[0] + 1 != pending_del_slot) {
2693 struct btrfs_key debug;
2694del_pending:
2695 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2696 pending_del_slot);
2697 ret = btrfs_del_items(trans, root, path,
2698 pending_del_slot,
2699 pending_del_nr);
2700 BUG_ON(ret);
2701 pending_del_nr = 0;
2702 btrfs_release_path(root, path);
06d9a8d7
CM		 2703 if (found_type == BTRFS_INODE_ITEM_KEY)
2704 break;
85e21bac
CM		 2705 goto search_again;
2706 }
39279cc3
CM		 2707 }
2708 ret = 0;
2709error:
85e21bac
CM		 2710 if (pending_del_nr) {
2711 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2712 pending_del_nr);
2713 }
39279cc3
CM		 2714 btrfs_free_path(path);
2715 inode->i_sb->s_dirt = 1;
2716 return ret;
2717}
2718
2719/*
2720 * taken from block_truncate_page, but does cow as it zeros out
2721 * any bytes left in the last page in the file.
2722 */
2723static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2724{
2725 struct inode *inode = mapping->host;
db94535d 2726 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2727 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2728 struct btrfs_ordered_extent *ordered;
2729 char *kaddr;
db94535d 2730 u32 blocksize = root->sectorsize;
39279cc3
CM		 2731 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2732 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2733 struct page *page;
39279cc3 2734 int ret = 0;
a52d9a80 2735 u64 page_start;
e6dcd2dc 2736 u64 page_end;
39279cc3
CM		 2737
2738 if ((offset & (blocksize - 1)) == 0)
2739 goto out;
2740
2741 ret = -ENOMEM;
211c17f5 2742again:
39279cc3
CM		 2743 page = grab_cache_page(mapping, index);
2744 if (!page)
2745 goto out;
e6dcd2dc
CM		 2746
2747 page_start = page_offset(page);
2748 page_end = page_start + PAGE_CACHE_SIZE - 1;
2749
39279cc3 2750 if (!PageUptodate(page)) {
9ebefb18 2751 ret = btrfs_readpage(NULL, page);
39279cc3 2752 lock_page(page);
211c17f5
CM		 2753 if (page->mapping != mapping) {
2754 unlock_page(page);
2755 page_cache_release(page);
2756 goto again;
2757 }
39279cc3
CM		 2758 if (!PageUptodate(page)) {
2759 ret = -EIO;
89642229 2760 goto out_unlock;
39279cc3
CM		 2761 }
2762 }
211c17f5 2763 wait_on_page_writeback(page);
e6dcd2dc
CM		 2764
2765 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2766 set_page_extent_mapped(page);
2767
2768 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2769 if (ordered) {
2770 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2771 unlock_page(page);
2772 page_cache_release(page);
eb84ae03 2773 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2774 btrfs_put_ordered_extent(ordered);
2775 goto again;
2776 }
2777
ea8c2819 2778 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2779 ret = 0;
2780 if (offset != PAGE_CACHE_SIZE) {
2781 kaddr = kmap(page);
2782 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2783 flush_dcache_page(page);
2784 kunmap(page);
2785 }
247e743c 2786 ClearPageChecked(page);
e6dcd2dc
CM		 2787 set_page_dirty(page);
2788 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2789
89642229 2790out_unlock:
39279cc3
CM		 2791 unlock_page(page);
2792 page_cache_release(page);
2793out:
2794 return ret;
2795}
2796
9036c102 2797int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2798{
9036c102
YZ		 2799 struct btrfs_trans_handle *trans;
2800 struct btrfs_root *root = BTRFS_I(inode)->root;
2801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2802 struct extent_map *em;
2803 u64 mask = root->sectorsize - 1;
2804 u64 hole_start = (inode->i_size + mask) & ~mask;
2805 u64 block_end = (size + mask) & ~mask;
2806 u64 last_byte;
2807 u64 cur_offset;
2808 u64 hole_size;
39279cc3
CM		 2809 int err;
2810
9036c102
YZ		 2811 if (size <= hole_start)
2812 return 0;
2813
6a63209f 2814 err = btrfs_check_metadata_free_space(root);
39279cc3
CM		 2815 if (err)
2816 return err;
2817
9036c102 2818 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2819
9036c102
YZ		 2820 while (1) {
2821 struct btrfs_ordered_extent *ordered;
2822 btrfs_wait_ordered_range(inode, hole_start,
2823 block_end - hole_start);
2824 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2825 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2826 if (!ordered)
2827 break;
2828 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2829 btrfs_put_ordered_extent(ordered);
2830 }
39279cc3 2831
9036c102
YZ		 2832 trans = btrfs_start_transaction(root, 1);
2833 btrfs_set_trans_block_group(trans, inode);
39279cc3 2834
9036c102
YZ		 2835 cur_offset = hole_start;
2836 while (1) {
2837 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2838 block_end - cur_offset, 0);
2839 BUG_ON(IS_ERR(em) || !em);
2840 last_byte = min(extent_map_end(em), block_end);
2841 last_byte = (last_byte + mask) & ~mask;
2842 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2843 u64 hint_byte = 0;
9036c102 2844 hole_size = last_byte - cur_offset;
771ed689
CM		 2845 err = btrfs_drop_extents(trans, root, inode,
2846 cur_offset,
2847 cur_offset + hole_size,
2848 cur_offset, &hint_byte);
2849 if (err)
2850 break;
9036c102
YZ		 2851 err = btrfs_insert_file_extent(trans, root,
2852 inode->i_ino, cur_offset, 0,
2853 0, hole_size, 0, hole_size,
2854 0, 0, 0);
2855 btrfs_drop_extent_cache(inode, hole_start,
2856 last_byte - 1, 0);
2857 }
2858 free_extent_map(em);
2859 cur_offset = last_byte;
2860 if (err || cur_offset >= block_end)
2861 break;
2862 }
1832a6d5 2863
9036c102
YZ		 2864 btrfs_end_transaction(trans, root);
2865 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2866 return err;
2867}
39279cc3 2868
9036c102
YZ		 2869static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2870{
2871 struct inode *inode = dentry->d_inode;
2872 int err;
39279cc3 2873
9036c102
YZ		 2874 err = inode_change_ok(inode, attr);
2875 if (err)
2876 return err;
2bf5a725 2877
9036c102
YZ		 2878 if (S_ISREG(inode->i_mode) &&
2879 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2880 err = btrfs_cont_expand(inode, attr->ia_size);
54aa1f4d
CM		 2881 if (err)
2882 return err;
39279cc3 2883 }
9036c102 2884
39279cc3 2885 err = inode_setattr(inode, attr);
33268eaf
JB		 2886
2887 if (!err && ((attr->ia_valid & ATTR_MODE)))
2888 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2889 return err;
2890}
61295eb8 2891
39279cc3
CM		 2892void btrfs_delete_inode(struct inode *inode)
2893{
2894 struct btrfs_trans_handle *trans;
2895 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2896 unsigned long nr;
39279cc3
CM		 2897 int ret;
2898
2899 truncate_inode_pages(&inode->i_data, 0);
2900 if (is_bad_inode(inode)) {
7b128766 2901 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 2902 goto no_delete;
2903 }
4a096752 2904 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 2905
dbe674a9 2906 btrfs_i_size_write(inode, 0);
180591bc 2907 trans = btrfs_join_transaction(root, 1);
5f39d397 2908
39279cc3 2909 btrfs_set_trans_block_group(trans, inode);
e02119d5 2910 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 2911 if (ret) {
2912 btrfs_orphan_del(NULL, inode);
54aa1f4d 2913 goto no_delete_lock;
7b128766
JB		 2914 }
2915
2916 btrfs_orphan_del(trans, inode);
85e21bac 2917
d3c2fdcf 2918 nr = trans->blocks_used;
85e21bac 2919 clear_inode(inode);
5f39d397 2920
39279cc3 2921 btrfs_end_transaction(trans, root);
d3c2fdcf 2922 btrfs_btree_balance_dirty(root, nr);
39279cc3 2923 return;
54aa1f4d
CM		 2924
2925no_delete_lock:
d3c2fdcf 2926 nr = trans->blocks_used;
54aa1f4d 2927 btrfs_end_transaction(trans, root);
d3c2fdcf 2928 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2929no_delete:
2930 clear_inode(inode);
2931}
2932
2933/*
2934 * this returns the key found in the dir entry in the location pointer.
2935 * If no dir entries were found, location->objectid is 0.
2936 */
2937static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2938 struct btrfs_key *location)
2939{
2940 const char *name = dentry->d_name.name;
2941 int namelen = dentry->d_name.len;
2942 struct btrfs_dir_item *di;
2943 struct btrfs_path *path;
2944 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 2945 int ret = 0;
39279cc3
CM		 2946
2947 path = btrfs_alloc_path();
2948 BUG_ON(!path);
3954401f 2949
39279cc3
CM		 2950 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2951 namelen, 0);
0d9f7f3e
Y		 2952 if (IS_ERR(di))
2953 ret = PTR_ERR(di);
d397712b
CM		 2954
2955 if (!di || IS_ERR(di))
3954401f 2956 goto out_err;
d397712b 2957
5f39d397 2958 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 2959out:
39279cc3
CM		 2960 btrfs_free_path(path);
2961 return ret;
3954401f
CM		 2962out_err:
2963 location->objectid = 0;
2964 goto out;
39279cc3
CM		 2965}
2966
2967/*
2968 * when we hit a tree root in a directory, the btrfs part of the inode
2969 * needs to be changed to reflect the root directory of the tree root. This
2970 * is kind of like crossing a mount point.
2971 */
2972static int fixup_tree_root_location(struct btrfs_root *root,
2973 struct btrfs_key *location,
58176a96
JB		 2974 struct btrfs_root **sub_root,
2975 struct dentry *dentry)
39279cc3 2976{
39279cc3
CM		 2977 struct btrfs_root_item *ri;
2978
2979 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2980 return 0;
2981 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2982 return 0;
2983
58176a96
JB		 2984 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2985 dentry->d_name.name,
2986 dentry->d_name.len);
39279cc3
CM		 2987 if (IS_ERR(*sub_root))
2988 return PTR_ERR(*sub_root);
2989
2990 ri = &(*sub_root)->root_item;
2991 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 2992 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2993 location->offset = 0;
2994
39279cc3
CM		 2995 return 0;
2996}
2997
e02119d5 2998static noinline void init_btrfs_i(struct inode *inode)
39279cc3 2999{
e02119d5
CM		 3000 struct btrfs_inode *bi = BTRFS_I(inode);
3001
3002 bi->i_acl = NULL;
3003 bi->i_default_acl = NULL;
3004
3005 bi->generation = 0;
c3027eb5 3006 bi->sequence = 0;
e02119d5
CM		 3007 bi->last_trans = 0;
3008 bi->logged_trans = 0;
3009 bi->delalloc_bytes = 0;
6a63209f 3010 bi->reserved_bytes = 0;
e02119d5
CM		 3011 bi->disk_i_size = 0;
3012 bi->flags = 0;
3013 bi->index_cnt = (u64)-1;
49eb7e46 3014 bi->log_dirty_trans = 0;
d1310b2e
CM		 3015 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3016 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 3017 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 3018 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3019 inode->i_mapping, GFP_NOFS);
ea8c2819 3020 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
ba1da2f4 3021 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 3022 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 3023 mutex_init(&BTRFS_I(inode)->log_mutex);
3024}
3025
3026static int btrfs_init_locked_inode(struct inode *inode, void *p)
3027{
3028 struct btrfs_iget_args *args = p;
3029 inode->i_ino = args->ino;
3030 init_btrfs_i(inode);
3031 BTRFS_I(inode)->root = args->root;
6a63209f 3032 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3033 return 0;
3034}
3035
3036static int btrfs_find_actor(struct inode *inode, void *opaque)
3037{
3038 struct btrfs_iget_args *args = opaque;
d397712b
CM		 3039 return args->ino == inode->i_ino &&
3040 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3041}
3042
5b21f2ed
ZY		 3043struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3044 struct btrfs_root *root, int wait)
3045{
3046 struct inode *inode;
3047 struct btrfs_iget_args args;
3048 args.ino = objectid;
3049 args.root = root;
3050
3051 if (wait) {
3052 inode = ilookup5(s, objectid, btrfs_find_actor,
3053 (void *)&args);
3054 } else {
3055 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3056 (void *)&args);
3057 }
3058 return inode;
3059}
3060
39279cc3
CM		 3061struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3062 struct btrfs_root *root)
3063{
3064 struct inode *inode;
3065 struct btrfs_iget_args args;
3066 args.ino = objectid;
3067 args.root = root;
3068
3069 inode = iget5_locked(s, objectid, btrfs_find_actor,
3070 btrfs_init_locked_inode,
3071 (void *)&args);
3072 return inode;
3073}
3074
1a54ef8c
BR		 3075/* Get an inode object given its location and corresponding root.
3076 * Returns in *is_new if the inode was read from disk
3077 */
3078struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3079 struct btrfs_root *root, int *is_new)
3080{
3081 struct inode *inode;
3082
3083 inode = btrfs_iget_locked(s, location->objectid, root);
3084 if (!inode)
3085 return ERR_PTR(-EACCES);
3086
3087 if (inode->i_state & I_NEW) {
3088 BTRFS_I(inode)->root = root;
3089 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3090 btrfs_read_locked_inode(inode);
3091 unlock_new_inode(inode);
3092 if (is_new)
3093 *is_new = 1;
3094 } else {
3095 if (is_new)
3096 *is_new = 0;
3097 }
3098
3099 return inode;
3100}
3101
3de4586c 3102struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3103{
d397712b 3104 struct inode *inode;
39279cc3
CM		 3105 struct btrfs_inode *bi = BTRFS_I(dir);
3106 struct btrfs_root *root = bi->root;
3107 struct btrfs_root *sub_root = root;
3108 struct btrfs_key location;
c146afad 3109 int ret, new;
39279cc3
CM		 3110
3111 if (dentry->d_name.len > BTRFS_NAME_LEN)
3112 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3113
39279cc3 3114 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3115
39279cc3
CM		 3116 if (ret < 0)
3117 return ERR_PTR(ret);
5f39d397 3118
39279cc3
CM		 3119 inode = NULL;
3120 if (location.objectid) {
58176a96
JB		 3121 ret = fixup_tree_root_location(root, &location, &sub_root,
3122 dentry);
39279cc3
CM		 3123 if (ret < 0)
3124 return ERR_PTR(ret);
3125 if (ret > 0)
3126 return ERR_PTR(-ENOENT);
1a54ef8c
BR		 3127 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3128 if (IS_ERR(inode))
3129 return ERR_CAST(inode);
39279cc3 3130 }
3de4586c
CM		 3131 return inode;
3132}
3133
3134static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3135 struct nameidata *nd)
3136{
3137 struct inode *inode;
3138
3139 if (dentry->d_name.len > BTRFS_NAME_LEN)
3140 return ERR_PTR(-ENAMETOOLONG);
3141
3142 inode = btrfs_lookup_dentry(dir, dentry);
3143 if (IS_ERR(inode))
3144 return ERR_CAST(inode);
7b128766 3145
39279cc3
CM		 3146 return d_splice_alias(inode, dentry);
3147}
3148
39279cc3
CM		 3149static unsigned char btrfs_filetype_table[] = {
3150 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3151};
3152
cbdf5a24
DW		 3153static int btrfs_real_readdir(struct file *filp, void *dirent,
3154 filldir_t filldir)
39279cc3 3155{
6da6abae 3156 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3157 struct btrfs_root *root = BTRFS_I(inode)->root;
3158 struct btrfs_item *item;
3159 struct btrfs_dir_item *di;
3160 struct btrfs_key key;
5f39d397 3161 struct btrfs_key found_key;
39279cc3
CM		 3162 struct btrfs_path *path;
3163 int ret;
3164 u32 nritems;
5f39d397 3165 struct extent_buffer *leaf;
39279cc3
CM		 3166 int slot;
3167 int advance;
3168 unsigned char d_type;
3169 int over = 0;
3170 u32 di_cur;
3171 u32 di_total;
3172 u32 di_len;
3173 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3174 char tmp_name[32];
3175 char *name_ptr;
3176 int name_len;
39279cc3
CM		 3177
3178 /* FIXME, use a real flag for deciding about the key type */
3179 if (root->fs_info->tree_root == root)
3180 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3181
3954401f
CM		 3182 /* special case for "." */
3183 if (filp->f_pos == 0) {
3184 over = filldir(dirent, ".", 1,
3185 1, inode->i_ino,
3186 DT_DIR);
3187 if (over)
3188 return 0;
3189 filp->f_pos = 1;
3190 }
3954401f
CM		 3191 /* special case for .., just use the back ref */
3192 if (filp->f_pos == 1) {
5ecc7e5d 3193 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3194 over = filldir(dirent, "..", 2,
5ecc7e5d 3195 2, pino, DT_DIR);
3954401f 3196 if (over)
49593bfa 3197 return 0;
3954401f
CM		 3198 filp->f_pos = 2;
3199 }
49593bfa
DW		 3200 path = btrfs_alloc_path();
3201 path->reada = 2;
3202
39279cc3
CM		 3203 btrfs_set_key_type(&key, key_type);
3204 key.offset = filp->f_pos;
49593bfa 3205 key.objectid = inode->i_ino;
5f39d397 3206
39279cc3
CM		 3207 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3208 if (ret < 0)
3209 goto err;
3210 advance = 0;
49593bfa
DW		 3211
3212 while (1) {
5f39d397
CM		 3213 leaf = path->nodes[0];
3214 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3215 slot = path->slots[0];
3216 if (advance || slot >= nritems) {
49593bfa 3217 if (slot >= nritems - 1) {
39279cc3
CM		 3218 ret = btrfs_next_leaf(root, path);
3219 if (ret)
3220 break;
5f39d397
CM		 3221 leaf = path->nodes[0];
3222 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3223 slot = path->slots[0];
3224 } else {
3225 slot++;
3226 path->slots[0]++;
3227 }
3228 }
3de4586c 3229
39279cc3 3230 advance = 1;
5f39d397
CM		 3231 item = btrfs_item_nr(leaf, slot);
3232 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3233
3234 if (found_key.objectid != key.objectid)
39279cc3 3235 break;
5f39d397 3236 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3237 break;
5f39d397 3238 if (found_key.offset < filp->f_pos)
39279cc3 3239 continue;
5f39d397
CM		 3240
3241 filp->f_pos = found_key.offset;
49593bfa 3242
39279cc3
CM		 3243 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3244 di_cur = 0;
5f39d397 3245 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3246
3247 while (di_cur < di_total) {
5f39d397
CM		 3248 struct btrfs_key location;
3249
3250 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3251 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3252 name_ptr = tmp_name;
3253 } else {
3254 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3255 if (!name_ptr) {
3256 ret = -ENOMEM;
3257 goto err;
3258 }
5f39d397
CM		 3259 }
3260 read_extent_buffer(leaf, name_ptr,
3261 (unsigned long)(di + 1), name_len);
3262
3263 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3264 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3265
3266 /* is this a reference to our own snapshot? If so
3267 * skip it
3268 */
3269 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3270 location.objectid == root->root_key.objectid) {
3271 over = 0;
3272 goto skip;
3273 }
5f39d397 3274 over = filldir(dirent, name_ptr, name_len,
49593bfa 3275 found_key.offset, location.objectid,
39279cc3 3276 d_type);
5f39d397 3277
3de4586c 3278skip:
5f39d397
CM		 3279 if (name_ptr != tmp_name)
3280 kfree(name_ptr);
3281
39279cc3
CM		 3282 if (over)
3283 goto nopos;
5103e947 3284 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3285 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3286 di_cur += di_len;
3287 di = (struct btrfs_dir_item *)((char *)di + di_len);
3288 }
3289 }
49593bfa
DW		 3290
3291 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 3292 if (key_type == BTRFS_DIR_INDEX_KEY)
89f135d8 3293 filp->f_pos = INT_LIMIT(off_t);
5e591a07
YZ		 3294 else
3295 filp->f_pos++;
39279cc3
CM		 3296nopos:
3297 ret = 0;
3298err:
39279cc3 3299 btrfs_free_path(path);
39279cc3
CM		 3300 return ret;
3301}
3302
3303int btrfs_write_inode(struct inode *inode, int wait)
3304{
3305 struct btrfs_root *root = BTRFS_I(inode)->root;
3306 struct btrfs_trans_handle *trans;
3307 int ret = 0;
3308
c146afad 3309 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3310 return 0;
3311
39279cc3 3312 if (wait) {
f9295749 3313 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3314 btrfs_set_trans_block_group(trans, inode);
3315 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3316 }
3317 return ret;
3318}
3319
3320/*
54aa1f4d 3321 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3322 * inode changes. But, it is most likely to find the inode in cache.
3323 * FIXME, needs more benchmarking...there are no reasons other than performance
3324 * to keep or drop this code.
3325 */
3326void btrfs_dirty_inode(struct inode *inode)
3327{
3328 struct btrfs_root *root = BTRFS_I(inode)->root;
3329 struct btrfs_trans_handle *trans;
3330
f9295749 3331 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3332 btrfs_set_trans_block_group(trans, inode);
3333 btrfs_update_inode(trans, root, inode);
3334 btrfs_end_transaction(trans, root);
39279cc3
CM		 3335}
3336
d352ac68
CM		 3337/*
3338 * find the highest existing sequence number in a directory
3339 * and then set the in-memory index_cnt variable to reflect
3340 * free sequence numbers
3341 */
aec7477b
JB		 3342static int btrfs_set_inode_index_count(struct inode *inode)
3343{
3344 struct btrfs_root *root = BTRFS_I(inode)->root;
3345 struct btrfs_key key, found_key;
3346 struct btrfs_path *path;
3347 struct extent_buffer *leaf;
3348 int ret;
3349
3350 key.objectid = inode->i_ino;
3351 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3352 key.offset = (u64)-1;
3353
3354 path = btrfs_alloc_path();
3355 if (!path)
3356 return -ENOMEM;
3357
3358 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3359 if (ret < 0)
3360 goto out;
3361 /* FIXME: we should be able to handle this */
3362 if (ret == 0)
3363 goto out;
3364 ret = 0;
3365
3366 /*
3367 * MAGIC NUMBER EXPLANATION:
3368 * since we search a directory based on f_pos we have to start at 2
3369 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3370 * else has to start at 2
3371 */
3372 if (path->slots[0] == 0) {
3373 BTRFS_I(inode)->index_cnt = 2;
3374 goto out;
3375 }
3376
3377 path->slots[0]--;
3378
3379 leaf = path->nodes[0];
3380 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3381
3382 if (found_key.objectid != inode->i_ino ||
3383 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3384 BTRFS_I(inode)->index_cnt = 2;
3385 goto out;
3386 }
3387
3388 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3389out:
3390 btrfs_free_path(path);
3391 return ret;
3392}
3393
d352ac68
CM		 3394/*
3395 * helper to find a free sequence number in a given directory. This current
3396 * code is very simple, later versions will do smarter things in the btree
3397 */
3de4586c 3398int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3399{
3400 int ret = 0;
3401
3402 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3403 ret = btrfs_set_inode_index_count(dir);
d397712b 3404 if (ret)
aec7477b
JB		 3405 return ret;
3406 }
3407
00e4e6b3 3408 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3409 BTRFS_I(dir)->index_cnt++;
3410
3411 return ret;
3412}
3413
39279cc3
CM		 3414static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3415 struct btrfs_root *root,
aec7477b 3416 struct inode *dir,
9c58309d 3417 const char *name, int name_len,
d2fb3437
YZ		 3418 u64 ref_objectid, u64 objectid,
3419 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3420{
3421 struct inode *inode;
5f39d397 3422 struct btrfs_inode_item *inode_item;
39279cc3 3423 struct btrfs_key *location;
5f39d397 3424 struct btrfs_path *path;
9c58309d
CM		 3425 struct btrfs_inode_ref *ref;
3426 struct btrfs_key key[2];
3427 u32 sizes[2];
3428 unsigned long ptr;
39279cc3
CM		 3429 int ret;
3430 int owner;
3431
5f39d397
CM		 3432 path = btrfs_alloc_path();
3433 BUG_ON(!path);
3434
39279cc3
CM		 3435 inode = new_inode(root->fs_info->sb);
3436 if (!inode)
3437 return ERR_PTR(-ENOMEM);
3438
aec7477b 3439 if (dir) {
3de4586c 3440 ret = btrfs_set_inode_index(dir, index);
aec7477b
JB		 3441 if (ret)
3442 return ERR_PTR(ret);
aec7477b
JB		 3443 }
3444 /*
3445 * index_cnt is ignored for everything but a dir,
3446 * btrfs_get_inode_index_count has an explanation for the magic
3447 * number
3448 */
e02119d5 3449 init_btrfs_i(inode);
aec7477b 3450 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3451 BTRFS_I(inode)->root = root;
e02119d5 3452 BTRFS_I(inode)->generation = trans->transid;
6a63209f 3453 btrfs_set_inode_space_info(root, inode);
b888db2b 3454
39279cc3
CM		 3455 if (mode & S_IFDIR)
3456 owner = 0;
3457 else
3458 owner = 1;
d2fb3437
YZ		 3459 BTRFS_I(inode)->block_group =
3460 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ		 3461 if ((mode & S_IFREG)) {
3462 if (btrfs_test_opt(root, NODATASUM))
3463 btrfs_set_flag(inode, NODATASUM);
3464 if (btrfs_test_opt(root, NODATACOW))
3465 btrfs_set_flag(inode, NODATACOW);
3466 }
9c58309d
CM		 3467
3468 key[0].objectid = objectid;
3469 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3470 key[0].offset = 0;
3471
3472 key[1].objectid = objectid;
3473 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3474 key[1].offset = ref_objectid;
3475
3476 sizes[0] = sizeof(struct btrfs_inode_item);
3477 sizes[1] = name_len + sizeof(*ref);
3478
b9473439 3479 path->leave_spinning = 1;
9c58309d
CM		 3480 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3481 if (ret != 0)
5f39d397
CM		 3482 goto fail;
3483
9c58309d
CM		 3484 if (objectid > root->highest_inode)
3485 root->highest_inode = objectid;
3486
79683f2d 3487 inode->i_uid = current_fsuid();
8c087b51 3488
42f15d77 3489 if (dir && (dir->i_mode & S_ISGID)) {
8c087b51
CB		 3490 inode->i_gid = dir->i_gid;
3491 if (S_ISDIR(mode))
3492 mode |= S_ISGID;
3493 } else
3494 inode->i_gid = current_fsgid();
3495
39279cc3
CM		 3496 inode->i_mode = mode;
3497 inode->i_ino = objectid;
a76a3cd4 3498 inode_set_bytes(inode, 0);
39279cc3 3499 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3500 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3501 struct btrfs_inode_item);
e02119d5 3502 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3503
3504 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3505 struct btrfs_inode_ref);
3506 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3507 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3508 ptr = (unsigned long)(ref + 1);
3509 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3510
5f39d397
CM		 3511 btrfs_mark_buffer_dirty(path->nodes[0]);
3512 btrfs_free_path(path);
3513
39279cc3
CM		 3514 location = &BTRFS_I(inode)->location;
3515 location->objectid = objectid;
39279cc3
CM		 3516 location->offset = 0;
3517 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3518
39279cc3
CM		 3519 insert_inode_hash(inode);
3520 return inode;
5f39d397 3521fail:
aec7477b
JB		 3522 if (dir)
3523 BTRFS_I(dir)->index_cnt--;
5f39d397
CM		 3524 btrfs_free_path(path);
3525 return ERR_PTR(ret);
39279cc3
CM		 3526}
3527
3528static inline u8 btrfs_inode_type(struct inode *inode)
3529{
3530 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3531}
3532
d352ac68
CM		 3533/*
3534 * utility function to add 'inode' into 'parent_inode' with
3535 * a give name and a given sequence number.
3536 * if 'add_backref' is true, also insert a backref from the
3537 * inode to the parent directory.
3538 */
e02119d5
CM		 3539int btrfs_add_link(struct btrfs_trans_handle *trans,
3540 struct inode *parent_inode, struct inode *inode,
3541 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3542{
3543 int ret;
3544 struct btrfs_key key;
e02119d5 3545 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3546
39279cc3 3547 key.objectid = inode->i_ino;
39279cc3
CM		 3548 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3549 key.offset = 0;
3550
e02119d5
CM		 3551 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3552 parent_inode->i_ino,
aec7477b 3553 &key, btrfs_inode_type(inode),
00e4e6b3 3554 index);
39279cc3 3555 if (ret == 0) {
9c58309d
CM		 3556 if (add_backref) {
3557 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3558 name, name_len,
3559 inode->i_ino,
3560 parent_inode->i_ino,
3561 index);
9c58309d 3562 }
dbe674a9 3563 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3564 name_len * 2);
79c44584 3565 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3566 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3567 }
3568 return ret;
3569}
3570
3571static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3572 struct dentry *dentry, struct inode *inode,
00e4e6b3 3573 int backref, u64 index)
39279cc3 3574{
e02119d5
CM		 3575 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3576 inode, dentry->d_name.name,
3577 dentry->d_name.len, backref, index);
39279cc3
CM		 3578 if (!err) {
3579 d_instantiate(dentry, inode);
3580 return 0;
3581 }
3582 if (err > 0)
3583 err = -EEXIST;
3584 return err;
3585}
3586
618e21d5
JB		 3587static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3588 int mode, dev_t rdev)
3589{
3590 struct btrfs_trans_handle *trans;
3591 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3592 struct inode *inode = NULL;
618e21d5
JB		 3593 int err;
3594 int drop_inode = 0;
3595 u64 objectid;
1832a6d5 3596 unsigned long nr = 0;
00e4e6b3 3597 u64 index = 0;
618e21d5
JB		 3598
3599 if (!new_valid_dev(rdev))
3600 return -EINVAL;
3601
6a63209f 3602 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3603 if (err)
3604 goto fail;
3605
618e21d5
JB		 3606 trans = btrfs_start_transaction(root, 1);
3607 btrfs_set_trans_block_group(trans, dir);
3608
3609 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3610 if (err) {
3611 err = -ENOSPC;
3612 goto out_unlock;
3613 }
3614
aec7477b 3615 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3616 dentry->d_name.len,
3617 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3618 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3619 err = PTR_ERR(inode);
3620 if (IS_ERR(inode))
3621 goto out_unlock;
3622
0279b4cd 3623 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3624 if (err) {
3625 drop_inode = 1;
3626 goto out_unlock;
3627 }
3628
618e21d5 3629 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3630 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3631 if (err)
3632 drop_inode = 1;
3633 else {
3634 inode->i_op = &btrfs_special_inode_operations;
3635 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3636 btrfs_update_inode(trans, root, inode);
618e21d5
JB		 3637 }
3638 dir->i_sb->s_dirt = 1;
3639 btrfs_update_inode_block_group(trans, inode);
3640 btrfs_update_inode_block_group(trans, dir);
3641out_unlock:
d3c2fdcf 3642 nr = trans->blocks_used;
89ce8a63 3643 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3644fail:
618e21d5
JB		 3645 if (drop_inode) {
3646 inode_dec_link_count(inode);
3647 iput(inode);
3648 }
d3c2fdcf 3649 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3650 return err;
3651}
3652
39279cc3
CM		 3653static int btrfs_create(struct inode *dir, struct dentry *dentry,
3654 int mode, struct nameidata *nd)
3655{
3656 struct btrfs_trans_handle *trans;
3657 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3658 struct inode *inode = NULL;
39279cc3
CM		 3659 int err;
3660 int drop_inode = 0;
1832a6d5 3661 unsigned long nr = 0;
39279cc3 3662 u64 objectid;
00e4e6b3 3663 u64 index = 0;
39279cc3 3664
6a63209f 3665 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3666 if (err)
3667 goto fail;
39279cc3
CM		 3668 trans = btrfs_start_transaction(root, 1);
3669 btrfs_set_trans_block_group(trans, dir);
3670
3671 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3672 if (err) {
3673 err = -ENOSPC;
3674 goto out_unlock;
3675 }
3676
aec7477b 3677 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3678 dentry->d_name.len,
3679 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3680 objectid, BTRFS_I(dir)->block_group, mode,
3681 &index);
39279cc3
CM		 3682 err = PTR_ERR(inode);
3683 if (IS_ERR(inode))
3684 goto out_unlock;
3685
0279b4cd 3686 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3687 if (err) {
3688 drop_inode = 1;
3689 goto out_unlock;
3690 }
3691
39279cc3 3692 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3693 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3694 if (err)
3695 drop_inode = 1;
3696 else {
3697 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3698 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3699 inode->i_fop = &btrfs_file_operations;
3700 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3701 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3702 }
3703 dir->i_sb->s_dirt = 1;
3704 btrfs_update_inode_block_group(trans, inode);
3705 btrfs_update_inode_block_group(trans, dir);
3706out_unlock:
d3c2fdcf 3707 nr = trans->blocks_used;
ab78c84d 3708 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3709fail:
39279cc3
CM		 3710 if (drop_inode) {
3711 inode_dec_link_count(inode);
3712 iput(inode);
3713 }
d3c2fdcf 3714 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3715 return err;
3716}
3717
3718static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3719 struct dentry *dentry)
3720{
3721 struct btrfs_trans_handle *trans;
3722 struct btrfs_root *root = BTRFS_I(dir)->root;
3723 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3724 u64 index;
1832a6d5 3725 unsigned long nr = 0;
39279cc3
CM		 3726 int err;
3727 int drop_inode = 0;
3728
3729 if (inode->i_nlink == 0)
3730 return -ENOENT;
3731
e02119d5 3732 btrfs_inc_nlink(inode);
6a63209f 3733 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3734 if (err)
3735 goto fail;
3de4586c 3736 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3737 if (err)
3738 goto fail;
3739
39279cc3 3740 trans = btrfs_start_transaction(root, 1);
5f39d397 3741
39279cc3
CM		 3742 btrfs_set_trans_block_group(trans, dir);
3743 atomic_inc(&inode->i_count);
aec7477b 3744
00e4e6b3 3745 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3746
39279cc3
CM		 3747 if (err)
3748 drop_inode = 1;
5f39d397 3749
39279cc3
CM		 3750 dir->i_sb->s_dirt = 1;
3751 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3752 err = btrfs_update_inode(trans, root, inode);
5f39d397 3753
54aa1f4d
CM		 3754 if (err)
3755 drop_inode = 1;
39279cc3 3756
d3c2fdcf 3757 nr = trans->blocks_used;
ab78c84d 3758 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3759fail:
39279cc3
CM		 3760 if (drop_inode) {
3761 inode_dec_link_count(inode);
3762 iput(inode);
3763 }
d3c2fdcf 3764 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3765 return err;
3766}
3767
39279cc3
CM		 3768static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3769{
b9d86667 3770 struct inode *inode = NULL;
39279cc3
CM		 3771 struct btrfs_trans_handle *trans;
3772 struct btrfs_root *root = BTRFS_I(dir)->root;
3773 int err = 0;
3774 int drop_on_err = 0;
b9d86667 3775 u64 objectid = 0;
00e4e6b3 3776 u64 index = 0;
d3c2fdcf 3777 unsigned long nr = 1;
39279cc3 3778
6a63209f 3779 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3780 if (err)
3781 goto out_unlock;
3782
39279cc3
CM		 3783 trans = btrfs_start_transaction(root, 1);
3784 btrfs_set_trans_block_group(trans, dir);
5f39d397 3785
39279cc3
CM		 3786 if (IS_ERR(trans)) {
3787 err = PTR_ERR(trans);
3788 goto out_unlock;
3789 }
3790
3791 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3792 if (err) {
3793 err = -ENOSPC;
3794 goto out_unlock;
3795 }
3796
aec7477b 3797 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3798 dentry->d_name.len,
3799 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3800 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3801 &index);
39279cc3
CM		 3802 if (IS_ERR(inode)) {
3803 err = PTR_ERR(inode);
3804 goto out_fail;
3805 }
5f39d397 3806
39279cc3 3807 drop_on_err = 1;
33268eaf 3808
0279b4cd 3809 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3810 if (err)
3811 goto out_fail;
3812
39279cc3
CM		 3813 inode->i_op = &btrfs_dir_inode_operations;
3814 inode->i_fop = &btrfs_dir_file_operations;
3815 btrfs_set_trans_block_group(trans, inode);
3816
dbe674a9 3817 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3818 err = btrfs_update_inode(trans, root, inode);
3819 if (err)
3820 goto out_fail;
5f39d397 3821
e02119d5
CM		 3822 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3823 inode, dentry->d_name.name,
3824 dentry->d_name.len, 0, index);
39279cc3
CM		 3825 if (err)
3826 goto out_fail;
5f39d397 3827
39279cc3
CM		 3828 d_instantiate(dentry, inode);
3829 drop_on_err = 0;
3830 dir->i_sb->s_dirt = 1;
3831 btrfs_update_inode_block_group(trans, inode);
3832 btrfs_update_inode_block_group(trans, dir);
3833
3834out_fail:
d3c2fdcf 3835 nr = trans->blocks_used;
ab78c84d 3836 btrfs_end_transaction_throttle(trans, root);
5f39d397 3837
39279cc3 3838out_unlock:
39279cc3
CM		 3839 if (drop_on_err)
3840 iput(inode);
d3c2fdcf 3841 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3842 return err;
3843}
3844
d352ac68
CM		 3845/* helper for btfs_get_extent. Given an existing extent in the tree,
3846 * and an extent that you want to insert, deal with overlap and insert
3847 * the new extent into the tree.
3848 */
3b951516
CM		 3849static int merge_extent_mapping(struct extent_map_tree *em_tree,
3850 struct extent_map *existing,
e6dcd2dc
CM		 3851 struct extent_map *em,
3852 u64 map_start, u64 map_len)
3b951516
CM		 3853{
3854 u64 start_diff;
3b951516 3855
e6dcd2dc
CM		 3856 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3857 start_diff = map_start - em->start;
3858 em->start = map_start;
3859 em->len = map_len;
c8b97818
CM		 3860 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3861 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3862 em->block_start += start_diff;
c8b97818
CM		 3863 em->block_len -= start_diff;
3864 }
e6dcd2dc 3865 return add_extent_mapping(em_tree, em);
3b951516
CM		 3866}
3867
c8b97818
CM		 3868static noinline int uncompress_inline(struct btrfs_path *path,
3869 struct inode *inode, struct page *page,
3870 size_t pg_offset, u64 extent_offset,
3871 struct btrfs_file_extent_item *item)
3872{
3873 int ret;
3874 struct extent_buffer *leaf = path->nodes[0];
3875 char *tmp;
3876 size_t max_size;
3877 unsigned long inline_size;
3878 unsigned long ptr;
3879
3880 WARN_ON(pg_offset != 0);
3881 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3882 inline_size = btrfs_file_extent_inline_item_len(leaf,
3883 btrfs_item_nr(leaf, path->slots[0]));
3884 tmp = kmalloc(inline_size, GFP_NOFS);
3885 ptr = btrfs_file_extent_inline_start(item);
3886
3887 read_extent_buffer(leaf, tmp, ptr, inline_size);
3888
5b050f04 3889 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 3890 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3891 inline_size, max_size);
3892 if (ret) {
3893 char *kaddr = kmap_atomic(page, KM_USER0);
3894 unsigned long copy_size = min_t(u64,
3895 PAGE_CACHE_SIZE - pg_offset,
3896 max_size - extent_offset);
3897 memset(kaddr + pg_offset, 0, copy_size);
3898 kunmap_atomic(kaddr, KM_USER0);
3899 }
3900 kfree(tmp);
3901 return 0;
3902}
3903
d352ac68
CM		 3904/*
3905 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 3906 * the ugly parts come from merging extents from the disk with the in-ram
3907 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 3908 * where the in-ram extents might be locked pending data=ordered completion.
3909 *
3910 * This also copies inline extents directly into the page.
3911 */
d397712b 3912
a52d9a80 3913struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 3914 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 3915 int create)
3916{
3917 int ret;
3918 int err = 0;
db94535d 3919 u64 bytenr;
a52d9a80
CM		 3920 u64 extent_start = 0;
3921 u64 extent_end = 0;
3922 u64 objectid = inode->i_ino;
3923 u32 found_type;
f421950f 3924 struct btrfs_path *path = NULL;
a52d9a80
CM		 3925 struct btrfs_root *root = BTRFS_I(inode)->root;
3926 struct btrfs_file_extent_item *item;
5f39d397
CM		 3927 struct extent_buffer *leaf;
3928 struct btrfs_key found_key;
a52d9a80
CM		 3929 struct extent_map *em = NULL;
3930 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 3931 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 3932 struct btrfs_trans_handle *trans = NULL;
c8b97818 3933 int compressed;
a52d9a80 3934
a52d9a80 3935again:
d1310b2e
CM		 3936 spin_lock(&em_tree->lock);
3937 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 3938 if (em)
3939 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 3940 spin_unlock(&em_tree->lock);
3941
a52d9a80 3942 if (em) {
e1c4b745
CM		 3943 if (em->start > start || em->start + em->len <= start)
3944 free_extent_map(em);
3945 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 3946 free_extent_map(em);
3947 else
3948 goto out;
a52d9a80 3949 }
d1310b2e 3950 em = alloc_extent_map(GFP_NOFS);
a52d9a80 3951 if (!em) {
d1310b2e
CM		 3952 err = -ENOMEM;
3953 goto out;
a52d9a80 3954 }
e6dcd2dc 3955 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 3956 em->start = EXTENT_MAP_HOLE;
445a6944 3957 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 3958 em->len = (u64)-1;
c8b97818 3959 em->block_len = (u64)-1;
f421950f
CM		 3960
3961 if (!path) {
3962 path = btrfs_alloc_path();
3963 BUG_ON(!path);
3964 }
3965
179e29e4
CM		 3966 ret = btrfs_lookup_file_extent(trans, root, path,
3967 objectid, start, trans != NULL);
a52d9a80
CM		 3968 if (ret < 0) {
3969 err = ret;
3970 goto out;
3971 }
3972
3973 if (ret != 0) {
3974 if (path->slots[0] == 0)
3975 goto not_found;
3976 path->slots[0]--;
3977 }
3978
5f39d397
CM		 3979 leaf = path->nodes[0];
3980 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 3981 struct btrfs_file_extent_item);
a52d9a80 3982 /* are we inside the extent that was found? */
5f39d397
CM		 3983 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3984 found_type = btrfs_key_type(&found_key);
3985 if (found_key.objectid != objectid ||
a52d9a80
CM		 3986 found_type != BTRFS_EXTENT_DATA_KEY) {
3987 goto not_found;
3988 }
3989
5f39d397
CM		 3990 found_type = btrfs_file_extent_type(leaf, item);
3991 extent_start = found_key.offset;
c8b97818 3992 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 3993 if (found_type == BTRFS_FILE_EXTENT_REG ||
3994 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 3995 extent_end = extent_start +
db94535d 3996 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 3997 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3998 size_t size;
3999 size = btrfs_file_extent_inline_len(leaf, item);
4000 extent_end = (extent_start + size + root->sectorsize - 1) &
4001 ~((u64)root->sectorsize - 1);
4002 }
4003
4004 if (start >= extent_end) {
4005 path->slots[0]++;
4006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4007 ret = btrfs_next_leaf(root, path);
4008 if (ret < 0) {
4009 err = ret;
4010 goto out;
a52d9a80 4011 }
9036c102
YZ		 4012 if (ret > 0)
4013 goto not_found;
4014 leaf = path->nodes[0];
a52d9a80 4015 }
9036c102
YZ		 4016 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4017 if (found_key.objectid != objectid ||
4018 found_key.type != BTRFS_EXTENT_DATA_KEY)
4019 goto not_found;
4020 if (start + len <= found_key.offset)
4021 goto not_found;
4022 em->start = start;
4023 em->len = found_key.offset - start;
4024 goto not_found_em;
4025 }
4026
d899e052
YZ		 4027 if (found_type == BTRFS_FILE_EXTENT_REG ||
4028 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 4029 em->start = extent_start;
4030 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 4031 em->orig_start = extent_start -
4032 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 4033 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4034 if (bytenr == 0) {
5f39d397 4035 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4036 goto insert;
4037 }
c8b97818
CM		 4038 if (compressed) {
4039 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4040 em->block_start = bytenr;
4041 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4042 item);
4043 } else {
4044 bytenr += btrfs_file_extent_offset(leaf, item);
4045 em->block_start = bytenr;
4046 em->block_len = em->len;
d899e052
YZ		 4047 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4048 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4049 }
a52d9a80
CM		 4050 goto insert;
4051 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4052 unsigned long ptr;
a52d9a80 4053 char *map;
3326d1b0
CM		 4054 size_t size;
4055 size_t extent_offset;
4056 size_t copy_size;
a52d9a80 4057
689f9346 4058 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4059 if (!page || create) {
689f9346 4060 em->start = extent_start;
9036c102 4061 em->len = extent_end - extent_start;
689f9346
Y		 4062 goto out;
4063 }
5f39d397 4064
9036c102
YZ		 4065 size = btrfs_file_extent_inline_len(leaf, item);
4066 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4067 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4068 size - extent_offset);
3326d1b0 4069 em->start = extent_start + extent_offset;
70dec807
CM		 4070 em->len = (copy_size + root->sectorsize - 1) &
4071 ~((u64)root->sectorsize - 1);
ff5b7ee3 4072 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 4073 if (compressed)
4074 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4075 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4076 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 4077 if (btrfs_file_extent_compression(leaf, item) ==
4078 BTRFS_COMPRESS_ZLIB) {
4079 ret = uncompress_inline(path, inode, page,
4080 pg_offset,
4081 extent_offset, item);
4082 BUG_ON(ret);
4083 } else {
4084 map = kmap(page);
4085 read_extent_buffer(leaf, map + pg_offset, ptr,
4086 copy_size);
4087 kunmap(page);
4088 }
179e29e4
CM		 4089 flush_dcache_page(page);
4090 } else if (create && PageUptodate(page)) {
4091 if (!trans) {
4092 kunmap(page);
4093 free_extent_map(em);
4094 em = NULL;
4095 btrfs_release_path(root, path);
f9295749 4096 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4097 goto again;
4098 }
c8b97818 4099 map = kmap(page);
70dec807 4100 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4101 copy_size);
c8b97818 4102 kunmap(page);
179e29e4 4103 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4104 }
d1310b2e
CM		 4105 set_extent_uptodate(io_tree, em->start,
4106 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4107 goto insert;
4108 } else {
d397712b 4109 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 4110 WARN_ON(1);
4111 }
4112not_found:
4113 em->start = start;
d1310b2e 4114 em->len = len;
a52d9a80 4115not_found_em:
5f39d397 4116 em->block_start = EXTENT_MAP_HOLE;
9036c102 4117 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4118insert:
4119 btrfs_release_path(root, path);
d1310b2e 4120 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 4121 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4122 "[%llu %llu]\n", (unsigned long long)em->start,
4123 (unsigned long long)em->len,
4124 (unsigned long long)start,
4125 (unsigned long long)len);
a52d9a80
CM		 4126 err = -EIO;
4127 goto out;
4128 }
d1310b2e
CM		 4129
4130 err = 0;
4131 spin_lock(&em_tree->lock);
a52d9a80 4132 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4133 /* it is possible that someone inserted the extent into the tree
4134 * while we had the lock dropped. It is also possible that
4135 * an overlapping map exists in the tree
4136 */
a52d9a80 4137 if (ret == -EEXIST) {
3b951516 4138 struct extent_map *existing;
e6dcd2dc
CM		 4139
4140 ret = 0;
4141
3b951516 4142 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4143 if (existing && (existing->start > start ||
4144 existing->start + existing->len <= start)) {
4145 free_extent_map(existing);
4146 existing = NULL;
4147 }
3b951516
CM		 4148 if (!existing) {
4149 existing = lookup_extent_mapping(em_tree, em->start,
4150 em->len);
4151 if (existing) {
4152 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4153 em, start,
4154 root->sectorsize);
3b951516
CM		 4155 free_extent_map(existing);
4156 if (err) {
4157 free_extent_map(em);
4158 em = NULL;
4159 }
4160 } else {
4161 err = -EIO;
3b951516
CM		 4162 free_extent_map(em);
4163 em = NULL;
4164 }
4165 } else {
4166 free_extent_map(em);
4167 em = existing;
e6dcd2dc 4168 err = 0;
a52d9a80 4169 }
a52d9a80 4170 }
d1310b2e 4171 spin_unlock(&em_tree->lock);
a52d9a80 4172out:
f421950f
CM		 4173 if (path)
4174 btrfs_free_path(path);
a52d9a80
CM		 4175 if (trans) {
4176 ret = btrfs_end_transaction(trans, root);
d397712b 4177 if (!err)
a52d9a80
CM		 4178 err = ret;
4179 }
a52d9a80
CM		 4180 if (err) {
4181 free_extent_map(em);
4182 WARN_ON(1);
4183 return ERR_PTR(err);
4184 }
4185 return em;
4186}
4187
16432985
CM		 4188static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4189 const struct iovec *iov, loff_t offset,
4190 unsigned long nr_segs)
4191{
e1c4b745 4192 return -EINVAL;
16432985
CM		 4193}
4194
1506fcc8
YS		 4195static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4196 __u64 start, __u64 len)
4197{
4198 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4199}
4200
a52d9a80 4201int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4202{
d1310b2e
CM		 4203 struct extent_io_tree *tree;
4204 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4205 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4206}
1832a6d5 4207
a52d9a80 4208static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4209{
d1310b2e 4210 struct extent_io_tree *tree;
b888db2b
CM		 4211
4212
4213 if (current->flags & PF_MEMALLOC) {
4214 redirty_page_for_writepage(wbc, page);
4215 unlock_page(page);
4216 return 0;
4217 }
d1310b2e 4218 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4219 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4220}
4221
f421950f
CM		 4222int btrfs_writepages(struct address_space *mapping,
4223 struct writeback_control *wbc)
b293f02e 4224{
d1310b2e 4225 struct extent_io_tree *tree;
771ed689 4226
d1310b2e 4227 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4228 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4229}
4230
3ab2fb5a
CM		 4231static int
4232btrfs_readpages(struct file *file, struct address_space *mapping,
4233 struct list_head *pages, unsigned nr_pages)
4234{
d1310b2e
CM		 4235 struct extent_io_tree *tree;
4236 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4237 return extent_readpages(tree, mapping, pages, nr_pages,
4238 btrfs_get_extent);
4239}
e6dcd2dc 4240static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4241{
d1310b2e
CM		 4242 struct extent_io_tree *tree;
4243 struct extent_map_tree *map;
a52d9a80 4244 int ret;
8c2383c3 4245
d1310b2e
CM		 4246 tree = &BTRFS_I(page->mapping->host)->io_tree;
4247 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4248 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4249 if (ret == 1) {
4250 ClearPagePrivate(page);
4251 set_page_private(page, 0);
4252 page_cache_release(page);
39279cc3 4253 }
a52d9a80 4254 return ret;
39279cc3
CM		 4255}
4256
e6dcd2dc
CM		 4257static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4258{
98509cfc
CM		 4259 if (PageWriteback(page) || PageDirty(page))
4260 return 0;
b335b003 4261 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 4262}
4263
a52d9a80 4264static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4265{
d1310b2e 4266 struct extent_io_tree *tree;
e6dcd2dc
CM		 4267 struct btrfs_ordered_extent *ordered;
4268 u64 page_start = page_offset(page);
4269 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4270
e6dcd2dc 4271 wait_on_page_writeback(page);
d1310b2e 4272 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4273 if (offset) {
4274 btrfs_releasepage(page, GFP_NOFS);
4275 return;
4276 }
4277
4278 lock_extent(tree, page_start, page_end, GFP_NOFS);
4279 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4280 page_offset(page));
4281 if (ordered) {
eb84ae03
CM		 4282 /*
4283 * IO on this page will never be started, so we need
4284 * to account for any ordered extents now
4285 */
e6dcd2dc
CM		 4286 clear_extent_bit(tree, page_start, page_end,
4287 EXTENT_DIRTY | EXTENT_DELALLOC |
4288 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4289 btrfs_finish_ordered_io(page->mapping->host,
4290 page_start, page_end);
e6dcd2dc
CM		 4291 btrfs_put_ordered_extent(ordered);
4292 lock_extent(tree, page_start, page_end, GFP_NOFS);
4293 }
4294 clear_extent_bit(tree, page_start, page_end,
4295 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4296 EXTENT_ORDERED,
4297 1, 1, GFP_NOFS);
4298 __btrfs_releasepage(page, GFP_NOFS);
4299
4a096752 4300 ClearPageChecked(page);
9ad6b7bc 4301 if (PagePrivate(page)) {
9ad6b7bc
CM		 4302 ClearPagePrivate(page);
4303 set_page_private(page, 0);
4304 page_cache_release(page);
4305 }
39279cc3
CM		 4306}
4307
9ebefb18
CM		 4308/*
4309 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4310 * called from a page fault handler when a page is first dirtied. Hence we must
4311 * be careful to check for EOF conditions here. We set the page up correctly
4312 * for a written page which means we get ENOSPC checking when writing into
4313 * holes and correct delalloc and unwritten extent mapping on filesystems that
4314 * support these features.
4315 *
4316 * We are not allowed to take the i_mutex here so we have to play games to
4317 * protect against truncate races as the page could now be beyond EOF. Because
4318 * vmtruncate() writes the inode size before removing pages, once we have the
4319 * page lock we can determine safely if the page is beyond EOF. If it is not
4320 * beyond EOF, then the page is guaranteed safe against truncation until we
4321 * unlock the page.
4322 */
4323int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4324{
6da6abae 4325 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4326 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4327 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4328 struct btrfs_ordered_extent *ordered;
4329 char *kaddr;
4330 unsigned long zero_start;
9ebefb18 4331 loff_t size;
1832a6d5 4332 int ret;
a52d9a80 4333 u64 page_start;
e6dcd2dc 4334 u64 page_end;
9ebefb18 4335
6a63209f 4336 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
1832a6d5
CM		 4337 if (ret)
4338 goto out;
4339
4340 ret = -EINVAL;
e6dcd2dc 4341again:
9ebefb18 4342 lock_page(page);
9ebefb18 4343 size = i_size_read(inode);
e6dcd2dc
CM		 4344 page_start = page_offset(page);
4345 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4346
9ebefb18 4347 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4348 (page_start >= size)) {
6a63209f 4349 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
9ebefb18
CM		 4350 /* page got truncated out from underneath us */
4351 goto out_unlock;
4352 }
e6dcd2dc
CM		 4353 wait_on_page_writeback(page);
4354
4355 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4356 set_page_extent_mapped(page);
4357
eb84ae03
CM		 4358 /*
4359 * we can't set the delalloc bits if there are pending ordered
4360 * extents. Drop our locks and wait for them to finish
4361 */
e6dcd2dc
CM		 4362 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4363 if (ordered) {
4364 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4365 unlock_page(page);
eb84ae03 4366 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4367 btrfs_put_ordered_extent(ordered);
4368 goto again;
4369 }
4370
ea8c2819 4371 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4372 ret = 0;
9ebefb18
CM		 4373
4374 /* page is wholly or partially inside EOF */
a52d9a80 4375 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4376 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4377 else
e6dcd2dc 4378 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4379
e6dcd2dc
CM		 4380 if (zero_start != PAGE_CACHE_SIZE) {
4381 kaddr = kmap(page);
4382 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4383 flush_dcache_page(page);
4384 kunmap(page);
4385 }
247e743c 4386 ClearPageChecked(page);
e6dcd2dc
CM		 4387 set_page_dirty(page);
4388 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4389
4390out_unlock:
4391 unlock_page(page);
1832a6d5 4392out:
9ebefb18
CM		 4393 return ret;
4394}
4395
39279cc3
CM		 4396static void btrfs_truncate(struct inode *inode)
4397{
4398 struct btrfs_root *root = BTRFS_I(inode)->root;
4399 int ret;
4400 struct btrfs_trans_handle *trans;
d3c2fdcf 4401 unsigned long nr;
dbe674a9 4402 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4403
4404 if (!S_ISREG(inode->i_mode))
4405 return;
4406 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4407 return;
4408
4409 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4410 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4411
39279cc3
CM		 4412 trans = btrfs_start_transaction(root, 1);
4413 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4414 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4415
7b128766
JB		 4416 ret = btrfs_orphan_add(trans, inode);
4417 if (ret)
4418 goto out;
39279cc3 4419 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4420 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4421 BTRFS_EXTENT_DATA_KEY);
39279cc3 4422 btrfs_update_inode(trans, root, inode);
5f39d397 4423
7b128766
JB		 4424 ret = btrfs_orphan_del(trans, inode);
4425 BUG_ON(ret);
4426
4427out:
4428 nr = trans->blocks_used;
89ce8a63 4429 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4430 BUG_ON(ret);
d3c2fdcf 4431 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4432}
4433
d352ac68
CM		 4434/*
4435 * create a new subvolume directory/inode (helper for the ioctl).
4436 */
d2fb3437
YZ		 4437int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4438 struct btrfs_root *new_root, struct dentry *dentry,
4439 u64 new_dirid, u64 alloc_hint)
39279cc3 4440{
39279cc3 4441 struct inode *inode;
cb8e7090 4442 int error;
00e4e6b3 4443 u64 index = 0;
39279cc3 4444
aec7477b 4445 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4446 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4447 if (IS_ERR(inode))
f46b5a66 4448 return PTR_ERR(inode);
39279cc3
CM		 4449 inode->i_op = &btrfs_dir_inode_operations;
4450 inode->i_fop = &btrfs_dir_file_operations;
4451
39279cc3 4452 inode->i_nlink = 1;
dbe674a9 4453 btrfs_i_size_write(inode, 0);
3b96362c 4454
cb8e7090
CH		 4455 error = btrfs_update_inode(trans, new_root, inode);
4456 if (error)
4457 return error;
4458
4459 d_instantiate(dentry, inode);
4460 return 0;
39279cc3
CM		 4461}
4462
d352ac68
CM		 4463/* helper function for file defrag and space balancing. This
4464 * forces readahead on a given range of bytes in an inode
4465 */
edbd8d4e 4466unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4467 struct file_ra_state *ra, struct file *file,
4468 pgoff_t offset, pgoff_t last_index)
4469{
8e7bf94f 4470 pgoff_t req_size = last_index - offset + 1;
86479a04 4471
86479a04
CM		 4472 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4473 return offset + req_size;
86479a04
CM		 4474}
4475
39279cc3
CM		 4476struct inode *btrfs_alloc_inode(struct super_block *sb)
4477{
4478 struct btrfs_inode *ei;
4479
4480 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4481 if (!ei)
4482 return NULL;
15ee9bc7 4483 ei->last_trans = 0;
e02119d5 4484 ei->logged_trans = 0;
e6dcd2dc 4485 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4486 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4487 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4488 INIT_LIST_HEAD(&ei->i_orphan);
39279cc3
CM		 4489 return &ei->vfs_inode;
4490}
4491
4492void btrfs_destroy_inode(struct inode *inode)
4493{
e6dcd2dc 4494 struct btrfs_ordered_extent *ordered;
39279cc3
CM		 4495 WARN_ON(!list_empty(&inode->i_dentry));
4496 WARN_ON(inode->i_data.nrpages);
4497
33268eaf
JB		 4498 if (BTRFS_I(inode)->i_acl &&
4499 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4500 posix_acl_release(BTRFS_I(inode)->i_acl);
4501 if (BTRFS_I(inode)->i_default_acl &&
4502 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4503 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4504
bcc63abb 4505 spin_lock(&BTRFS_I(inode)->root->list_lock);
7b128766
JB		 4506 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4507 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4508 " list\n", inode->i_ino);
4509 dump_stack();
4510 }
bcc63abb 4511 spin_unlock(&BTRFS_I(inode)->root->list_lock);
7b128766 4512
d397712b 4513 while (1) {
e6dcd2dc
CM		 4514 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4515 if (!ordered)
4516 break;
4517 else {
d397712b
CM		 4518 printk(KERN_ERR "btrfs found ordered "
4519 "extent %llu %llu on inode cleanup\n",
4520 (unsigned long long)ordered->file_offset,
4521 (unsigned long long)ordered->len);
e6dcd2dc
CM		 4522 btrfs_remove_ordered_extent(inode, ordered);
4523 btrfs_put_ordered_extent(ordered);
4524 btrfs_put_ordered_extent(ordered);
4525 }
4526 }
5b21f2ed 4527 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4528 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4529}
4530
0ee0fda0 4531static void init_once(void *foo)
39279cc3
CM		 4532{
4533 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4534
4535 inode_init_once(&ei->vfs_inode);
4536}
4537
4538void btrfs_destroy_cachep(void)
4539{
4540 if (btrfs_inode_cachep)
4541 kmem_cache_destroy(btrfs_inode_cachep);
4542 if (btrfs_trans_handle_cachep)
4543 kmem_cache_destroy(btrfs_trans_handle_cachep);
4544 if (btrfs_transaction_cachep)
4545 kmem_cache_destroy(btrfs_transaction_cachep);
4546 if (btrfs_bit_radix_cachep)
4547 kmem_cache_destroy(btrfs_bit_radix_cachep);
4548 if (btrfs_path_cachep)
4549 kmem_cache_destroy(btrfs_path_cachep);
4550}
4551
86479a04 4552struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
92fee66d 4553 unsigned long extra_flags,
2b1f55b0 4554 void (*ctor)(void *))
92fee66d
CM		 4555{
4556 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2b1f55b0 4557 SLAB_MEM_SPREAD | extra_flags), ctor);
92fee66d
CM		 4558}
4559
39279cc3
CM		 4560int btrfs_init_cachep(void)
4561{
86479a04 4562 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
92fee66d
CM		 4563 sizeof(struct btrfs_inode),
4564 0, init_once);
39279cc3
CM		 4565 if (!btrfs_inode_cachep)
4566 goto fail;
86479a04
CM		 4567 btrfs_trans_handle_cachep =
4568 btrfs_cache_create("btrfs_trans_handle_cache",
4569 sizeof(struct btrfs_trans_handle),
4570 0, NULL);
39279cc3
CM		 4571 if (!btrfs_trans_handle_cachep)
4572 goto fail;
86479a04 4573 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
39279cc3 4574 sizeof(struct btrfs_transaction),
92fee66d 4575 0, NULL);
39279cc3
CM		 4576 if (!btrfs_transaction_cachep)
4577 goto fail;
86479a04 4578 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
23223584 4579 sizeof(struct btrfs_path),
92fee66d 4580 0, NULL);
39279cc3
CM		 4581 if (!btrfs_path_cachep)
4582 goto fail;
86479a04 4583 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
92fee66d 4584 SLAB_DESTROY_BY_RCU, NULL);
39279cc3
CM		 4585 if (!btrfs_bit_radix_cachep)
4586 goto fail;
4587 return 0;
4588fail:
4589 btrfs_destroy_cachep();
4590 return -ENOMEM;
4591}
4592
4593static int btrfs_getattr(struct vfsmount *mnt,
4594 struct dentry *dentry, struct kstat *stat)
4595{
4596 struct inode *inode = dentry->d_inode;
4597 generic_fillattr(inode, stat);
3394e160 4598 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4599 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4600 stat->blocks = (inode_get_bytes(inode) +
4601 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4602 return 0;
4603}
4604
d397712b
CM		 4605static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4606 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 4607{
4608 struct btrfs_trans_handle *trans;
4609 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4610 struct inode *new_inode = new_dentry->d_inode;
4611 struct inode *old_inode = old_dentry->d_inode;
4612 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4613 u64 index = 0;
39279cc3
CM		 4614 int ret;
4615
3394e160
CM		 4616 /* we're not allowed to rename between subvolumes */
4617 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4618 BTRFS_I(new_dir)->root->root_key.objectid)
4619 return -EXDEV;
4620
39279cc3
CM		 4621 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4622 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4623 return -ENOTEMPTY;
4624 }
5f39d397 4625
0660b5af
CM		 4626 /* to rename a snapshot or subvolume, we need to juggle the
4627 * backrefs. This isn't coded yet
4628 */
4629 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4630 return -EXDEV;
4631
6a63209f 4632 ret = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4633 if (ret)
4634 goto out_unlock;
4635
39279cc3 4636 trans = btrfs_start_transaction(root, 1);
5f39d397 4637
39279cc3 4638 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4639
e02119d5 4640 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4641 old_dir->i_ctime = old_dir->i_mtime = ctime;
4642 new_dir->i_ctime = new_dir->i_mtime = ctime;
4643 old_inode->i_ctime = ctime;
5f39d397 4644
e02119d5
CM		 4645 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4646 old_dentry->d_name.name,
4647 old_dentry->d_name.len);
39279cc3
CM		 4648 if (ret)
4649 goto out_fail;
4650
4651 if (new_inode) {
4652 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4653 ret = btrfs_unlink_inode(trans, root, new_dir,
4654 new_dentry->d_inode,
4655 new_dentry->d_name.name,
4656 new_dentry->d_name.len);
39279cc3
CM		 4657 if (ret)
4658 goto out_fail;
7b128766 4659 if (new_inode->i_nlink == 0) {
e02119d5 4660 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4661 if (ret)
4662 goto out_fail;
4663 }
e02119d5 4664
39279cc3 4665 }
3de4586c 4666 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4667 if (ret)
4668 goto out_fail;
4669
e02119d5
CM		 4670 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4671 old_inode, new_dentry->d_name.name,
4672 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4673 if (ret)
4674 goto out_fail;
4675
4676out_fail:
ab78c84d 4677 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4678out_unlock:
39279cc3
CM		 4679 return ret;
4680}
4681
d352ac68
CM		 4682/*
4683 * some fairly slow code that needs optimization. This walks the list
4684 * of all the inodes with pending delalloc and forces them to disk.
4685 */
ea8c2819
CM		 4686int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4687{
4688 struct list_head *head = &root->fs_info->delalloc_inodes;
4689 struct btrfs_inode *binode;
5b21f2ed 4690 struct inode *inode;
ea8c2819 4691
c146afad
YZ		 4692 if (root->fs_info->sb->s_flags & MS_RDONLY)
4693 return -EROFS;
4694
75eff68e 4695 spin_lock(&root->fs_info->delalloc_lock);
d397712b 4696 while (!list_empty(head)) {
ea8c2819
CM		 4697 binode = list_entry(head->next, struct btrfs_inode,
4698 delalloc_inodes);
5b21f2ed
ZY		 4699 inode = igrab(&binode->vfs_inode);
4700 if (!inode)
4701 list_del_init(&binode->delalloc_inodes);
75eff68e 4702 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 4703 if (inode) {
8c8bee1d 4704 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4705 iput(inode);
4706 }
4707 cond_resched();
75eff68e 4708 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 4709 }
75eff68e 4710 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 4711
4712 /* the filemap_flush will queue IO into the worker threads, but
4713 * we have to make sure the IO is actually started and that
4714 * ordered extents get created before we return
4715 */
4716 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 4717 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 4718 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4719 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4720 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4721 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4722 }
4723 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4724 return 0;
4725}
4726
39279cc3
CM		 4727static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4728 const char *symname)
4729{
4730 struct btrfs_trans_handle *trans;
4731 struct btrfs_root *root = BTRFS_I(dir)->root;
4732 struct btrfs_path *path;
4733 struct btrfs_key key;
1832a6d5 4734 struct inode *inode = NULL;
39279cc3
CM		 4735 int err;
4736 int drop_inode = 0;
4737 u64 objectid;
00e4e6b3 4738 u64 index = 0 ;
39279cc3
CM		 4739 int name_len;
4740 int datasize;
5f39d397 4741 unsigned long ptr;
39279cc3 4742 struct btrfs_file_extent_item *ei;
5f39d397 4743 struct extent_buffer *leaf;
1832a6d5 4744 unsigned long nr = 0;
39279cc3
CM		 4745
4746 name_len = strlen(symname) + 1;
4747 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4748 return -ENAMETOOLONG;
1832a6d5 4749
6a63209f 4750 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4751 if (err)
4752 goto out_fail;
4753
39279cc3
CM		 4754 trans = btrfs_start_transaction(root, 1);
4755 btrfs_set_trans_block_group(trans, dir);
4756
4757 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4758 if (err) {
4759 err = -ENOSPC;
4760 goto out_unlock;
4761 }
4762
aec7477b 4763 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4764 dentry->d_name.len,
4765 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4766 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4767 &index);
39279cc3
CM		 4768 err = PTR_ERR(inode);
4769 if (IS_ERR(inode))
4770 goto out_unlock;
4771
0279b4cd 4772 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 4773 if (err) {
4774 drop_inode = 1;
4775 goto out_unlock;
4776 }
4777
39279cc3 4778 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4779 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4780 if (err)
4781 drop_inode = 1;
4782 else {
4783 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4784 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4785 inode->i_fop = &btrfs_file_operations;
4786 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4787 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 4788 }
4789 dir->i_sb->s_dirt = 1;
4790 btrfs_update_inode_block_group(trans, inode);
4791 btrfs_update_inode_block_group(trans, dir);
4792 if (drop_inode)
4793 goto out_unlock;
4794
4795 path = btrfs_alloc_path();
4796 BUG_ON(!path);
4797 key.objectid = inode->i_ino;
4798 key.offset = 0;
39279cc3
CM		 4799 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4800 datasize = btrfs_file_extent_calc_inline_size(name_len);
4801 err = btrfs_insert_empty_item(trans, root, path, &key,
4802 datasize);
54aa1f4d
CM		 4803 if (err) {
4804 drop_inode = 1;
4805 goto out_unlock;
4806 }
5f39d397
CM		 4807 leaf = path->nodes[0];
4808 ei = btrfs_item_ptr(leaf, path->slots[0],
4809 struct btrfs_file_extent_item);
4810 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4811 btrfs_set_file_extent_type(leaf, ei,
39279cc3 4812 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 4813 btrfs_set_file_extent_encryption(leaf, ei, 0);
4814 btrfs_set_file_extent_compression(leaf, ei, 0);
4815 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4816 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4817
39279cc3 4818 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 4819 write_extent_buffer(leaf, symname, ptr, name_len);
4820 btrfs_mark_buffer_dirty(leaf);
39279cc3 4821 btrfs_free_path(path);
5f39d397 4822
39279cc3
CM		 4823 inode->i_op = &btrfs_symlink_inode_operations;
4824 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 4825 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 4826 inode_set_bytes(inode, name_len);
dbe674a9 4827 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 4828 err = btrfs_update_inode(trans, root, inode);
4829 if (err)
4830 drop_inode = 1;
39279cc3
CM		 4831
4832out_unlock:
d3c2fdcf 4833 nr = trans->blocks_used;
ab78c84d 4834 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4835out_fail:
39279cc3
CM		 4836 if (drop_inode) {
4837 inode_dec_link_count(inode);
4838 iput(inode);
4839 }
d3c2fdcf 4840 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4841 return err;
4842}
16432985 4843
d899e052
YZ		 4844static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4845 u64 alloc_hint, int mode)
4846{
4847 struct btrfs_trans_handle *trans;
4848 struct btrfs_root *root = BTRFS_I(inode)->root;
4849 struct btrfs_key ins;
4850 u64 alloc_size;
4851 u64 cur_offset = start;
4852 u64 num_bytes = end - start;
4853 int ret = 0;
4854
4855 trans = btrfs_join_transaction(root, 1);
4856 BUG_ON(!trans);
4857 btrfs_set_trans_block_group(trans, inode);
4858
4859 while (num_bytes > 0) {
4860 alloc_size = min(num_bytes, root->fs_info->max_extent);
4861 ret = btrfs_reserve_extent(trans, root, alloc_size,
4862 root->sectorsize, 0, alloc_hint,
4863 (u64)-1, &ins, 1);
4864 if (ret) {
4865 WARN_ON(1);
4866 goto out;
4867 }
4868 ret = insert_reserved_file_extent(trans, inode,
4869 cur_offset, ins.objectid,
4870 ins.offset, ins.offset,
4871 ins.offset, 0, 0, 0,
4872 BTRFS_FILE_EXTENT_PREALLOC);
4873 BUG_ON(ret);
4874 num_bytes -= ins.offset;
4875 cur_offset += ins.offset;
4876 alloc_hint = ins.objectid + ins.offset;
4877 }
4878out:
4879 if (cur_offset > start) {
4880 inode->i_ctime = CURRENT_TIME;
4881 btrfs_set_flag(inode, PREALLOC);
4882 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4883 cur_offset > i_size_read(inode))
4884 btrfs_i_size_write(inode, cur_offset);
4885 ret = btrfs_update_inode(trans, root, inode);
4886 BUG_ON(ret);
4887 }
4888
4889 btrfs_end_transaction(trans, root);
4890 return ret;
4891}
4892
4893static long btrfs_fallocate(struct inode *inode, int mode,
4894 loff_t offset, loff_t len)
4895{
4896 u64 cur_offset;
4897 u64 last_byte;
4898 u64 alloc_start;
4899 u64 alloc_end;
4900 u64 alloc_hint = 0;
4901 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4902 struct extent_map *em;
4903 int ret;
4904
4905 alloc_start = offset & ~mask;
4906 alloc_end = (offset + len + mask) & ~mask;
4907
4908 mutex_lock(&inode->i_mutex);
4909 if (alloc_start > inode->i_size) {
4910 ret = btrfs_cont_expand(inode, alloc_start);
4911 if (ret)
4912 goto out;
4913 }
4914
4915 while (1) {
4916 struct btrfs_ordered_extent *ordered;
4917 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4918 alloc_end - 1, GFP_NOFS);
4919 ordered = btrfs_lookup_first_ordered_extent(inode,
4920 alloc_end - 1);
4921 if (ordered &&
4922 ordered->file_offset + ordered->len > alloc_start &&
4923 ordered->file_offset < alloc_end) {
4924 btrfs_put_ordered_extent(ordered);
4925 unlock_extent(&BTRFS_I(inode)->io_tree,
4926 alloc_start, alloc_end - 1, GFP_NOFS);
4927 btrfs_wait_ordered_range(inode, alloc_start,
4928 alloc_end - alloc_start);
4929 } else {
4930 if (ordered)
4931 btrfs_put_ordered_extent(ordered);
4932 break;
4933 }
4934 }
4935
4936 cur_offset = alloc_start;
4937 while (1) {
4938 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4939 alloc_end - cur_offset, 0);
4940 BUG_ON(IS_ERR(em) || !em);
4941 last_byte = min(extent_map_end(em), alloc_end);
4942 last_byte = (last_byte + mask) & ~mask;
4943 if (em->block_start == EXTENT_MAP_HOLE) {
4944 ret = prealloc_file_range(inode, cur_offset,
4945 last_byte, alloc_hint, mode);
4946 if (ret < 0) {
4947 free_extent_map(em);
4948 break;
4949 }
4950 }
4951 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4952 alloc_hint = em->block_start;
4953 free_extent_map(em);
4954
4955 cur_offset = last_byte;
4956 if (cur_offset >= alloc_end) {
4957 ret = 0;
4958 break;
4959 }
4960 }
4961 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4962 GFP_NOFS);
4963out:
4964 mutex_unlock(&inode->i_mutex);
4965 return ret;
4966}
4967
e6dcd2dc
CM		 4968static int btrfs_set_page_dirty(struct page *page)
4969{
e6dcd2dc
CM		 4970 return __set_page_dirty_nobuffers(page);
4971}
4972
0ee0fda0 4973static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd
Y		 4974{
4975 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4976 return -EACCES;
33268eaf 4977 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 4978}
39279cc3
CM		 4979
4980static struct inode_operations btrfs_dir_inode_operations = {
3394e160 4981 .getattr = btrfs_getattr,
39279cc3
CM		 4982 .lookup = btrfs_lookup,
4983 .create = btrfs_create,
4984 .unlink = btrfs_unlink,
4985 .link = btrfs_link,
4986 .mkdir = btrfs_mkdir,
4987 .rmdir = btrfs_rmdir,
4988 .rename = btrfs_rename,
4989 .symlink = btrfs_symlink,
4990 .setattr = btrfs_setattr,
618e21d5 4991 .mknod = btrfs_mknod,
95819c05
CH		 4992 .setxattr = btrfs_setxattr,
4993 .getxattr = btrfs_getxattr,
5103e947 4994 .listxattr = btrfs_listxattr,
95819c05 4995 .removexattr = btrfs_removexattr,
fdebe2bd 4996 .permission = btrfs_permission,
39279cc3 4997};
39279cc3
CM		 4998static struct inode_operations btrfs_dir_ro_inode_operations = {
4999 .lookup = btrfs_lookup,
fdebe2bd 5000 .permission = btrfs_permission,
39279cc3 5001};
39279cc3
CM		 5002static struct file_operations btrfs_dir_file_operations = {
5003 .llseek = generic_file_llseek,
5004 .read = generic_read_dir,
cbdf5a24 5005 .readdir = btrfs_real_readdir,
34287aa3 5006 .unlocked_ioctl = btrfs_ioctl,
39279cc3 5007#ifdef CONFIG_COMPAT
34287aa3 5008 .compat_ioctl = btrfs_ioctl,
39279cc3 5009#endif
6bf13c0c 5010 .release = btrfs_release_file,
e02119d5 5011 .fsync = btrfs_sync_file,
39279cc3
CM		 5012};
5013
d1310b2e 5014static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 5015 .fill_delalloc = run_delalloc_range,
065631f6 5016 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 5017 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 5018 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 5019 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 5020 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 5021 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 5022 .set_bit_hook = btrfs_set_bit_hook,
5023 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 5024};
5025
35054394
CM		 5026/*
5027 * btrfs doesn't support the bmap operation because swapfiles
5028 * use bmap to make a mapping of extents in the file. They assume
5029 * these extents won't change over the life of the file and they
5030 * use the bmap result to do IO directly to the drive.
5031 *
5032 * the btrfs bmap call would return logical addresses that aren't
5033 * suitable for IO and they also will change frequently as COW
5034 * operations happen. So, swapfile + btrfs == corruption.
5035 *
5036 * For now we're avoiding this by dropping bmap.
5037 */
39279cc3
CM		 5038static struct address_space_operations btrfs_aops = {
5039 .readpage = btrfs_readpage,
5040 .writepage = btrfs_writepage,
b293f02e 5041 .writepages = btrfs_writepages,
3ab2fb5a 5042 .readpages = btrfs_readpages,
39279cc3 5043 .sync_page = block_sync_page,
16432985 5044 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 5045 .invalidatepage = btrfs_invalidatepage,
5046 .releasepage = btrfs_releasepage,
e6dcd2dc 5047 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 5048};
5049
5050static struct address_space_operations btrfs_symlink_aops = {
5051 .readpage = btrfs_readpage,
5052 .writepage = btrfs_writepage,
2bf5a725
CM		 5053 .invalidatepage = btrfs_invalidatepage,
5054 .releasepage = btrfs_releasepage,
39279cc3
CM		 5055};
5056
5057static struct inode_operations btrfs_file_inode_operations = {
5058 .truncate = btrfs_truncate,
5059 .getattr = btrfs_getattr,
5060 .setattr = btrfs_setattr,
95819c05
CH		 5061 .setxattr = btrfs_setxattr,
5062 .getxattr = btrfs_getxattr,
5103e947 5063 .listxattr = btrfs_listxattr,
95819c05 5064 .removexattr = btrfs_removexattr,
fdebe2bd 5065 .permission = btrfs_permission,
d899e052 5066 .fallocate = btrfs_fallocate,
1506fcc8 5067 .fiemap = btrfs_fiemap,
39279cc3 5068};
618e21d5
JB		 5069static struct inode_operations btrfs_special_inode_operations = {
5070 .getattr = btrfs_getattr,
5071 .setattr = btrfs_setattr,
fdebe2bd 5072 .permission = btrfs_permission,
95819c05
CH		 5073 .setxattr = btrfs_setxattr,
5074 .getxattr = btrfs_getxattr,
33268eaf 5075 .listxattr = btrfs_listxattr,
95819c05 5076 .removexattr = btrfs_removexattr,
618e21d5 5077};
39279cc3
CM		 5078static struct inode_operations btrfs_symlink_inode_operations = {
5079 .readlink = generic_readlink,
5080 .follow_link = page_follow_link_light,
5081 .put_link = page_put_link,
fdebe2bd 5082 .permission = btrfs_permission,
0279b4cd
JO		 5083 .setxattr = btrfs_setxattr,
5084 .getxattr = btrfs_getxattr,
5085 .listxattr = btrfs_listxattr,
5086 .removexattr = btrfs_removexattr,
39279cc3 5087};
Linux 6.x block layer and io_uring tree(s)