projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
powerpc: Don't do generic calibrate_delay()
[deliverable/linux.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"
c8b97818 51#include "compression.h"
b4ce94de 52#include "locking.h"
39279cc3
CM		 53
54struct btrfs_iget_args {
55 u64 ino;
56 struct btrfs_root *root;
57};
58
59static struct inode_operations btrfs_dir_inode_operations;
60static struct inode_operations btrfs_symlink_inode_operations;
61static struct inode_operations btrfs_dir_ro_inode_operations;
618e21d5 62static struct inode_operations btrfs_special_inode_operations;
39279cc3
CM		 63static struct inode_operations btrfs_file_inode_operations;
64static struct address_space_operations btrfs_aops;
65static struct address_space_operations btrfs_symlink_aops;
66static struct file_operations btrfs_dir_file_operations;
d1310b2e 67static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 68
69static struct kmem_cache *btrfs_inode_cachep;
70struct kmem_cache *btrfs_trans_handle_cachep;
71struct kmem_cache *btrfs_transaction_cachep;
39279cc3
CM		 72struct kmem_cache *btrfs_path_cachep;
73
74#define S_SHIFT 12
75static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
83};
84
7b128766 85static void btrfs_truncate(struct inode *inode);
c8b97818 86static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
771ed689
CM		 87static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
7b128766 91
0279b4cd
JO		 92static int btrfs_init_inode_security(struct inode *inode, struct inode *dir)
93{
94 int err;
95
96 err = btrfs_init_acl(inode, dir);
97 if (!err)
98 err = btrfs_xattr_security_init(inode, dir);
99 return err;
100}
101
c8b97818
CM		 102/*
103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
106 */
d397712b 107static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 108 struct btrfs_root *root, struct inode *inode,
109 u64 start, size_t size, size_t compressed_size,
110 struct page **compressed_pages)
111{
112 struct btrfs_key key;
113 struct btrfs_path *path;
114 struct extent_buffer *leaf;
115 struct page *page = NULL;
116 char *kaddr;
117 unsigned long ptr;
118 struct btrfs_file_extent_item *ei;
119 int err = 0;
120 int ret;
121 size_t cur_size = size;
122 size_t datasize;
123 unsigned long offset;
124 int use_compress = 0;
125
126 if (compressed_size && compressed_pages) {
127 use_compress = 1;
128 cur_size = compressed_size;
129 }
130
d397712b
CM		 131 path = btrfs_alloc_path();
132 if (!path)
c8b97818
CM		 133 return -ENOMEM;
134
b9473439 135 path->leave_spinning = 1;
c8b97818
CM		 136 btrfs_set_trans_block_group(trans, inode);
137
138 key.objectid = inode->i_ino;
139 key.offset = start;
140 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 141 datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143 inode_add_bytes(inode, size);
144 ret = btrfs_insert_empty_item(trans, root, path, &key,
145 datasize);
146 BUG_ON(ret);
147 if (ret) {
148 err = ret;
c8b97818
CM		 149 goto fail;
150 }
151 leaf = path->nodes[0];
152 ei = btrfs_item_ptr(leaf, path->slots[0],
153 struct btrfs_file_extent_item);
154 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156 btrfs_set_file_extent_encryption(leaf, ei, 0);
157 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159 ptr = btrfs_file_extent_inline_start(ei);
160
161 if (use_compress) {
162 struct page *cpage;
163 int i = 0;
d397712b 164 while (compressed_size > 0) {
c8b97818 165 cpage = compressed_pages[i];
5b050f04 166 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 167 PAGE_CACHE_SIZE);
168
b9473439 169 kaddr = kmap_atomic(cpage, KM_USER0);
c8b97818 170 write_extent_buffer(leaf, kaddr, ptr, cur_size);
b9473439 171 kunmap_atomic(kaddr, KM_USER0);
c8b97818
CM		 172
173 i++;
174 ptr += cur_size;
175 compressed_size -= cur_size;
176 }
177 btrfs_set_file_extent_compression(leaf, ei,
178 BTRFS_COMPRESS_ZLIB);
179 } else {
180 page = find_get_page(inode->i_mapping,
181 start >> PAGE_CACHE_SHIFT);
182 btrfs_set_file_extent_compression(leaf, ei, 0);
183 kaddr = kmap_atomic(page, KM_USER0);
184 offset = start & (PAGE_CACHE_SIZE - 1);
185 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186 kunmap_atomic(kaddr, KM_USER0);
187 page_cache_release(page);
188 }
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_free_path(path);
191
192 BTRFS_I(inode)->disk_i_size = inode->i_size;
193 btrfs_update_inode(trans, root, inode);
194 return 0;
195fail:
196 btrfs_free_path(path);
197 return err;
198}
199
200
201/*
202 * conditionally insert an inline extent into the file. This
203 * does the checks required to make sure the data is small enough
204 * to fit as an inline extent.
205 */
7f366cfe 206static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 207 struct btrfs_root *root,
208 struct inode *inode, u64 start, u64 end,
209 size_t compressed_size,
210 struct page **compressed_pages)
211{
212 u64 isize = i_size_read(inode);
213 u64 actual_end = min(end + 1, isize);
214 u64 inline_len = actual_end - start;
215 u64 aligned_end = (end + root->sectorsize - 1) &
216 ~((u64)root->sectorsize - 1);
217 u64 hint_byte;
218 u64 data_len = inline_len;
219 int ret;
220
221 if (compressed_size)
222 data_len = compressed_size;
223
224 if (start > 0 ||
70b99e69 225 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 226 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227 (!compressed_size &&
228 (actual_end & (root->sectorsize - 1)) == 0) ||
229 end + 1 < isize ||
230 data_len > root->fs_info->max_inline) {
231 return 1;
232 }
233
c8b97818 234 ret = btrfs_drop_extents(trans, root, inode, start,
e980b50c 235 aligned_end, aligned_end, start, &hint_byte);
c8b97818
CM		 236 BUG_ON(ret);
237
238 if (isize > actual_end)
239 inline_len = min_t(u64, isize, actual_end);
240 ret = insert_inline_extent(trans, root, inode, start,
241 inline_len, compressed_size,
242 compressed_pages);
243 BUG_ON(ret);
244 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
c8b97818
CM		 245 return 0;
246}
247
771ed689
CM		 248struct async_extent {
249 u64 start;
250 u64 ram_size;
251 u64 compressed_size;
252 struct page **pages;
253 unsigned long nr_pages;
254 struct list_head list;
255};
256
257struct async_cow {
258 struct inode *inode;
259 struct btrfs_root *root;
260 struct page *locked_page;
261 u64 start;
262 u64 end;
263 struct list_head extents;
264 struct btrfs_work work;
265};
266
267static noinline int add_async_extent(struct async_cow *cow,
268 u64 start, u64 ram_size,
269 u64 compressed_size,
270 struct page **pages,
271 unsigned long nr_pages)
272{
273 struct async_extent *async_extent;
274
275 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276 async_extent->start = start;
277 async_extent->ram_size = ram_size;
278 async_extent->compressed_size = compressed_size;
279 async_extent->pages = pages;
280 async_extent->nr_pages = nr_pages;
281 list_add_tail(&async_extent->list, &cow->extents);
282 return 0;
283}
284
d352ac68 285/*
771ed689
CM		 286 * we create compressed extents in two phases. The first
287 * phase compresses a range of pages that have already been
288 * locked (both pages and state bits are locked).
c8b97818 289 *
771ed689
CM		 290 * This is done inside an ordered work queue, and the compression
291 * is spread across many cpus. The actual IO submission is step
292 * two, and the ordered work queue takes care of making sure that
293 * happens in the same order things were put onto the queue by
294 * writepages and friends.
c8b97818 295 *
771ed689
CM		 296 * If this code finds it can't get good compression, it puts an
297 * entry onto the work queue to write the uncompressed bytes. This
298 * makes sure that both compressed inodes and uncompressed inodes
299 * are written in the same order that pdflush sent them down.
d352ac68 300 */
771ed689
CM		 301static noinline int compress_file_range(struct inode *inode,
302 struct page *locked_page,
303 u64 start, u64 end,
304 struct async_cow *async_cow,
305 int *num_added)
b888db2b
CM		 306{
307 struct btrfs_root *root = BTRFS_I(inode)->root;
308 struct btrfs_trans_handle *trans;
db94535d 309 u64 num_bytes;
c8b97818
CM		 310 u64 orig_start;
311 u64 disk_num_bytes;
db94535d 312 u64 blocksize = root->sectorsize;
c8b97818 313 u64 actual_end;
42dc7bab 314 u64 isize = i_size_read(inode);
e6dcd2dc 315 int ret = 0;
c8b97818
CM		 316 struct page **pages = NULL;
317 unsigned long nr_pages;
318 unsigned long nr_pages_ret = 0;
319 unsigned long total_compressed = 0;
320 unsigned long total_in = 0;
321 unsigned long max_compressed = 128 * 1024;
771ed689 322 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 323 int i;
324 int will_compress;
b888db2b 325
c8b97818
CM		 326 orig_start = start;
327
42dc7bab 328 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 329again:
330 will_compress = 0;
331 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 333
f03d9301
CM		 334 /*
335 * we don't want to send crud past the end of i_size through
336 * compression, that's just a waste of CPU time. So, if the
337 * end of the file is before the start of our current
338 * requested range of bytes, we bail out to the uncompressed
339 * cleanup code that can deal with all of this.
340 *
341 * It isn't really the fastest way to fix things, but this is a
342 * very uncommon corner.
343 */
344 if (actual_end <= start)
345 goto cleanup_and_bail_uncompressed;
346
c8b97818
CM		 347 total_compressed = actual_end - start;
348
349 /* we want to make sure that amount of ram required to uncompress
350 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 351 * of a compressed extent to 128k. This is a crucial number
352 * because it also controls how easily we can spread reads across
353 * cpus for decompression.
354 *
355 * We also want to make sure the amount of IO required to do
356 * a random read is reasonably small, so we limit the size of
357 * a compressed extent to 128k.
c8b97818
CM		 358 */
359 total_compressed = min(total_compressed, max_uncompressed);
db94535d 360 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 361 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 362 disk_num_bytes = num_bytes;
363 total_in = 0;
364 ret = 0;
db94535d 365
771ed689
CM		 366 /*
367 * we do compression for mount -o compress and when the
368 * inode has not been flagged as nocompress. This flag can
369 * change at any time if we discover bad compression ratios.
c8b97818 370 */
6cbff00f 371 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
c8b97818
CM		 372 btrfs_test_opt(root, COMPRESS)) {
373 WARN_ON(pages);
cfbc246e 374 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
c8b97818 375
c8b97818
CM		 376 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377 total_compressed, pages,
378 nr_pages, &nr_pages_ret,
379 &total_in,
380 &total_compressed,
381 max_compressed);
382
383 if (!ret) {
384 unsigned long offset = total_compressed &
385 (PAGE_CACHE_SIZE - 1);
386 struct page *page = pages[nr_pages_ret - 1];
387 char *kaddr;
388
389 /* zero the tail end of the last page, we might be
390 * sending it down to disk
391 */
392 if (offset) {
393 kaddr = kmap_atomic(page, KM_USER0);
394 memset(kaddr + offset, 0,
395 PAGE_CACHE_SIZE - offset);
396 kunmap_atomic(kaddr, KM_USER0);
397 }
398 will_compress = 1;
399 }
400 }
401 if (start == 0) {
771ed689
CM		 402 trans = btrfs_join_transaction(root, 1);
403 BUG_ON(!trans);
404 btrfs_set_trans_block_group(trans, inode);
405
c8b97818 406 /* lets try to make an inline extent */
771ed689 407 if (ret || total_in < (actual_end - start)) {
c8b97818 408 /* we didn't compress the entire range, try
771ed689 409 * to make an uncompressed inline extent.
c8b97818
CM		 410 */
411 ret = cow_file_range_inline(trans, root, inode,
412 start, end, 0, NULL);
413 } else {
771ed689 414 /* try making a compressed inline extent */
c8b97818
CM		 415 ret = cow_file_range_inline(trans, root, inode,
416 start, end,
417 total_compressed, pages);
418 }
771ed689 419 btrfs_end_transaction(trans, root);
c8b97818 420 if (ret == 0) {
771ed689
CM		 421 /*
422 * inline extent creation worked, we don't need
423 * to create any more async work items. Unlock
424 * and free up our temp pages.
425 */
c8b97818
CM		 426 extent_clear_unlock_delalloc(inode,
427 &BTRFS_I(inode)->io_tree,
771ed689
CM		 428 start, end, NULL, 1, 0,
429 0, 1, 1, 1);
c8b97818
CM		 430 ret = 0;
431 goto free_pages_out;
432 }
433 }
434
435 if (will_compress) {
436 /*
437 * we aren't doing an inline extent round the compressed size
438 * up to a block size boundary so the allocator does sane
439 * things
440 */
441 total_compressed = (total_compressed + blocksize - 1) &
442 ~(blocksize - 1);
443
444 /*
445 * one last check to make sure the compression is really a
446 * win, compare the page count read with the blocks on disk
447 */
448 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
449 ~(PAGE_CACHE_SIZE - 1);
450 if (total_compressed >= total_in) {
451 will_compress = 0;
452 } else {
453 disk_num_bytes = total_compressed;
454 num_bytes = total_in;
455 }
456 }
457 if (!will_compress && pages) {
458 /*
459 * the compression code ran but failed to make things smaller,
460 * free any pages it allocated and our page pointer array
461 */
462 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 463 WARN_ON(pages[i]->mapping);
c8b97818
CM		 464 page_cache_release(pages[i]);
465 }
466 kfree(pages);
467 pages = NULL;
468 total_compressed = 0;
469 nr_pages_ret = 0;
470
471 /* flag the file so we don't compress in the future */
6cbff00f 472 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
c8b97818 473 }
771ed689
CM		 474 if (will_compress) {
475 *num_added += 1;
c8b97818 476
771ed689
CM		 477 /* the async work queues will take care of doing actual
478 * allocation on disk for these compressed pages,
479 * and will submit them to the elevator.
480 */
481 add_async_extent(async_cow, start, num_bytes,
482 total_compressed, pages, nr_pages_ret);
179e29e4 483
42dc7bab 484 if (start + num_bytes < end && start + num_bytes < actual_end) {
771ed689
CM		 485 start += num_bytes;
486 pages = NULL;
487 cond_resched();
488 goto again;
489 }
490 } else {
f03d9301 491cleanup_and_bail_uncompressed:
771ed689
CM		 492 /*
493 * No compression, but we still need to write the pages in
494 * the file we've been given so far. redirty the locked
495 * page if it corresponds to our extent and set things up
496 * for the async work queue to run cow_file_range to do
497 * the normal delalloc dance
498 */
499 if (page_offset(locked_page) >= start &&
500 page_offset(locked_page) <= end) {
501 __set_page_dirty_nobuffers(locked_page);
502 /* unlocked later on in the async handlers */
503 }
504 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
505 *num_added += 1;
506 }
3b951516 507
771ed689
CM		 508out:
509 return 0;
510
511free_pages_out:
512 for (i = 0; i < nr_pages_ret; i++) {
513 WARN_ON(pages[i]->mapping);
514 page_cache_release(pages[i]);
515 }
d397712b 516 kfree(pages);
771ed689
CM		 517
518 goto out;
519}
520
521/*
522 * phase two of compressed writeback. This is the ordered portion
523 * of the code, which only gets called in the order the work was
524 * queued. We walk all the async extents created by compress_file_range
525 * and send them down to the disk.
526 */
527static noinline int submit_compressed_extents(struct inode *inode,
528 struct async_cow *async_cow)
529{
530 struct async_extent *async_extent;
531 u64 alloc_hint = 0;
532 struct btrfs_trans_handle *trans;
533 struct btrfs_key ins;
534 struct extent_map *em;
535 struct btrfs_root *root = BTRFS_I(inode)->root;
536 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537 struct extent_io_tree *io_tree;
538 int ret;
539
540 if (list_empty(&async_cow->extents))
541 return 0;
542
543 trans = btrfs_join_transaction(root, 1);
544
d397712b 545 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 546 async_extent = list_entry(async_cow->extents.next,
547 struct async_extent, list);
548 list_del(&async_extent->list);
c8b97818 549
771ed689
CM		 550 io_tree = &BTRFS_I(inode)->io_tree;
551
552 /* did the compression code fall back to uncompressed IO? */
553 if (!async_extent->pages) {
554 int page_started = 0;
555 unsigned long nr_written = 0;
556
557 lock_extent(io_tree, async_extent->start,
d397712b
CM		 558 async_extent->start +
559 async_extent->ram_size - 1, GFP_NOFS);
771ed689
CM		 560
561 /* allocate blocks */
562 cow_file_range(inode, async_cow->locked_page,
563 async_extent->start,
564 async_extent->start +
565 async_extent->ram_size - 1,
566 &page_started, &nr_written, 0);
567
568 /*
569 * if page_started, cow_file_range inserted an
570 * inline extent and took care of all the unlocking
571 * and IO for us. Otherwise, we need to submit
572 * all those pages down to the drive.
573 */
574 if (!page_started)
575 extent_write_locked_range(io_tree,
576 inode, async_extent->start,
d397712b 577 async_extent->start +
771ed689
CM		 578 async_extent->ram_size - 1,
579 btrfs_get_extent,
580 WB_SYNC_ALL);
581 kfree(async_extent);
582 cond_resched();
583 continue;
584 }
585
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start + async_extent->ram_size - 1,
588 GFP_NOFS);
c8b97818 589 /*
771ed689
CM		 590 * here we're doing allocation and writeback of the
591 * compressed pages
c8b97818 592 */
771ed689
CM		 593 btrfs_drop_extent_cache(inode, async_extent->start,
594 async_extent->start +
595 async_extent->ram_size - 1, 0);
596
597 ret = btrfs_reserve_extent(trans, root,
598 async_extent->compressed_size,
599 async_extent->compressed_size,
600 0, alloc_hint,
601 (u64)-1, &ins, 1);
602 BUG_ON(ret);
603 em = alloc_extent_map(GFP_NOFS);
604 em->start = async_extent->start;
605 em->len = async_extent->ram_size;
445a6944 606 em->orig_start = em->start;
c8b97818 607
771ed689
CM		 608 em->block_start = ins.objectid;
609 em->block_len = ins.offset;
610 em->bdev = root->fs_info->fs_devices->latest_bdev;
611 set_bit(EXTENT_FLAG_PINNED, &em->flags);
612 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
613
d397712b 614 while (1) {
771ed689
CM		 615 spin_lock(&em_tree->lock);
616 ret = add_extent_mapping(em_tree, em);
617 spin_unlock(&em_tree->lock);
618 if (ret != -EEXIST) {
619 free_extent_map(em);
620 break;
621 }
622 btrfs_drop_extent_cache(inode, async_extent->start,
623 async_extent->start +
624 async_extent->ram_size - 1, 0);
625 }
626
627 ret = btrfs_add_ordered_extent(inode, async_extent->start,
628 ins.objectid,
629 async_extent->ram_size,
630 ins.offset,
631 BTRFS_ORDERED_COMPRESSED);
632 BUG_ON(ret);
633
634 btrfs_end_transaction(trans, root);
635
636 /*
637 * clear dirty, set writeback and unlock the pages.
638 */
639 extent_clear_unlock_delalloc(inode,
640 &BTRFS_I(inode)->io_tree,
641 async_extent->start,
642 async_extent->start +
643 async_extent->ram_size - 1,
644 NULL, 1, 1, 0, 1, 1, 0);
645
646 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 647 async_extent->start,
648 async_extent->ram_size,
649 ins.objectid,
650 ins.offset, async_extent->pages,
651 async_extent->nr_pages);
771ed689
CM		 652
653 BUG_ON(ret);
654 trans = btrfs_join_transaction(root, 1);
655 alloc_hint = ins.objectid + ins.offset;
656 kfree(async_extent);
657 cond_resched();
658 }
659
660 btrfs_end_transaction(trans, root);
661 return 0;
662}
663
664/*
665 * when extent_io.c finds a delayed allocation range in the file,
666 * the call backs end up in this code. The basic idea is to
667 * allocate extents on disk for the range, and create ordered data structs
668 * in ram to track those extents.
669 *
670 * locked_page is the page that writepage had locked already. We use
671 * it to make sure we don't do extra locks or unlocks.
672 *
673 * *page_started is set to one if we unlock locked_page and do everything
674 * required to start IO on it. It may be clean and already done with
675 * IO when we return.
676 */
677static noinline int cow_file_range(struct inode *inode,
678 struct page *locked_page,
679 u64 start, u64 end, int *page_started,
680 unsigned long *nr_written,
681 int unlock)
682{
683 struct btrfs_root *root = BTRFS_I(inode)->root;
684 struct btrfs_trans_handle *trans;
685 u64 alloc_hint = 0;
686 u64 num_bytes;
687 unsigned long ram_size;
688 u64 disk_num_bytes;
689 u64 cur_alloc_size;
690 u64 blocksize = root->sectorsize;
691 u64 actual_end;
42dc7bab 692 u64 isize = i_size_read(inode);
771ed689
CM		 693 struct btrfs_key ins;
694 struct extent_map *em;
695 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
696 int ret = 0;
697
698 trans = btrfs_join_transaction(root, 1);
699 BUG_ON(!trans);
700 btrfs_set_trans_block_group(trans, inode);
701
42dc7bab 702 actual_end = min_t(u64, isize, end + 1);
771ed689
CM		 703
704 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
705 num_bytes = max(blocksize, num_bytes);
706 disk_num_bytes = num_bytes;
707 ret = 0;
708
709 if (start == 0) {
710 /* lets try to make an inline extent */
711 ret = cow_file_range_inline(trans, root, inode,
712 start, end, 0, NULL);
713 if (ret == 0) {
714 extent_clear_unlock_delalloc(inode,
715 &BTRFS_I(inode)->io_tree,
716 start, end, NULL, 1, 1,
717 1, 1, 1, 1);
718 *nr_written = *nr_written +
719 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
720 *page_started = 1;
721 ret = 0;
722 goto out;
723 }
724 }
725
726 BUG_ON(disk_num_bytes >
727 btrfs_super_total_bytes(&root->fs_info->super_copy));
728
729 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
730
d397712b 731 while (disk_num_bytes > 0) {
c8b97818 732 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
e6dcd2dc 733 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 734 root->sectorsize, 0, alloc_hint,
e6dcd2dc 735 (u64)-1, &ins, 1);
d397712b
CM		 736 BUG_ON(ret);
737
e6dcd2dc
CM		 738 em = alloc_extent_map(GFP_NOFS);
739 em->start = start;
445a6944 740 em->orig_start = em->start;
c8b97818 741
771ed689
CM		 742 ram_size = ins.offset;
743 em->len = ins.offset;
c8b97818 744
e6dcd2dc 745 em->block_start = ins.objectid;
c8b97818 746 em->block_len = ins.offset;
e6dcd2dc 747 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 748 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 749
d397712b 750 while (1) {
e6dcd2dc
CM		 751 spin_lock(&em_tree->lock);
752 ret = add_extent_mapping(em_tree, em);
753 spin_unlock(&em_tree->lock);
754 if (ret != -EEXIST) {
755 free_extent_map(em);
756 break;
757 }
758 btrfs_drop_extent_cache(inode, start,
c8b97818 759 start + ram_size - 1, 0);
e6dcd2dc
CM		 760 }
761
98d20f67 762 cur_alloc_size = ins.offset;
e6dcd2dc 763 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 764 ram_size, cur_alloc_size, 0);
e6dcd2dc 765 BUG_ON(ret);
c8b97818 766
17d217fe
YZ		 767 if (root->root_key.objectid ==
768 BTRFS_DATA_RELOC_TREE_OBJECTID) {
769 ret = btrfs_reloc_clone_csums(inode, start,
770 cur_alloc_size);
771 BUG_ON(ret);
772 }
773
d397712b 774 if (disk_num_bytes < cur_alloc_size)
3b951516 775 break;
d397712b 776
c8b97818
CM		 777 /* we're not doing compressed IO, don't unlock the first
778 * page (which the caller expects to stay locked), don't
779 * clear any dirty bits and don't set any writeback bits
780 */
781 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
782 start, start + ram_size - 1,
771ed689
CM		 783 locked_page, unlock, 1,
784 1, 0, 0, 0);
c8b97818 785 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 786 num_bytes -= cur_alloc_size;
787 alloc_hint = ins.objectid + ins.offset;
788 start += cur_alloc_size;
b888db2b 789 }
b888db2b 790out:
771ed689 791 ret = 0;
b888db2b 792 btrfs_end_transaction(trans, root);
c8b97818 793
be20aa9d 794 return ret;
771ed689 795}
c8b97818 796
771ed689
CM		 797/*
798 * work queue call back to started compression on a file and pages
799 */
800static noinline void async_cow_start(struct btrfs_work *work)
801{
802 struct async_cow *async_cow;
803 int num_added = 0;
804 async_cow = container_of(work, struct async_cow, work);
805
806 compress_file_range(async_cow->inode, async_cow->locked_page,
807 async_cow->start, async_cow->end, async_cow,
808 &num_added);
809 if (num_added == 0)
810 async_cow->inode = NULL;
811}
812
813/*
814 * work queue call back to submit previously compressed pages
815 */
816static noinline void async_cow_submit(struct btrfs_work *work)
817{
818 struct async_cow *async_cow;
819 struct btrfs_root *root;
820 unsigned long nr_pages;
821
822 async_cow = container_of(work, struct async_cow, work);
823
824 root = async_cow->root;
825 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
826 PAGE_CACHE_SHIFT;
827
828 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
829
830 if (atomic_read(&root->fs_info->async_delalloc_pages) <
831 5 * 1042 * 1024 &&
832 waitqueue_active(&root->fs_info->async_submit_wait))
833 wake_up(&root->fs_info->async_submit_wait);
834
d397712b 835 if (async_cow->inode)
771ed689 836 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 837}
c8b97818 838
771ed689
CM		 839static noinline void async_cow_free(struct btrfs_work *work)
840{
841 struct async_cow *async_cow;
842 async_cow = container_of(work, struct async_cow, work);
843 kfree(async_cow);
844}
845
846static int cow_file_range_async(struct inode *inode, struct page *locked_page,
847 u64 start, u64 end, int *page_started,
848 unsigned long *nr_written)
849{
850 struct async_cow *async_cow;
851 struct btrfs_root *root = BTRFS_I(inode)->root;
852 unsigned long nr_pages;
853 u64 cur_end;
854 int limit = 10 * 1024 * 1042;
855
771ed689
CM		 856 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
857 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
d397712b 858 while (start < end) {
771ed689
CM		 859 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
860 async_cow->inode = inode;
861 async_cow->root = root;
862 async_cow->locked_page = locked_page;
863 async_cow->start = start;
864
6cbff00f 865 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 866 cur_end = end;
867 else
868 cur_end = min(end, start + 512 * 1024 - 1);
869
870 async_cow->end = cur_end;
871 INIT_LIST_HEAD(&async_cow->extents);
872
873 async_cow->work.func = async_cow_start;
874 async_cow->work.ordered_func = async_cow_submit;
875 async_cow->work.ordered_free = async_cow_free;
876 async_cow->work.flags = 0;
877
771ed689
CM		 878 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
879 PAGE_CACHE_SHIFT;
880 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
881
882 btrfs_queue_worker(&root->fs_info->delalloc_workers,
883 &async_cow->work);
884
885 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
886 wait_event(root->fs_info->async_submit_wait,
887 (atomic_read(&root->fs_info->async_delalloc_pages) <
888 limit));
889 }
890
d397712b 891 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 892 atomic_read(&root->fs_info->async_delalloc_pages)) {
893 wait_event(root->fs_info->async_submit_wait,
894 (atomic_read(&root->fs_info->async_delalloc_pages) ==
895 0));
896 }
897
898 *nr_written += nr_pages;
899 start = cur_end + 1;
900 }
901 *page_started = 1;
902 return 0;
be20aa9d
CM		 903}
904
d397712b 905static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 906 u64 bytenr, u64 num_bytes)
907{
908 int ret;
909 struct btrfs_ordered_sum *sums;
910 LIST_HEAD(list);
911
07d400a6
YZ		 912 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
913 bytenr + num_bytes - 1, &list);
17d217fe
YZ		 914 if (ret == 0 && list_empty(&list))
915 return 0;
916
917 while (!list_empty(&list)) {
918 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
919 list_del(&sums->list);
920 kfree(sums);
921 }
922 return 1;
923}
924
d352ac68
CM		 925/*
926 * when nowcow writeback call back. This checks for snapshots or COW copies
927 * of the extents that exist in the file, and COWs the file as required.
928 *
929 * If no cow copies or snapshots exist, we write directly to the existing
930 * blocks on disk
931 */
7f366cfe
CM		 932static noinline int run_delalloc_nocow(struct inode *inode,
933 struct page *locked_page,
771ed689
CM		 934 u64 start, u64 end, int *page_started, int force,
935 unsigned long *nr_written)
be20aa9d 936{
be20aa9d 937 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 938 struct btrfs_trans_handle *trans;
be20aa9d 939 struct extent_buffer *leaf;
be20aa9d 940 struct btrfs_path *path;
80ff3856 941 struct btrfs_file_extent_item *fi;
be20aa9d 942 struct btrfs_key found_key;
80ff3856
YZ		 943 u64 cow_start;
944 u64 cur_offset;
945 u64 extent_end;
5d4f98a2 946 u64 extent_offset;
80ff3856
YZ		 947 u64 disk_bytenr;
948 u64 num_bytes;
949 int extent_type;
950 int ret;
d899e052 951 int type;
80ff3856
YZ		 952 int nocow;
953 int check_prev = 1;
be20aa9d
CM		 954
955 path = btrfs_alloc_path();
956 BUG_ON(!path);
7ea394f1
YZ		 957 trans = btrfs_join_transaction(root, 1);
958 BUG_ON(!trans);
be20aa9d 959
80ff3856
YZ		 960 cow_start = (u64)-1;
961 cur_offset = start;
962 while (1) {
963 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
964 cur_offset, 0);
965 BUG_ON(ret < 0);
966 if (ret > 0 && path->slots[0] > 0 && check_prev) {
967 leaf = path->nodes[0];
968 btrfs_item_key_to_cpu(leaf, &found_key,
969 path->slots[0] - 1);
970 if (found_key.objectid == inode->i_ino &&
971 found_key.type == BTRFS_EXTENT_DATA_KEY)
972 path->slots[0]--;
973 }
974 check_prev = 0;
975next_slot:
976 leaf = path->nodes[0];
977 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
978 ret = btrfs_next_leaf(root, path);
979 if (ret < 0)
980 BUG_ON(1);
981 if (ret > 0)
982 break;
983 leaf = path->nodes[0];
984 }
be20aa9d 985
80ff3856
YZ		 986 nocow = 0;
987 disk_bytenr = 0;
17d217fe 988 num_bytes = 0;
80ff3856
YZ		 989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
990
991 if (found_key.objectid > inode->i_ino ||
992 found_key.type > BTRFS_EXTENT_DATA_KEY ||
993 found_key.offset > end)
994 break;
995
996 if (found_key.offset > cur_offset) {
997 extent_end = found_key.offset;
998 goto out_check;
999 }
1000
1001 fi = btrfs_item_ptr(leaf, path->slots[0],
1002 struct btrfs_file_extent_item);
1003 extent_type = btrfs_file_extent_type(leaf, fi);
1004
d899e052
YZ		 1005 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1006 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1007 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1008 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 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 1025 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5d4f98a2
YZ		 1026 found_key.offset -
1027 extent_offset, disk_bytenr))
17d217fe 1028 goto out_check;
5d4f98a2 1029 disk_bytenr += extent_offset;
17d217fe
YZ		 1030 disk_bytenr += cur_offset - found_key.offset;
1031 num_bytes = min(end + 1, extent_end) - cur_offset;
1032 /*
1033 * force cow if csum exists in the range.
1034 * this ensure that csum for a given extent are
1035 * either valid or do not exist.
1036 */
1037 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1038 goto out_check;
80ff3856
YZ		 1039 nocow = 1;
1040 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1041 extent_end = found_key.offset +
1042 btrfs_file_extent_inline_len(leaf, fi);
1043 extent_end = ALIGN(extent_end, root->sectorsize);
1044 } else {
1045 BUG_ON(1);
1046 }
1047out_check:
1048 if (extent_end <= start) {
1049 path->slots[0]++;
1050 goto next_slot;
1051 }
1052 if (!nocow) {
1053 if (cow_start == (u64)-1)
1054 cow_start = cur_offset;
1055 cur_offset = extent_end;
1056 if (cur_offset > end)
1057 break;
1058 path->slots[0]++;
1059 goto next_slot;
7ea394f1
YZ		 1060 }
1061
1062 btrfs_release_path(root, path);
80ff3856
YZ		 1063 if (cow_start != (u64)-1) {
1064 ret = cow_file_range(inode, locked_page, cow_start,
771ed689
CM		 1065 found_key.offset - 1, page_started,
1066 nr_written, 1);
80ff3856
YZ		 1067 BUG_ON(ret);
1068 cow_start = (u64)-1;
7ea394f1 1069 }
80ff3856 1070
d899e052
YZ		 1071 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1072 struct extent_map *em;
1073 struct extent_map_tree *em_tree;
1074 em_tree = &BTRFS_I(inode)->extent_tree;
1075 em = alloc_extent_map(GFP_NOFS);
1076 em->start = cur_offset;
445a6944 1077 em->orig_start = em->start;
d899e052
YZ		 1078 em->len = num_bytes;
1079 em->block_len = num_bytes;
1080 em->block_start = disk_bytenr;
1081 em->bdev = root->fs_info->fs_devices->latest_bdev;
1082 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1083 while (1) {
1084 spin_lock(&em_tree->lock);
1085 ret = add_extent_mapping(em_tree, em);
1086 spin_unlock(&em_tree->lock);
1087 if (ret != -EEXIST) {
1088 free_extent_map(em);
1089 break;
1090 }
1091 btrfs_drop_extent_cache(inode, em->start,
1092 em->start + em->len - 1, 0);
1093 }
1094 type = BTRFS_ORDERED_PREALLOC;
1095 } else {
1096 type = BTRFS_ORDERED_NOCOW;
1097 }
80ff3856
YZ		 1098
1099 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052
YZ		 1100 num_bytes, num_bytes, type);
1101 BUG_ON(ret);
771ed689 1102
d899e052
YZ		 1103 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1104 cur_offset, cur_offset + num_bytes - 1,
771ed689 1105 locked_page, 1, 1, 1, 0, 0, 0);
80ff3856
YZ		 1106 cur_offset = extent_end;
1107 if (cur_offset > end)
1108 break;
be20aa9d 1109 }
80ff3856
YZ		 1110 btrfs_release_path(root, path);
1111
1112 if (cur_offset <= end && cow_start == (u64)-1)
1113 cow_start = cur_offset;
1114 if (cow_start != (u64)-1) {
1115 ret = cow_file_range(inode, locked_page, cow_start, end,
771ed689 1116 page_started, nr_written, 1);
80ff3856
YZ		 1117 BUG_ON(ret);
1118 }
1119
1120 ret = btrfs_end_transaction(trans, root);
1121 BUG_ON(ret);
7ea394f1 1122 btrfs_free_path(path);
80ff3856 1123 return 0;
be20aa9d
CM		 1124}
1125
d352ac68
CM		 1126/*
1127 * extent_io.c call back to do delayed allocation processing
1128 */
c8b97818 1129static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1130 u64 start, u64 end, int *page_started,
1131 unsigned long *nr_written)
be20aa9d 1132{
be20aa9d 1133 int ret;
7f366cfe 1134 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1135
6cbff00f 1136 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
c8b97818 1137 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1138 page_started, 1, nr_written);
6cbff00f 1139 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
d899e052 1140 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1141 page_started, 0, nr_written);
7f366cfe
CM		 1142 else if (!btrfs_test_opt(root, COMPRESS))
1143 ret = cow_file_range(inode, locked_page, start, end,
1144 page_started, nr_written, 1);
be20aa9d 1145 else
771ed689 1146 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1147 page_started, nr_written);
b888db2b
CM		 1148 return ret;
1149}
1150
d352ac68
CM		 1151/*
1152 * extent_io.c set_bit_hook, used to track delayed allocation
1153 * bytes in this file, and to maintain the list of inodes that
1154 * have pending delalloc work to be done.
1155 */
b2950863 1156static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1157 unsigned long old, unsigned long bits)
291d673e 1158{
75eff68e
CM		 1159 /*
1160 * set_bit and clear bit hooks normally require _irqsave/restore
1161 * but in this case, we are only testeing for the DELALLOC
1162 * bit, which is only set or cleared with irqs on
1163 */
b0c68f8b 1164 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1165 struct btrfs_root *root = BTRFS_I(inode)->root;
6a63209f 1166 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
75eff68e 1167 spin_lock(&root->fs_info->delalloc_lock);
9069218d 1168 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
291d673e 1169 root->fs_info->delalloc_bytes += end - start + 1;
ea8c2819
CM		 1170 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1171 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1172 &root->fs_info->delalloc_inodes);
1173 }
75eff68e 1174 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1175 }
1176 return 0;
1177}
1178
d352ac68
CM		 1179/*
1180 * extent_io.c clear_bit_hook, see set_bit_hook for why
1181 */
b2950863 1182static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
b0c68f8b 1183 unsigned long old, unsigned long bits)
291d673e 1184{
75eff68e
CM		 1185 /*
1186 * set_bit and clear bit hooks normally require _irqsave/restore
1187 * but in this case, we are only testeing for the DELALLOC
1188 * bit, which is only set or cleared with irqs on
1189 */
b0c68f8b 1190 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
291d673e 1191 struct btrfs_root *root = BTRFS_I(inode)->root;
bcbfce8a 1192
75eff68e 1193 spin_lock(&root->fs_info->delalloc_lock);
b0c68f8b 1194 if (end - start + 1 > root->fs_info->delalloc_bytes) {
d397712b
CM		 1195 printk(KERN_INFO "btrfs warning: delalloc account "
1196 "%llu %llu\n",
1197 (unsigned long long)end - start + 1,
1198 (unsigned long long)
1199 root->fs_info->delalloc_bytes);
6a63209f 1200 btrfs_delalloc_free_space(root, inode, (u64)-1);
b0c68f8b 1201 root->fs_info->delalloc_bytes = 0;
9069218d 1202 BTRFS_I(inode)->delalloc_bytes = 0;
b0c68f8b 1203 } else {
6a63209f
JB		 1204 btrfs_delalloc_free_space(root, inode,
1205 end - start + 1);
b0c68f8b 1206 root->fs_info->delalloc_bytes -= end - start + 1;
9069218d 1207 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
b0c68f8b 1208 }
ea8c2819
CM		 1209 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1210 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1211 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1212 }
75eff68e 1213 spin_unlock(&root->fs_info->delalloc_lock);
291d673e
CM		 1214 }
1215 return 0;
1216}
1217
d352ac68
CM		 1218/*
1219 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1220 * we don't create bios that span stripes or chunks
1221 */
239b14b3 1222int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1223 size_t size, struct bio *bio,
1224 unsigned long bio_flags)
239b14b3
CM		 1225{
1226 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1227 struct btrfs_mapping_tree *map_tree;
a62b9401 1228 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1229 u64 length = 0;
1230 u64 map_length;
239b14b3
CM		 1231 int ret;
1232
771ed689
CM		 1233 if (bio_flags & EXTENT_BIO_COMPRESSED)
1234 return 0;
1235
f2d8d74d 1236 length = bio->bi_size;
239b14b3
CM		 1237 map_tree = &root->fs_info->mapping_tree;
1238 map_length = length;
cea9e445 1239 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1240 &map_length, NULL, 0);
cea9e445 1241
d397712b 1242 if (map_length < length + size)
239b14b3 1243 return 1;
239b14b3
CM		 1244 return 0;
1245}
1246
d352ac68
CM		 1247/*
1248 * in order to insert checksums into the metadata in large chunks,
1249 * we wait until bio submission time. All the pages in the bio are
1250 * checksummed and sums are attached onto the ordered extent record.
1251 *
1252 * At IO completion time the cums attached on the ordered extent record
1253 * are inserted into the btree
1254 */
d397712b
CM		 1255static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1256 struct bio *bio, int mirror_num,
1257 unsigned long bio_flags)
065631f6 1258{
065631f6 1259 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1260 int ret = 0;
e015640f 1261
d20f7043 1262 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
44b8bd7e 1263 BUG_ON(ret);
4a69a410
CM		 1264 return 0;
1265}
e015640f 1266
4a69a410
CM		 1267/*
1268 * in order to insert checksums into the metadata in large chunks,
1269 * we wait until bio submission time. All the pages in the bio are
1270 * checksummed and sums are attached onto the ordered extent record.
1271 *
1272 * At IO completion time the cums attached on the ordered extent record
1273 * are inserted into the btree
1274 */
b2950863 1275static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
4a69a410
CM		 1276 int mirror_num, unsigned long bio_flags)
1277{
1278 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1279 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1280}
1281
d352ac68 1282/*
cad321ad
CM		 1283 * extent_io.c submission hook. This does the right thing for csum calculation
1284 * on write, or reading the csums from the tree before a read
d352ac68 1285 */
b2950863 1286static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 1287 int mirror_num, unsigned long bio_flags)
44b8bd7e
CM		 1288{
1289 struct btrfs_root *root = BTRFS_I(inode)->root;
1290 int ret = 0;
19b9bdb0 1291 int skip_sum;
44b8bd7e 1292
6cbff00f 1293 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1294
e6dcd2dc
CM		 1295 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1296 BUG_ON(ret);
065631f6 1297
4d1b5fb4 1298 if (!(rw & (1 << BIO_RW))) {
d20f7043 1299 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1300 return btrfs_submit_compressed_read(inode, bio,
1301 mirror_num, bio_flags);
d20f7043
CM		 1302 } else if (!skip_sum)
1303 btrfs_lookup_bio_sums(root, inode, bio, NULL);
4d1b5fb4 1304 goto mapit;
19b9bdb0 1305 } else if (!skip_sum) {
17d217fe
YZ		 1306 /* csum items have already been cloned */
1307 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1308 goto mapit;
19b9bdb0
CM		 1309 /* we're doing a write, do the async checksumming */
1310 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1311 inode, rw, bio, mirror_num,
4a69a410
CM		 1312 bio_flags, __btrfs_submit_bio_start,
1313 __btrfs_submit_bio_done);
19b9bdb0
CM		 1314 }
1315
0b86a832 1316mapit:
8b712842 1317 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1318}
6885f308 1319
d352ac68
CM		 1320/*
1321 * given a list of ordered sums record them in the inode. This happens
1322 * at IO completion time based on sums calculated at bio submission time.
1323 */
ba1da2f4 1324static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1325 struct inode *inode, u64 file_offset,
1326 struct list_head *list)
1327{
e6dcd2dc
CM		 1328 struct btrfs_ordered_sum *sum;
1329
1330 btrfs_set_trans_block_group(trans, inode);
c6e30871
QF		 1331
1332 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1333 btrfs_csum_file_blocks(trans,
1334 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1335 }
1336 return 0;
1337}
1338
ea8c2819
CM		 1339int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340{
d397712b 1341 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
771ed689 1342 WARN_ON(1);
ea8c2819
CM		 1343 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344 GFP_NOFS);
1345}
1346
d352ac68 1347/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1348struct btrfs_writepage_fixup {
1349 struct page *page;
1350 struct btrfs_work work;
1351};
1352
b2950863 1353static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1354{
1355 struct btrfs_writepage_fixup *fixup;
1356 struct btrfs_ordered_extent *ordered;
1357 struct page *page;
1358 struct inode *inode;
1359 u64 page_start;
1360 u64 page_end;
1361
1362 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363 page = fixup->page;
4a096752 1364again:
247e743c
CM		 1365 lock_page(page);
1366 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367 ClearPageChecked(page);
1368 goto out_page;
1369 }
1370
1371 inode = page->mapping->host;
1372 page_start = page_offset(page);
1373 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
4a096752
CM		 1376
1377 /* already ordered? We're done */
1378 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379 EXTENT_ORDERED, 0)) {
247e743c 1380 goto out;
4a096752
CM		 1381 }
1382
1383 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384 if (ordered) {
1385 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386 page_end, GFP_NOFS);
1387 unlock_page(page);
1388 btrfs_start_ordered_extent(inode, ordered, 1);
1389 goto again;
1390 }
247e743c 1391
ea8c2819 1392 btrfs_set_extent_delalloc(inode, page_start, page_end);
247e743c
CM		 1393 ClearPageChecked(page);
1394out:
1395 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396out_page:
1397 unlock_page(page);
1398 page_cache_release(page);
1399}
1400
1401/*
1402 * There are a few paths in the higher layers of the kernel that directly
1403 * set the page dirty bit without asking the filesystem if it is a
1404 * good idea. This causes problems because we want to make sure COW
1405 * properly happens and the data=ordered rules are followed.
1406 *
c8b97818 1407 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1408 * hasn't been properly setup for IO. We kick off an async process
1409 * to fix it up. The async helper will wait for ordered extents, set
1410 * the delalloc bit and make it safe to write the page.
1411 */
b2950863 1412static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1413{
1414 struct inode *inode = page->mapping->host;
1415 struct btrfs_writepage_fixup *fixup;
1416 struct btrfs_root *root = BTRFS_I(inode)->root;
1417 int ret;
1418
1419 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420 EXTENT_ORDERED, 0);
1421 if (ret)
1422 return 0;
1423
1424 if (PageChecked(page))
1425 return -EAGAIN;
1426
1427 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428 if (!fixup)
1429 return -EAGAIN;
f421950f 1430
247e743c
CM		 1431 SetPageChecked(page);
1432 page_cache_get(page);
1433 fixup->work.func = btrfs_writepage_fixup_worker;
1434 fixup->page = page;
1435 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436 return -EAGAIN;
1437}
1438
d899e052
YZ		 1439static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440 struct inode *inode, u64 file_pos,
1441 u64 disk_bytenr, u64 disk_num_bytes,
1442 u64 num_bytes, u64 ram_bytes,
e980b50c 1443 u64 locked_end,
d899e052
YZ		 1444 u8 compression, u8 encryption,
1445 u16 other_encoding, int extent_type)
1446{
1447 struct btrfs_root *root = BTRFS_I(inode)->root;
1448 struct btrfs_file_extent_item *fi;
1449 struct btrfs_path *path;
1450 struct extent_buffer *leaf;
1451 struct btrfs_key ins;
1452 u64 hint;
1453 int ret;
1454
1455 path = btrfs_alloc_path();
1456 BUG_ON(!path);
1457
b9473439 1458 path->leave_spinning = 1;
d899e052 1459 ret = btrfs_drop_extents(trans, root, inode, file_pos,
e980b50c
CM		 1460 file_pos + num_bytes, locked_end,
1461 file_pos, &hint);
d899e052
YZ		 1462 BUG_ON(ret);
1463
1464 ins.objectid = inode->i_ino;
1465 ins.offset = file_pos;
1466 ins.type = BTRFS_EXTENT_DATA_KEY;
1467 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1468 BUG_ON(ret);
1469 leaf = path->nodes[0];
1470 fi = btrfs_item_ptr(leaf, path->slots[0],
1471 struct btrfs_file_extent_item);
1472 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1473 btrfs_set_file_extent_type(leaf, fi, extent_type);
1474 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1475 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1476 btrfs_set_file_extent_offset(leaf, fi, 0);
1477 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1478 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1479 btrfs_set_file_extent_compression(leaf, fi, compression);
1480 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1481 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439
CM		 1482
1483 btrfs_unlock_up_safe(path, 1);
1484 btrfs_set_lock_blocking(leaf);
1485
d899e052
YZ		 1486 btrfs_mark_buffer_dirty(leaf);
1487
1488 inode_add_bytes(inode, num_bytes);
1489 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1490
1491 ins.objectid = disk_bytenr;
1492 ins.offset = disk_num_bytes;
1493 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1494 ret = btrfs_alloc_reserved_file_extent(trans, root,
1495 root->root_key.objectid,
1496 inode->i_ino, file_pos, &ins);
d899e052 1497 BUG_ON(ret);
d899e052 1498 btrfs_free_path(path);
b9473439 1499
d899e052
YZ		 1500 return 0;
1501}
1502
5d13a98f
CM		 1503/*
1504 * helper function for btrfs_finish_ordered_io, this
1505 * just reads in some of the csum leaves to prime them into ram
1506 * before we start the transaction. It limits the amount of btree
1507 * reads required while inside the transaction.
1508 */
1509static noinline void reada_csum(struct btrfs_root *root,
1510 struct btrfs_path *path,
1511 struct btrfs_ordered_extent *ordered_extent)
1512{
1513 struct btrfs_ordered_sum *sum;
1514 u64 bytenr;
1515
1516 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1517 list);
1518 bytenr = sum->sums[0].bytenr;
1519
1520 /*
1521 * we don't care about the results, the point of this search is
1522 * just to get the btree leaves into ram
1523 */
1524 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1525}
1526
d352ac68
CM		 1527/* as ordered data IO finishes, this gets called so we can finish
1528 * an ordered extent if the range of bytes in the file it covers are
1529 * fully written.
1530 */
211f90e6 1531static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
e6dcd2dc 1532{
e6dcd2dc
CM		 1533 struct btrfs_root *root = BTRFS_I(inode)->root;
1534 struct btrfs_trans_handle *trans;
5d13a98f 1535 struct btrfs_ordered_extent *ordered_extent = NULL;
e6dcd2dc 1536 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
b7ec40d7 1537 struct btrfs_path *path;
d899e052 1538 int compressed = 0;
e6dcd2dc
CM		 1539 int ret;
1540
1541 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
ba1da2f4 1542 if (!ret)
e6dcd2dc 1543 return 0;
e6dcd2dc 1544
b7ec40d7
CM		 1545 /*
1546 * before we join the transaction, try to do some of our IO.
1547 * This will limit the amount of IO that we have to do with
1548 * the transaction running. We're unlikely to need to do any
1549 * IO if the file extents are new, the disk_i_size checks
1550 * covers the most common case.
1551 */
1552 if (start < BTRFS_I(inode)->disk_i_size) {
1553 path = btrfs_alloc_path();
1554 if (path) {
1555 ret = btrfs_lookup_file_extent(NULL, root, path,
1556 inode->i_ino,
1557 start, 0);
5d13a98f
CM		 1558 ordered_extent = btrfs_lookup_ordered_extent(inode,
1559 start);
1560 if (!list_empty(&ordered_extent->list)) {
1561 btrfs_release_path(root, path);
1562 reada_csum(root, path, ordered_extent);
1563 }
b7ec40d7
CM		 1564 btrfs_free_path(path);
1565 }
1566 }
1567
f9295749 1568 trans = btrfs_join_transaction(root, 1);
e6dcd2dc 1569
5d13a98f
CM		 1570 if (!ordered_extent)
1571 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
e6dcd2dc 1572 BUG_ON(!ordered_extent);
7ea394f1
YZ		 1573 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1574 goto nocow;
e6dcd2dc
CM		 1575
1576 lock_extent(io_tree, ordered_extent->file_offset,
1577 ordered_extent->file_offset + ordered_extent->len - 1,
1578 GFP_NOFS);
1579
c8b97818 1580 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
d899e052
YZ		 1581 compressed = 1;
1582 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1583 BUG_ON(compressed);
1584 ret = btrfs_mark_extent_written(trans, root, inode,
1585 ordered_extent->file_offset,
1586 ordered_extent->file_offset +
1587 ordered_extent->len);
1588 BUG_ON(ret);
1589 } else {
1590 ret = insert_reserved_file_extent(trans, inode,
1591 ordered_extent->file_offset,
1592 ordered_extent->start,
1593 ordered_extent->disk_len,
1594 ordered_extent->len,
1595 ordered_extent->len,
e980b50c
CM		 1596 ordered_extent->file_offset +
1597 ordered_extent->len,
d899e052
YZ		 1598 compressed, 0, 0,
1599 BTRFS_FILE_EXTENT_REG);
1600 BUG_ON(ret);
1601 }
e6dcd2dc
CM		 1602 unlock_extent(io_tree, ordered_extent->file_offset,
1603 ordered_extent->file_offset + ordered_extent->len - 1,
1604 GFP_NOFS);
7ea394f1 1605nocow:
e6dcd2dc
CM		 1606 add_pending_csums(trans, inode, ordered_extent->file_offset,
1607 &ordered_extent->list);
1608
34353029 1609 mutex_lock(&BTRFS_I(inode)->extent_mutex);
dbe674a9 1610 btrfs_ordered_update_i_size(inode, ordered_extent);
e02119d5 1611 btrfs_update_inode(trans, root, inode);
e6dcd2dc 1612 btrfs_remove_ordered_extent(inode, ordered_extent);
34353029 1613 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
7f3c74fb 1614
e6dcd2dc
CM		 1615 /* once for us */
1616 btrfs_put_ordered_extent(ordered_extent);
1617 /* once for the tree */
1618 btrfs_put_ordered_extent(ordered_extent);
1619
e6dcd2dc
CM		 1620 btrfs_end_transaction(trans, root);
1621 return 0;
1622}
1623
b2950863 1624static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1625 struct extent_state *state, int uptodate)
1626{
1627 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1628}
1629
d352ac68
CM		 1630/*
1631 * When IO fails, either with EIO or csum verification fails, we
1632 * try other mirrors that might have a good copy of the data. This
1633 * io_failure_record is used to record state as we go through all the
1634 * mirrors. If another mirror has good data, the page is set up to date
1635 * and things continue. If a good mirror can't be found, the original
1636 * bio end_io callback is called to indicate things have failed.
1637 */
7e38326f
CM		 1638struct io_failure_record {
1639 struct page *page;
1640 u64 start;
1641 u64 len;
1642 u64 logical;
d20f7043 1643 unsigned long bio_flags;
7e38326f
CM		 1644 int last_mirror;
1645};
1646
b2950863 1647static int btrfs_io_failed_hook(struct bio *failed_bio,
1259ab75
CM		 1648 struct page *page, u64 start, u64 end,
1649 struct extent_state *state)
7e38326f
CM		 1650{
1651 struct io_failure_record *failrec = NULL;
1652 u64 private;
1653 struct extent_map *em;
1654 struct inode *inode = page->mapping->host;
1655 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
3b951516 1656 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7e38326f
CM		 1657 struct bio *bio;
1658 int num_copies;
1659 int ret;
1259ab75 1660 int rw;
7e38326f
CM		 1661 u64 logical;
1662
1663 ret = get_state_private(failure_tree, start, &private);
1664 if (ret) {
7e38326f
CM		 1665 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1666 if (!failrec)
1667 return -ENOMEM;
1668 failrec->start = start;
1669 failrec->len = end - start + 1;
1670 failrec->last_mirror = 0;
d20f7043 1671 failrec->bio_flags = 0;
7e38326f 1672
3b951516
CM		 1673 spin_lock(&em_tree->lock);
1674 em = lookup_extent_mapping(em_tree, start, failrec->len);
1675 if (em->start > start || em->start + em->len < start) {
1676 free_extent_map(em);
1677 em = NULL;
1678 }
1679 spin_unlock(&em_tree->lock);
7e38326f
CM		 1680
1681 if (!em || IS_ERR(em)) {
1682 kfree(failrec);
1683 return -EIO;
1684 }
1685 logical = start - em->start;
1686 logical = em->block_start + logical;
d20f7043
CM		 1687 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1688 logical = em->block_start;
1689 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1690 }
7e38326f
CM		 1691 failrec->logical = logical;
1692 free_extent_map(em);
1693 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1694 EXTENT_DIRTY, GFP_NOFS);
587f7704
CM		 1695 set_state_private(failure_tree, start,
1696 (u64)(unsigned long)failrec);
7e38326f 1697 } else {
587f7704 1698 failrec = (struct io_failure_record *)(unsigned long)private;
7e38326f
CM		 1699 }
1700 num_copies = btrfs_num_copies(
1701 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1702 failrec->logical, failrec->len);
1703 failrec->last_mirror++;
1704 if (!state) {
cad321ad 1705 spin_lock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1706 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1707 failrec->start,
1708 EXTENT_LOCKED);
1709 if (state && state->start != failrec->start)
1710 state = NULL;
cad321ad 1711 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
7e38326f
CM		 1712 }
1713 if (!state || failrec->last_mirror > num_copies) {
1714 set_state_private(failure_tree, failrec->start, 0);
1715 clear_extent_bits(failure_tree, failrec->start,
1716 failrec->start + failrec->len - 1,
1717 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1718 kfree(failrec);
1719 return -EIO;
1720 }
1721 bio = bio_alloc(GFP_NOFS, 1);
1722 bio->bi_private = state;
1723 bio->bi_end_io = failed_bio->bi_end_io;
1724 bio->bi_sector = failrec->logical >> 9;
1725 bio->bi_bdev = failed_bio->bi_bdev;
e1c4b745 1726 bio->bi_size = 0;
d20f7043 1727
7e38326f 1728 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1259ab75
CM		 1729 if (failed_bio->bi_rw & (1 << BIO_RW))
1730 rw = WRITE;
1731 else
1732 rw = READ;
1733
1734 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
c8b97818 1735 failrec->last_mirror,
d20f7043 1736 failrec->bio_flags);
1259ab75
CM		 1737 return 0;
1738}
1739
d352ac68
CM		 1740/*
1741 * each time an IO finishes, we do a fast check in the IO failure tree
1742 * to see if we need to process or clean up an io_failure_record
1743 */
b2950863 1744static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1259ab75
CM		 1745{
1746 u64 private;
1747 u64 private_failure;
1748 struct io_failure_record *failure;
1749 int ret;
1750
1751 private = 0;
1752 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1753 (u64)-1, 1, EXTENT_DIRTY)) {
1754 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1755 start, &private_failure);
1756 if (ret == 0) {
1757 failure = (struct io_failure_record *)(unsigned long)
1758 private_failure;
1759 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1760 failure->start, 0);
1761 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1762 failure->start,
1763 failure->start + failure->len - 1,
1764 EXTENT_DIRTY | EXTENT_LOCKED,
1765 GFP_NOFS);
1766 kfree(failure);
1767 }
1768 }
7e38326f
CM		 1769 return 0;
1770}
1771
d352ac68
CM		 1772/*
1773 * when reads are done, we need to check csums to verify the data is correct
1774 * if there's a match, we allow the bio to finish. If not, we go through
1775 * the io_failure_record routines to find good copies
1776 */
b2950863 1777static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
70dec807 1778 struct extent_state *state)
07157aac 1779{
35ebb934 1780 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 1781 struct inode *inode = page->mapping->host;
d1310b2e 1782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 1783 char *kaddr;
aadfeb6e 1784 u64 private = ~(u32)0;
07157aac 1785 int ret;
ff79f819
CM		 1786 struct btrfs_root *root = BTRFS_I(inode)->root;
1787 u32 csum = ~(u32)0;
d1310b2e 1788
d20f7043
CM		 1789 if (PageChecked(page)) {
1790 ClearPageChecked(page);
1791 goto good;
1792 }
6cbff00f
CH		 1793
1794 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
17d217fe
YZ		 1795 return 0;
1796
1797 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1798 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1799 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1800 GFP_NOFS);
b6cda9bc 1801 return 0;
17d217fe 1802 }
d20f7043 1803
c2e639f0 1804 if (state && state->start == start) {
70dec807
CM		 1805 private = state->private;
1806 ret = 0;
1807 } else {
1808 ret = get_state_private(io_tree, start, &private);
1809 }
9ab86c8e 1810 kaddr = kmap_atomic(page, KM_USER0);
d397712b 1811 if (ret)
07157aac 1812 goto zeroit;
d397712b 1813
ff79f819
CM		 1814 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1815 btrfs_csum_final(csum, (char *)&csum);
d397712b 1816 if (csum != private)
07157aac 1817 goto zeroit;
d397712b 1818
9ab86c8e 1819 kunmap_atomic(kaddr, KM_USER0);
d20f7043 1820good:
7e38326f
CM		 1821 /* if the io failure tree for this inode is non-empty,
1822 * check to see if we've recovered from a failed IO
1823 */
1259ab75 1824 btrfs_clean_io_failures(inode, start);
07157aac
CM		 1825 return 0;
1826
1827zeroit:
193f284d
CM		 1828 if (printk_ratelimit()) {
1829 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1830 "private %llu\n", page->mapping->host->i_ino,
1831 (unsigned long long)start, csum,
1832 (unsigned long long)private);
1833 }
db94535d
CM		 1834 memset(kaddr + offset, 1, end - start + 1);
1835 flush_dcache_page(page);
9ab86c8e 1836 kunmap_atomic(kaddr, KM_USER0);
3b951516
CM		 1837 if (private == 0)
1838 return 0;
7e38326f 1839 return -EIO;
07157aac 1840}
b888db2b 1841
7b128766
JB		 1842/*
1843 * This creates an orphan entry for the given inode in case something goes
1844 * wrong in the middle of an unlink/truncate.
1845 */
1846int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1847{
1848 struct btrfs_root *root = BTRFS_I(inode)->root;
1849 int ret = 0;
1850
bcc63abb 1851 spin_lock(&root->list_lock);
7b128766
JB		 1852
1853 /* already on the orphan list, we're good */
1854 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1855 spin_unlock(&root->list_lock);
7b128766
JB		 1856 return 0;
1857 }
1858
1859 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1860
bcc63abb 1861 spin_unlock(&root->list_lock);
7b128766
JB		 1862
1863 /*
1864 * insert an orphan item to track this unlinked/truncated file
1865 */
1866 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1867
1868 return ret;
1869}
1870
1871/*
1872 * We have done the truncate/delete so we can go ahead and remove the orphan
1873 * item for this particular inode.
1874 */
1875int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1876{
1877 struct btrfs_root *root = BTRFS_I(inode)->root;
1878 int ret = 0;
1879
bcc63abb 1880 spin_lock(&root->list_lock);
7b128766
JB		 1881
1882 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
bcc63abb 1883 spin_unlock(&root->list_lock);
7b128766
JB		 1884 return 0;
1885 }
1886
1887 list_del_init(&BTRFS_I(inode)->i_orphan);
1888 if (!trans) {
bcc63abb 1889 spin_unlock(&root->list_lock);
7b128766
JB		 1890 return 0;
1891 }
1892
bcc63abb 1893 spin_unlock(&root->list_lock);
7b128766
JB		 1894
1895 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1896
1897 return ret;
1898}
1899
1900/*
1901 * this cleans up any orphans that may be left on the list from the last use
1902 * of this root.
1903 */
1904void btrfs_orphan_cleanup(struct btrfs_root *root)
1905{
1906 struct btrfs_path *path;
1907 struct extent_buffer *leaf;
1908 struct btrfs_item *item;
1909 struct btrfs_key key, found_key;
1910 struct btrfs_trans_handle *trans;
1911 struct inode *inode;
1912 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1913
7b128766
JB		 1914 path = btrfs_alloc_path();
1915 if (!path)
1916 return;
1917 path->reada = -1;
1918
1919 key.objectid = BTRFS_ORPHAN_OBJECTID;
1920 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1921 key.offset = (u64)-1;
1922
7b128766
JB		 1923
1924 while (1) {
1925 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1926 if (ret < 0) {
1927 printk(KERN_ERR "Error searching slot for orphan: %d"
1928 "\n", ret);
1929 break;
1930 }
1931
1932 /*
1933 * if ret == 0 means we found what we were searching for, which
1934 * is weird, but possible, so only screw with path if we didnt
1935 * find the key and see if we have stuff that matches
1936 */
1937 if (ret > 0) {
1938 if (path->slots[0] == 0)
1939 break;
1940 path->slots[0]--;
1941 }
1942
1943 /* pull out the item */
1944 leaf = path->nodes[0];
1945 item = btrfs_item_nr(leaf, path->slots[0]);
1946 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1947
1948 /* make sure the item matches what we want */
1949 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1950 break;
1951 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1952 break;
1953
1954 /* release the path since we're done with it */
1955 btrfs_release_path(root, path);
1956
1957 /*
1958 * this is where we are basically btrfs_lookup, without the
1959 * crossing root thing. we store the inode number in the
1960 * offset of the orphan item.
1961 */
5d4f98a2
YZ		 1962 found_key.objectid = found_key.offset;
1963 found_key.type = BTRFS_INODE_ITEM_KEY;
1964 found_key.offset = 0;
1965 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1966 if (IS_ERR(inode))
7b128766
JB		 1967 break;
1968
7b128766
JB		 1969 /*
1970 * add this inode to the orphan list so btrfs_orphan_del does
1971 * the proper thing when we hit it
1972 */
bcc63abb 1973 spin_lock(&root->list_lock);
7b128766 1974 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
bcc63abb 1975 spin_unlock(&root->list_lock);
7b128766
JB		 1976
1977 /*
1978 * if this is a bad inode, means we actually succeeded in
1979 * removing the inode, but not the orphan record, which means
1980 * we need to manually delete the orphan since iput will just
1981 * do a destroy_inode
1982 */
1983 if (is_bad_inode(inode)) {
5b21f2ed 1984 trans = btrfs_start_transaction(root, 1);
7b128766 1985 btrfs_orphan_del(trans, inode);
5b21f2ed 1986 btrfs_end_transaction(trans, root);
7b128766
JB		 1987 iput(inode);
1988 continue;
1989 }
1990
1991 /* if we have links, this was a truncate, lets do that */
1992 if (inode->i_nlink) {
1993 nr_truncate++;
1994 btrfs_truncate(inode);
1995 } else {
1996 nr_unlink++;
1997 }
1998
1999 /* this will do delete_inode and everything for us */
2000 iput(inode);
2001 }
2002
2003 if (nr_unlink)
2004 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2005 if (nr_truncate)
2006 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2007
2008 btrfs_free_path(path);
7b128766
JB		 2009}
2010
46a53cca
CM		 2011/*
2012 * very simple check to peek ahead in the leaf looking for xattrs. If we
2013 * don't find any xattrs, we know there can't be any acls.
2014 *
2015 * slot is the slot the inode is in, objectid is the objectid of the inode
2016 */
2017static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2018 int slot, u64 objectid)
2019{
2020 u32 nritems = btrfs_header_nritems(leaf);
2021 struct btrfs_key found_key;
2022 int scanned = 0;
2023
2024 slot++;
2025 while (slot < nritems) {
2026 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2027
2028 /* we found a different objectid, there must not be acls */
2029 if (found_key.objectid != objectid)
2030 return 0;
2031
2032 /* we found an xattr, assume we've got an acl */
2033 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2034 return 1;
2035
2036 /*
2037 * we found a key greater than an xattr key, there can't
2038 * be any acls later on
2039 */
2040 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2041 return 0;
2042
2043 slot++;
2044 scanned++;
2045
2046 /*
2047 * it goes inode, inode backrefs, xattrs, extents,
2048 * so if there are a ton of hard links to an inode there can
2049 * be a lot of backrefs. Don't waste time searching too hard,
2050 * this is just an optimization
2051 */
2052 if (scanned >= 8)
2053 break;
2054 }
2055 /* we hit the end of the leaf before we found an xattr or
2056 * something larger than an xattr. We have to assume the inode
2057 * has acls
2058 */
2059 return 1;
2060}
2061
d352ac68
CM		 2062/*
2063 * read an inode from the btree into the in-memory inode
2064 */
5d4f98a2 2065static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2066{
2067 struct btrfs_path *path;
5f39d397 2068 struct extent_buffer *leaf;
39279cc3 2069 struct btrfs_inode_item *inode_item;
0b86a832 2070 struct btrfs_timespec *tspec;
39279cc3
CM		 2071 struct btrfs_root *root = BTRFS_I(inode)->root;
2072 struct btrfs_key location;
46a53cca 2073 int maybe_acls;
39279cc3 2074 u64 alloc_group_block;
618e21d5 2075 u32 rdev;
39279cc3
CM		 2076 int ret;
2077
2078 path = btrfs_alloc_path();
2079 BUG_ON(!path);
39279cc3 2080 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2081
39279cc3 2082 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2083 if (ret)
39279cc3 2084 goto make_bad;
39279cc3 2085
5f39d397
CM		 2086 leaf = path->nodes[0];
2087 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2088 struct btrfs_inode_item);
2089
2090 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2091 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2092 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2093 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
dbe674a9 2094 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2095
2096 tspec = btrfs_inode_atime(inode_item);
2097 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2098 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2099
2100 tspec = btrfs_inode_mtime(inode_item);
2101 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2102 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2103
2104 tspec = btrfs_inode_ctime(inode_item);
2105 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2106 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2107
a76a3cd4 2108 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2109 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
c3027eb5 2110 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2111 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2112 inode->i_rdev = 0;
5f39d397
CM		 2113 rdev = btrfs_inode_rdev(leaf, inode_item);
2114
aec7477b 2115 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2116 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
aec7477b 2117
5f39d397 2118 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
b4ce94de 2119
46a53cca
CM		 2120 /*
2121 * try to precache a NULL acl entry for files that don't have
2122 * any xattrs or acls
2123 */
2124 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2125 if (!maybe_acls) {
2126 BTRFS_I(inode)->i_acl = NULL;
2127 BTRFS_I(inode)->i_default_acl = NULL;
2128 }
2129
d2fb3437
YZ		 2130 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2131 alloc_group_block, 0);
39279cc3
CM		 2132 btrfs_free_path(path);
2133 inode_item = NULL;
2134
39279cc3 2135 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2136 case S_IFREG:
2137 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2138 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2139 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2140 inode->i_fop = &btrfs_file_operations;
2141 inode->i_op = &btrfs_file_inode_operations;
2142 break;
2143 case S_IFDIR:
2144 inode->i_fop = &btrfs_dir_file_operations;
2145 if (root == root->fs_info->tree_root)
2146 inode->i_op = &btrfs_dir_ro_inode_operations;
2147 else
2148 inode->i_op = &btrfs_dir_inode_operations;
2149 break;
2150 case S_IFLNK:
2151 inode->i_op = &btrfs_symlink_inode_operations;
2152 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2153 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2154 break;
618e21d5 2155 default:
0279b4cd 2156 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2157 init_special_inode(inode, inode->i_mode, rdev);
2158 break;
39279cc3 2159 }
6cbff00f
CH		 2160
2161 btrfs_update_iflags(inode);
39279cc3
CM		 2162 return;
2163
2164make_bad:
39279cc3 2165 btrfs_free_path(path);
39279cc3
CM		 2166 make_bad_inode(inode);
2167}
2168
d352ac68
CM		 2169/*
2170 * given a leaf and an inode, copy the inode fields into the leaf
2171 */
e02119d5
CM		 2172static void fill_inode_item(struct btrfs_trans_handle *trans,
2173 struct extent_buffer *leaf,
5f39d397 2174 struct btrfs_inode_item *item,
39279cc3
CM		 2175 struct inode *inode)
2176{
5f39d397
CM		 2177 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2178 btrfs_set_inode_gid(leaf, item, inode->i_gid);
dbe674a9 2179 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2180 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2181 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2182
2183 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2184 inode->i_atime.tv_sec);
2185 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2186 inode->i_atime.tv_nsec);
2187
2188 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2189 inode->i_mtime.tv_sec);
2190 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2191 inode->i_mtime.tv_nsec);
2192
2193 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2194 inode->i_ctime.tv_sec);
2195 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2196 inode->i_ctime.tv_nsec);
2197
a76a3cd4 2198 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2199 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
c3027eb5 2200 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
e02119d5 2201 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2202 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2203 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d2fb3437 2204 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
39279cc3
CM		 2205}
2206
d352ac68
CM		 2207/*
2208 * copy everything in the in-memory inode into the btree.
2209 */
d397712b
CM		 2210noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2211 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2212{
2213 struct btrfs_inode_item *inode_item;
2214 struct btrfs_path *path;
5f39d397 2215 struct extent_buffer *leaf;
39279cc3
CM		 2216 int ret;
2217
2218 path = btrfs_alloc_path();
2219 BUG_ON(!path);
b9473439 2220 path->leave_spinning = 1;
39279cc3
CM		 2221 ret = btrfs_lookup_inode(trans, root, path,
2222 &BTRFS_I(inode)->location, 1);
2223 if (ret) {
2224 if (ret > 0)
2225 ret = -ENOENT;
2226 goto failed;
2227 }
2228
b4ce94de 2229 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2230 leaf = path->nodes[0];
2231 inode_item = btrfs_item_ptr(leaf, path->slots[0],
39279cc3
CM		 2232 struct btrfs_inode_item);
2233
e02119d5 2234 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2235 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2236 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2237 ret = 0;
2238failed:
39279cc3
CM		 2239 btrfs_free_path(path);
2240 return ret;
2241}
2242
2243
d352ac68
CM		 2244/*
2245 * unlink helper that gets used here in inode.c and in the tree logging
2246 * recovery code. It remove a link in a directory with a given name, and
2247 * also drops the back refs in the inode to the directory
2248 */
e02119d5
CM		 2249int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2250 struct btrfs_root *root,
2251 struct inode *dir, struct inode *inode,
2252 const char *name, int name_len)
39279cc3
CM		 2253{
2254 struct btrfs_path *path;
39279cc3 2255 int ret = 0;
5f39d397 2256 struct extent_buffer *leaf;
39279cc3 2257 struct btrfs_dir_item *di;
5f39d397 2258 struct btrfs_key key;
aec7477b 2259 u64 index;
39279cc3
CM		 2260
2261 path = btrfs_alloc_path();
54aa1f4d
CM		 2262 if (!path) {
2263 ret = -ENOMEM;
2264 goto err;
2265 }
2266
b9473439 2267 path->leave_spinning = 1;
39279cc3
CM		 2268 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2269 name, name_len, -1);
2270 if (IS_ERR(di)) {
2271 ret = PTR_ERR(di);
2272 goto err;
2273 }
2274 if (!di) {
2275 ret = -ENOENT;
2276 goto err;
2277 }
5f39d397
CM		 2278 leaf = path->nodes[0];
2279 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2280 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2281 if (ret)
2282 goto err;
39279cc3
CM		 2283 btrfs_release_path(root, path);
2284
aec7477b 2285 ret = btrfs_del_inode_ref(trans, root, name, name_len,
e02119d5
CM		 2286 inode->i_ino,
2287 dir->i_ino, &index);
aec7477b 2288 if (ret) {
d397712b 2289 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
aec7477b 2290 "inode %lu parent %lu\n", name_len, name,
e02119d5 2291 inode->i_ino, dir->i_ino);
aec7477b
JB		 2292 goto err;
2293 }
2294
39279cc3 2295 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
aec7477b 2296 index, name, name_len, -1);
39279cc3
CM		 2297 if (IS_ERR(di)) {
2298 ret = PTR_ERR(di);
2299 goto err;
2300 }
2301 if (!di) {
2302 ret = -ENOENT;
2303 goto err;
2304 }
2305 ret = btrfs_delete_one_dir_name(trans, root, path, di);
925baedd 2306 btrfs_release_path(root, path);
39279cc3 2307
e02119d5
CM		 2308 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2309 inode, dir->i_ino);
49eb7e46 2310 BUG_ON(ret != 0 && ret != -ENOENT);
e02119d5
CM		 2311
2312 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2313 dir, index);
2314 BUG_ON(ret);
39279cc3
CM		 2315err:
2316 btrfs_free_path(path);
e02119d5
CM		 2317 if (ret)
2318 goto out;
2319
2320 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2321 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2322 btrfs_update_inode(trans, root, dir);
2323 btrfs_drop_nlink(inode);
2324 ret = btrfs_update_inode(trans, root, inode);
e02119d5 2325out:
39279cc3
CM		 2326 return ret;
2327}
2328
2329static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2330{
2331 struct btrfs_root *root;
2332 struct btrfs_trans_handle *trans;
7b128766 2333 struct inode *inode = dentry->d_inode;
39279cc3 2334 int ret;
1832a6d5 2335 unsigned long nr = 0;
39279cc3
CM		 2336
2337 root = BTRFS_I(dir)->root;
1832a6d5 2338
39279cc3 2339 trans = btrfs_start_transaction(root, 1);
5f39d397 2340
39279cc3 2341 btrfs_set_trans_block_group(trans, dir);
12fcfd22
CM		 2342
2343 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2344
e02119d5
CM		 2345 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2346 dentry->d_name.name, dentry->d_name.len);
7b128766
JB		 2347
2348 if (inode->i_nlink == 0)
2349 ret = btrfs_orphan_add(trans, inode);
2350
d3c2fdcf 2351 nr = trans->blocks_used;
5f39d397 2352
89ce8a63 2353 btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2354 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 2355 return ret;
2356}
2357
2358static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2359{
2360 struct inode *inode = dentry->d_inode;
1832a6d5 2361 int err = 0;
39279cc3
CM		 2362 int ret;
2363 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 2364 struct btrfs_trans_handle *trans;
1832a6d5 2365 unsigned long nr = 0;
39279cc3 2366
3394e160
CM		 2367 /*
2368 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2369 * the root of a subvolume or snapshot
2370 */
2371 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2372 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
134d4512 2373 return -ENOTEMPTY;
925baedd 2374 }
134d4512 2375
39279cc3
CM		 2376 trans = btrfs_start_transaction(root, 1);
2377 btrfs_set_trans_block_group(trans, dir);
39279cc3 2378
7b128766
JB		 2379 err = btrfs_orphan_add(trans, inode);
2380 if (err)
2381 goto fail_trans;
2382
39279cc3 2383 /* now the directory is empty */
e02119d5
CM		 2384 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2385 dentry->d_name.name, dentry->d_name.len);
d397712b 2386 if (!err)
dbe674a9 2387 btrfs_i_size_write(inode, 0);
3954401f 2388
7b128766 2389fail_trans:
d3c2fdcf 2390 nr = trans->blocks_used;
89ce8a63 2391 ret = btrfs_end_transaction_throttle(trans, root);
d3c2fdcf 2392 btrfs_btree_balance_dirty(root, nr);
3954401f 2393
39279cc3
CM		 2394 if (ret && !err)
2395 err = ret;
2396 return err;
2397}
2398
d20f7043 2399#if 0
323ac95b
CM		 2400/*
2401 * when truncating bytes in a file, it is possible to avoid reading
2402 * the leaves that contain only checksum items. This can be the
2403 * majority of the IO required to delete a large file, but it must
2404 * be done carefully.
2405 *
2406 * The keys in the level just above the leaves are checked to make sure
2407 * the lowest key in a given leaf is a csum key, and starts at an offset
2408 * after the new size.
2409 *
2410 * Then the key for the next leaf is checked to make sure it also has
2411 * a checksum item for the same file. If it does, we know our target leaf
2412 * contains only checksum items, and it can be safely freed without reading
2413 * it.
2414 *
2415 * This is just an optimization targeted at large files. It may do
2416 * nothing. It will return 0 unless things went badly.
2417 */
2418static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2419 struct btrfs_root *root,
2420 struct btrfs_path *path,
2421 struct inode *inode, u64 new_size)
2422{
2423 struct btrfs_key key;
2424 int ret;
2425 int nritems;
2426 struct btrfs_key found_key;
2427 struct btrfs_key other_key;
5b84e8d6
YZ		 2428 struct btrfs_leaf_ref *ref;
2429 u64 leaf_gen;
2430 u64 leaf_start;
323ac95b
CM		 2431
2432 path->lowest_level = 1;
2433 key.objectid = inode->i_ino;
2434 key.type = BTRFS_CSUM_ITEM_KEY;
2435 key.offset = new_size;
2436again:
2437 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2438 if (ret < 0)
2439 goto out;
2440
2441 if (path->nodes[1] == NULL) {
2442 ret = 0;
2443 goto out;
2444 }
2445 ret = 0;
2446 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2447 nritems = btrfs_header_nritems(path->nodes[1]);
2448
2449 if (!nritems)
2450 goto out;
2451
2452 if (path->slots[1] >= nritems)
2453 goto next_node;
2454
2455 /* did we find a key greater than anything we want to delete? */
2456 if (found_key.objectid > inode->i_ino ||
2457 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2458 goto out;
2459
2460 /* we check the next key in the node to make sure the leave contains
2461 * only checksum items. This comparison doesn't work if our
2462 * leaf is the last one in the node
2463 */
2464 if (path->slots[1] + 1 >= nritems) {
2465next_node:
2466 /* search forward from the last key in the node, this
2467 * will bring us into the next node in the tree
2468 */
2469 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2470
2471 /* unlikely, but we inc below, so check to be safe */
2472 if (found_key.offset == (u64)-1)
2473 goto out;
2474
2475 /* search_forward needs a path with locks held, do the
2476 * search again for the original key. It is possible
2477 * this will race with a balance and return a path that
2478 * we could modify, but this drop is just an optimization
2479 * and is allowed to miss some leaves.
2480 */
2481 btrfs_release_path(root, path);
2482 found_key.offset++;
2483
2484 /* setup a max key for search_forward */
2485 other_key.offset = (u64)-1;
2486 other_key.type = key.type;
2487 other_key.objectid = key.objectid;
2488
2489 path->keep_locks = 1;
2490 ret = btrfs_search_forward(root, &found_key, &other_key,
2491 path, 0, 0);
2492 path->keep_locks = 0;
2493 if (ret || found_key.objectid != key.objectid ||
2494 found_key.type != key.type) {
2495 ret = 0;
2496 goto out;
2497 }
2498
2499 key.offset = found_key.offset;
2500 btrfs_release_path(root, path);
2501 cond_resched();
2502 goto again;
2503 }
2504
2505 /* we know there's one more slot after us in the tree,
2506 * read that key so we can verify it is also a checksum item
2507 */
2508 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2509
2510 if (found_key.objectid < inode->i_ino)
2511 goto next_key;
2512
2513 if (found_key.type != key.type || found_key.offset < new_size)
2514 goto next_key;
2515
2516 /*
2517 * if the key for the next leaf isn't a csum key from this objectid,
2518 * we can't be sure there aren't good items inside this leaf.
2519 * Bail out
2520 */
2521 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2522 goto out;
2523
5b84e8d6
YZ		 2524 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2525 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
323ac95b
CM		 2526 /*
2527 * it is safe to delete this leaf, it contains only
2528 * csum items from this inode at an offset >= new_size
2529 */
5b84e8d6 2530 ret = btrfs_del_leaf(trans, root, path, leaf_start);
323ac95b
CM		 2531 BUG_ON(ret);
2532
5b84e8d6
YZ		 2533 if (root->ref_cows && leaf_gen < trans->transid) {
2534 ref = btrfs_alloc_leaf_ref(root, 0);
2535 if (ref) {
2536 ref->root_gen = root->root_key.offset;
2537 ref->bytenr = leaf_start;
2538 ref->owner = 0;
2539 ref->generation = leaf_gen;
2540 ref->nritems = 0;
2541
bd56b302
CM		 2542 btrfs_sort_leaf_ref(ref);
2543
5b84e8d6
YZ		 2544 ret = btrfs_add_leaf_ref(root, ref, 0);
2545 WARN_ON(ret);
2546 btrfs_free_leaf_ref(root, ref);
2547 } else {
2548 WARN_ON(1);
2549 }
2550 }
323ac95b
CM		 2551next_key:
2552 btrfs_release_path(root, path);
2553
2554 if (other_key.objectid == inode->i_ino &&
2555 other_key.type == key.type && other_key.offset > key.offset) {
2556 key.offset = other_key.offset;
2557 cond_resched();
2558 goto again;
2559 }
2560 ret = 0;
2561out:
2562 /* fixup any changes we've made to the path */
2563 path->lowest_level = 0;
2564 path->keep_locks = 0;
2565 btrfs_release_path(root, path);
2566 return ret;
2567}
2568
d20f7043
CM		 2569#endif
2570
39279cc3
CM		 2571/*
2572 * this can truncate away extent items, csum items and directory items.
2573 * It starts at a high offset and removes keys until it can't find
d352ac68 2574 * any higher than new_size
39279cc3
CM		 2575 *
2576 * csum items that cross the new i_size are truncated to the new size
2577 * as well.
7b128766
JB		 2578 *
2579 * min_type is the minimum key type to truncate down to. If set to 0, this
2580 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 2581 */
e02119d5
CM		 2582noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2583 struct btrfs_root *root,
2584 struct inode *inode,
2585 u64 new_size, u32 min_type)
39279cc3
CM		 2586{
2587 int ret;
2588 struct btrfs_path *path;
2589 struct btrfs_key key;
5f39d397 2590 struct btrfs_key found_key;
06d9a8d7 2591 u32 found_type = (u8)-1;
5f39d397 2592 struct extent_buffer *leaf;
39279cc3
CM		 2593 struct btrfs_file_extent_item *fi;
2594 u64 extent_start = 0;
db94535d 2595 u64 extent_num_bytes = 0;
5d4f98a2 2596 u64 extent_offset = 0;
39279cc3
CM		 2597 u64 item_end = 0;
2598 int found_extent;
2599 int del_item;
85e21bac
CM		 2600 int pending_del_nr = 0;
2601 int pending_del_slot = 0;
179e29e4 2602 int extent_type = -1;
771ed689 2603 int encoding;
3b951516 2604 u64 mask = root->sectorsize - 1;
39279cc3 2605
e02119d5 2606 if (root->ref_cows)
5b21f2ed 2607 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
39279cc3 2608 path = btrfs_alloc_path();
3c69faec 2609 path->reada = -1;
39279cc3 2610 BUG_ON(!path);
5f39d397 2611
39279cc3
CM		 2612 /* FIXME, add redo link to tree so we don't leak on crash */
2613 key.objectid = inode->i_ino;
2614 key.offset = (u64)-1;
5f39d397
CM		 2615 key.type = (u8)-1;
2616
85e21bac 2617search_again:
b9473439 2618 path->leave_spinning = 1;
85e21bac 2619 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
d397712b 2620 if (ret < 0)
85e21bac 2621 goto error;
d397712b 2622
85e21bac 2623 if (ret > 0) {
e02119d5
CM		 2624 /* there are no items in the tree for us to truncate, we're
2625 * done
2626 */
2627 if (path->slots[0] == 0) {
2628 ret = 0;
2629 goto error;
2630 }
85e21bac
CM		 2631 path->slots[0]--;
2632 }
2633
d397712b 2634 while (1) {
39279cc3 2635 fi = NULL;
5f39d397
CM		 2636 leaf = path->nodes[0];
2637 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2638 found_type = btrfs_key_type(&found_key);
771ed689 2639 encoding = 0;
39279cc3 2640
5f39d397 2641 if (found_key.objectid != inode->i_ino)
39279cc3 2642 break;
5f39d397 2643
85e21bac 2644 if (found_type < min_type)
39279cc3
CM		 2645 break;
2646
5f39d397 2647 item_end = found_key.offset;
39279cc3 2648 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 2649 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 2650 struct btrfs_file_extent_item);
179e29e4 2651 extent_type = btrfs_file_extent_type(leaf, fi);
771ed689
CM		 2652 encoding = btrfs_file_extent_compression(leaf, fi);
2653 encoding |= btrfs_file_extent_encryption(leaf, fi);
2654 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2655
179e29e4 2656 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 2657 item_end +=
db94535d 2658 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 2659 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 2660 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 2661 fi);
39279cc3 2662 }
008630c1 2663 item_end--;
39279cc3 2664 }
e02119d5 2665 if (item_end < new_size) {
d397712b 2666 if (found_type == BTRFS_DIR_ITEM_KEY)
b888db2b 2667 found_type = BTRFS_INODE_ITEM_KEY;
d397712b 2668 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
d20f7043 2669 found_type = BTRFS_EXTENT_DATA_KEY;
d397712b 2670 else if (found_type == BTRFS_EXTENT_DATA_KEY)
85e21bac 2671 found_type = BTRFS_XATTR_ITEM_KEY;
d397712b 2672 else if (found_type == BTRFS_XATTR_ITEM_KEY)
85e21bac 2673 found_type = BTRFS_INODE_REF_KEY;
d397712b 2674 else if (found_type)
b888db2b 2675 found_type--;
d397712b 2676 else
b888db2b 2677 break;
a61721d5 2678 btrfs_set_key_type(&key, found_type);
85e21bac 2679 goto next;
39279cc3 2680 }
e02119d5 2681 if (found_key.offset >= new_size)
39279cc3
CM		 2682 del_item = 1;
2683 else
2684 del_item = 0;
2685 found_extent = 0;
2686
2687 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 2688 if (found_type != BTRFS_EXTENT_DATA_KEY)
2689 goto delete;
2690
2691 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 2692 u64 num_dec;
db94535d 2693 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
771ed689 2694 if (!del_item && !encoding) {
db94535d
CM		 2695 u64 orig_num_bytes =
2696 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 2697 extent_num_bytes = new_size -
5f39d397 2698 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 2699 extent_num_bytes = extent_num_bytes &
2700 ~((u64)root->sectorsize - 1);
db94535d
CM		 2701 btrfs_set_file_extent_num_bytes(leaf, fi,
2702 extent_num_bytes);
2703 num_dec = (orig_num_bytes -
9069218d 2704 extent_num_bytes);
e02119d5 2705 if (root->ref_cows && extent_start != 0)
a76a3cd4 2706 inode_sub_bytes(inode, num_dec);
5f39d397 2707 btrfs_mark_buffer_dirty(leaf);
39279cc3 2708 } else {
db94535d
CM		 2709 extent_num_bytes =
2710 btrfs_file_extent_disk_num_bytes(leaf,
2711 fi);
5d4f98a2
YZ		 2712 extent_offset = found_key.offset -
2713 btrfs_file_extent_offset(leaf, fi);
2714
39279cc3 2715 /* FIXME blocksize != 4096 */
9069218d 2716 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 2717 if (extent_start != 0) {
2718 found_extent = 1;
e02119d5 2719 if (root->ref_cows)
a76a3cd4 2720 inode_sub_bytes(inode, num_dec);
e02119d5 2721 }
39279cc3 2722 }
9069218d 2723 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 2724 /*
2725 * we can't truncate inline items that have had
2726 * special encodings
2727 */
2728 if (!del_item &&
2729 btrfs_file_extent_compression(leaf, fi) == 0 &&
2730 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2731 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 2732 u32 size = new_size - found_key.offset;
2733
2734 if (root->ref_cows) {
a76a3cd4
YZ		 2735 inode_sub_bytes(inode, item_end + 1 -
2736 new_size);
e02119d5
CM		 2737 }
2738 size =
2739 btrfs_file_extent_calc_inline_size(size);
9069218d 2740 ret = btrfs_truncate_item(trans, root, path,
e02119d5 2741 size, 1);
9069218d 2742 BUG_ON(ret);
e02119d5 2743 } else if (root->ref_cows) {
a76a3cd4
YZ		 2744 inode_sub_bytes(inode, item_end + 1 -
2745 found_key.offset);
9069218d 2746 }
39279cc3 2747 }
179e29e4 2748delete:
39279cc3 2749 if (del_item) {
85e21bac
CM		 2750 if (!pending_del_nr) {
2751 /* no pending yet, add ourselves */
2752 pending_del_slot = path->slots[0];
2753 pending_del_nr = 1;
2754 } else if (pending_del_nr &&
2755 path->slots[0] + 1 == pending_del_slot) {
2756 /* hop on the pending chunk */
2757 pending_del_nr++;
2758 pending_del_slot = path->slots[0];
2759 } else {
d397712b 2760 BUG();
85e21bac 2761 }
39279cc3
CM		 2762 } else {
2763 break;
2764 }
5d4f98a2 2765 if (found_extent && root->ref_cows) {
b9473439 2766 btrfs_set_path_blocking(path);
39279cc3 2767 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 2768 extent_num_bytes, 0,
2769 btrfs_header_owner(leaf),
2770 inode->i_ino, extent_offset);
39279cc3
CM		 2771 BUG_ON(ret);
2772 }
85e21bac
CM		 2773next:
2774 if (path->slots[0] == 0) {
2775 if (pending_del_nr)
2776 goto del_pending;
2777 btrfs_release_path(root, path);
06d9a8d7
CM		 2778 if (found_type == BTRFS_INODE_ITEM_KEY)
2779 break;
85e21bac
CM		 2780 goto search_again;
2781 }
2782
2783 path->slots[0]--;
2784 if (pending_del_nr &&
2785 path->slots[0] + 1 != pending_del_slot) {
2786 struct btrfs_key debug;
2787del_pending:
2788 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2789 pending_del_slot);
2790 ret = btrfs_del_items(trans, root, path,
2791 pending_del_slot,
2792 pending_del_nr);
2793 BUG_ON(ret);
2794 pending_del_nr = 0;
2795 btrfs_release_path(root, path);
06d9a8d7
CM		 2796 if (found_type == BTRFS_INODE_ITEM_KEY)
2797 break;
85e21bac
CM		 2798 goto search_again;
2799 }
39279cc3
CM		 2800 }
2801 ret = 0;
2802error:
85e21bac
CM		 2803 if (pending_del_nr) {
2804 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2805 pending_del_nr);
2806 }
39279cc3 2807 btrfs_free_path(path);
39279cc3
CM		 2808 return ret;
2809}
2810
2811/*
2812 * taken from block_truncate_page, but does cow as it zeros out
2813 * any bytes left in the last page in the file.
2814 */
2815static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2816{
2817 struct inode *inode = mapping->host;
db94535d 2818 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 2819 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2820 struct btrfs_ordered_extent *ordered;
2821 char *kaddr;
db94535d 2822 u32 blocksize = root->sectorsize;
39279cc3
CM		 2823 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2824 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2825 struct page *page;
39279cc3 2826 int ret = 0;
a52d9a80 2827 u64 page_start;
e6dcd2dc 2828 u64 page_end;
39279cc3
CM		 2829
2830 if ((offset & (blocksize - 1)) == 0)
2831 goto out;
2832
2833 ret = -ENOMEM;
211c17f5 2834again:
39279cc3
CM		 2835 page = grab_cache_page(mapping, index);
2836 if (!page)
2837 goto out;
e6dcd2dc
CM		 2838
2839 page_start = page_offset(page);
2840 page_end = page_start + PAGE_CACHE_SIZE - 1;
2841
39279cc3 2842 if (!PageUptodate(page)) {
9ebefb18 2843 ret = btrfs_readpage(NULL, page);
39279cc3 2844 lock_page(page);
211c17f5
CM		 2845 if (page->mapping != mapping) {
2846 unlock_page(page);
2847 page_cache_release(page);
2848 goto again;
2849 }
39279cc3
CM		 2850 if (!PageUptodate(page)) {
2851 ret = -EIO;
89642229 2852 goto out_unlock;
39279cc3
CM		 2853 }
2854 }
211c17f5 2855 wait_on_page_writeback(page);
e6dcd2dc
CM		 2856
2857 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2858 set_page_extent_mapped(page);
2859
2860 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2861 if (ordered) {
2862 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2863 unlock_page(page);
2864 page_cache_release(page);
eb84ae03 2865 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 2866 btrfs_put_ordered_extent(ordered);
2867 goto again;
2868 }
2869
ea8c2819 2870 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc
CM		 2871 ret = 0;
2872 if (offset != PAGE_CACHE_SIZE) {
2873 kaddr = kmap(page);
2874 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2875 flush_dcache_page(page);
2876 kunmap(page);
2877 }
247e743c 2878 ClearPageChecked(page);
e6dcd2dc
CM		 2879 set_page_dirty(page);
2880 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
39279cc3 2881
89642229 2882out_unlock:
39279cc3
CM		 2883 unlock_page(page);
2884 page_cache_release(page);
2885out:
2886 return ret;
2887}
2888
9036c102 2889int btrfs_cont_expand(struct inode *inode, loff_t size)
39279cc3 2890{
9036c102
YZ		 2891 struct btrfs_trans_handle *trans;
2892 struct btrfs_root *root = BTRFS_I(inode)->root;
2893 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2894 struct extent_map *em;
2895 u64 mask = root->sectorsize - 1;
2896 u64 hole_start = (inode->i_size + mask) & ~mask;
2897 u64 block_end = (size + mask) & ~mask;
2898 u64 last_byte;
2899 u64 cur_offset;
2900 u64 hole_size;
39279cc3
CM		 2901 int err;
2902
9036c102
YZ		 2903 if (size <= hole_start)
2904 return 0;
2905
6a63209f 2906 err = btrfs_check_metadata_free_space(root);
39279cc3
CM		 2907 if (err)
2908 return err;
2909
9036c102 2910 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2bf5a725 2911
9036c102
YZ		 2912 while (1) {
2913 struct btrfs_ordered_extent *ordered;
2914 btrfs_wait_ordered_range(inode, hole_start,
2915 block_end - hole_start);
2916 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2917 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2918 if (!ordered)
2919 break;
2920 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2921 btrfs_put_ordered_extent(ordered);
2922 }
39279cc3 2923
9036c102
YZ		 2924 trans = btrfs_start_transaction(root, 1);
2925 btrfs_set_trans_block_group(trans, inode);
39279cc3 2926
9036c102
YZ		 2927 cur_offset = hole_start;
2928 while (1) {
2929 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2930 block_end - cur_offset, 0);
2931 BUG_ON(IS_ERR(em) || !em);
2932 last_byte = min(extent_map_end(em), block_end);
2933 last_byte = (last_byte + mask) & ~mask;
2934 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
771ed689 2935 u64 hint_byte = 0;
9036c102 2936 hole_size = last_byte - cur_offset;
771ed689
CM		 2937 err = btrfs_drop_extents(trans, root, inode,
2938 cur_offset,
2939 cur_offset + hole_size,
e980b50c 2940 block_end,
771ed689
CM		 2941 cur_offset, &hint_byte);
2942 if (err)
2943 break;
9036c102
YZ		 2944 err = btrfs_insert_file_extent(trans, root,
2945 inode->i_ino, cur_offset, 0,
2946 0, hole_size, 0, hole_size,
2947 0, 0, 0);
2948 btrfs_drop_extent_cache(inode, hole_start,
2949 last_byte - 1, 0);
2950 }
2951 free_extent_map(em);
2952 cur_offset = last_byte;
2953 if (err || cur_offset >= block_end)
2954 break;
2955 }
1832a6d5 2956
9036c102
YZ		 2957 btrfs_end_transaction(trans, root);
2958 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2959 return err;
2960}
39279cc3 2961
9036c102
YZ		 2962static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2963{
2964 struct inode *inode = dentry->d_inode;
2965 int err;
39279cc3 2966
9036c102
YZ		 2967 err = inode_change_ok(inode, attr);
2968 if (err)
2969 return err;
2bf5a725 2970
5a3f23d5
CM		 2971 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2972 if (attr->ia_size > inode->i_size) {
2973 err = btrfs_cont_expand(inode, attr->ia_size);
2974 if (err)
2975 return err;
2976 } else if (inode->i_size > 0 &&
2977 attr->ia_size == 0) {
2978
2979 /* we're truncating a file that used to have good
2980 * data down to zero. Make sure it gets into
2981 * the ordered flush list so that any new writes
2982 * get down to disk quickly.
2983 */
2984 BTRFS_I(inode)->ordered_data_close = 1;
2985 }
39279cc3 2986 }
9036c102 2987
39279cc3 2988 err = inode_setattr(inode, attr);
33268eaf
JB		 2989
2990 if (!err && ((attr->ia_valid & ATTR_MODE)))
2991 err = btrfs_acl_chmod(inode);
39279cc3
CM		 2992 return err;
2993}
61295eb8 2994
39279cc3
CM		 2995void btrfs_delete_inode(struct inode *inode)
2996{
2997 struct btrfs_trans_handle *trans;
2998 struct btrfs_root *root = BTRFS_I(inode)->root;
d3c2fdcf 2999 unsigned long nr;
39279cc3
CM		 3000 int ret;
3001
3002 truncate_inode_pages(&inode->i_data, 0);
3003 if (is_bad_inode(inode)) {
7b128766 3004 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3005 goto no_delete;
3006 }
4a096752 3007 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3008
dbe674a9 3009 btrfs_i_size_write(inode, 0);
180591bc 3010 trans = btrfs_join_transaction(root, 1);
5f39d397 3011
39279cc3 3012 btrfs_set_trans_block_group(trans, inode);
e02119d5 3013 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
7b128766
JB		 3014 if (ret) {
3015 btrfs_orphan_del(NULL, inode);
54aa1f4d 3016 goto no_delete_lock;
7b128766
JB		 3017 }
3018
3019 btrfs_orphan_del(trans, inode);
85e21bac 3020
d3c2fdcf 3021 nr = trans->blocks_used;
85e21bac 3022 clear_inode(inode);
5f39d397 3023
39279cc3 3024 btrfs_end_transaction(trans, root);
d3c2fdcf 3025 btrfs_btree_balance_dirty(root, nr);
39279cc3 3026 return;
54aa1f4d
CM		 3027
3028no_delete_lock:
d3c2fdcf 3029 nr = trans->blocks_used;
54aa1f4d 3030 btrfs_end_transaction(trans, root);
d3c2fdcf 3031 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3032no_delete:
3033 clear_inode(inode);
3034}
3035
3036/*
3037 * this returns the key found in the dir entry in the location pointer.
3038 * If no dir entries were found, location->objectid is 0.
3039 */
3040static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3041 struct btrfs_key *location)
3042{
3043 const char *name = dentry->d_name.name;
3044 int namelen = dentry->d_name.len;
3045 struct btrfs_dir_item *di;
3046 struct btrfs_path *path;
3047 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3048 int ret = 0;
39279cc3
CM		 3049
3050 path = btrfs_alloc_path();
3051 BUG_ON(!path);
3954401f 3052
39279cc3
CM		 3053 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3054 namelen, 0);
0d9f7f3e
Y		 3055 if (IS_ERR(di))
3056 ret = PTR_ERR(di);
d397712b
CM		 3057
3058 if (!di || IS_ERR(di))
3954401f 3059 goto out_err;
d397712b 3060
5f39d397 3061 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3062out:
39279cc3
CM		 3063 btrfs_free_path(path);
3064 return ret;
3954401f
CM		 3065out_err:
3066 location->objectid = 0;
3067 goto out;
39279cc3
CM		 3068}
3069
3070/*
3071 * when we hit a tree root in a directory, the btrfs part of the inode
3072 * needs to be changed to reflect the root directory of the tree root. This
3073 * is kind of like crossing a mount point.
3074 */
3075static int fixup_tree_root_location(struct btrfs_root *root,
3076 struct btrfs_key *location,
58176a96
JB		 3077 struct btrfs_root **sub_root,
3078 struct dentry *dentry)
39279cc3 3079{
39279cc3
CM		 3080 struct btrfs_root_item *ri;
3081
3082 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3083 return 0;
3084 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3085 return 0;
3086
58176a96
JB		 3087 *sub_root = btrfs_read_fs_root(root->fs_info, location,
3088 dentry->d_name.name,
3089 dentry->d_name.len);
39279cc3
CM		 3090 if (IS_ERR(*sub_root))
3091 return PTR_ERR(*sub_root);
3092
3093 ri = &(*sub_root)->root_item;
3094 location->objectid = btrfs_root_dirid(ri);
39279cc3
CM		 3095 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3096 location->offset = 0;
3097
39279cc3
CM		 3098 return 0;
3099}
3100
5d4f98a2
YZ		 3101static void inode_tree_add(struct inode *inode)
3102{
3103 struct btrfs_root *root = BTRFS_I(inode)->root;
3104 struct btrfs_inode *entry;
3105 struct rb_node **p = &root->inode_tree.rb_node;
3106 struct rb_node *parent = NULL;
3107
3108 spin_lock(&root->inode_lock);
3109 while (*p) {
3110 parent = *p;
3111 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3112
3113 if (inode->i_ino < entry->vfs_inode.i_ino)
3114 p = &(*p)->rb_left;
3115 else if (inode->i_ino > entry->vfs_inode.i_ino)
3116 p = &(*p)->rb_right;
3117 else {
3118 WARN_ON(!(entry->vfs_inode.i_state &
3119 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3120 break;
3121 }
3122 }
3123 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3124 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3125 spin_unlock(&root->inode_lock);
3126}
3127
3128static void inode_tree_del(struct inode *inode)
3129{
3130 struct btrfs_root *root = BTRFS_I(inode)->root;
3131
3132 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3133 spin_lock(&root->inode_lock);
3134 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3135 spin_unlock(&root->inode_lock);
3136 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3137 }
3138}
3139
e02119d5 3140static noinline void init_btrfs_i(struct inode *inode)
39279cc3 3141{
e02119d5
CM		 3142 struct btrfs_inode *bi = BTRFS_I(inode);
3143
7b1a14bb
CM		 3144 bi->i_acl = BTRFS_ACL_NOT_CACHED;
3145 bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
e02119d5
CM		 3146
3147 bi->generation = 0;
c3027eb5 3148 bi->sequence = 0;
e02119d5
CM		 3149 bi->last_trans = 0;
3150 bi->logged_trans = 0;
3151 bi->delalloc_bytes = 0;
6a63209f 3152 bi->reserved_bytes = 0;
e02119d5
CM		 3153 bi->disk_i_size = 0;
3154 bi->flags = 0;
3155 bi->index_cnt = (u64)-1;
12fcfd22 3156 bi->last_unlink_trans = 0;
2757495c 3157 bi->ordered_data_close = 0;
d1310b2e
CM		 3158 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3159 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
b888db2b 3160 inode->i_mapping, GFP_NOFS);
7e38326f
CM		 3161 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3162 inode->i_mapping, GFP_NOFS);
ea8c2819 3163 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
5a3f23d5 3164 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
5d4f98a2 3165 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
ba1da2f4 3166 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
ee6e6504 3167 mutex_init(&BTRFS_I(inode)->extent_mutex);
e02119d5
CM		 3168 mutex_init(&BTRFS_I(inode)->log_mutex);
3169}
3170
3171static int btrfs_init_locked_inode(struct inode *inode, void *p)
3172{
3173 struct btrfs_iget_args *args = p;
3174 inode->i_ino = args->ino;
3175 init_btrfs_i(inode);
3176 BTRFS_I(inode)->root = args->root;
6a63209f 3177 btrfs_set_inode_space_info(args->root, inode);
39279cc3
CM		 3178 return 0;
3179}
3180
3181static int btrfs_find_actor(struct inode *inode, void *opaque)
3182{
3183 struct btrfs_iget_args *args = opaque;
d397712b
CM		 3184 return args->ino == inode->i_ino &&
3185 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 3186}
3187
5d4f98a2
YZ		 3188static struct inode *btrfs_iget_locked(struct super_block *s,
3189 u64 objectid,
3190 struct btrfs_root *root)
39279cc3
CM		 3191{
3192 struct inode *inode;
3193 struct btrfs_iget_args args;
3194 args.ino = objectid;
3195 args.root = root;
3196
3197 inode = iget5_locked(s, objectid, btrfs_find_actor,
3198 btrfs_init_locked_inode,
3199 (void *)&args);
3200 return inode;
3201}
3202
1a54ef8c
BR		 3203/* Get an inode object given its location and corresponding root.
3204 * Returns in *is_new if the inode was read from disk
3205 */
3206struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5d4f98a2 3207 struct btrfs_root *root)
1a54ef8c
BR		 3208{
3209 struct inode *inode;
3210
3211 inode = btrfs_iget_locked(s, location->objectid, root);
3212 if (!inode)
5d4f98a2 3213 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 3214
3215 if (inode->i_state & I_NEW) {
3216 BTRFS_I(inode)->root = root;
3217 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3218 btrfs_read_locked_inode(inode);
5d4f98a2
YZ		 3219
3220 inode_tree_add(inode);
1a54ef8c 3221 unlock_new_inode(inode);
1a54ef8c
BR		 3222 }
3223
3224 return inode;
3225}
3226
3de4586c 3227struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 3228{
d397712b 3229 struct inode *inode;
39279cc3
CM		 3230 struct btrfs_inode *bi = BTRFS_I(dir);
3231 struct btrfs_root *root = bi->root;
3232 struct btrfs_root *sub_root = root;
3233 struct btrfs_key location;
5d4f98a2 3234 int ret;
39279cc3
CM		 3235
3236 if (dentry->d_name.len > BTRFS_NAME_LEN)
3237 return ERR_PTR(-ENAMETOOLONG);
5f39d397 3238
39279cc3 3239 ret = btrfs_inode_by_name(dir, dentry, &location);
5f39d397 3240
39279cc3
CM		 3241 if (ret < 0)
3242 return ERR_PTR(ret);
5f39d397 3243
39279cc3
CM		 3244 inode = NULL;
3245 if (location.objectid) {
58176a96
JB		 3246 ret = fixup_tree_root_location(root, &location, &sub_root,
3247 dentry);
39279cc3
CM		 3248 if (ret < 0)
3249 return ERR_PTR(ret);
3250 if (ret > 0)
3251 return ERR_PTR(-ENOENT);
5d4f98a2 3252 inode = btrfs_iget(dir->i_sb, &location, sub_root);
1a54ef8c
BR		 3253 if (IS_ERR(inode))
3254 return ERR_CAST(inode);
39279cc3 3255 }
3de4586c
CM		 3256 return inode;
3257}
3258
3259static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3260 struct nameidata *nd)
3261{
3262 struct inode *inode;
3263
3264 if (dentry->d_name.len > BTRFS_NAME_LEN)
3265 return ERR_PTR(-ENAMETOOLONG);
3266
3267 inode = btrfs_lookup_dentry(dir, dentry);
3268 if (IS_ERR(inode))
3269 return ERR_CAST(inode);
7b128766 3270
39279cc3
CM		 3271 return d_splice_alias(inode, dentry);
3272}
3273
39279cc3
CM		 3274static unsigned char btrfs_filetype_table[] = {
3275 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3276};
3277
cbdf5a24
DW		 3278static int btrfs_real_readdir(struct file *filp, void *dirent,
3279 filldir_t filldir)
39279cc3 3280{
6da6abae 3281 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 3282 struct btrfs_root *root = BTRFS_I(inode)->root;
3283 struct btrfs_item *item;
3284 struct btrfs_dir_item *di;
3285 struct btrfs_key key;
5f39d397 3286 struct btrfs_key found_key;
39279cc3
CM		 3287 struct btrfs_path *path;
3288 int ret;
3289 u32 nritems;
5f39d397 3290 struct extent_buffer *leaf;
39279cc3
CM		 3291 int slot;
3292 int advance;
3293 unsigned char d_type;
3294 int over = 0;
3295 u32 di_cur;
3296 u32 di_total;
3297 u32 di_len;
3298 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 3299 char tmp_name[32];
3300 char *name_ptr;
3301 int name_len;
39279cc3
CM		 3302
3303 /* FIXME, use a real flag for deciding about the key type */
3304 if (root->fs_info->tree_root == root)
3305 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 3306
3954401f
CM		 3307 /* special case for "." */
3308 if (filp->f_pos == 0) {
3309 over = filldir(dirent, ".", 1,
3310 1, inode->i_ino,
3311 DT_DIR);
3312 if (over)
3313 return 0;
3314 filp->f_pos = 1;
3315 }
3954401f
CM		 3316 /* special case for .., just use the back ref */
3317 if (filp->f_pos == 1) {
5ecc7e5d 3318 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 3319 over = filldir(dirent, "..", 2,
5ecc7e5d 3320 2, pino, DT_DIR);
3954401f 3321 if (over)
49593bfa 3322 return 0;
3954401f
CM		 3323 filp->f_pos = 2;
3324 }
49593bfa
DW		 3325 path = btrfs_alloc_path();
3326 path->reada = 2;
3327
39279cc3
CM		 3328 btrfs_set_key_type(&key, key_type);
3329 key.offset = filp->f_pos;
49593bfa 3330 key.objectid = inode->i_ino;
5f39d397 3331
39279cc3
CM		 3332 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3333 if (ret < 0)
3334 goto err;
3335 advance = 0;
49593bfa
DW		 3336
3337 while (1) {
5f39d397
CM		 3338 leaf = path->nodes[0];
3339 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3340 slot = path->slots[0];
3341 if (advance || slot >= nritems) {
49593bfa 3342 if (slot >= nritems - 1) {
39279cc3
CM		 3343 ret = btrfs_next_leaf(root, path);
3344 if (ret)
3345 break;
5f39d397
CM		 3346 leaf = path->nodes[0];
3347 nritems = btrfs_header_nritems(leaf);
39279cc3
CM		 3348 slot = path->slots[0];
3349 } else {
3350 slot++;
3351 path->slots[0]++;
3352 }
3353 }
3de4586c 3354
39279cc3 3355 advance = 1;
5f39d397
CM		 3356 item = btrfs_item_nr(leaf, slot);
3357 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3358
3359 if (found_key.objectid != key.objectid)
39279cc3 3360 break;
5f39d397 3361 if (btrfs_key_type(&found_key) != key_type)
39279cc3 3362 break;
5f39d397 3363 if (found_key.offset < filp->f_pos)
39279cc3 3364 continue;
5f39d397
CM		 3365
3366 filp->f_pos = found_key.offset;
49593bfa 3367
39279cc3
CM		 3368 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3369 di_cur = 0;
5f39d397 3370 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 3371
3372 while (di_cur < di_total) {
5f39d397
CM		 3373 struct btrfs_key location;
3374
3375 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 3376 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 3377 name_ptr = tmp_name;
3378 } else {
3379 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 3380 if (!name_ptr) {
3381 ret = -ENOMEM;
3382 goto err;
3383 }
5f39d397
CM		 3384 }
3385 read_extent_buffer(leaf, name_ptr,
3386 (unsigned long)(di + 1), name_len);
3387
3388 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3389 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c
CM		 3390
3391 /* is this a reference to our own snapshot? If so
3392 * skip it
3393 */
3394 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3395 location.objectid == root->root_key.objectid) {
3396 over = 0;
3397 goto skip;
3398 }
5f39d397 3399 over = filldir(dirent, name_ptr, name_len,
49593bfa 3400 found_key.offset, location.objectid,
39279cc3 3401 d_type);
5f39d397 3402
3de4586c 3403skip:
5f39d397
CM		 3404 if (name_ptr != tmp_name)
3405 kfree(name_ptr);
3406
39279cc3
CM		 3407 if (over)
3408 goto nopos;
5103e947 3409 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 3410 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 3411 di_cur += di_len;
3412 di = (struct btrfs_dir_item *)((char *)di + di_len);
3413 }
3414 }
49593bfa
DW		 3415
3416 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 3417 if (key_type == BTRFS_DIR_INDEX_KEY)
89f135d8 3418 filp->f_pos = INT_LIMIT(off_t);
5e591a07
YZ		 3419 else
3420 filp->f_pos++;
39279cc3
CM		 3421nopos:
3422 ret = 0;
3423err:
39279cc3 3424 btrfs_free_path(path);
39279cc3
CM		 3425 return ret;
3426}
3427
3428int btrfs_write_inode(struct inode *inode, int wait)
3429{
3430 struct btrfs_root *root = BTRFS_I(inode)->root;
3431 struct btrfs_trans_handle *trans;
3432 int ret = 0;
3433
c146afad 3434 if (root->fs_info->btree_inode == inode)
4ca8b41e
CM		 3435 return 0;
3436
39279cc3 3437 if (wait) {
f9295749 3438 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3439 btrfs_set_trans_block_group(trans, inode);
3440 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 3441 }
3442 return ret;
3443}
3444
3445/*
54aa1f4d 3446 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 3447 * inode changes. But, it is most likely to find the inode in cache.
3448 * FIXME, needs more benchmarking...there are no reasons other than performance
3449 * to keep or drop this code.
3450 */
3451void btrfs_dirty_inode(struct inode *inode)
3452{
3453 struct btrfs_root *root = BTRFS_I(inode)->root;
3454 struct btrfs_trans_handle *trans;
3455
f9295749 3456 trans = btrfs_join_transaction(root, 1);
39279cc3
CM		 3457 btrfs_set_trans_block_group(trans, inode);
3458 btrfs_update_inode(trans, root, inode);
3459 btrfs_end_transaction(trans, root);
39279cc3
CM		 3460}
3461
d352ac68
CM		 3462/*
3463 * find the highest existing sequence number in a directory
3464 * and then set the in-memory index_cnt variable to reflect
3465 * free sequence numbers
3466 */
aec7477b
JB		 3467static int btrfs_set_inode_index_count(struct inode *inode)
3468{
3469 struct btrfs_root *root = BTRFS_I(inode)->root;
3470 struct btrfs_key key, found_key;
3471 struct btrfs_path *path;
3472 struct extent_buffer *leaf;
3473 int ret;
3474
3475 key.objectid = inode->i_ino;
3476 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3477 key.offset = (u64)-1;
3478
3479 path = btrfs_alloc_path();
3480 if (!path)
3481 return -ENOMEM;
3482
3483 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3484 if (ret < 0)
3485 goto out;
3486 /* FIXME: we should be able to handle this */
3487 if (ret == 0)
3488 goto out;
3489 ret = 0;
3490
3491 /*
3492 * MAGIC NUMBER EXPLANATION:
3493 * since we search a directory based on f_pos we have to start at 2
3494 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3495 * else has to start at 2
3496 */
3497 if (path->slots[0] == 0) {
3498 BTRFS_I(inode)->index_cnt = 2;
3499 goto out;
3500 }
3501
3502 path->slots[0]--;
3503
3504 leaf = path->nodes[0];
3505 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3506
3507 if (found_key.objectid != inode->i_ino ||
3508 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3509 BTRFS_I(inode)->index_cnt = 2;
3510 goto out;
3511 }
3512
3513 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3514out:
3515 btrfs_free_path(path);
3516 return ret;
3517}
3518
d352ac68
CM		 3519/*
3520 * helper to find a free sequence number in a given directory. This current
3521 * code is very simple, later versions will do smarter things in the btree
3522 */
3de4586c 3523int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 3524{
3525 int ret = 0;
3526
3527 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3528 ret = btrfs_set_inode_index_count(dir);
d397712b 3529 if (ret)
aec7477b
JB		 3530 return ret;
3531 }
3532
00e4e6b3 3533 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 3534 BTRFS_I(dir)->index_cnt++;
3535
3536 return ret;
3537}
3538
39279cc3
CM		 3539static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3540 struct btrfs_root *root,
aec7477b 3541 struct inode *dir,
9c58309d 3542 const char *name, int name_len,
d2fb3437
YZ		 3543 u64 ref_objectid, u64 objectid,
3544 u64 alloc_hint, int mode, u64 *index)
39279cc3
CM		 3545{
3546 struct inode *inode;
5f39d397 3547 struct btrfs_inode_item *inode_item;
39279cc3 3548 struct btrfs_key *location;
5f39d397 3549 struct btrfs_path *path;
9c58309d
CM		 3550 struct btrfs_inode_ref *ref;
3551 struct btrfs_key key[2];
3552 u32 sizes[2];
3553 unsigned long ptr;
39279cc3
CM		 3554 int ret;
3555 int owner;
3556
5f39d397
CM		 3557 path = btrfs_alloc_path();
3558 BUG_ON(!path);
3559
39279cc3
CM		 3560 inode = new_inode(root->fs_info->sb);
3561 if (!inode)
3562 return ERR_PTR(-ENOMEM);
3563
aec7477b 3564 if (dir) {
3de4586c 3565 ret = btrfs_set_inode_index(dir, index);
09771430
SF		 3566 if (ret) {
3567 iput(inode);
aec7477b 3568 return ERR_PTR(ret);
09771430 3569 }
aec7477b
JB		 3570 }
3571 /*
3572 * index_cnt is ignored for everything but a dir,
3573 * btrfs_get_inode_index_count has an explanation for the magic
3574 * number
3575 */
e02119d5 3576 init_btrfs_i(inode);
aec7477b 3577 BTRFS_I(inode)->index_cnt = 2;
39279cc3 3578 BTRFS_I(inode)->root = root;
e02119d5 3579 BTRFS_I(inode)->generation = trans->transid;
6a63209f 3580 btrfs_set_inode_space_info(root, inode);
b888db2b 3581
39279cc3
CM		 3582 if (mode & S_IFDIR)
3583 owner = 0;
3584 else
3585 owner = 1;
d2fb3437
YZ		 3586 BTRFS_I(inode)->block_group =
3587 btrfs_find_block_group(root, 0, alloc_hint, owner);
17d217fe
YZ		 3588 if ((mode & S_IFREG)) {
3589 if (btrfs_test_opt(root, NODATASUM))
6cbff00f 3590 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
17d217fe 3591 if (btrfs_test_opt(root, NODATACOW))
6cbff00f 3592 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
17d217fe 3593 }
9c58309d
CM		 3594
3595 key[0].objectid = objectid;
3596 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3597 key[0].offset = 0;
3598
3599 key[1].objectid = objectid;
3600 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3601 key[1].offset = ref_objectid;
3602
3603 sizes[0] = sizeof(struct btrfs_inode_item);
3604 sizes[1] = name_len + sizeof(*ref);
3605
b9473439 3606 path->leave_spinning = 1;
9c58309d
CM		 3607 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3608 if (ret != 0)
5f39d397
CM		 3609 goto fail;
3610
9c58309d
CM		 3611 if (objectid > root->highest_inode)
3612 root->highest_inode = objectid;
3613
79683f2d 3614 inode->i_uid = current_fsuid();
8c087b51 3615
42f15d77 3616 if (dir && (dir->i_mode & S_ISGID)) {
8c087b51
CB		 3617 inode->i_gid = dir->i_gid;
3618 if (S_ISDIR(mode))
3619 mode |= S_ISGID;
3620 } else
3621 inode->i_gid = current_fsgid();
3622
39279cc3
CM		 3623 inode->i_mode = mode;
3624 inode->i_ino = objectid;
a76a3cd4 3625 inode_set_bytes(inode, 0);
39279cc3 3626 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 3627 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3628 struct btrfs_inode_item);
e02119d5 3629 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 3630
3631 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3632 struct btrfs_inode_ref);
3633 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 3634 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 3635 ptr = (unsigned long)(ref + 1);
3636 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3637
5f39d397
CM		 3638 btrfs_mark_buffer_dirty(path->nodes[0]);
3639 btrfs_free_path(path);
3640
39279cc3
CM		 3641 location = &BTRFS_I(inode)->location;
3642 location->objectid = objectid;
39279cc3
CM		 3643 location->offset = 0;
3644 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3645
6cbff00f
CH		 3646 btrfs_inherit_iflags(inode, dir);
3647
39279cc3 3648 insert_inode_hash(inode);
5d4f98a2 3649 inode_tree_add(inode);
39279cc3 3650 return inode;
5f39d397 3651fail:
aec7477b
JB		 3652 if (dir)
3653 BTRFS_I(dir)->index_cnt--;
5f39d397 3654 btrfs_free_path(path);
09771430 3655 iput(inode);
5f39d397 3656 return ERR_PTR(ret);
39279cc3
CM		 3657}
3658
3659static inline u8 btrfs_inode_type(struct inode *inode)
3660{
3661 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3662}
3663
d352ac68
CM		 3664/*
3665 * utility function to add 'inode' into 'parent_inode' with
3666 * a give name and a given sequence number.
3667 * if 'add_backref' is true, also insert a backref from the
3668 * inode to the parent directory.
3669 */
e02119d5
CM		 3670int btrfs_add_link(struct btrfs_trans_handle *trans,
3671 struct inode *parent_inode, struct inode *inode,
3672 const char *name, int name_len, int add_backref, u64 index)
39279cc3
CM		 3673{
3674 int ret;
3675 struct btrfs_key key;
e02119d5 3676 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5f39d397 3677
39279cc3 3678 key.objectid = inode->i_ino;
39279cc3
CM		 3679 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3680 key.offset = 0;
3681
e02119d5
CM		 3682 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3683 parent_inode->i_ino,
aec7477b 3684 &key, btrfs_inode_type(inode),
00e4e6b3 3685 index);
39279cc3 3686 if (ret == 0) {
9c58309d
CM		 3687 if (add_backref) {
3688 ret = btrfs_insert_inode_ref(trans, root,
e02119d5
CM		 3689 name, name_len,
3690 inode->i_ino,
3691 parent_inode->i_ino,
3692 index);
9c58309d 3693 }
dbe674a9 3694 btrfs_i_size_write(parent_inode, parent_inode->i_size +
e02119d5 3695 name_len * 2);
79c44584 3696 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
e02119d5 3697 ret = btrfs_update_inode(trans, root, parent_inode);
39279cc3
CM		 3698 }
3699 return ret;
3700}
3701
3702static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
9c58309d 3703 struct dentry *dentry, struct inode *inode,
00e4e6b3 3704 int backref, u64 index)
39279cc3 3705{
e02119d5
CM		 3706 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3707 inode, dentry->d_name.name,
3708 dentry->d_name.len, backref, index);
39279cc3
CM		 3709 if (!err) {
3710 d_instantiate(dentry, inode);
3711 return 0;
3712 }
3713 if (err > 0)
3714 err = -EEXIST;
3715 return err;
3716}
3717
618e21d5
JB		 3718static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3719 int mode, dev_t rdev)
3720{
3721 struct btrfs_trans_handle *trans;
3722 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3723 struct inode *inode = NULL;
618e21d5
JB		 3724 int err;
3725 int drop_inode = 0;
3726 u64 objectid;
1832a6d5 3727 unsigned long nr = 0;
00e4e6b3 3728 u64 index = 0;
618e21d5
JB		 3729
3730 if (!new_valid_dev(rdev))
3731 return -EINVAL;
3732
6a63209f 3733 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3734 if (err)
3735 goto fail;
3736
618e21d5
JB		 3737 trans = btrfs_start_transaction(root, 1);
3738 btrfs_set_trans_block_group(trans, dir);
3739
3740 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3741 if (err) {
3742 err = -ENOSPC;
3743 goto out_unlock;
3744 }
3745
aec7477b 3746 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3747 dentry->d_name.len,
3748 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3 3749 BTRFS_I(dir)->block_group, mode, &index);
618e21d5
JB		 3750 err = PTR_ERR(inode);
3751 if (IS_ERR(inode))
3752 goto out_unlock;
3753
0279b4cd 3754 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3755 if (err) {
3756 drop_inode = 1;
3757 goto out_unlock;
3758 }
3759
618e21d5 3760 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3761 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
618e21d5
JB		 3762 if (err)
3763 drop_inode = 1;
3764 else {
3765 inode->i_op = &btrfs_special_inode_operations;
3766 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 3767 btrfs_update_inode(trans, root, inode);
618e21d5 3768 }
618e21d5
JB		 3769 btrfs_update_inode_block_group(trans, inode);
3770 btrfs_update_inode_block_group(trans, dir);
3771out_unlock:
d3c2fdcf 3772 nr = trans->blocks_used;
89ce8a63 3773 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3774fail:
618e21d5
JB		 3775 if (drop_inode) {
3776 inode_dec_link_count(inode);
3777 iput(inode);
3778 }
d3c2fdcf 3779 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 3780 return err;
3781}
3782
39279cc3
CM		 3783static int btrfs_create(struct inode *dir, struct dentry *dentry,
3784 int mode, struct nameidata *nd)
3785{
3786 struct btrfs_trans_handle *trans;
3787 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 3788 struct inode *inode = NULL;
39279cc3
CM		 3789 int err;
3790 int drop_inode = 0;
1832a6d5 3791 unsigned long nr = 0;
39279cc3 3792 u64 objectid;
00e4e6b3 3793 u64 index = 0;
39279cc3 3794
6a63209f 3795 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3796 if (err)
3797 goto fail;
39279cc3
CM		 3798 trans = btrfs_start_transaction(root, 1);
3799 btrfs_set_trans_block_group(trans, dir);
3800
3801 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3802 if (err) {
3803 err = -ENOSPC;
3804 goto out_unlock;
3805 }
3806
aec7477b 3807 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3808 dentry->d_name.len,
3809 dentry->d_parent->d_inode->i_ino,
00e4e6b3
CM		 3810 objectid, BTRFS_I(dir)->block_group, mode,
3811 &index);
39279cc3
CM		 3812 err = PTR_ERR(inode);
3813 if (IS_ERR(inode))
3814 goto out_unlock;
3815
0279b4cd 3816 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3817 if (err) {
3818 drop_inode = 1;
3819 goto out_unlock;
3820 }
3821
39279cc3 3822 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 3823 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 3824 if (err)
3825 drop_inode = 1;
3826 else {
3827 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3828 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 3829 inode->i_fop = &btrfs_file_operations;
3830 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 3831 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 3832 }
39279cc3
CM		 3833 btrfs_update_inode_block_group(trans, inode);
3834 btrfs_update_inode_block_group(trans, dir);
3835out_unlock:
d3c2fdcf 3836 nr = trans->blocks_used;
ab78c84d 3837 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3838fail:
39279cc3
CM		 3839 if (drop_inode) {
3840 inode_dec_link_count(inode);
3841 iput(inode);
3842 }
d3c2fdcf 3843 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3844 return err;
3845}
3846
3847static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3848 struct dentry *dentry)
3849{
3850 struct btrfs_trans_handle *trans;
3851 struct btrfs_root *root = BTRFS_I(dir)->root;
3852 struct inode *inode = old_dentry->d_inode;
00e4e6b3 3853 u64 index;
1832a6d5 3854 unsigned long nr = 0;
39279cc3
CM		 3855 int err;
3856 int drop_inode = 0;
3857
3858 if (inode->i_nlink == 0)
3859 return -ENOENT;
3860
e02119d5 3861 btrfs_inc_nlink(inode);
6a63209f 3862 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3863 if (err)
3864 goto fail;
3de4586c 3865 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 3866 if (err)
3867 goto fail;
3868
39279cc3 3869 trans = btrfs_start_transaction(root, 1);
5f39d397 3870
39279cc3
CM		 3871 btrfs_set_trans_block_group(trans, dir);
3872 atomic_inc(&inode->i_count);
aec7477b 3873
00e4e6b3 3874 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
5f39d397 3875
39279cc3
CM		 3876 if (err)
3877 drop_inode = 1;
5f39d397 3878
39279cc3 3879 btrfs_update_inode_block_group(trans, dir);
54aa1f4d 3880 err = btrfs_update_inode(trans, root, inode);
5f39d397 3881
54aa1f4d
CM		 3882 if (err)
3883 drop_inode = 1;
39279cc3 3884
d3c2fdcf 3885 nr = trans->blocks_used;
12fcfd22
CM		 3886
3887 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
ab78c84d 3888 btrfs_end_transaction_throttle(trans, root);
1832a6d5 3889fail:
39279cc3
CM		 3890 if (drop_inode) {
3891 inode_dec_link_count(inode);
3892 iput(inode);
3893 }
d3c2fdcf 3894 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3895 return err;
3896}
3897
39279cc3
CM		 3898static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3899{
b9d86667 3900 struct inode *inode = NULL;
39279cc3
CM		 3901 struct btrfs_trans_handle *trans;
3902 struct btrfs_root *root = BTRFS_I(dir)->root;
3903 int err = 0;
3904 int drop_on_err = 0;
b9d86667 3905 u64 objectid = 0;
00e4e6b3 3906 u64 index = 0;
d3c2fdcf 3907 unsigned long nr = 1;
39279cc3 3908
6a63209f 3909 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 3910 if (err)
3911 goto out_unlock;
3912
39279cc3
CM		 3913 trans = btrfs_start_transaction(root, 1);
3914 btrfs_set_trans_block_group(trans, dir);
5f39d397 3915
39279cc3
CM		 3916 if (IS_ERR(trans)) {
3917 err = PTR_ERR(trans);
3918 goto out_unlock;
3919 }
3920
3921 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3922 if (err) {
3923 err = -ENOSPC;
3924 goto out_unlock;
3925 }
3926
aec7477b 3927 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 3928 dentry->d_name.len,
3929 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 3930 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3931 &index);
39279cc3
CM		 3932 if (IS_ERR(inode)) {
3933 err = PTR_ERR(inode);
3934 goto out_fail;
3935 }
5f39d397 3936
39279cc3 3937 drop_on_err = 1;
33268eaf 3938
0279b4cd 3939 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 3940 if (err)
3941 goto out_fail;
3942
39279cc3
CM		 3943 inode->i_op = &btrfs_dir_inode_operations;
3944 inode->i_fop = &btrfs_dir_file_operations;
3945 btrfs_set_trans_block_group(trans, inode);
3946
dbe674a9 3947 btrfs_i_size_write(inode, 0);
39279cc3
CM		 3948 err = btrfs_update_inode(trans, root, inode);
3949 if (err)
3950 goto out_fail;
5f39d397 3951
e02119d5
CM		 3952 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3953 inode, dentry->d_name.name,
3954 dentry->d_name.len, 0, index);
39279cc3
CM		 3955 if (err)
3956 goto out_fail;
5f39d397 3957
39279cc3
CM		 3958 d_instantiate(dentry, inode);
3959 drop_on_err = 0;
39279cc3
CM		 3960 btrfs_update_inode_block_group(trans, inode);
3961 btrfs_update_inode_block_group(trans, dir);
3962
3963out_fail:
d3c2fdcf 3964 nr = trans->blocks_used;
ab78c84d 3965 btrfs_end_transaction_throttle(trans, root);
5f39d397 3966
39279cc3 3967out_unlock:
39279cc3
CM		 3968 if (drop_on_err)
3969 iput(inode);
d3c2fdcf 3970 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3971 return err;
3972}
3973
d352ac68
CM		 3974/* helper for btfs_get_extent. Given an existing extent in the tree,
3975 * and an extent that you want to insert, deal with overlap and insert
3976 * the new extent into the tree.
3977 */
3b951516
CM		 3978static int merge_extent_mapping(struct extent_map_tree *em_tree,
3979 struct extent_map *existing,
e6dcd2dc
CM		 3980 struct extent_map *em,
3981 u64 map_start, u64 map_len)
3b951516
CM		 3982{
3983 u64 start_diff;
3b951516 3984
e6dcd2dc
CM		 3985 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3986 start_diff = map_start - em->start;
3987 em->start = map_start;
3988 em->len = map_len;
c8b97818
CM		 3989 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3990 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 3991 em->block_start += start_diff;
c8b97818
CM		 3992 em->block_len -= start_diff;
3993 }
e6dcd2dc 3994 return add_extent_mapping(em_tree, em);
3b951516
CM		 3995}
3996
c8b97818
CM		 3997static noinline int uncompress_inline(struct btrfs_path *path,
3998 struct inode *inode, struct page *page,
3999 size_t pg_offset, u64 extent_offset,
4000 struct btrfs_file_extent_item *item)
4001{
4002 int ret;
4003 struct extent_buffer *leaf = path->nodes[0];
4004 char *tmp;
4005 size_t max_size;
4006 unsigned long inline_size;
4007 unsigned long ptr;
4008
4009 WARN_ON(pg_offset != 0);
4010 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4011 inline_size = btrfs_file_extent_inline_item_len(leaf,
4012 btrfs_item_nr(leaf, path->slots[0]));
4013 tmp = kmalloc(inline_size, GFP_NOFS);
4014 ptr = btrfs_file_extent_inline_start(item);
4015
4016 read_extent_buffer(leaf, tmp, ptr, inline_size);
4017
5b050f04 4018 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
c8b97818
CM		 4019 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4020 inline_size, max_size);
4021 if (ret) {
4022 char *kaddr = kmap_atomic(page, KM_USER0);
4023 unsigned long copy_size = min_t(u64,
4024 PAGE_CACHE_SIZE - pg_offset,
4025 max_size - extent_offset);
4026 memset(kaddr + pg_offset, 0, copy_size);
4027 kunmap_atomic(kaddr, KM_USER0);
4028 }
4029 kfree(tmp);
4030 return 0;
4031}
4032
d352ac68
CM		 4033/*
4034 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 4035 * the ugly parts come from merging extents from the disk with the in-ram
4036 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 4037 * where the in-ram extents might be locked pending data=ordered completion.
4038 *
4039 * This also copies inline extents directly into the page.
4040 */
d397712b 4041
a52d9a80 4042struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 4043 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 4044 int create)
4045{
4046 int ret;
4047 int err = 0;
db94535d 4048 u64 bytenr;
a52d9a80
CM		 4049 u64 extent_start = 0;
4050 u64 extent_end = 0;
4051 u64 objectid = inode->i_ino;
4052 u32 found_type;
f421950f 4053 struct btrfs_path *path = NULL;
a52d9a80
CM		 4054 struct btrfs_root *root = BTRFS_I(inode)->root;
4055 struct btrfs_file_extent_item *item;
5f39d397
CM		 4056 struct extent_buffer *leaf;
4057 struct btrfs_key found_key;
a52d9a80
CM		 4058 struct extent_map *em = NULL;
4059 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 4060 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 4061 struct btrfs_trans_handle *trans = NULL;
c8b97818 4062 int compressed;
a52d9a80 4063
a52d9a80 4064again:
d1310b2e
CM		 4065 spin_lock(&em_tree->lock);
4066 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 4067 if (em)
4068 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 4069 spin_unlock(&em_tree->lock);
4070
a52d9a80 4071 if (em) {
e1c4b745
CM		 4072 if (em->start > start || em->start + em->len <= start)
4073 free_extent_map(em);
4074 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 4075 free_extent_map(em);
4076 else
4077 goto out;
a52d9a80 4078 }
d1310b2e 4079 em = alloc_extent_map(GFP_NOFS);
a52d9a80 4080 if (!em) {
d1310b2e
CM		 4081 err = -ENOMEM;
4082 goto out;
a52d9a80 4083 }
e6dcd2dc 4084 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 4085 em->start = EXTENT_MAP_HOLE;
445a6944 4086 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 4087 em->len = (u64)-1;
c8b97818 4088 em->block_len = (u64)-1;
f421950f
CM		 4089
4090 if (!path) {
4091 path = btrfs_alloc_path();
4092 BUG_ON(!path);
4093 }
4094
179e29e4
CM		 4095 ret = btrfs_lookup_file_extent(trans, root, path,
4096 objectid, start, trans != NULL);
a52d9a80
CM		 4097 if (ret < 0) {
4098 err = ret;
4099 goto out;
4100 }
4101
4102 if (ret != 0) {
4103 if (path->slots[0] == 0)
4104 goto not_found;
4105 path->slots[0]--;
4106 }
4107
5f39d397
CM		 4108 leaf = path->nodes[0];
4109 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 4110 struct btrfs_file_extent_item);
a52d9a80 4111 /* are we inside the extent that was found? */
5f39d397
CM		 4112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4113 found_type = btrfs_key_type(&found_key);
4114 if (found_key.objectid != objectid ||
a52d9a80
CM		 4115 found_type != BTRFS_EXTENT_DATA_KEY) {
4116 goto not_found;
4117 }
4118
5f39d397
CM		 4119 found_type = btrfs_file_extent_type(leaf, item);
4120 extent_start = found_key.offset;
c8b97818 4121 compressed = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 4122 if (found_type == BTRFS_FILE_EXTENT_REG ||
4123 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 4124 extent_end = extent_start +
db94535d 4125 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 4126 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4127 size_t size;
4128 size = btrfs_file_extent_inline_len(leaf, item);
4129 extent_end = (extent_start + size + root->sectorsize - 1) &
4130 ~((u64)root->sectorsize - 1);
4131 }
4132
4133 if (start >= extent_end) {
4134 path->slots[0]++;
4135 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4136 ret = btrfs_next_leaf(root, path);
4137 if (ret < 0) {
4138 err = ret;
4139 goto out;
a52d9a80 4140 }
9036c102
YZ		 4141 if (ret > 0)
4142 goto not_found;
4143 leaf = path->nodes[0];
a52d9a80 4144 }
9036c102
YZ		 4145 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4146 if (found_key.objectid != objectid ||
4147 found_key.type != BTRFS_EXTENT_DATA_KEY)
4148 goto not_found;
4149 if (start + len <= found_key.offset)
4150 goto not_found;
4151 em->start = start;
4152 em->len = found_key.offset - start;
4153 goto not_found_em;
4154 }
4155
d899e052
YZ		 4156 if (found_type == BTRFS_FILE_EXTENT_REG ||
4157 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 4158 em->start = extent_start;
4159 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 4160 em->orig_start = extent_start -
4161 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 4162 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4163 if (bytenr == 0) {
5f39d397 4164 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 4165 goto insert;
4166 }
c8b97818
CM		 4167 if (compressed) {
4168 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4169 em->block_start = bytenr;
4170 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4171 item);
4172 } else {
4173 bytenr += btrfs_file_extent_offset(leaf, item);
4174 em->block_start = bytenr;
4175 em->block_len = em->len;
d899e052
YZ		 4176 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4177 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 4178 }
a52d9a80
CM		 4179 goto insert;
4180 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4181 unsigned long ptr;
a52d9a80 4182 char *map;
3326d1b0
CM		 4183 size_t size;
4184 size_t extent_offset;
4185 size_t copy_size;
a52d9a80 4186
689f9346 4187 em->block_start = EXTENT_MAP_INLINE;
c8b97818 4188 if (!page || create) {
689f9346 4189 em->start = extent_start;
9036c102 4190 em->len = extent_end - extent_start;
689f9346
Y		 4191 goto out;
4192 }
5f39d397 4193
9036c102
YZ		 4194 size = btrfs_file_extent_inline_len(leaf, item);
4195 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 4196 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 4197 size - extent_offset);
3326d1b0 4198 em->start = extent_start + extent_offset;
70dec807
CM		 4199 em->len = (copy_size + root->sectorsize - 1) &
4200 ~((u64)root->sectorsize - 1);
ff5b7ee3 4201 em->orig_start = EXTENT_MAP_INLINE;
c8b97818
CM		 4202 if (compressed)
4203 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
689f9346 4204 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 4205 if (create == 0 && !PageUptodate(page)) {
c8b97818
CM		 4206 if (btrfs_file_extent_compression(leaf, item) ==
4207 BTRFS_COMPRESS_ZLIB) {
4208 ret = uncompress_inline(path, inode, page,
4209 pg_offset,
4210 extent_offset, item);
4211 BUG_ON(ret);
4212 } else {
4213 map = kmap(page);
4214 read_extent_buffer(leaf, map + pg_offset, ptr,
4215 copy_size);
4216 kunmap(page);
4217 }
179e29e4
CM		 4218 flush_dcache_page(page);
4219 } else if (create && PageUptodate(page)) {
4220 if (!trans) {
4221 kunmap(page);
4222 free_extent_map(em);
4223 em = NULL;
4224 btrfs_release_path(root, path);
f9295749 4225 trans = btrfs_join_transaction(root, 1);
179e29e4
CM		 4226 goto again;
4227 }
c8b97818 4228 map = kmap(page);
70dec807 4229 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 4230 copy_size);
c8b97818 4231 kunmap(page);
179e29e4 4232 btrfs_mark_buffer_dirty(leaf);
a52d9a80 4233 }
d1310b2e
CM		 4234 set_extent_uptodate(io_tree, em->start,
4235 extent_map_end(em) - 1, GFP_NOFS);
a52d9a80
CM		 4236 goto insert;
4237 } else {
d397712b 4238 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 4239 WARN_ON(1);
4240 }
4241not_found:
4242 em->start = start;
d1310b2e 4243 em->len = len;
a52d9a80 4244not_found_em:
5f39d397 4245 em->block_start = EXTENT_MAP_HOLE;
9036c102 4246 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80
CM		 4247insert:
4248 btrfs_release_path(root, path);
d1310b2e 4249 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 4250 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4251 "[%llu %llu]\n", (unsigned long long)em->start,
4252 (unsigned long long)em->len,
4253 (unsigned long long)start,
4254 (unsigned long long)len);
a52d9a80
CM		 4255 err = -EIO;
4256 goto out;
4257 }
d1310b2e
CM		 4258
4259 err = 0;
4260 spin_lock(&em_tree->lock);
a52d9a80 4261 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 4262 /* it is possible that someone inserted the extent into the tree
4263 * while we had the lock dropped. It is also possible that
4264 * an overlapping map exists in the tree
4265 */
a52d9a80 4266 if (ret == -EEXIST) {
3b951516 4267 struct extent_map *existing;
e6dcd2dc
CM		 4268
4269 ret = 0;
4270
3b951516 4271 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 4272 if (existing && (existing->start > start ||
4273 existing->start + existing->len <= start)) {
4274 free_extent_map(existing);
4275 existing = NULL;
4276 }
3b951516
CM		 4277 if (!existing) {
4278 existing = lookup_extent_mapping(em_tree, em->start,
4279 em->len);
4280 if (existing) {
4281 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 4282 em, start,
4283 root->sectorsize);
3b951516
CM		 4284 free_extent_map(existing);
4285 if (err) {
4286 free_extent_map(em);
4287 em = NULL;
4288 }
4289 } else {
4290 err = -EIO;
3b951516
CM		 4291 free_extent_map(em);
4292 em = NULL;
4293 }
4294 } else {
4295 free_extent_map(em);
4296 em = existing;
e6dcd2dc 4297 err = 0;
a52d9a80 4298 }
a52d9a80 4299 }
d1310b2e 4300 spin_unlock(&em_tree->lock);
a52d9a80 4301out:
f421950f
CM		 4302 if (path)
4303 btrfs_free_path(path);
a52d9a80
CM		 4304 if (trans) {
4305 ret = btrfs_end_transaction(trans, root);
d397712b 4306 if (!err)
a52d9a80
CM		 4307 err = ret;
4308 }
a52d9a80
CM		 4309 if (err) {
4310 free_extent_map(em);
a52d9a80
CM		 4311 return ERR_PTR(err);
4312 }
4313 return em;
4314}
4315
16432985
CM		 4316static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4317 const struct iovec *iov, loff_t offset,
4318 unsigned long nr_segs)
4319{
e1c4b745 4320 return -EINVAL;
16432985
CM		 4321}
4322
1506fcc8
YS		 4323static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4324 __u64 start, __u64 len)
4325{
4326 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4327}
4328
a52d9a80 4329int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 4330{
d1310b2e
CM		 4331 struct extent_io_tree *tree;
4332 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4333 return extent_read_full_page(tree, page, btrfs_get_extent);
9ebefb18 4334}
1832a6d5 4335
a52d9a80 4336static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 4337{
d1310b2e 4338 struct extent_io_tree *tree;
b888db2b
CM		 4339
4340
4341 if (current->flags & PF_MEMALLOC) {
4342 redirty_page_for_writepage(wbc, page);
4343 unlock_page(page);
4344 return 0;
4345 }
d1310b2e 4346 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 4347 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 4348}
4349
f421950f
CM		 4350int btrfs_writepages(struct address_space *mapping,
4351 struct writeback_control *wbc)
b293f02e 4352{
d1310b2e 4353 struct extent_io_tree *tree;
771ed689 4354
d1310b2e 4355 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 4356 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4357}
4358
3ab2fb5a
CM		 4359static int
4360btrfs_readpages(struct file *file, struct address_space *mapping,
4361 struct list_head *pages, unsigned nr_pages)
4362{
d1310b2e
CM		 4363 struct extent_io_tree *tree;
4364 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 4365 return extent_readpages(tree, mapping, pages, nr_pages,
4366 btrfs_get_extent);
4367}
e6dcd2dc 4368static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 4369{
d1310b2e
CM		 4370 struct extent_io_tree *tree;
4371 struct extent_map_tree *map;
a52d9a80 4372 int ret;
8c2383c3 4373
d1310b2e
CM		 4374 tree = &BTRFS_I(page->mapping->host)->io_tree;
4375 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 4376 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 4377 if (ret == 1) {
4378 ClearPagePrivate(page);
4379 set_page_private(page, 0);
4380 page_cache_release(page);
39279cc3 4381 }
a52d9a80 4382 return ret;
39279cc3
CM		 4383}
4384
e6dcd2dc
CM		 4385static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4386{
98509cfc
CM		 4387 if (PageWriteback(page) || PageDirty(page))
4388 return 0;
b335b003 4389 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 4390}
4391
a52d9a80 4392static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 4393{
d1310b2e 4394 struct extent_io_tree *tree;
e6dcd2dc
CM		 4395 struct btrfs_ordered_extent *ordered;
4396 u64 page_start = page_offset(page);
4397 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 4398
e6dcd2dc 4399 wait_on_page_writeback(page);
d1310b2e 4400 tree = &BTRFS_I(page->mapping->host)->io_tree;
e6dcd2dc
CM		 4401 if (offset) {
4402 btrfs_releasepage(page, GFP_NOFS);
4403 return;
4404 }
4405
4406 lock_extent(tree, page_start, page_end, GFP_NOFS);
4407 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4408 page_offset(page));
4409 if (ordered) {
eb84ae03
CM		 4410 /*
4411 * IO on this page will never be started, so we need
4412 * to account for any ordered extents now
4413 */
e6dcd2dc
CM		 4414 clear_extent_bit(tree, page_start, page_end,
4415 EXTENT_DIRTY | EXTENT_DELALLOC |
4416 EXTENT_LOCKED, 1, 0, GFP_NOFS);
211f90e6
CM		 4417 btrfs_finish_ordered_io(page->mapping->host,
4418 page_start, page_end);
e6dcd2dc
CM		 4419 btrfs_put_ordered_extent(ordered);
4420 lock_extent(tree, page_start, page_end, GFP_NOFS);
4421 }
4422 clear_extent_bit(tree, page_start, page_end,
4423 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4424 EXTENT_ORDERED,
4425 1, 1, GFP_NOFS);
4426 __btrfs_releasepage(page, GFP_NOFS);
4427
4a096752 4428 ClearPageChecked(page);
9ad6b7bc 4429 if (PagePrivate(page)) {
9ad6b7bc
CM		 4430 ClearPagePrivate(page);
4431 set_page_private(page, 0);
4432 page_cache_release(page);
4433 }
39279cc3
CM		 4434}
4435
9ebefb18
CM		 4436/*
4437 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4438 * called from a page fault handler when a page is first dirtied. Hence we must
4439 * be careful to check for EOF conditions here. We set the page up correctly
4440 * for a written page which means we get ENOSPC checking when writing into
4441 * holes and correct delalloc and unwritten extent mapping on filesystems that
4442 * support these features.
4443 *
4444 * We are not allowed to take the i_mutex here so we have to play games to
4445 * protect against truncate races as the page could now be beyond EOF. Because
4446 * vmtruncate() writes the inode size before removing pages, once we have the
4447 * page lock we can determine safely if the page is beyond EOF. If it is not
4448 * beyond EOF, then the page is guaranteed safe against truncation until we
4449 * unlock the page.
4450 */
c2ec175c 4451int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 4452{
c2ec175c 4453 struct page *page = vmf->page;
6da6abae 4454 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 4455 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4456 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4457 struct btrfs_ordered_extent *ordered;
4458 char *kaddr;
4459 unsigned long zero_start;
9ebefb18 4460 loff_t size;
1832a6d5 4461 int ret;
a52d9a80 4462 u64 page_start;
e6dcd2dc 4463 u64 page_end;
9ebefb18 4464
6a63209f 4465 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
56a76f82
NP		 4466 if (ret) {
4467 if (ret == -ENOMEM)
4468 ret = VM_FAULT_OOM;
4469 else /* -ENOSPC, -EIO, etc */
4470 ret = VM_FAULT_SIGBUS;
1832a6d5 4471 goto out;
56a76f82 4472 }
1832a6d5 4473
56a76f82 4474 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 4475again:
9ebefb18 4476 lock_page(page);
9ebefb18 4477 size = i_size_read(inode);
e6dcd2dc
CM		 4478 page_start = page_offset(page);
4479 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 4480
9ebefb18 4481 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 4482 (page_start >= size)) {
6a63209f 4483 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
9ebefb18
CM		 4484 /* page got truncated out from underneath us */
4485 goto out_unlock;
4486 }
e6dcd2dc
CM		 4487 wait_on_page_writeback(page);
4488
4489 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4490 set_page_extent_mapped(page);
4491
eb84ae03
CM		 4492 /*
4493 * we can't set the delalloc bits if there are pending ordered
4494 * extents. Drop our locks and wait for them to finish
4495 */
e6dcd2dc
CM		 4496 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4497 if (ordered) {
4498 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4499 unlock_page(page);
eb84ae03 4500 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4501 btrfs_put_ordered_extent(ordered);
4502 goto again;
4503 }
4504
ea8c2819 4505 btrfs_set_extent_delalloc(inode, page_start, page_end);
e6dcd2dc 4506 ret = 0;
9ebefb18
CM		 4507
4508 /* page is wholly or partially inside EOF */
a52d9a80 4509 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 4510 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 4511 else
e6dcd2dc 4512 zero_start = PAGE_CACHE_SIZE;
9ebefb18 4513
e6dcd2dc
CM		 4514 if (zero_start != PAGE_CACHE_SIZE) {
4515 kaddr = kmap(page);
4516 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4517 flush_dcache_page(page);
4518 kunmap(page);
4519 }
247e743c 4520 ClearPageChecked(page);
e6dcd2dc 4521 set_page_dirty(page);
5a3f23d5
CM		 4522
4523 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
e6dcd2dc 4524 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
9ebefb18
CM		 4525
4526out_unlock:
4527 unlock_page(page);
1832a6d5 4528out:
9ebefb18
CM		 4529 return ret;
4530}
4531
39279cc3
CM		 4532static void btrfs_truncate(struct inode *inode)
4533{
4534 struct btrfs_root *root = BTRFS_I(inode)->root;
4535 int ret;
4536 struct btrfs_trans_handle *trans;
d3c2fdcf 4537 unsigned long nr;
dbe674a9 4538 u64 mask = root->sectorsize - 1;
39279cc3
CM		 4539
4540 if (!S_ISREG(inode->i_mode))
4541 return;
4542 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4543 return;
4544
4545 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4a096752 4546 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
39279cc3 4547
39279cc3 4548 trans = btrfs_start_transaction(root, 1);
5a3f23d5
CM		 4549
4550 /*
4551 * setattr is responsible for setting the ordered_data_close flag,
4552 * but that is only tested during the last file release. That
4553 * could happen well after the next commit, leaving a great big
4554 * window where new writes may get lost if someone chooses to write
4555 * to this file after truncating to zero
4556 *
4557 * The inode doesn't have any dirty data here, and so if we commit
4558 * this is a noop. If someone immediately starts writing to the inode
4559 * it is very likely we'll catch some of their writes in this
4560 * transaction, and the commit will find this file on the ordered
4561 * data list with good things to send down.
4562 *
4563 * This is a best effort solution, there is still a window where
4564 * using truncate to replace the contents of the file will
4565 * end up with a zero length file after a crash.
4566 */
4567 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4568 btrfs_add_ordered_operation(trans, root, inode);
4569
39279cc3 4570 btrfs_set_trans_block_group(trans, inode);
dbe674a9 4571 btrfs_i_size_write(inode, inode->i_size);
39279cc3 4572
7b128766
JB		 4573 ret = btrfs_orphan_add(trans, inode);
4574 if (ret)
4575 goto out;
39279cc3 4576 /* FIXME, add redo link to tree so we don't leak on crash */
e02119d5 4577 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
85e21bac 4578 BTRFS_EXTENT_DATA_KEY);
39279cc3 4579 btrfs_update_inode(trans, root, inode);
5f39d397 4580
7b128766
JB		 4581 ret = btrfs_orphan_del(trans, inode);
4582 BUG_ON(ret);
4583
4584out:
4585 nr = trans->blocks_used;
89ce8a63 4586 ret = btrfs_end_transaction_throttle(trans, root);
39279cc3 4587 BUG_ON(ret);
d3c2fdcf 4588 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 4589}
4590
d352ac68
CM		 4591/*
4592 * create a new subvolume directory/inode (helper for the ioctl).
4593 */
d2fb3437
YZ		 4594int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4595 struct btrfs_root *new_root, struct dentry *dentry,
4596 u64 new_dirid, u64 alloc_hint)
39279cc3 4597{
39279cc3 4598 struct inode *inode;
cb8e7090 4599 int error;
00e4e6b3 4600 u64 index = 0;
39279cc3 4601
aec7477b 4602 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
d2fb3437 4603 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
54aa1f4d 4604 if (IS_ERR(inode))
f46b5a66 4605 return PTR_ERR(inode);
39279cc3
CM		 4606 inode->i_op = &btrfs_dir_inode_operations;
4607 inode->i_fop = &btrfs_dir_file_operations;
4608
39279cc3 4609 inode->i_nlink = 1;
dbe674a9 4610 btrfs_i_size_write(inode, 0);
3b96362c 4611
cb8e7090
CH		 4612 error = btrfs_update_inode(trans, new_root, inode);
4613 if (error)
4614 return error;
4615
4616 d_instantiate(dentry, inode);
4617 return 0;
39279cc3
CM		 4618}
4619
d352ac68
CM		 4620/* helper function for file defrag and space balancing. This
4621 * forces readahead on a given range of bytes in an inode
4622 */
edbd8d4e 4623unsigned long btrfs_force_ra(struct address_space *mapping,
86479a04
CM		 4624 struct file_ra_state *ra, struct file *file,
4625 pgoff_t offset, pgoff_t last_index)
4626{
8e7bf94f 4627 pgoff_t req_size = last_index - offset + 1;
86479a04 4628
86479a04
CM		 4629 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4630 return offset + req_size;
86479a04
CM		 4631}
4632
39279cc3
CM		 4633struct inode *btrfs_alloc_inode(struct super_block *sb)
4634{
4635 struct btrfs_inode *ei;
4636
4637 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4638 if (!ei)
4639 return NULL;
15ee9bc7 4640 ei->last_trans = 0;
e02119d5 4641 ei->logged_trans = 0;
e6dcd2dc 4642 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
33268eaf
JB		 4643 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4644 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
7b128766 4645 INIT_LIST_HEAD(&ei->i_orphan);
5a3f23d5 4646 INIT_LIST_HEAD(&ei->ordered_operations);
39279cc3
CM		 4647 return &ei->vfs_inode;
4648}
4649
4650void btrfs_destroy_inode(struct inode *inode)
4651{
e6dcd2dc 4652 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 4653 struct btrfs_root *root = BTRFS_I(inode)->root;
4654
39279cc3
CM		 4655 WARN_ON(!list_empty(&inode->i_dentry));
4656 WARN_ON(inode->i_data.nrpages);
4657
33268eaf
JB		 4658 if (BTRFS_I(inode)->i_acl &&
4659 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4660 posix_acl_release(BTRFS_I(inode)->i_acl);
4661 if (BTRFS_I(inode)->i_default_acl &&
4662 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4663 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4664
5a3f23d5
CM		 4665 /*
4666 * Make sure we're properly removed from the ordered operation
4667 * lists.
4668 */
4669 smp_mb();
4670 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4671 spin_lock(&root->fs_info->ordered_extent_lock);
4672 list_del_init(&BTRFS_I(inode)->ordered_operations);
4673 spin_unlock(&root->fs_info->ordered_extent_lock);
4674 }
4675
4676 spin_lock(&root->list_lock);
7b128766
JB		 4677 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4678 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4679 " list\n", inode->i_ino);
4680 dump_stack();
4681 }
5a3f23d5 4682 spin_unlock(&root->list_lock);
7b128766 4683
d397712b 4684 while (1) {
e6dcd2dc
CM		 4685 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4686 if (!ordered)
4687 break;
4688 else {
d397712b
CM		 4689 printk(KERN_ERR "btrfs found ordered "
4690 "extent %llu %llu on inode cleanup\n",
4691 (unsigned long long)ordered->file_offset,
4692 (unsigned long long)ordered->len);
e6dcd2dc
CM		 4693 btrfs_remove_ordered_extent(inode, ordered);
4694 btrfs_put_ordered_extent(ordered);
4695 btrfs_put_ordered_extent(ordered);
4696 }
4697 }
5d4f98a2 4698 inode_tree_del(inode);
5b21f2ed 4699 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
39279cc3
CM		 4700 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4701}
4702
0ee0fda0 4703static void init_once(void *foo)
39279cc3
CM		 4704{
4705 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4706
4707 inode_init_once(&ei->vfs_inode);
4708}
4709
4710void btrfs_destroy_cachep(void)
4711{
4712 if (btrfs_inode_cachep)
4713 kmem_cache_destroy(btrfs_inode_cachep);
4714 if (btrfs_trans_handle_cachep)
4715 kmem_cache_destroy(btrfs_trans_handle_cachep);
4716 if (btrfs_transaction_cachep)
4717 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 4718 if (btrfs_path_cachep)
4719 kmem_cache_destroy(btrfs_path_cachep);
4720}
4721
4722int btrfs_init_cachep(void)
4723{
9601e3f6
CH		 4724 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4725 sizeof(struct btrfs_inode), 0,
4726 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 4727 if (!btrfs_inode_cachep)
4728 goto fail;
9601e3f6
CH		 4729
4730 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4731 sizeof(struct btrfs_trans_handle), 0,
4732 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 4733 if (!btrfs_trans_handle_cachep)
4734 goto fail;
9601e3f6
CH		 4735
4736 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4737 sizeof(struct btrfs_transaction), 0,
4738 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 4739 if (!btrfs_transaction_cachep)
4740 goto fail;
9601e3f6
CH		 4741
4742 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4743 sizeof(struct btrfs_path), 0,
4744 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 4745 if (!btrfs_path_cachep)
4746 goto fail;
9601e3f6 4747
39279cc3
CM		 4748 return 0;
4749fail:
4750 btrfs_destroy_cachep();
4751 return -ENOMEM;
4752}
4753
4754static int btrfs_getattr(struct vfsmount *mnt,
4755 struct dentry *dentry, struct kstat *stat)
4756{
4757 struct inode *inode = dentry->d_inode;
4758 generic_fillattr(inode, stat);
3394e160 4759 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
d6667462 4760 stat->blksize = PAGE_CACHE_SIZE;
a76a3cd4
YZ		 4761 stat->blocks = (inode_get_bytes(inode) +
4762 BTRFS_I(inode)->delalloc_bytes) >> 9;
39279cc3
CM		 4763 return 0;
4764}
4765
d397712b
CM		 4766static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4767 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 4768{
4769 struct btrfs_trans_handle *trans;
4770 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4771 struct inode *new_inode = new_dentry->d_inode;
4772 struct inode *old_inode = old_dentry->d_inode;
4773 struct timespec ctime = CURRENT_TIME;
00e4e6b3 4774 u64 index = 0;
39279cc3
CM		 4775 int ret;
4776
3394e160
CM		 4777 /* we're not allowed to rename between subvolumes */
4778 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4779 BTRFS_I(new_dir)->root->root_key.objectid)
4780 return -EXDEV;
4781
39279cc3
CM		 4782 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4783 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4784 return -ENOTEMPTY;
4785 }
5f39d397 4786
0660b5af
CM		 4787 /* to rename a snapshot or subvolume, we need to juggle the
4788 * backrefs. This isn't coded yet
4789 */
4790 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4791 return -EXDEV;
4792
6a63209f 4793 ret = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4794 if (ret)
4795 goto out_unlock;
4796
5a3f23d5
CM		 4797 /*
4798 * we're using rename to replace one file with another.
4799 * and the replacement file is large. Start IO on it now so
4800 * we don't add too much work to the end of the transaction
4801 */
4802 if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4803 new_inode->i_size &&
4804 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4805 filemap_flush(old_inode->i_mapping);
4806
39279cc3 4807 trans = btrfs_start_transaction(root, 1);
5f39d397 4808
5a3f23d5
CM		 4809 /*
4810 * make sure the inode gets flushed if it is replacing
4811 * something.
4812 */
4813 if (new_inode && new_inode->i_size &&
4814 old_inode && S_ISREG(old_inode->i_mode)) {
4815 btrfs_add_ordered_operation(trans, root, old_inode);
4816 }
4817
12fcfd22
CM		 4818 /*
4819 * this is an ugly little race, but the rename is required to make
4820 * sure that if we crash, the inode is either at the old name
4821 * or the new one. pinning the log transaction lets us make sure
4822 * we don't allow a log commit to come in after we unlink the
4823 * name but before we add the new name back in.
4824 */
4825 btrfs_pin_log_trans(root);
4826
39279cc3 4827 btrfs_set_trans_block_group(trans, new_dir);
39279cc3 4828
e02119d5 4829 btrfs_inc_nlink(old_dentry->d_inode);
39279cc3
CM		 4830 old_dir->i_ctime = old_dir->i_mtime = ctime;
4831 new_dir->i_ctime = new_dir->i_mtime = ctime;
4832 old_inode->i_ctime = ctime;
5f39d397 4833
12fcfd22
CM		 4834 if (old_dentry->d_parent != new_dentry->d_parent)
4835 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4836
e02119d5
CM		 4837 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4838 old_dentry->d_name.name,
4839 old_dentry->d_name.len);
39279cc3
CM		 4840 if (ret)
4841 goto out_fail;
4842
4843 if (new_inode) {
4844 new_inode->i_ctime = CURRENT_TIME;
e02119d5
CM		 4845 ret = btrfs_unlink_inode(trans, root, new_dir,
4846 new_dentry->d_inode,
4847 new_dentry->d_name.name,
4848 new_dentry->d_name.len);
39279cc3
CM		 4849 if (ret)
4850 goto out_fail;
7b128766 4851 if (new_inode->i_nlink == 0) {
e02119d5 4852 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7b128766
JB		 4853 if (ret)
4854 goto out_fail;
4855 }
e02119d5 4856
39279cc3 4857 }
3de4586c 4858 ret = btrfs_set_inode_index(new_dir, &index);
aec7477b
JB		 4859 if (ret)
4860 goto out_fail;
4861
e02119d5
CM		 4862 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4863 old_inode, new_dentry->d_name.name,
4864 new_dentry->d_name.len, 1, index);
39279cc3
CM		 4865 if (ret)
4866 goto out_fail;
4867
12fcfd22
CM		 4868 btrfs_log_new_name(trans, old_inode, old_dir,
4869 new_dentry->d_parent);
39279cc3 4870out_fail:
12fcfd22
CM		 4871
4872 /* this btrfs_end_log_trans just allows the current
4873 * log-sub transaction to complete
4874 */
4875 btrfs_end_log_trans(root);
ab78c84d 4876 btrfs_end_transaction_throttle(trans, root);
1832a6d5 4877out_unlock:
39279cc3
CM		 4878 return ret;
4879}
4880
d352ac68
CM		 4881/*
4882 * some fairly slow code that needs optimization. This walks the list
4883 * of all the inodes with pending delalloc and forces them to disk.
4884 */
ea8c2819
CM		 4885int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4886{
4887 struct list_head *head = &root->fs_info->delalloc_inodes;
4888 struct btrfs_inode *binode;
5b21f2ed 4889 struct inode *inode;
ea8c2819 4890
c146afad
YZ		 4891 if (root->fs_info->sb->s_flags & MS_RDONLY)
4892 return -EROFS;
4893
75eff68e 4894 spin_lock(&root->fs_info->delalloc_lock);
d397712b 4895 while (!list_empty(head)) {
ea8c2819
CM		 4896 binode = list_entry(head->next, struct btrfs_inode,
4897 delalloc_inodes);
5b21f2ed
ZY		 4898 inode = igrab(&binode->vfs_inode);
4899 if (!inode)
4900 list_del_init(&binode->delalloc_inodes);
75eff68e 4901 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 4902 if (inode) {
8c8bee1d 4903 filemap_flush(inode->i_mapping);
5b21f2ed
ZY		 4904 iput(inode);
4905 }
4906 cond_resched();
75eff68e 4907 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 4908 }
75eff68e 4909 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 4910
4911 /* the filemap_flush will queue IO into the worker threads, but
4912 * we have to make sure the IO is actually started and that
4913 * ordered extents get created before we return
4914 */
4915 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 4916 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 4917 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 4918 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 4919 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4920 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 4921 }
4922 atomic_dec(&root->fs_info->async_submit_draining);
ea8c2819
CM		 4923 return 0;
4924}
4925
39279cc3
CM		 4926static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4927 const char *symname)
4928{
4929 struct btrfs_trans_handle *trans;
4930 struct btrfs_root *root = BTRFS_I(dir)->root;
4931 struct btrfs_path *path;
4932 struct btrfs_key key;
1832a6d5 4933 struct inode *inode = NULL;
39279cc3
CM		 4934 int err;
4935 int drop_inode = 0;
4936 u64 objectid;
00e4e6b3 4937 u64 index = 0 ;
39279cc3
CM		 4938 int name_len;
4939 int datasize;
5f39d397 4940 unsigned long ptr;
39279cc3 4941 struct btrfs_file_extent_item *ei;
5f39d397 4942 struct extent_buffer *leaf;
1832a6d5 4943 unsigned long nr = 0;
39279cc3
CM		 4944
4945 name_len = strlen(symname) + 1;
4946 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4947 return -ENAMETOOLONG;
1832a6d5 4948
6a63209f 4949 err = btrfs_check_metadata_free_space(root);
1832a6d5
CM		 4950 if (err)
4951 goto out_fail;
4952
39279cc3
CM		 4953 trans = btrfs_start_transaction(root, 1);
4954 btrfs_set_trans_block_group(trans, dir);
4955
4956 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4957 if (err) {
4958 err = -ENOSPC;
4959 goto out_unlock;
4960 }
4961
aec7477b 4962 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9c58309d
CM		 4963 dentry->d_name.len,
4964 dentry->d_parent->d_inode->i_ino, objectid,
00e4e6b3
CM		 4965 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4966 &index);
39279cc3
CM		 4967 err = PTR_ERR(inode);
4968 if (IS_ERR(inode))
4969 goto out_unlock;
4970
0279b4cd 4971 err = btrfs_init_inode_security(inode, dir);
33268eaf
JB		 4972 if (err) {
4973 drop_inode = 1;
4974 goto out_unlock;
4975 }
4976
39279cc3 4977 btrfs_set_trans_block_group(trans, inode);
00e4e6b3 4978 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
39279cc3
CM		 4979 if (err)
4980 drop_inode = 1;
4981 else {
4982 inode->i_mapping->a_ops = &btrfs_aops;
04160088 4983 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3
CM		 4984 inode->i_fop = &btrfs_file_operations;
4985 inode->i_op = &btrfs_file_inode_operations;
d1310b2e 4986 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 4987 }
39279cc3
CM		 4988 btrfs_update_inode_block_group(trans, inode);
4989 btrfs_update_inode_block_group(trans, dir);
4990 if (drop_inode)
4991 goto out_unlock;
4992
4993 path = btrfs_alloc_path();
4994 BUG_ON(!path);
4995 key.objectid = inode->i_ino;
4996 key.offset = 0;
39279cc3
CM		 4997 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4998 datasize = btrfs_file_extent_calc_inline_size(name_len);
4999 err = btrfs_insert_empty_item(trans, root, path, &key,
5000 datasize);
54aa1f4d
CM		 5001 if (err) {
5002 drop_inode = 1;
5003 goto out_unlock;
5004 }
5f39d397
CM		 5005 leaf = path->nodes[0];
5006 ei = btrfs_item_ptr(leaf, path->slots[0],
5007 struct btrfs_file_extent_item);
5008 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5009 btrfs_set_file_extent_type(leaf, ei,
39279cc3 5010 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 5011 btrfs_set_file_extent_encryption(leaf, ei, 0);
5012 btrfs_set_file_extent_compression(leaf, ei, 0);
5013 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5014 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5015
39279cc3 5016 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 5017 write_extent_buffer(leaf, symname, ptr, name_len);
5018 btrfs_mark_buffer_dirty(leaf);
39279cc3 5019 btrfs_free_path(path);
5f39d397 5020
39279cc3
CM		 5021 inode->i_op = &btrfs_symlink_inode_operations;
5022 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 5023 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 5024 inode_set_bytes(inode, name_len);
dbe674a9 5025 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 5026 err = btrfs_update_inode(trans, root, inode);
5027 if (err)
5028 drop_inode = 1;
39279cc3
CM		 5029
5030out_unlock:
d3c2fdcf 5031 nr = trans->blocks_used;
ab78c84d 5032 btrfs_end_transaction_throttle(trans, root);
1832a6d5 5033out_fail:
39279cc3
CM		 5034 if (drop_inode) {
5035 inode_dec_link_count(inode);
5036 iput(inode);
5037 }
d3c2fdcf 5038 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5039 return err;
5040}
16432985 5041
546888da
CM		 5042static int prealloc_file_range(struct btrfs_trans_handle *trans,
5043 struct inode *inode, u64 start, u64 end,
e980b50c 5044 u64 locked_end, u64 alloc_hint, int mode)
d899e052 5045{
d899e052
YZ		 5046 struct btrfs_root *root = BTRFS_I(inode)->root;
5047 struct btrfs_key ins;
5048 u64 alloc_size;
5049 u64 cur_offset = start;
5050 u64 num_bytes = end - start;
5051 int ret = 0;
5052
d899e052
YZ		 5053 while (num_bytes > 0) {
5054 alloc_size = min(num_bytes, root->fs_info->max_extent);
5055 ret = btrfs_reserve_extent(trans, root, alloc_size,
5056 root->sectorsize, 0, alloc_hint,
5057 (u64)-1, &ins, 1);
5058 if (ret) {
5059 WARN_ON(1);
5060 goto out;
5061 }
5062 ret = insert_reserved_file_extent(trans, inode,
5063 cur_offset, ins.objectid,
5064 ins.offset, ins.offset,
e980b50c
CM		 5065 ins.offset, locked_end,
5066 0, 0, 0,
d899e052
YZ		 5067 BTRFS_FILE_EXTENT_PREALLOC);
5068 BUG_ON(ret);
5069 num_bytes -= ins.offset;
5070 cur_offset += ins.offset;
5071 alloc_hint = ins.objectid + ins.offset;
5072 }
5073out:
5074 if (cur_offset > start) {
5075 inode->i_ctime = CURRENT_TIME;
6cbff00f 5076 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052
YZ		 5077 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5078 cur_offset > i_size_read(inode))
5079 btrfs_i_size_write(inode, cur_offset);
5080 ret = btrfs_update_inode(trans, root, inode);
5081 BUG_ON(ret);
5082 }
5083
d899e052
YZ		 5084 return ret;
5085}
5086
5087static long btrfs_fallocate(struct inode *inode, int mode,
5088 loff_t offset, loff_t len)
5089{
5090 u64 cur_offset;
5091 u64 last_byte;
5092 u64 alloc_start;
5093 u64 alloc_end;
5094 u64 alloc_hint = 0;
e980b50c 5095 u64 locked_end;
d899e052
YZ		 5096 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5097 struct extent_map *em;
546888da 5098 struct btrfs_trans_handle *trans;
d899e052
YZ		 5099 int ret;
5100
5101 alloc_start = offset & ~mask;
5102 alloc_end = (offset + len + mask) & ~mask;
5103
546888da
CM		 5104 /*
5105 * wait for ordered IO before we have any locks. We'll loop again
5106 * below with the locks held.
5107 */
5108 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5109
d899e052
YZ		 5110 mutex_lock(&inode->i_mutex);
5111 if (alloc_start > inode->i_size) {
5112 ret = btrfs_cont_expand(inode, alloc_start);
5113 if (ret)
5114 goto out;
5115 }
5116
e980b50c 5117 locked_end = alloc_end - 1;
d899e052
YZ		 5118 while (1) {
5119 struct btrfs_ordered_extent *ordered;
546888da
CM		 5120
5121 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5122 if (!trans) {
5123 ret = -EIO;
5124 goto out;
5125 }
5126
5127 /* the extent lock is ordered inside the running
5128 * transaction
5129 */
e980b50c
CM		 5130 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5131 GFP_NOFS);
d899e052
YZ		 5132 ordered = btrfs_lookup_first_ordered_extent(inode,
5133 alloc_end - 1);
5134 if (ordered &&
5135 ordered->file_offset + ordered->len > alloc_start &&
5136 ordered->file_offset < alloc_end) {
5137 btrfs_put_ordered_extent(ordered);
5138 unlock_extent(&BTRFS_I(inode)->io_tree,
e980b50c 5139 alloc_start, locked_end, GFP_NOFS);
546888da
CM		 5140 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5141
5142 /*
5143 * we can't wait on the range with the transaction
5144 * running or with the extent lock held
5145 */
d899e052
YZ		 5146 btrfs_wait_ordered_range(inode, alloc_start,
5147 alloc_end - alloc_start);
5148 } else {
5149 if (ordered)
5150 btrfs_put_ordered_extent(ordered);
5151 break;
5152 }
5153 }
5154
5155 cur_offset = alloc_start;
5156 while (1) {
5157 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5158 alloc_end - cur_offset, 0);
5159 BUG_ON(IS_ERR(em) || !em);
5160 last_byte = min(extent_map_end(em), alloc_end);
5161 last_byte = (last_byte + mask) & ~mask;
5162 if (em->block_start == EXTENT_MAP_HOLE) {
546888da 5163 ret = prealloc_file_range(trans, inode, cur_offset,
e980b50c
CM		 5164 last_byte, locked_end + 1,
5165 alloc_hint, mode);
d899e052
YZ		 5166 if (ret < 0) {
5167 free_extent_map(em);
5168 break;
5169 }
5170 }
5171 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5172 alloc_hint = em->block_start;
5173 free_extent_map(em);
5174
5175 cur_offset = last_byte;
5176 if (cur_offset >= alloc_end) {
5177 ret = 0;
5178 break;
5179 }
5180 }
e980b50c 5181 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
d899e052 5182 GFP_NOFS);
546888da
CM		 5183
5184 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
d899e052
YZ		 5185out:
5186 mutex_unlock(&inode->i_mutex);
5187 return ret;
5188}
5189
e6dcd2dc
CM		 5190static int btrfs_set_page_dirty(struct page *page)
5191{
e6dcd2dc
CM		 5192 return __set_page_dirty_nobuffers(page);
5193}
5194
0ee0fda0 5195static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 5196{
6cbff00f 5197 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
fdebe2bd 5198 return -EACCES;
33268eaf 5199 return generic_permission(inode, mask, btrfs_check_acl);
fdebe2bd 5200}
39279cc3
CM		 5201
5202static struct inode_operations btrfs_dir_inode_operations = {
3394e160 5203 .getattr = btrfs_getattr,
39279cc3
CM		 5204 .lookup = btrfs_lookup,
5205 .create = btrfs_create,
5206 .unlink = btrfs_unlink,
5207 .link = btrfs_link,
5208 .mkdir = btrfs_mkdir,
5209 .rmdir = btrfs_rmdir,
5210 .rename = btrfs_rename,
5211 .symlink = btrfs_symlink,
5212 .setattr = btrfs_setattr,
618e21d5 5213 .mknod = btrfs_mknod,
95819c05
CH		 5214 .setxattr = btrfs_setxattr,
5215 .getxattr = btrfs_getxattr,
5103e947 5216 .listxattr = btrfs_listxattr,
95819c05 5217 .removexattr = btrfs_removexattr,
fdebe2bd 5218 .permission = btrfs_permission,
39279cc3 5219};
39279cc3
CM		 5220static struct inode_operations btrfs_dir_ro_inode_operations = {
5221 .lookup = btrfs_lookup,
fdebe2bd 5222 .permission = btrfs_permission,
39279cc3 5223};
39279cc3
CM		 5224static struct file_operations btrfs_dir_file_operations = {
5225 .llseek = generic_file_llseek,
5226 .read = generic_read_dir,
cbdf5a24 5227 .readdir = btrfs_real_readdir,
34287aa3 5228 .unlocked_ioctl = btrfs_ioctl,
39279cc3 5229#ifdef CONFIG_COMPAT
34287aa3 5230 .compat_ioctl = btrfs_ioctl,
39279cc3 5231#endif
6bf13c0c 5232 .release = btrfs_release_file,
e02119d5 5233 .fsync = btrfs_sync_file,
39279cc3
CM		 5234};
5235
d1310b2e 5236static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 5237 .fill_delalloc = run_delalloc_range,
065631f6 5238 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 5239 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 5240 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 5241 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 5242 .writepage_start_hook = btrfs_writepage_start_hook,
1259ab75 5243 .readpage_io_failed_hook = btrfs_io_failed_hook,
b0c68f8b
CM		 5244 .set_bit_hook = btrfs_set_bit_hook,
5245 .clear_bit_hook = btrfs_clear_bit_hook,
07157aac
CM		 5246};
5247
35054394
CM		 5248/*
5249 * btrfs doesn't support the bmap operation because swapfiles
5250 * use bmap to make a mapping of extents in the file. They assume
5251 * these extents won't change over the life of the file and they
5252 * use the bmap result to do IO directly to the drive.
5253 *
5254 * the btrfs bmap call would return logical addresses that aren't
5255 * suitable for IO and they also will change frequently as COW
5256 * operations happen. So, swapfile + btrfs == corruption.
5257 *
5258 * For now we're avoiding this by dropping bmap.
5259 */
39279cc3
CM		 5260static struct address_space_operations btrfs_aops = {
5261 .readpage = btrfs_readpage,
5262 .writepage = btrfs_writepage,
b293f02e 5263 .writepages = btrfs_writepages,
3ab2fb5a 5264 .readpages = btrfs_readpages,
39279cc3 5265 .sync_page = block_sync_page,
16432985 5266 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 5267 .invalidatepage = btrfs_invalidatepage,
5268 .releasepage = btrfs_releasepage,
e6dcd2dc 5269 .set_page_dirty = btrfs_set_page_dirty,
39279cc3
CM		 5270};
5271
5272static struct address_space_operations btrfs_symlink_aops = {
5273 .readpage = btrfs_readpage,
5274 .writepage = btrfs_writepage,
2bf5a725
CM		 5275 .invalidatepage = btrfs_invalidatepage,
5276 .releasepage = btrfs_releasepage,
39279cc3
CM		 5277};
5278
5279static struct inode_operations btrfs_file_inode_operations = {
5280 .truncate = btrfs_truncate,
5281 .getattr = btrfs_getattr,
5282 .setattr = btrfs_setattr,
95819c05
CH		 5283 .setxattr = btrfs_setxattr,
5284 .getxattr = btrfs_getxattr,
5103e947 5285 .listxattr = btrfs_listxattr,
95819c05 5286 .removexattr = btrfs_removexattr,
fdebe2bd 5287 .permission = btrfs_permission,
d899e052 5288 .fallocate = btrfs_fallocate,
1506fcc8 5289 .fiemap = btrfs_fiemap,
39279cc3 5290};
618e21d5
JB		 5291static struct inode_operations btrfs_special_inode_operations = {
5292 .getattr = btrfs_getattr,
5293 .setattr = btrfs_setattr,
fdebe2bd 5294 .permission = btrfs_permission,
95819c05
CH		 5295 .setxattr = btrfs_setxattr,
5296 .getxattr = btrfs_getxattr,
33268eaf 5297 .listxattr = btrfs_listxattr,
95819c05 5298 .removexattr = btrfs_removexattr,
618e21d5 5299};
39279cc3
CM		 5300static struct inode_operations btrfs_symlink_inode_operations = {
5301 .readlink = generic_readlink,
5302 .follow_link = page_follow_link_light,
5303 .put_link = page_put_link,
fdebe2bd 5304 .permission = btrfs_permission,
0279b4cd
JO		 5305 .setxattr = btrfs_setxattr,
5306 .getxattr = btrfs_getxattr,
5307 .listxattr = btrfs_listxattr,
5308 .removexattr = btrfs_removexattr,
39279cc3 5309};
Linux kernel - Deliverables
This page took 0.417748 seconds and 5 git commands to generate.